NAFTA
TECHNICAL WORKING GROUP
ON PESTICIDES (TWG)
DEVELOPMENTAL NEUROTOXICITY STUDY (DNT)
GUIDANCE DOCUMENT

-------
The Developmental Neurotoxicity study (DNT) assesses behavioural and neurobiological parameters
in order to ascertain the effects of chemicals on the developing animal. When required, the results from
this study play an important role in the determination/identification of toxicological endpoints for
regulatory purposes.
The rodent DNT study paradigm has evolved over time with the most recent test guidelines updated in
2007 with the introduction of OECD guideline 426.1 Despite this, there has been a continued interest in
developing additional guidance for evaluators responsible for assessing chemicals that trigger this kind
of an assessment. From a regulatory perspective, such guidance would serve as a tool that could be
referred to when evaluating and interpreting the study results.
As a NAFTA Technical Working Group (TWG) on Pesticides-inspired multi-governmental initiative,
the development of this guidance has involved consultation with both governmental and non-
governmental stakeholders, which includes experts in the area of DNT. The internal guidance
document (IGD) itself is a result of a collaborative effort from an intergovernmental working group
with the following members from the United States Environmental Protection Agency (US EPA) and
Health Canada, respectively:
•	Dr. Virginia Moser, Dr. Kathleen Raffaele, Dr. Kevin Crofton, and Dr. Mary Gilbert from US
EPA; and
•	Dr. Wayne Bowers and Mr. Francis Bailey from Health Canada.
The IGD is not meant to be prescriptive; however, it has been developed to provide background and
the identification of design and test details for detailed clinical observations, motor activity, acoustic/
auditory startle response, and learning and memory data. While the underlying goal is to have a DNT
IGD at the disposal of regulatory evaluators (US, Mexico and Canada), the information provided in
this guidance could also be useful in the conduct of other research initiatives, whether retrospective or
prospective, and relevant technical guidelines, such as the recently published OECD Extended 1-
generation reproduction study guideline.
If you are interested in attaining a copy of the DNT IGD, it is available from the NAFTA TWG
on Pesticides website.
This report has been subjected to the U.S. Environmental Protection Agency, Office of Research
and Development's peer and administrative review process and has been approved for publication.
1 http://www.oecd-ilibrary.org/environment/test-no-426-developmental-neurotoxicity-study_9789264067394-
en;jsessionid=7bt5uldknlfjn.x-oecd-live-01

-------
North American Free Trade Agreement (NAFTA)
Technical Working Group on Pesticides (TWG)
Developmental Neurotoxicity Study
Guidance Document
December, 2016
Contributors
Dr. Virginia Moser, United States Environmental Protection Agency (USEPA)
Francis Bailey, Health Canada - Pest Management Regulatory Agency
Dr. Wayne Bowers, Health Canada
Dr. Kathleen Raffaele, USEPA
Dr. Kevin Crofton, USEPA
Dr. Mary Gilbert, USEPA

-------
Contents
Introduction	p. 1
Module A - Observations	p. A -1
Module B - Motor Activity	p. B - 1
Module C - Acoustic/Auditory Startle Response	p. C -1
Module D - Learning and Memory	p. D - 1
Module E - Data Considerations and Integration	p. E -1

-------
Developmental neurotoxicity tests: Introduction
Background
The developing nervous system is especially vulnerable to certain chemicals (Bondy and
Campbell, 2005; Rodier, 1995), and exposures may result in altered neural development with
consequences that may be quite unlike the chemical's effects in an adult nervous system (Barone
et al., 2000). For these reasons, regulatory agencies (OECD, 2007; US EPA, 1998a) have
promulgated testing guidelines for developmental neurotoxicity (DNT). DNT refers to any
adverse effect of exposure to a toxic substance on the normal development of nervous system
structures and/or functions (US EPA, 1998b). The basic purpose of DNT guideline testing is to
act as an initial assessment and screen for the potential of chemicals to cause adverse
neurodevelopmental outcomes.
Experimental Guidelines
The full history of the emergence of regulatory DNT testing can be found in Makris et al. (2009).
Briefly, the basic design and test specifics of the US EPA test guidelines were developed at a
workshop held in 1989, following which the specific guideline (US EPA, 1998a) were developed
and eventually finalized in 1998. The OECD test guideline was based on that of the US EPA,
but included enhancements developed through discussion and international agreement. The final
guideline represents compromises in some areas of the need for and specific conduct of some
tests (OECD, 2007).
Once put into practice there were some aspects of the guidelines that were modified to improve
their sensitivity. The US EPA has issued guidance on specific aspects of testing, although the
1998 guideline has not been formally revised. These changes include: 1) increasing the dosing
period to weaning, given the considerable nervous system development that is occurring up to
that time; 2) increasing the sample size for neuropathology for better quantification; and 3) in
certain cases, including acetylcholinesterase measurements in the dam and pups.
The OECD extended one-generation guideline was most recently developed, with the goal of
allowing neurotoxicological assessments of a subset of offspring (neurotoxicity cohort) as part of
a reproductive toxicity study. The need for including the neurotoxicity group is based on
"existing knowledge" and "needs of various regulatory authorities", and its inclusion in all
studies is doubtful. Since there are many fewer pups, the number of behavioral tests included is
severely limited. It is not intended to provide a complete assessment of developmental
neurotoxicity or a replacement for more detailed studies (OECD, 2011).
1

-------
The guideline-specific requirements for each test measure are described in subsequent modules.
The following table presents aspects of general experimental design across these guidelines.
Table 1 - Guideline requirements for DNT anc
extended one-generation reproduction studies

US EPA OPPTS
870.6300 (1998)
OECD 426 (2007)
OECD 443 (2011)
Neurotoxicity
cohort
Animal selection
Rat, do not use Fischer
344
Rat, justify other species
Rat, justify other
species
Number of
litters/ animals
20 litters/dose
recommended
At least 20/dose
10 litters/dose out
of 20/litter in larger
study
Testing
assignments
1 male or 1 female from
each of 10 litters/dose for
specific behavioral tests,
allocated such that
testing histories do not
confound subsequent
measures
6/sex/dose for
neuropathology*
Unofficial guidance to
increase to 10/sex/dose
1 male and 1 female from
each of 20 litters/dose for
specific behavioral tests,
allocated such that testing
histories do not confound
subsequent measures
10/sex/dose for
neuropathology
1 male and 1 female
from each of 10
litters/dose, all pups
get all tests
Dosing period
GD6-LD10*
Unofficial guidance to
extend dosing to LD21
GD0=sperm positive
GD6-LD21
GD0=sperm positive
2 wk premating, 2
wk during mating,
through LD21
Dose
administration
Oral to dams
Most relevant route,
usually oral, to dams
Consider direct dosing to
pups where warranted
Most relevant route,
usually oral, to
dams
Dose selection
High dose should
produce some toxicity
(decrease weight gain
<20%)
Low dose should not
produce effects
High dose should produce
some toxicity (decrease
weight gain <10%)
Low dose should not
produce effects
Base on TK data
Culling of litters
PND4, aim for 4 male
and 4 female
On or before PND4, aim
for equal number males
and females
PND4, aim for 5
males and 5
females/litter
Age
nomenclature

PND0=day of birth but
prefer post-coital to
postnatal age

2

-------
Standard Evaluation Goals
Health Canada and the US EPA developed this document on the review and interpretation of
submitted DNT data to provide guidance on how to evaluate the quality, the conduct, and
resulting data derived from the behavioral methods employed in the OECD and/or EPA DNT
Guidelines. The reviewer may be exposed to data collected under any of the current guidelines
(OECD, 2007; OECD, 2011; US EPA, 1998a), as guideline selection depends on when the
studies were conducted, and whether the study was specific for the US EPA or other international
organizations; this guidance is applicable to all. This guidance is for use by for regulatory
agency scientists reviewing these DNT data, especially those who may not be experts in
neurotoxicity or developmental neurotoxicity. The guidance was generated by an international
collaboration between Health Canada and the US EPA. The overall goal of the guidance is to
foster better and more consistent reviews of DNT behavioral data between these two countries.
This guidance may also be useful for other international regulatory agencies or those interpreting
data generated under the auspices of the OECD or US EPA DNT guidelines.
References
Barone S Jr, Das KP, Lassiter TL, White LD. (2000). Vulnerable processes of nervous system
development: A review of markers and methods. Neurotoxicology 21:15-36.
Bondy SC, Campbell A (2005). Developmental neurotoxicology. J. Neurosci. Res. 81:605-612.
Makris SL, Raffaele K, Allen S, Bowers WJ, Hass U, Alleva E, Calamandrei G, Sheets L, Amcoff
P, Delrue N, Crofton KM. (2009). A retrospective performance assessment of the
developmental neurotoxicity study in support of the OECD test guideline 426. Environ.
Health Perspec. 117:17-25
OECD Guideline for the Testing of Chemicals, Developmental Neurotoxicity Study No. 426,
2007.
OECD Guideline for the Testing of Chemicals, Extended One-Generation Reproductive Toxicity
Study N. 443,2011.
Rodier PM. (1995) Developing brain as a target of toxicity. Environ. Health Perspect. 103 (suppl
6):73-76.
US EPA. (1998a). Health effects guidelines OPPTS 870.6300 Developmental Neurotoxicity Study,
EPA/ 712/c-98/239.Office of Prevention Pesticides and Toxic Substances.
US EPA. (1998b). Guidelines for neurotoxicity risk assessment. EPA/630/R-95/001F. Washington,
DC:US EPA.
3

-------
MODULE - A
EVALUATION OF OBSERVATIONS
AND FUNCTIONAL TEST DATA
Observations
A- 1

-------
1.	Introduction	3
2.	Test Description	3
3.	Guideline requirements	4
4.	Observational Testing Procedures	5
4.1	Specific Protocols and Tests	5
4.2	Observers	6
4.3	Experimental Control	6
4.4	Test Subject	7
5.	Data Reporting	7
5.1	Data checks	7
5.2	Positive control data	8
6.	Data analysis	9
7.	Interpretation	9
8.	References	11
Observations	A - 2

-------
MODULE A - INTERPRETATION OF OBSERVATIONS
AND FUNCTIONAL TEST DATA
1.	Introduction
The use of observational and functional tests for screening in toxicology assessments gained
popularity after several expert panels and scientists recommended such in the late 1970s and
early 1980s. Their use is based on the premise that behavior represents the integrated sum of
activities mediated by the nervous system, and is a sensitive marker of nervous system
dysfunction. Taken from the Irwin screening battery that is widely used in drug development
(Irwin, 1962, 1968), test batteries for behavioral evaluations in neurotoxicology, broadly known
as functional observational batteries (FOB; Moser, 1989; McDaniel and Moser, 1993), were
developed and validated in subsequent years. Based on the widespread and accepted use of such
test batteries in adult neurotoxicity screening in rodents, observational and functional tests were
included as part of DNT testing to evaluate overall neurological function in both the dam and
offspring. However, FOB testing, per se, is not required in the EPA or OECD (426) DNT
guidelines.
2.	Test Description
When conducted optimally, observational test batteries provide a systematic and detailed
evaluation of the animal's behavior and function. Measurements of complex movements, e.g.,
gait, reflect multiple neuronal functions, contribute to sensitivity of the approach but suffer with
lack of specificity. On the other hand, tests of simple reflex behaviors, e.g., simple sensory
responses, may be more specific but only be altered by a few neurotoxic agents. In general,
screening batteries are based on noninvasive observations and manipulations, and rely heavily
on careful evaluation of the individual animal. Some tests can include appropriate test devices
(for example, automated motor activity chambers, strain gauges for grip strength) to provide
quantitative, objective data. On the other hand, many measures are often subjective.
There is a range of approaches for these observations, from very simplified to more expanded
clinical observations. These clinical observations vary widely across laboratories and there are no
published protocols or guidelines. A lack of standardization in how and what data are collected
negatively influences the ability to interpret or compare data across laboratories.
More specific screening batteries, such as the FOB, include a broad range of assessments. To
accurately be considered an FOB, the protocol should focus on detailed observations as well as
specific tests of reflexes, responses, and abilities. Using an FOB, several neurological functions
must be assessed, including autonomic, neuromuscular, and sensory, as well as levels of activity
and excitability. Numerous FOB protocols have been published, but even more are unpublished
(e.g., those used in testing laboratories). As with clinical observations, there is a lack of
standardization across laboratories.
3.	Guideline requirements
All of the DNT test guidelines require close evaluations of both dams (during gestation and
lactation) and offspring (from an early age to adulthood). The verbiage in the DNT guidelines
Observations
A - 3

-------
that describe the clinical observations (see Table 1) is, however, lifted from parts of the US
EPA and OECD adult neurotoxicity test guidelines. There is a wide variety of test protocols
being used in testing laboratories that generally fulfill the DNT guideline requirements.
It is important to understand that the EPA and OECD (426) DNT guidelines do not specify or
even mention the use of the FOB. Unfortunately, many researchers and reviewers do not
understand this distinction. On the other hand, the OECD extended one-generation study (443)
does require observations in the main study, and does actually specify the FOB in the DNT
cohort.
The test requirements are listed in Table 1. It is notable that the most recent guideline (OECD
443) actually has the most explicit description and requirements, whereas the earliest guideline
(US EPA 870.6200) is the least specific.
Table 1. Requirements for observational and functional assessments in DNT test guidelines.

US EPA 870.6300
OECD 426
OECD 443
Test subject and age
Dams at least twice
during the gestational
period, at least twice
during lactation
10/dose group
Dams at least twice
during the gestational
period, at least twice
during lactation
At least 10/dose group
Dams on a weekly
basis
Offspring on PNDs
4, 11,21,35,45, 60
10/sex/dose group
Offspring weekly
preweaning, at least
every two weeks in
adolescence and as
adults
At least 1/sex/litter
All F1 animals on a
weekly basis after
weaning
DNT cohort:
10/sex/dose between
PND63-75
Test apparatus
Outside the home
cage
Outside the home cage
Home cage and
observation arena
outside the cage
Specific observations
Autonomic function:
lacrimation,
salivation,
piloerection,
exophthalmos,
urination, defecation,
pupillary function;
Convulsions,
tremors, abnormal
movements; Posture,
gait abnormalities;
Unusual or abnormal
behaviors
Changes in skin, fur,
eyes, mucous
membranes, secretions,
autonomic activity;
Unusual responses in
body position, activity,
coordination, gait,
posture; Reactivity to
handling, placing,
other stimuli; Clonic
or tonic movements,
convulsions, tremors;
Unusual or abnormal
behaviors
Fl: Changes in skin,
fur, eyes, mucous
membranes,
secretions, autonomic
activity; Unusual
responses in gait,
posture; Reactivity to
handling; Clonic or
tonic movements;
Unusual or abnormal
behaviors
DNT cohort:
Requires FOB,
Appendix lists
specific endpoints
under categories of
Observations
A - 4

-------



home cage and open
field, manipulative,
and physiologic
(taken from several
FOB publications)

Observers
Trained technicians,
unaware of
treatment.
Demonstrate
interob server
reliability if not same
technician
throughout
Trained technicians,
unaware of treatment.
Advisable to have
same technician
throughout
Trained technicians,
unaware of treatment.
Advisable to have
same technician
throughout
Protocols
Standardized
procedures
Standardized
procedures
Explicit operationally
defined scales and
scoring criteria
Objective quantitative
measures where
possible
4. Observational Testing Procedures
4.1. Specific Protocols and Tests
There are differences in guideline requirements, but common to all guidelines are assessments of
autonomic (e.g., salivation, lacrimation) and motor function (gait, posture), convulsive
behaviors, as well as a catch-all phrase "unusual or abnormal behaviors". The OECD 426 adds
mention of sensory responsiveness, and the OECD 443 specifically requires the more extensive
FOB. It is likely, however, that individual testing laboratories use or will use the same protocol
for all DNT studies.
In reality, testing laboratories often have one protocol of clinical observations that is used
regardless of the specific requirements (i.e., for adult and DNT studies). The focus is on
autonomic dysfunction, abnormal posture or movements, and unusual behaviors or appearance.
In some protocols, measures of reactivity in response to handling or removal from cage are
added. Responses to sensory stimuli, specific reflexes, or measures of grip strength are most
often not included, and these clinical observations would not be classified as a valid FOB. Thus,
even if a report states that a FOB was used, that is not the case unless the full battery of tests is
used. The reviewer should understand these differences between clinical observations, which are
most common, and a full FOB, which will likely not be used.
Regardless of which endpoints are included, clear defined protocols are critical to good
experimentation. The sequence of tests should progress from the least (e.g., observations in the
home cage and open field) to the most invasive (e.g., handling assessments) to minimize the
influence of stress on subsequent measures. Since these are often subjective evaluations,
Observations
A - 5

-------
explicitly defined scores and criteria should be used: anything less is subject to observer bias.
Measures that are ranked provide more information on treatment effects than do binary, or all-or-
nothing, measures. If used, binary measures should include a clear description of what
constitutes "other than normal." With lethality, for example, "yes" and "no" are easily
distinguished. However, the case for other evaluations such as activity is not so obvious, where
"increased" vs "decreased", or "normal" vs "abnormal", must have associated with them
operational definitions or else the measure is meaningless and large observer differences may
occur. A ranking, or scale, describing different levels of activity would improve consistency
across observers and would allow a reasonable evaluation of the data by the reviewer. The
laboratory's standard operating procedure (SOP) should provide descriptions of each test
measure, order of testing, and evaluation scales. The reviewer can request this protocol when
necessary to understand better the submitted data.
4.2	Observers
Unlike the many automated and quantitative tests used in a DNT study, the observer is often the
"instrument" collecting the data for these functional and observational measures. The guidelines
highlight the need for trained technicians. Furthermore, the same technician throughout the
study contributes to consistency, and information regarding comparability of technicians (inter-
observer reliability) is required if the same technician is not used. Such information should be
included in the main study report to assure that this was addressed.
Observational procedures are not difficult, but do require a significant level of comprehension
and technique on the part of the observer. It is crucial that the observer be unaware ("blind") of
the subject's treatment, in order to prevent deliberate or subconscious bias from affecting the
data. The availability of a training video and manual for the FOB (available from US EPA) has
done much to train observers in many types of settings, including academic, government, and
commercial laboratories. The use of positive-control studies is also useful to demonstrate
sensitivity of the protocol as well as expertise of the technicians. If conducted, these positive-
control studies should encompass all test ages (as young as one week of age) with multiple
behavioral outcomes.
4.3	Experimental Control
Good experimental practices of counterbalancing testing procedures, minimizing interfering
factors, etc., are a necessary part of conducting behavioral tests. The response of the whole
animal is measured; therefore, there is a multitude of possible influences from various sources.
Extraneous factors (e.g., noise, light, odors, etc.) must be rigidly controlled as they may have a
significant effect on the behaviors being monitored. Many of the specifics of conducting these
tests have been summarized (Slikker et al., 2005). Since no training of the subject is involved,
there is no explicit control over observed behaviors and variability may be high, suggesting a
need for larger sample size. In general, behaviors that include the subject's innate responses
(e.g., righting reflex, pain response) or a function that is important or critical to its survival (e.g.,
pupil response) tend to be less variable than spontaneous behaviors (e.g., activity, rearing).
Strain and sex of the subjects may also influence the results (e.g., Moser, 1996).
Observations
A - 6

-------
4.4 Test Subject
The test guidelines include evaluation of both dams and offspring. While a single protocol may
not be appropriate for evaluations at all ages, this seems to be a common practice in many
testing laboratories. Explanations of modifications to adjust for these different situations are
desirable. Such modifications should account for the immature reflexes and motor function of
young animals, especially true at the youngest ages.
Testing pregnant dams requires care and minimal manipulations, especially late in gestation.
Also, testing dams during lactation usually means separating her from the litter. The dams
should only be removed for a short period of time (less than 20-30 min) or else the pups should
be provided some means of thermoregulation. Likewise, when pups are removed for
observations, thermoregulatory controls must be provided if they are away from the dam for
very long. These situations can cause differences in behavior that are not the case for standard
adult rodents, and therefore data may not be easily extrapolated.
The US EPA guidelines require these observations be conducted on at least 10 pups/sex/dose,
and the OECD guidelines state 1 pup/sex/litter (which would provide 10/sex/dose if 20
litters/dose are used). As with all the DNT tests, the assignment of pups to observational testing
should be one male and one female from each litter, or else one male or one female should be
drawn from separate litters. For dams, at least 10/dose should be examined. It is desirable to
use the same subject at each testing time, even though this requires identification of individual
pups. There should be a random selection for which subjects are tested. Specific information
for these issues should be included in the report.
5. Data Reporting
There are many dependent variables in these test batteries, making data summarization a
challenge. At the minimum, tables of summary data across all endpoints should be provided.
Where treatment effects are noted, individual animal data and/or more detailed information
should be provided.
5.1 Data checks
There are specific checks that can be used to evaluate the adequacy of submitted observational
data. Up front, there should be:
•	Listing and description of what endpoints were included in the protocol and the scoring
criteria or explicit descriptions used for each endpoint. Simply referencing a standard
operating procedure is not appropriate
•	Modifications made to adjust the observations at different stages of the study, including
age of the test subject
•	Training of the technician and whether the same technician was used throughout. If more
than one technician was involved, inter-observer reliability should be described
•	Procedures for assuring that the technician was unaware of the treatment of each test
subject
Observations
A - 7

-------
Another data check is to evaluate the degree of variability. While this discussion is mostly
applicable for controls, it should be noted that treatment may produce more or less variability
and such information is also useful. While it may be counterintuitive, for some measures a lack
of variability suggests inappropriate evaluations or scoring criteria. Behavioral measures that are
not under experimenter control, e.g., reactivity or activity, will vary across animals. Scoring
criteria without sufficient resolution to document these differences will not be sensitive to
anything but the most obvious toxicity. In cases where all subjects receive the same score,
across all treatments and test times, it is obvious that the criteria are not adequate. Often, this
includes a phrase such as "normal" that does not explicitly define anything. Simply put, such
evaluations have little value for detecting subtle to moderate behavioral changes. In addition, at
some ages there may be more variability due to differences in the rate of complete nervous
system maturation. Examples of measures for which some variability across animals should be
expected are:
•	Reactivity to handling, placing
•	Activity and/or rearing in an open field or test arena
•	Responses to sensory stimuli
•	Gait, posture - to the extent that some rodents walk differently
•	Urination, defecation
On the other hand, some functions are biologically controlled and show little variability. Any
rare occurrence of these signs in control animals should be addressed. For example, lacrimation
or salivation indicate autonomic dysfunction, and should not be seen in control animals.
Reflexes such as the righting response are stereotyped neurological actions and show very little
variability; however, some of these reflexes are developing during the preweaning period and
there could be differences in the subjects during this time. Specific observations reflecting poor
health should not be seen in controls. Note that the wording of some required observations listed
in the guidelines are vague, such as "changes in skin, fur, mucous membranes", and "unusual
or abnormal behaviors", but these may be indicators of nonspecific illness. Some of the
endpoints that should not vary in controls are listed below.
•	Salivation, lacrimation
•	Pupil response to light
•	Piloerection, exophthalmos
•	Convulsions, tremor - although young pups will display very fine "tremors"
•	Major gait or postural changes
•	Changes in righting reflex
5.2 Positive control data
Positive control data are useful for determining the range of neurological effects that can be
characterized by the testing laboratory. Often, however, these studies are conducted in adult
rodents with acute, high doses of neurotoxic chemicals that produce overt toxicity. These types
of studies are less useful for assuring that the laboratory can detect subtle neurotoxicity,
especially in very young pups. The DNT guidelines do not require positive control data with
Observations
A - 8

-------
clinical observations.
6.	Data analysis
The use of rating scales or descriptions makes statistical analyses of observational data more
complex, since it is not appropriate to treat such data as continuous, parametric data.
Transformations of the data may be necessary as well as consideration of non-parametric
analyses. In addition, some data may show very low frequencies (e.g., 0, 1, or 2), making both
analysis and interpretation more difficult. Typically, data with high variability should be
examined for outliers or data errors; in such cases, evaluation of individual data is critical. For
example, extreme values may be due to technician or instrument error, differences in individual
sensitivity, or other factor. It is also important to note that numerous measures are collected on
each individual subject. When littermates are tested, this should also be included in the
analyses. Some appropriate approaches for analyzing behavioral data are described by Holson
and colleagues (2008).
7.	Interpretation
Interpretation is based on the information from individual endpoints and where possible, the
profile or pattern of effects observed; for example, motor deficits would be evidenced as changes
in gait, posture, righting, or other neuromuscular measures. When evaluating changes in specific
endpoints, it is important to have an understanding of the variability, specificity, and influences
on that measure. There is greater confidence in data collected quantitatively over observational
data. The US EPA has published guidance for the evaluation of neurobehavioral data, including
screening batteries (US EPA, 1998). Neurotoxic adverse effects are defined as any change in the
structure or function of the central and/or peripheral nervous system. This includes alterations in
either direction (i.e., increases or decreases) from baseline or normal conditions as well as effects
that are transient, occur only at specific times during development, or appear as changes in the
ontogeny of developmental processes.
Observed changes in the dams on study indicate acute (or subchronic) toxicity of the chemical
itself. It is important to note the time of testing in relation to chemical administration if it is
given directly. For example, with oral gavage dosing, testing after that day's dose could be
assessing acute but reversible effects from that dose. Time of testing is less important in studies
where the chemical is administered via feed or water. Given that the dose ranges used in DNT
studies often include a high dose with some systemic toxicity or effects of body weight, some
significant changes might be expected. On the other hand, severe toxicity, e.g., convulsions,
observed in the dam suggest that findings in the pups could be confounded by the maternal
toxicity.
Effects observed in pups during the preweanling period may likewise reflect acute toxicity. This
is especially true in studies where the pup is directly dosed with the test chemical. As with the
dams, considerations regarding time of testing and severity of effects apply. For example,
convulsions in pups indicate highly toxic doses and, if the pups survive, behavioral changes into
adulthood may be expected. This should be considered when evaluating other DNT endpoints.
Most of the observations specified in DNT studies have been shown to be sensitive to direct
Observations
A - 9

-------
chemical effects following acute or chronic dosing, and it is unlikely that such effects will be
detected in adult offspring since by this age the animals are no longer receiving direct chemical
exposures. For example, direct effects of chemicals that inhibit acetylcholinesterase include
autonomic dysfunction, but this has never been seen in adult offspring that were developmentally
exposed to those chemicals. The types of persistent effects that may occur as a result of
exposure to developmentally neurotoxic chemicals, e.g., changes in reactivity, sensory
responses, or neuromotor function, are mostly not assessed with the current guideline structure:
the sole exception is the neurotoxicity cohort of the recent OECD 443 study. Consequently, the
absence of findings in the adult offspring should not be used to discount effects in the young
animal.
Observations
A - 10

-------
8. References
Holson RR, Freshwater L, Maurissen JPJ, Moser VC, and Phang W. (2008) Statistical issues
and techniques appropriate for developmental neurotoxicity testing: A report from the ILSI
Research Foundation/Risk Science Institute expert working group on neurodevelopmental
endpoints. Neurotoxicol. Teratol. 30, 326-348.
Irwin S. (1962) Drug screening and evaluative procedures. Science 136, 123-128
Irwin S. (1968) Comprehensive observational assessment: la. A systematic, quantitative
procedure for assessing the behavioral and physiologic state of the mouse.
Psychopharmacologia (Berl) 13, 222-257
McDaniel KL, Moser VC. (1993) Utility of a neurobehavioral screening battery for
differentiating the effects of two pyrethroids, permethrin and cypermethrin. Neurotoxicol.
Teratol. 15, 71-83
Moser VC (1989) Screening approaches for neurotoxicity - a functional observational battery.
J. Am. Coll. Toxicol. 8, 85-93.
Slikker, Jr. W, Acuff K, Boyes W, Chelonis J, Crofton KM, Dearlove G, Li A, Moser VC,
Newland C, Rossi J, Schantz S, Sette W, Sheets L, Stanton M, Tyl S, and Sobotka TJ.
(2005). Behavioral test methods workshop: A summary. Neurotoxicol. Teratol. 27, 417-
427.
US EPA (1998) Guidelines for neurotoxicity risk assessment. Federal Register 63, 26926-26954.
Observations
A- 11

-------
MODULE -B
EVALUATION OF MOTOR ACTIVITY DATA
Motor Activity
B - 1

-------
1.	Introduction	3
2.	Test Description	3
3.	Guideline Requirements	6
4.	Motor Activity Testing Systems	6
4.1.	Test Chambers	6
4.2.	Detection Systems	7
4.3.	Activity Measurements	7
5.	Test Procedures	9
6.	Data Reporting	10
6.1	Dependent Variables	10
6.2	Reported Data	11
7.	Data Analysis	11
7.1 Statistical Models	12
Main effects	12
Interactions	13
8.	Interpretation of Motor Activity Data	14
8.1	Properties of motor activity control data	14
8.2	Variability in motor activity data	15
8.3	Evaluation of Treatment effects	15
8.4	Treatment-related effects on ontogeny	16
8.5.	Statistical versus biological significance	17
8.6.	Dose-response results	17
8.7.	Habituation results	18
9.	General interpretation	19
10.	References	20
Motor Activity	B - 2

-------
MODULE B - EVALUATION OF MOTOR ACTIVITY DATA
1.	Introduction
Overall study design plays an important role in determining the quality, reliability, and
interpretation of motor activity data in DNT studies. Described herein are factors more
specifically associated with design and conduct of motor activity testing in DNT studies and
considerations in the interpretation of motor activity data. This document is not meant to be
prescriptive but to provide background and the identification of design and test details for motor
activity data that need to be considered in the interpretation of motor activity test data.
2.	Test Description
As defined in the US EPA Test Guidelines (870.6200), motor activity is "any movement of the
experimental animal". Although this is a very broad definition, motor activity is more typically
considered to be locomotor movements in a horizontal direction (ambulation) as well as other
directions (e.g., vertical, rearing). Smaller, fine movements (e.g., sniffing, grooming) may be
included as motor activity but are not typically considered ambulatory, or locomotor, activity.
Motor activity is an apical behavior that reflects a number of underlying processes including
motor capacity, sensory functioning, emotional processing, non-associative learning
(habituation), and integrative (cognitive) processes (Denenberg, 1969; Kelley, 1993; MacPhail
et al., 1989; Maurissen and Mattsson, 1989; Ross, 2001). Decades of research in psychology,
pharmacology, and toxicology have established the ability of motor activity measurements to
provide insight in normal or altered nervous system function and development.
Motor activity tests are designed primarily to assess locomotor activity although modern
automated test equipment provide a wide array of activity measures including ambulatory
activity (movement between locations), exploratory movements (rearing or vertical movements)
as well as a variety of fine motor movements (e.g., sniffing, scratching, grooming). It is
important to be aware that automated equipment can provide measures of these actions
separately or as an integrated measure of all activity movements (referred to as "total activity" in
this document). An understanding of the measures that are operationally defined by the software
or protocol is necessary to determine precisely what measures are actually reported. Since fine
motor movements such as sniffing, scratching, etc., do not provide a measure of locomotor or
ambulatory activity, care should be taken to ensure that activity test data clearly distinguish
among these various activity measures and clearly indicate the type of the activity measures
reported.
In DNT studies, motor activity measurements serve three main purposes. First, motor activity
levels provide a sensitive measure of apical nervous system function. Second, the ontogeny of
motor activity follows a developmental pattern that reflects the development and maturation of
the nervous system. Third, motor activity studies provide a measure of non-associative learning
(habituation), a basic form of learning essential to adaptive behavior and critical for normal
interaction of animals with their environment. In the case of activity tests, habituation is
normally measured as a decrease in locomotor movements over the course of the test session,
although decreases in total motor activity (all measures of all motor movement) may also be
used to assess habituation.
Motor Activity
B -3

-------
The ontogeny of motor function has been well investigated. The development of normal
locomotor capacity in rodents begins around PND13 (Bolles and Woods, 1964; Bronstein, 1972;
Shriner et al., 2009) although the required postural development is not fully developed at this
age. Adult-form locomotion typically occurs after PND15-16 in rats but does not appear to be
fully developed until around PND21 (Altman and Sudarshan, 1975; Muir, 2000). Indeed, both
the neuronal and neuromuscular development required for adult-form locomotion occurs
primarily in the third week of postnatal development in rodents (Clarac et al., 2004; Vinay et al.,
2000, 2002).
Early studies on the development of motor activity show that in pre-weanling rodents, motor
activity increases between PND12-13 to PND16-18 and decreases between PND20 and 23
(Campbell and Mabry, 1972; Moorcroft, 1971; Oakley and Plotkin, 1975; Shaywitz et al., 1979).
In one laboratory, the peak period of spontaneous ambulation in Sprague-Dawley rats was
between 19 and 22 days of age (Shaywitz et al., 1979). There are some results suggesting a role
for environmental novelty in developmental patterns of motor activity. For example, Ruppert et
al. (1985) reported that when the same Long-Evans rat pups tested repeatedly from PND13 to 21
they display an increase in activity from PND14 and 16. When pups were tested only on PND15,
18 or 21 this increase in activity was not observed (Ruppert et al., 1985); however, this study did
not test PND13 and precluded the possibility of detecting the increase between PND13 and
PND15-18. Others have suggested that peak ambulation between PND14 and 16 only occurs
when the test environment is novel (Campbell and Raskin, 1978).
More recently, however, guideline studies from a number of laboratories have shown a pattern of
increased locomotor activity at PND17 compared to PND13, and decreased or similar locomotion
levels at PND21 compared to PND17 (Raffaele et al., 2003). An example of typical patterns of
changing overall activity levels as well as development of habituation with age is shown in
Figure 1. There have been some suggestions that the developmental profiles are relatively flat,
while monotonically increasing patterns may also be evident, depending on the strain of animal,
activity device, frequency of testing, and other experimental conditions; this has not been fully
characterized. The EPA DNT guideline requirements imply that the same animals must be
tested in activity tests in the pre-weaning period, which is important since experimental history
and intra-session habituation which may influence subsequent activity levels. The OECD
guidelines indicate that the same animals must be tested in the pre-weaning period if motor
activity data are used to assess behavioral ontogeny.
Motor Activity
B -4

-------
PND 13 Females
PND 17 Females
20 30 40 50
PND 21 Females
20 30 40 50
PND 60 Females
30 40
Time (min)
300
250
200
150
100
50
0
PND 13 Males
10 20 30 40 50
PND 17 Males
10 20 30 40 50
PND 21 Males
20 30 40 50
PND 60 Males
30 40
Time (min)
Figure 1. An example of motor activity counts across the sessions for a single group of male
and female rats tested on PND 13, 17, 21, and 60. Activity is very low at PND 13 and higher at
PND17 but habituation is not evident. At PND21 and 60, habituation is evident and in addition,
female activity levels are often higher than males at PND60.
Motor Activity
B -5

-------
3. Guideline Requirements
While regulatory test requirements for the OECD and the US EPA are comparable for motor
activity testing as part of DNT studies, there are some differences that should be considered.
These are described in the table below. Note that the OECD TG 443 extended one-generation
reproductive toxicity study that includes optional testing for motor activity employs the same
test procedures as described below.
Table 1. Test requirements for OECD and US EPA DNT motor activity tests.

US EPA OPPTS 870.6300
OECD 426

Age of
testing
PND13, 17, 21, and 60 (±2 days)
1-3 times preweaning, once as young
adults (PND60-70),
Use of preweaning motor activity as
measure of behavioral ontogeny
"strongly recommended"21
Test duration
Duration sufficient to approach
asymptotic activity levels "by the
last 20% of the session" for
controls
All test sessions have same
duration
Duration sufficient for detection of
habituation in control animals
Data
collection
Activity counts collected in
blocks of < 10 min
No mention
Study design
Testing of treatment groups should
time of day and each animal tested i
)e counterbalanced over test device,
ndividually
a Motor activity tests onPND13, 17, 21 fulfills OECD's requirement for behavioral ontogeny evaluation
4. Motor Activity Testing Systems
4.1. Test Chambers
Assessment of motor activity can be measured using a variety of techniques and equipment that
will vary between laboratories. The shape, size, movement detection system, and data
processing (software) may all affect the measured level and type of motor activity reported. It is
important to be aware of these influences, especially in comparing motor activity results between
studies and laboratories or even within laboratories when activity test equipment or procedures
are changed. While there is a wide array of potential test arenas for assessing motor activity,
most activity testing is conducted in one of four main types of chambers:
•	Open-field type activity chambers (usually square or rectangular) of various sizes;
•	Shaped chambers, such as figure-8 or doughnut configurations;
Motor Activity
B -6

-------
•	Home cage activity monitors (usually photocell or toggle switch systems mounted onto
home cages);
•	Running wheels, usually in home cages.
4.2.	Detection Systems
Both the OECD and the US EPA test guidelines require that motor activity testing be conducted
in automated test chambers. Like the architecture of test chambers, a number of systems have
been developed to automatically record motor activity. Most automated motor activity systems
fall into one of the following categories for movement detection, although the last two types of
systems are rarely, if ever, used in DNT studies:
•	Photocell systems: these systems record the number of photocell beam breaks, with more
complex systems recording the location and timing of photocell beam breaks;
•	Video systems: these systems record and digitize the location of animals within the test
arena and use software to characterize movement in the arena;
•	Infrared sensors: these systems record changes in temperature associated with animals in
the test arena and use software to characterize movement within the test arena;
•	Jiggle or tilt-meters: these systems record the number of contacts of switches that detects
changes in vertical force as animals move in the test arena and provide measures of gross
body movements;
•	Contact switches to measure rotations (for running wheels): these systems monitor the
movement of a rotating wheel as animals run within a running wheel and are typically
located in home cages of animals.
Because of these variations in the types of systems used to collect motor activity data, there are a
number of potential motor activity metrics that may be reported. For photocell or contact switch
systems, only counts are directly measured. Other parameters such as ambulatory distance and
time are generated by software algorithms using photocell counts, changes in sequences of
photocell counts, and timing of changes in photocell counts. Measures such as ambulatory
counts, ambulatory distance, ambulatory time, etc., are generated with software algorithms that
incorporate beam breaks, timing of beam breaks and location of beam breaks. For video-based
systems, only distance moved and time in movement are actually detected. Software algorithms
are used to generate measures such as ambulatory distance, time in ambulation, time immobile,
etc. based on changes in location, speed, and nature of changes in location of the animals. In
most automated activity measuring systems, ambulatory counts, ambulatory time and ambulatory
distance are highly correlated. It should be noted that most photocell and video-based systems
employ software settings to eliminate small movements (e.g., grooming, sniffing) from inclusion
in ambulatory time estimates. Jiggle or tilt meters provide less detailed information on
movement and typically provide information only on the number of switch contacts that indicate
the number of gross movements.
4.3.	Activity Measurements
Potential motor activity metrics include:
• Ambulatory counts: such as photocell counts, wheel rotations, tilt-meter contacts;
Motor Activity
B -7

-------
•	Ambulatory distance: such as computed distance moved (e.g., based on software
algorithm or based on number of photocells beams broken);
•	Ambulatory time: such as amount of time animal is engaged in ambulatory movement;
may vary depending on algorithm used to detect ambulatory episodes;
•	Vertical movements (rearing): photocell based systems that incorporate a second set of
photocell mounted at a height above the normal height of a prone animal will detect
vertical movements of the animals. Video-based systems can also measure vertical
movements based on changes in the size of the video image of the animal as it rears and
thus changes size and shape. Infrared, jiggle or tilt-meters are not effective in measuring
vertical movement of rodents;
•	Small or fine motor movements: usually refers to discrete non-ambulatory movements
such as sniffing, grooming, and stereotypy;
•	Total activity: typically refers to the sum of all movements of the animal in the test
chamber, and may include ambulatory, vertical (rearing) as well as small non-ambulatory
movements (grooming, sniffing). Total activity should be clearly distinguished from total
session activity which refers to a specific activity measure (e.g., ambulatory counts)
summed or collapsed across the entire session. In this document, the terms total activity
indicates all activity measured incorporated into one measure (e.g., ambulatory, vertical
and small movements added together to generate one single activity measure), while total
session activity refers to a single activity measures totaled over and time blocks in the test
session (e.g., sum of ambulatory counts for all time blocks). The distinction between
these metrics is important, even though some laboratory reports may not be clear on this.
Software settings for defining the type and threshold for activity units can be critical for
computing measures of ambulatory activity. For example, some systems permit the setting of
thresholds of the number of sequential photocell breaks required to trigger an ambulatory count.
Because the size of the animal (e.g., young versus mature animals) and the distance between
photocells will influence thresholds for recording ambulatory counts, it is important that these
software threshold parameters be reported. For example, low thresholds for ambulatory triggers
may result in high sensitivity for fine movements (e.g., twitching, sniffing, scratching, and head
movement) that may be included in ambulatory counts and thus provide an inaccurate measure of
true locomotor movement. Some software can adjust for the size of animals and can provide
options for setting thresholds to trigger locomotor counts. Vertical positioning of photocell
sensors can be critical for detecting vertical (rearing) movements, especially in young rodents
since incorrect vertical location of photocell may preclude the possibility of detecting vertical
movements in small animals. It should also be considered that in photocell systems, because of
constant distance between photocells, breaking 4 photocells in sequence for a small animal (e.g.,
30 gm) reflects greater locomotor distance than the same number of photocell breaks for an adult
animal (e.g., 400 gm). In addition, some chambers may be more suited for specific ages, leading
to the use of completely different systems for young and adult rats. For these many reasons, direct
comparison of measures of activity between young and adult animals should be conducted with
caution.
Similar to photocell-based systems, video-based activity systems also employ key parameters in
software computed measures of ambulation, small movements, etc. Thresholds for both the
amount of movement required as well as the speed of movement are employed in calculations of
ambulatory distance and ambulatory time. As with photocell-based systems, differences in the
size of young and mature animals will affect how these threshold criteria should be set since the
Motor Activity
B - 8

-------
same criteria parameters will have different sensitivities for detecting ambulatory movements in
small and large animals. In addition, video tracking systems are based on contrast, and are
therefore prone to artifacts caused by reflections and other situations where the tracker does not
accurately follow the rat. Reviewing video files would allow assessment of this problem, but
doing so may not be feasible.
Consideration of measured parameters is also important for infrared systems where the size
(weight) of the animal, distance from the sensor, as well as position relative to the sensor can all
affect the signal. Chamber size may play a more important role in sensitivity for infrared
systems than photocell or video systems.
In order to permit critical evaluation of motor activity data reported in DNT studies, reports
should include sufficient detail to ensure that it is clear how activity data were collected and
computed. This includes:
•	Providing details of recording instrumentation (e.g., number and height of photocell
beams, beam locations and distance between beams, or details of recording
procedures for other detection methods such as video systems);
•	Providing details on software versions and settings used for computing dependent
variables. For example, photocell systems require software setting of thresholds for
the number of photocell breaks required to trigger an ambulatory count, number of
sequential photocell breaks (determined by size of animals). These settings will
determine when an ambulatory count is recorded and should be set relative to size of
animals. For video-based systems, software parameters for establishing thresholds to
trigger ambulatory counts should be reported.
These settings affect the raw data generated, as well as the software-computed activity measures,
and in turn will impact comparisons across studies and/or laboratories and comparisons across
ages (e.g., preweaning, adult) within a study.
5. Test Procedures
Besides the multitude of environmental factors (e.g., lighting, noise) that may alter behavior in
general (described in Slikker et al., 2005), there are a number of factors that specifically
influence motor activity. Procedures to control these factors should be described in the protocol
or experimental design and/or reported in the results. These include:
•	Environmental factors such as sound levels, temperature in test area, lighting conditions,
time of testing relative to light-dark cycle;
•	Testing time in relation to daily dosing when either the dam or pup are being directly
dosed (acute effects of dosing);
•	Testing of treatment groups should be counterbalanced over test device, time of day and
each animal tested individually;
•	Standardized procedures for cleaning activity chambers, including procedures
(chemicals and drying time) used between testing animals and between testing days;
•	Identify how animals are loaded into activity chambers and how individual test sessions
are started. If animals are stagger started, then identify precautions used to minimize
disturbance of animals already in the test chambers while other animals are loaded into
test chambers. If started simultaneously, identify procedures used to place animals in test
Motor Activity
B -9

-------
chambers and then start activity recording in all animals.
6. Data Reporting
The main results that are of interest from DNT motor activity testing are:
•	The effects of chemicals on overall level of motor activity (total session activity);
•	Effects of chemicals on habituation in motor activity;
•	Effects of chemicals on the ontogeny of motor activity in pre-weaning animals;
•	Sex differences in effects of chemicals on motor activity.
In order to evaluate the effects of chemicals on these parameters, it is essential that DNT motor
activity reports provide clear information on study design, the measures (dependent variables),
and the statistical procedures used to evaluate the main results. As noted in section 4, a number of
possible measures can be generated from automated motor activity systems depending on the
specific test equipment and software parameter settings employed. In the reporting and analysis
of motor activity data, it should be clearly stated whether the total of all activity measures have
been aggregated into a single measure (e.g., sum of ambulatory counts, vertical counts, and small
movements). Likewise, if motor activity data are reported as only ambulatory counts, then this
too should be clearly stated.
6.1 Dependent Variables
As described in section 4.2, there are a number of dependent measures that can be used
depending on equipment setup and software parameters, and it is important that the dependent
variables recorded and analyzed are unambiguously described. For most test systems, dependent
measures reported may be one or more of the following:
•	Ambulatory counts;
•	Ambulatory distance;
•	Ambulatory time;
•	Average speed of movement
•	Vertical movements (rearing);
•	Small or fine motor movements;
•	Total activity (some aggregation of multiple separate activity measures);
•	Habituation.
Assessment of treatment effects on habituation of motor activity is one of the main results from
DNT activity tests. However, habituation it is not a directly-measured dependent variable; it is
actually an inferred result based on within-session decreases in activity measures.
While habituation can be assessed in a variety of ways, the main characteristic of habituation is a
decrease in a response following repeated exposure to a stimulus that does not involve sensory
adaptation/fatigue or motor fatigue (Grissom and Bhatnagar, 2009; Leussis and Bolivar, 2006;
Rankin et al., 2009; Thompson, 2009; Thompson and Spencer, 1966). In the case of motor
activity tests, habituation to the test environment is measured as a decrease in motor activity over
the course of a single test session. Test sessions should be of sufficient duration that asymptotic
Motor Activity
B - 10

-------
activity levels in control rats are reached before the end of the session (US EPA guidelines specify
the last 20% of the session); the time necessary for habituation may vary across test arenas. The
precise method used to determine habituation in DNT motor activity study results should clearly
indicate how habituation is calculated. One measure is a calculation of the ratio of activity over
the first and second halves of the test session. Another approach is testing for a significant effect
of decreasing activity level over the time blocks for the duration of the test session (e.g., trend
analysis). Other habituation metrics may also be used (e.g., first time block relative to the last
time block). Regardless, the specific method to calculate habituation must be clearly described.
Many automated motor activity systems provide more than one measure of motor activity and
may report them separately as well as include all measures in a single aggregated total activity
measure. Since the total activity is correlated with the measures contained within it, these
variables are not independent and therefore cannot be evaluated independently. In some cases,
analyzing only total activity may be appropriate, but in other cases a more detailed analysis of
separate measures (e.g., both horizontally and vertically-directed activity) may provide more
detailed information. For example, it is possible that specific activity measures (e.g., vertical
counts) may be affected by chemical exposure when total activity is not affected by chemical
exposure. Justification for the choice of measures to report and/or analyze should be described.
6.2 Reported Data
In order to evaluate treatment effects, the following level of data detail should be reported:
•	Activity measures should be described broken down by dose group, sex and time
blocks at each test age. Note that the US EPA DNT guidelines require that each time
block be no greater than 10 minutes;
•	Activity measures should be reported as means ± standard deviations for each time block,
treatment group and sex;
•	Total session activity (i.e., collapsed across time blocks) should be reported as mean ±
standard deviation for each measure of motor activity (e.g., ambulatory counts, vertical
counts). Where total activity (i.e., all separate activity measures aggregated into a single
activity measure) is reported, these should be reported by time block, sex and treatment
group (means ± standard deviations) as well as collapsed across time blocks (means ±
standard deviations).
Standard errors may also be reported as an estimate of population variability, but note that this
measure may not be appropriate with very different sample sizes across groups.
7. Data Analysis
Measurement endpoints must be statistically analyzed using statistical methods appropriate for
repeated-measures data. Statistical analyses specific for DNT data are described by Holson and
colleagues (2008). As with all developmental data, the unit of analysis is the litter. Where a
male and female are sampled from the same litter, the activity must be analyzed with sex nested
within the litter, using litter as a random factor, or a hierarchical analysis with sex as a matching
factor. While there are a number of approaches for the analysis of activity data, repeated-
measures ANOVA is the most generally accepted approach. Data analysis should include an
overall ANOVA that includes treatment, sex and time blocks; one-way analyses of each sex can
Motor Activity
B - 11

-------
only be conducted in the face of significant interactions between sex and treatment. Furthermore,
demonstrating within-session habituation requires repeated-measures analyses in order to show
an effect of time block. Interval-by-interval analysis of dose effects at each time block is
absolutely not appropriate in the absence of a statistically significant interaction between time
and dose group from repeated-measures analyses.
A number of relatively common problems are evident in the presentation and analysis of DNT
data, and these limit the interpretation of study results and may produce misleading results.
These include:
•	Sex is often not included in the overall data analysis, with results for males and females
analyzed separately. This approach not only precludes direct testing of sex effects and
treatment-by-sex interactions, it also reduces statistical power to detect treatment effects;
•	Motor activity results are analyzed separately at each time block. This approach precludes
direct testing of a time-block effect, and any interactions associated with time blocks.
Because a significant time-block effect is normally required to support demonstration of
habituation, this approach precludes direct demonstration of habituation, or any treatment
effects on habituation;
•	Time block is not included in data analysis in cases where motor activity results are
reported only for total session activity (i.e., collapsed across time blocks). Total session
activity data reported as mean values collapsed across time blocks may mask treatment-
related effects, as differences can exist at separate time blocks (i.e., treatment-by-time
interaction).
In some laboratories, motor activity data can be highly variable with coefficients of variation
(CV) ranging from 20-100% or more, depending on age, test conditions and the specific measures
used. Since CV measures can provide an indication of the overall variability (and uncontrolled
variation) associated with the test situation and the inherent variability of the behavior, it is not
appropriate to use a fixed percent change in response (e.g., 20-25% change) in the absence of a
statistically significant difference as an index of a treatment effect. With mixed-effects-modeling,
a covariance matrix that best fits the data can be selected from many available types of covariance
matrices. For analysis of results within a specific age group, factors include sex, dose and time
blocks as the repeated-measure factor, keeping in mind that litter should also be included where
there is more than one animal from each litter. Generally, interactions should be examined first, as
significant interactions indicate that main effects are modified by a second factor. Interactions that
could be examined include treatment-by-time blocks, treatment-by-sex, and treatment-by-dose-
by-sex. Depending on the nature of the interactions (e.g., non-crossed interactions), main effects
may also be examined.
7.1 Statistical Models
Below is a brief description of the types of statistical effects that may be obtained from analyses
of DNT motor activity data:
•	Main effects:
o Treatment: indicates that chemical exposure affects the overall level of activity collapsed
across all other factors (i.e., sex, time blocks, and litter). This is effectively a direct
statistical test of the effect of treatment on total session activity since time blocks and sex
Motor Activity
B - 12

-------
are ignored in this statistical test. A significant treatment effect indicates that the exposure
affects total session activity;
o Sex: indicates a difference in response between males and females while all other
factors (treatment, time block, litter) are held constant;
o Intersession Time Block: indicates that activity levels vary across time blocks in the test
session. A significant effect of time block is required to demonstrate habituation. Note
that a significant effect of time is necessary but not sufficient to demonstrate habituation.
A significant effect of time may indicate an increase in activity levels over time blocks,
while habituation requires a decrease in activity levels over time blocks.
• Interactions:
o Treatment-by-time blocks: indicates only that the effect of time blocks varies between
treatment groups (or alternately, the effect of treatment varies between time blocks).
Normally this interaction is required to demonstrate that chemical exposure affects
habituation. However, the precise nature of the interaction must be evaluated to
determine how treatment groups differ over time blocks since, for example, it is possible
to have a significant treatment-by-time interaction that does not indicate habituation
(e.g., treatment increases activity levels over time in a test session but the size of the
increase varies between treatment groups). Potential differences in patterns of
habituation are illustrated in Figure 2. Acceptable approaches to assess the interaction
include simple effects tests or specific contrasts, at each time block. Another approach
is to treat time block as a continuous variable instead of as a categorical variable in the
model, and the habituation is described by the slope (i.e., coefficient) of time block. A
significant treatment-by-time block interaction indicates that habituation patterns are
different between the dose groups;
o Treatment-by-sex: indicates only that the effect of treatment differs between males and
females. Note that the nature of the interaction must be evaluated, usually with simple
effects or contrast interaction tests, or specific contrasts. It is possible to have a significant
treatment-by-sex interaction that reflects that males and females are both significantly
affected by the treatment in the same direction but the magnitude of the change differs
between males and females;
o Time blocks-by-sex: indicates that the pattern of activity across time blocks differs
between males and females;
o Treatment-by-time blocks-by-sex: indicates that the interaction between sex and
treatment varies over time blocks (or, alternately, that the effect of treatment varies
between males and females over time blocks). This interaction is normally required to
demonstrate that the effect of treatment on habituation differs between sexes. As with
other interactions, the nature of the interaction must be evaluated using contrast
interaction tests, simple- or slice-effects, or specific contrast. It is necessary to show that
the change in activity levels over the test session is influenced by treatment and that this
change in activity levels is affected differently between sexes.
In addition to ANOVA results described above, the actual magnitude of change should be
considered since effects of a large magnitude are sometimes not statistically significant due to
high variability, but could be of substantial biological significance (also see discussion of
biological significance in section 8.5). This is especially useful where critical significance values
(e.g., p-values) are inconsistent (e.g., results between age groups) due to variability in data, while
effects size measures may be comparable. Lastly, actual significance values of statistical tests
Motor Activity
B - 13

-------
should be reported rather than critical p-values like p<0.05 since all modern statistical software
programs report actual p-values. Reporting results of statistical testing using only critical p-
values such as p<0.05 can result in qualitatively different interpretation of test results where
p=0.048 and p=0.051 when there is no solid basis for treating these two statistical test results
substantially different.
For the analysis and statistical treatment of motor activity ontogeny in the preweaning period
(i.e., comparison of motor activity between preweaning test days), it is most useful to analyze
total session activity rather than activity-by-time blocks, since habituation has not developed at
younger ages (PND13/17). Thus, this analysis would involve a two-way ANOVA with total
session activity as the dependent measure, and age and treatment as independent variables. As
motor activity is typically not influenced by sex at these ages, sex may be excluded in the
analysis. In some cases, separate statistical analyses for each age period may be justified. These
include circumstances where there are significant differences in the test systems (equipment),
relative measurement scales, data variability, non-linearity, etc. While the pattern of ontogeny
may be evident, total-session activity data may not always achieve statistical significance due to
the large variability seen at younger ages. Despite this, qualitative evaluation of these results
should be conducted.
8. Interpretation of Motor Activity Data
8.1 Properties of motor activity control data
Regardless of the system used to measure motor activity, there are a number of behavioral
properties in the measured data that should be evident and should serve as a reference point for
evaluating reported results. Because of differences among the automated test systems
commercially available, direct comparisons of absolute ambulatory counts between equipment
types or laboratories are not likely to be useful. However, properties of behavioral measures of
activity that should be present in control animals include:
•	Control animals should demonstrate age-appropriate levels of activity;
•	Control animals should show age-appropriate levels of variability in activity measures;
for example, Figure 1 shows how the variability decreases with age;
•	Habituation of ambulatory activity should be evident in animals tested at or after about three
weeks of age.
The absence of these properties in control animals indicates that the normal activity pattern is not
present in control animals and may indicate some unspecified problem in the test procedures.
Unambiguous interpretation of treatment effects on overall activity and habituation under these
conditions is not possible since control animals are not exhibiting the behavioral profile that is
being evaluated for a treatment effect. Historical control data can provide perspective regarding
pattern of control activity at different ages based on equipment and laboratory procedures.
Motor activity positive control data should demonstrate the ability of the test equipment to detect
both increases and decreases in motor activity, and are therefore important for evaluating the
relative proficiency of a testing laboratory in detecting chemical-induced changes. Positive
control data help verify the ability of laboratory personnel in a realistic environment, and can help
Motor Activity
B - 14

-------
to characterize intra-laboratory and inter-laboratory reliability of a test method. The methods
should be completely described, and must be the same as those used in the study being evaluated
(e.g., the same equipment should be used, motor activity sessions should be of the same duration,
the same number and strain of animals should be used). The data presentation should be
sufficiently complete to permit evaluation of the sensitivity of the method, including individual
data and measures of variability. The positive control data should have been collected within a
reasonable time frame before the current study (e.g., the last few years). New positive control
data should also be collected when personnel or other critical laboratory elements change. This
new positive control data, while important for interpretation of treatment-related effects, can also
be used to determine whether a shift has occurred in the control data collected by the testing
laboratory. The lack of adequate positive control data can result in low confidence in negative
data which in turn increases the possibility that negative results may in fact be a false negative
finding.
8.2	Variability in motor activity data
Variability of control data is a key to understanding the laboratory's experimental control of the
study. Historical control data from laboratories conducting the US EPA DNT guideline studies
consistently demonstrate that motor activity is much more variable at PND13 and 17 than at later
ages (Raffaele et al., 2003, 2008). Mean CV's for 14 laboratories was approximately 50-140% at
PND13; 40-100% at PND17; 20-60% at PND21 and 18-30% at PND60 (Raffaele et al., 2008;
data based on inspection of graphs). It is possible that some of this variability could be reduced
by improving environmental experimental conditions (Raffaele et al., 2008). The high CV's in
data collected at PND13 and 17 may be due, in part, to lower levels of motor activity as well as
higher intrinsic variability during a period of rapid development of motor function when eyes are
beginning to open. Other (non-DNT) studies in adult rats reported study means of 20- 25% CV
(Crofton et al., 1991), whereas Moser et al. (1997) reported mean CVs of 20-53% across studies
from laboratories worldwide.
Excessive variation limits the ability to detect statistically significant treatment effects by
increasing error variance. Where high levels of variability are reported, checking raw data values
for unusual or extreme measures can help determine if treatment effects are masked by odd or
unusual measures from individual animals that inflate variability and thus decrease power to
detect treatment effects. Another potential reason for high CVs is that the data are not normally
distributed, warranting evaluation of the underlying distribution or transformation of the data
before analyzing. Review of the laboratory's historical control data (see section 8.1) will also
provide a mechanism for assessing the variability in submitted studies. Increased variance in
treated groups may also affect statistical power to detect significant differences but this increased
variance may also be an important indicator of a treatment effect.
8.3	Evaluation of Treatment effects
A review of the data for each treatment group is necessary to look for unusual values,
measurement errors, etc. Variability in treatment groups may be higher than in controls since
there can be individual differences in sensitivity to the treatment, resulting in a range of
responses. It is also possible that a single outlier value could reflect a chamber failure or some
other non-treatment related factor. Thus, a survey of the actual values within each treatment
group is an important first step in analyzing the data. For simplicity, it may be useful to examine
Motor Activity
B - 15

-------
the total session data before the habituation data.
Activity data should be appropriately analyzed across all treatment groups, and in the face of
significant main effects or interactions, post-hoc tests may be conducted to either determine
which groups are different from control (e.g., Dunnett's) or have multiple comparisons across all
dose groups (e.g., Tukey-Kramer, or other methods that protect for alpha inflation from multiple
comparisons). In studies where there are multiple levels of treatment (i.e., doses or
concentration), changes in activity are more straightforward when there is a graded change (i.e.,
dose-response). There may be studies in which only the highest dose produces statistically
significant changes, which could also be considered a form of dose-response. More difficult,
however, are cases in which there are significant differences from control at only the low or
middle doses, or some other non-linear response.
8.4 Treatment-related effects on ontogeny
A shift, or alteration, of the pattern of motor activity development may be an indication of
treatment effects that is somewhat different from evaluating only habituation or total session
data. For example, effects may not occur until a certain age, then persist, whereas other effects
may only be evident early and then revert to control levels. An example of different patterns of
changes was presented by Ruppert et al. (1985), where all three metals under study produced
hyperactivity by PND21 but the patterns of change differed in timing, onset, and development of
this effect. These specific changes are described in Table 2. Clear patterns such as these,
however, are more difficult when only three days are sampled (as in current DNT protocols)
compared to the daily sampling conducted in this study.
Table 2. Example of chemical-induced changes in the ontogeny of motor activity in pre-weaning
rats.
Metal
Age at testing*



PND13-15
PND16-18
PND19-21
Cadmium
Decreased
activity
Decreased activity but only on
PND16; PND17 & 18 at control
levels
Increased activity, peak
at PND20
Triethyl tin
No change
Increased activity, peak at PND18
Increased activity
Trimethyl
tin
No change
Increased activity, peak at PND16
Increased activity
* note that in this study, pups were tested daily from PND13 to 21. The data have been
collapsed across 2-3 days at a time for presentation here.
Motor Activity
B - 16

-------
8.5.	Statistical versus biological significance
The conclusion of whether or not a chemical exposure has resulted in adverse outcomes on
neurodevelopment should not rely solely on evidence of a statistically significant treatment-
related effect. Holson and colleagues (2008) provide a well-balanced and in-depth review of
appropriate statistical models for use with DNT studies. The use of appropriate statistical models
for data analysis is critical to support conclusions. Beyond the discussion presented in section 7,
a reviewer not conversant in statistical models may want to seek professional assistance before
making any decisions about the nature of findings in a DNT study.
Issues to consider when assessing the biological significance of effects in DNT studies include
variability in the reported data and the implications of this on the potential for false-negative
(Type II error, more likely) or else false-positive (type I error) results. In addition, statistical
significance of treatment effects does not necessarily mean that the statistically significant effect
is biological relevant. Because of the large number of potential endpoints in DNT studies, care
must be exercised in ensuring that experiment-wise errors rates are appropriately controlled to
minimize false positives (Holson et al. 2008). However, guideline DNT studies are conducted to
screen chemicals for possible adverse effects on the developing nervous system and are often the
only study examining all of these endpoints. Thus, consideration of a higher false positive rate
rather than a lower false negative rate may be a more conservative approach in some cases.
Most importantly, biologically significant, treatment-related findings may occur in the absence of
statistical significance for a variety of reasons. Large variance in the measured behavior may
occur due to poor control over testing environment or testing procedures, improper
implementation of testing equipment, or inherent variability in the behavior. If reported
variances in a given behavioral test compared with other reports from the same laboratory
(historical controls) or with other laboratories suggest excessive variance within a study, the
study itself may be flawed, at least for the parameter under consideration. This test would be
considered in the context of the other measures in the study, at which point the particular
parameter may not be used, or this could warrant a repeat of the study with better experimental
control. It is critical to remember that statistical models are one of many tools that assist in data
interpretation, not a sole replacement for expert judgment and careful scientific interpretation.
8.6.	Dose-response results
Motor activity data typically show decreased levels as a result of treatment; however, both
increased activity and inverted U-shape curves (increased activity only at a middle or lower
dose) are not unusual. Increased activity in acute studies represents stimulatory effects on the
nervous system. Persistent increases in activity may also reflect connectivity alterations in
specific brain regions, but such associations have not been carefully characterized. In studies of
acute effects (not typical in DNT studies), inverted U-shapes can represent decreases in
inhibitory systems producing excitation, followed by generalized decreases (depression).
Biphasic response data have also been reported in offspring following perinatal exposure to
neurotoxicants (Gilbert et al., 1999; Rasmussen and Newland, 2001). Furthermore, it is not
uncommon for only lower doses to be effective on behavioral measures, which may indicate that
compensation or other mechanisms of toxicity may play a role higher doses (Robertson and
Motor Activity
B - 17

-------
Grutsch Jr., 1987). Thus, while monotonic decreases in activity are common, increased activity
as well as non-linear dose-response data may also be relevant and should be closely evaluated.
8.7. Habituation results
Demonstrating a significant treatment effect on habituation requires that control animals display
normal habituation patterns observed around weaning or in adults. Generally, this would be
indicated by a decrease in motor activity in control animals over the course of a test session;
statistically this would be supported by a significant effect of time blocks in control animals.
The demonstration of treatment-related effects on habituation would normally be reflected by
alteration in the rate of change in motor activity levels over the course of the test session relative
to the rate of change in motor activity levels in control animals. In Figure 2, different patterns are
shown that may be observed as a function of dose of a single chemical; however, such patterns
may also be produced by different chemicals. For example, both Dose 2 and Dose 3 show
alterations in the rate of change in ambulation over the test session compared to control animals.
Note that it is possible for treatments to either increase or decrease habituation. While
habituation would be reduced to comparable levels in both Doses 2 and 3 (note parallel response
curves), Dose 2 differs from Control in the last half of the test while Dose 3 differs from Control
in the first half of the test. Dose 4 provides an example of potentiated or enhanced habituation,
where ambulation counts are lower than Control levels in the second half of the test but not the
first half of the test. Figure 2 also illustrates that reporting only total session counts may mask
alterations in habituation. For instance, Dose 3 and 4 show reduced and enhanced habituation
compared to Control, respectively, but total session counts for Dose 3 and 4 are not different
from Control.
Ambulatory Counts by Time Blocks
Total Session Ambulatory Counts
¦—¦ Control
¦—¦ Dosel
A 4 Dose 2
¦ ¦ Dose 3
•—¦ Dose 4
Control
Dose 2
Dose 3
Dose 4
Motor Activity
B - 18

-------
Figure 2: Examples of habituation data. Normal habituation (decreased ambulation over the test
session; left panel) in adult rats is shown in Control and Dose 1, although Dose 1 shows overall
lower ambulatory activity (right panel). Dose 2 and Dose 3 have comparable habituation patterns
but have a reduced rate of habituation compared to Control and Dose 1. Dose 4 exhibits an
enhanced rate of habituation compared to all other groups as illustrated by the more rapid
decrease in ambulatory counts in the second half of the test session. The right panel illustrates
total session activity collapsed across time blocks. As can be seen, without data presented as time
blocks, it is not possible to evaluate habituation. In addition, Dose 3 and Dose 4 show
comparable total session activity despite exhibiting opposite patterns of habituation. Similarly,
Dose 2 and Dose 3 have different levels of total session activity despite having comparable
patterns of habituation. In the absence of time block data, the conclusion that Dose 3 and Dose 4
are comparable (i.e., no treatment differences) would be erroneous. Similarly, while total session
activity is comparable between Control and Dose 3 and Dose 4, using only total session activity
would mask the fact that Dose 3 and 4 exhibit opposite alterations in habituation patterns and
both differ from the Control habituation pattern.
9. General interpretation
Because motor activity is an apical behavior, non-specific or indirect effects on the nervous
system from abnormal maternal behavior, malnutrition, growth retardation and/or other more
general effects on development can also result in changes in motor activity (Li 2005).
Alterations in motor activity may also reflect other effects such as altered sensory function and
altered reactivity (emotionality) that result in behaviors that are inconsistent with motor activity
(e.g., altered reactivity producing freezing). When supported by the appropriate statistical
analyses, sex-specific alterations in activity (e.g., effects in one sex only) may suggest endocrine-
mediated effects or differences in toxicokinetic parameters.
Although these alterations in motor activity may reflect neurotoxic effects, they may not
necessarily reflect specific neurotoxic effects on motor function per se. While this is a
disadvantage in specifying the nature of the neurotoxic effects, motor activity does have the
advantage of monitoring a range of neurotoxic effects in addition to motor function. Because
alterations in motor activity can be influenced by a variety of factors (e.g., effects on body
weight), it is important for changes in motor activity to be evaluated in the context of other
effects detected in DNT and other toxicity studies. It is also important to recognize that effects
not specific to neurotoxicity could co-occur with treatment-related alterations in nervous system
function or development. Changes in motor activity should not be dismissed or considered
secondary to these non-specific toxicity findings.
Motor Activity
B - 19

-------
10. References
Altman J and Sudarshan K. (1975) Postnatal development of locomotion in the laboratory rat.
Animal Behavior, 23, 896-920.
Bolles RC and Woods PJ. (1964) The ontogeny of behaviour in the albino rat. Animal Behavior,
12, 427-441.
Bronstein PM. (1972) Open field behaviour of the rat as a function of age. J. Comp. Physiol.
Psychol., 80, 335-341.
Campbell BA and Mabry PD. (1972) Ontogeny of behavioral arousal. A comparative study. J.
Comp. Physiol. Psychol., 81, 371-379.
Campbell RA and Raskin LA. (1978) Ontogeny of behavioral arousal: the role of environmental
stimuli. J. Comp. Physiol. Psychol., 92, 176-184.
Clarac F, Brocard F, and Vinay, L. (2004) The maturation of locomotor networks. In: Progress
in Brain Research: Brain Mechanisms for the Integration of Posture and Movement (Ed. by
D.G.S.Shigemi Mori), pp. 57-66. Elsevier.
Crofton KM, Howard JL, Moser VC, Gill MW, Reiter LW, Til son HA, and MacPhail RC.
(1991) Interlaboratory comparison of motor activity experiments: implications for
neurotoxicological assessments. Neurotoxicol. Teratol., 13, 599-609.
Denenberg VH. (1969) Open-field behavior in the rat: what does it mean? Ann. NY Acad. Sci.
169, 852-859.
Gilbert ME, Mack CM, and Lasley SM. (1999) Chronic developmental lead exposure and
hippocampal long-term potentiation: biphasic dose-response relationship. Neurotoxicol. 20,
71-82.
Grissom N and Bhatnagar S. (2009) Habituation to repeated stress: Get used to it. Neurobiol.
Learning Memory, 92, 215-224.
Holson RR, Freshwater L, Maurissen JPJ, Moser VC and Phang W. (2008) Statistical issues and
techniques appropriate for developmental neurotoxicity testing: A report from the ILSI
Research Foundation/Risk Science Institute expert working group on neurodevelopmental
endpoints. Neurotoxicol. Teratol., 30, 326-348.
Kelley AE. (1993) Locomotor activity and exploration. In: Techniques in Behavioral and Neural
Sciences: Methods in Behavioral Pharmacology (vol 10) (Ed. by F.van Haaren), pp. 499-
518. Elsevier, New York.
Leussis MP and Bolivar VJ. (2006) Habituation in rodents: A review of behavior, neurobiology,
and genetics. Neurosci. Biobehav. Rev., 30, 1045-1064.
Li AA (2005) Regulatory developmental neurotoxicology testing: data evaluation for risk
assessment purposes. linviron. Toxicol. Pharmacol. 19, 727-733.
MacPhail RC, Peele DB, and Crofton KM. (1989) Motor activity and screening for
neurotoxicity. J. Am. Coll. Toxicol., 8, 117-125.
Maurissen JP and Mattsson JL. (1989) Critical assessment of motor activity as a screen for
neurotoxicity. Toxicol. Ind. Health, 5, 195-202.
Moorcroft WH. (1971) Ontogeny of forebrain inhibition of behavioral arousal in the rat. Brain
Res., 35, 513-522.
Moser VC, Becking GC, Cuomo V, Frantik E, Kulig BM, MacPhail RC, Tilson HA, Winneke G,
Brightwell WS, De-Salvia MA, Gill MW, Haggerty GC, Hornychova M, Lammers J, Larsen
JJ, McDaniel KL, Nelson BK, and Ostergaard G. (1997) The IPCS Collaborative Study on
Neurobehavioral Screening Methods: IV. Control data .Neurotoxicol., 18, 947-968.
Motor Activity
B -20

-------
Muir G. (2000) Early ontogeny of locomotor behaviour: a comparison between altricial and
precocial animals. Brain Res. Bull. 53, 719-726.
Oakley DA and Plotkin HC. (1975) Ontogeny of spontaneous locomotor activity in rabbit, rat,
and guinea pig. J. Comp. Physiol. Psychol. 89, 267-273.
Raffaele KC, Fisher, Jr JE, Hancock S, Hazelden K and Sobrian SK. (2008) Determining normal
variability in a developmental neurotoxicity test: a report from the ILSI Research
Foundation/Risk Science Institute expert working group on neurodevelopmental endpoints.
Neurotoxicol. Teratol. 30, 288-325.
Raffaele KC, Sette WF, Makris SL, Moser VC, Crofton KM. Motor activity in developmental
neurotoxicity testing: A cross-laboratory comparison of control data. Presented at the
annual Society of Toxicology meeting, 2003.
Rankin CH, Abrams T, Barry RJ, Bhatnagar S, Clayton DF, Colombo J, Coppola G, Geyer MA,
Glanzman DL, Marsland S, McSweeney FK, Wilson DA, Wu CF, and Thompson RF.
(2009) Habituation revisited: An updated and revised description of the behavioral
characteristics of habituation. Neurobiol. Learn. Mem. 92, 135-138.
Rasmussen EB and Newland MC. (2001) Developmental exposure to methylmercury alters
behavioral sensitivity to D-amphetamine and pentobarbital in adult rats. Neurotoxicol.
Teratol. 23, 45-55.
Robertson AD and Grutsch, Jr JF. (1987) Biphasic responses, quantal signals and cellular
behaviour. J. Theor. Biol., 125, 41-60.
Ross JF. (2001) Tier I neurological assessment in regulated animal safety studies. In: Handbook
o/Neurotoxicology (Ed. by E.J.Massarro), pp. 461-506. Humana Press, New Jersey.
Ruppert PH, Dean KF, and Reiter LW. (1985) Development of locomotor activity of rat pups
exposed to heavy metals. Toxicol. Appl. Pharmacol. 78, 69-77.
Shaywitz BA, Gordon JW, Klopper JH, Zelterman DA, and Irvine J. (1979) Ontogenesis of
spontaneous activity and habituation of activity in the rat pup. Dev. Psychobiol., 12, 359-
367.
Shriner AM, Drever FR, and Metz GA. (2009) The development of skilled walking in the rat.
Behav. Brain Res. 205, 426-435.
Slikker, Jr W, Acuff K, Boyes WK, Chelonis J, Crofton KM, Dearlove GE, Li A, Moser VC, and
Newland C. (2005) Behavioral test methods workshop. Neurotoxicol. Teratol. 27, 417-427.
Thompson RF and Spencer WA. (1966) Habituation: a model phenomenon for the study of
neuronal substrates of behavior. Psychol. Rev., 73, 16-43.
Thompson RF (2009) Habituation: A history. Neurobiol. Learn. Mem. 92, 127-134.
Vinay L, Brocard F, Clarac F, Norreel JC, Pearlstein E, and Pflieger JF. (2002) Development of
posture and locomotion: an interplay of endogenously generated activities and neurotrophic
actions by descending pathways. Brain Res. Brain Res. Rev, 40, 118-129.
Vinay L, Brocard F, Pflieger J, Simeoni-Alias J, and Clarac F. (2000) Perinatal development of
lumbar motoneurons and their inputs in the rat. Brain Res. Bull. 53, 635-647.
Motor Activity
B -21

-------
MODULE C
EVALUATION OF AUDITORY STARTLE
RESPONSE DATA
Auditory Startle
C- 1

-------
1.	Introduction	3
2.	Test Description	3
3.	Guideline Requirements	4
4.	Auditory Startle Response Testing Systems	6
4.1 Testing Equipment	6
5.	Test Procedures	9
6.	Data Reporting	9
6.1.	Dependent Variables	11
6.2.	Reported Data	14
7.	Data Analysis	14
7.1. Statistical Models	16
8.	Interpretation of Startle Response Results	18
8.1.	Properties of startle control data	18
8.2.	Variability in auditory startle data	21
8.3.	Positive Controls	22
8.4.	Statistical versus biological significance	23
8.5.	Dose-response results	23
8.6.	Treatment-related Habituation results	25
9.	General Interpretation	28
10.	References	31
Auditory Startle	C - 2

-------
MODULE C - EVALUATION OF AUDITORY STARTLE RESPONSE DATA
1.	Introduction
The design of the DNT study plays a critical role in determining the overall quality, reliability,
and interpretation of startle data in DNT studies. Described herein are factors more specifically
associated with the design and conduct of auditory, or acoustic, startle response testing in DNT
studies as well as considerations in interpreting auditory startle data. This document is not
intended to be prescriptive but to provide background and the identification of design and test
details for auditory startle studies that need to be considered in the interpretation of test data.
2.	Test Description
The US EPA Test Guidelines (TG) (870.6300) and the OECD TG 443 both require a test of
auditory startle. The OECD TG 426 is less specific, only requiring testing of "motor and sensory
testing". TG 426 allows flexibility in how to measure motor and sensory testing and provides the
following examples: extensor thrust response, righting reflex, and auditory startle habituation.
The startle response is a characteristic set of reflexive muscle movements elicited by a sudden
intense sensory stimulus (Davis and Eaton, 1984). In toxicology studies, acoustic stimuli are
most commonly used to elicit the startle response, but visual and somatosensory stimuli have
also been used (Ison 1984). The auditory startle response, elicited by sudden intense acoustic
stimuli, is mediated by a simple neuronal circuit of four to six neurons between the cochlea, the
brainstem, and spinal ventral horn motor neurons (Davis and Eaton, 1984; Koch, 1999). For a
detailed review of the startle response the reader is directed to reviews of the physiology,
psychophysiology and anatomy of the response, and its use in toxicology (Crofton, 1992; Davis,
1980; Davis et al., 1982; Fechter et al., 1986; Geyer and Swerdlow, 1998; Hoffman and Ison,
1980).
In DNT studies, the auditory startle response test is used to assess the integrity of a sensory-
evoked motor response. In addition, the use of repeated test trials allows estimation of
habituation to the startle stimulus (Geyer and Swerdlow, 1998). Habituation is a basic form of
learning essential to adaptive behavior and critical for normal interaction of animals with their
environment. Habituation is normally measured as a decrease in response amplitude after
repeated startle stimulus presentation during the test session. Although not required or
commonly used, some DNT studies will also include tests of reflex modification (Crofton, 1992;
Ison and Hoffman, 1983; Li et al., 2009). Reflex modification is the change in a reflex response
due to a perceptible and antecedent change in the sensory environment (Ison and Hoffman,
1983). Auditory pre-pulse inhibition of the startle response is a specific form of reflex
modification. In DNT testing this typically involves the use of two auditory stimuli. One
stimulus is a supra-threshold stimulus that reliably elicits a startle response (i.e., eliciting
stimulus) and normally exceeds 110 decibels. The second stimulus (pre-pulse stimulus) is
presented at a sound pressure level that is subthreshold (i.e., does not produce a startle response
by itself).
Auditory Startle
C - 3

-------
For example, a prepulse stimulus of 20-50 msec duration may be presented 30-500 msec prior to
the presentation of the startle stimulus (Swerdlow and Geyer, 1998). An effective prepulse
stimulus will decrease the amplitude of the startle response, despite the fact that the prepulse
stimulus will normally have no impact when presented alone. Reflex modification, while rarely
used in regulatory studies, can provide valuable information to discriminate between a
dysfunction in the sensory or motor component of the auditory startle response (Crofton, 1992;
Young and Fechter, 1983).
When used in guideline DNT studies, auditory startle testing is conducted at an early post-
weaning age, as well as a young adult age near the end of the study. The ontogeny of auditory
startle response as well as reflex modification has been well studied. The startle develops late in
the second week of life in rats and mice. The startle response is tightly linked to maturation of the
cochlea and opening of external auditory meatus, with the hearing and musculature with
responses beginning within the second postnatal week (Brunjes and Alberts, 1981; Parisi and
Ison, 1979; Sheets et al., 1988; Shnerson and Willott, 1980). Development of auditory reflex
modification begins around 15 days of age and shows an adult pattern by PND21 (Parisi and
Ison, 1979). Sheets et al. (1988) studied the ontogeny of both the amplitude and latency of the
startle response, as well as the impact of background noise levels in rats from 13 to 21 days of
age. This work demonstrated a decline in response latency as age increased, but the behavior had
not completely reached adult status by PND21. Ontogeny of habituation of the response has not
been as well studied and is thought to mature somewhere between PND18 and 25. Early work by
Williams et al. (1975) found no habituation at PND17 and 'profound' decrement across repeated
trials at PND36. Other reports have shown habituation occurring by PND20 or PND23 (Goldey et
al., 1994; Wise et al., 1997), but not to the extent found in adults. It can therefore be concluded
that the startle response and habituation thereof, while not fully mature, are measureable within
days of weaning.
Startle testing is normally conducted using automated test equipment that can provide a wide
array of response characteristics, the most important of which are measures of amplitude and
latency (see section 4 below). Critical to proper interpretation of results in DNT studies is an
understanding of the test system and the measured outputs that are operationally defined by the
software used to process the response. This includes both the equipment used to produce, as
well as calibrate, the response measuring device as well as the acoustic stimuli used to elicit the
response.
3. Guideline Requirements
Neither the US EPA nor the OECD provide specific guidance on the types of test systems that
should be used (Table 1). The EPA guideline specifies the use of 5 blocks of 10 trials per session,
and provides a reference for specific details (Adams et al., 1985b) that outlines the use of an
automated testing system with specific acoustic stimuli, number of trials, and data collection
procedures. OECD 426 provides no specifics on testing methods for auditory startle, but does
have a series of references for both auditory habituation and reflex modification procedures.
OECD 443, similar to the EPA guideline, states that testing should include 50 trials, consisting of
data blocked into 5 blocks of ten trials, but also indicates that test conditions should be optimized
to produce intra-session habituation.
Auditory Startle
C - 4

-------
The remainder of this section is intended to provide a brief outline of the kinds of experimental
equipment needed to test auditory startle response habituation. Since reflex modification is a
more complex procedure, and is rarely used, the reader is directed to published reviews as stated
above for more information.
Table 1. Auditory startle response testing requirements listed in the US EPA and the OECD
guidelines.

US EPA
OPPTS 870.6300
OECD 426
OECD 443
Age of
testing
"around the time of
weaning and
around day 60"
Recommended ages for
adolescent testing are: learning
and memory = PND25±2;
motor and sensory function =
PND25±2. Recommended ages
for testing young adults is
PND60-70. "the minimum
number of times when
measurements should be
performed. Depending on the
anticipated effects, and the
results of the initial
measurements, it may be
advisable to add additional time
points (e.g., aged animals) or to
perform the measurements in
other developmental stages"
An auditory startle test
should be performed on
PND24 (±1 day) using
animals in cohort 2A.
The day of testing
should be
counterbalanced across
treated and control
groups.
Assignment
of pups
"one male or one
female from each
litter (total of 10
males and 10
females per dose
group)"
"20/sex (1/sex/litter)"
"the same or separate pairs of
male and female animals may
be assigned to different
behavioral tests."
"Ten male and 10
female cohort 2A
animals and 10 male
and 10 female cohort 2B
animals, from each
treatment group (for
each cohort: 1 male or 1
female per litter; all
litters represented by at
least 1 pup; randomly
selected) should be used
for neurotoxicity
assessments. Cohort 2A
animals should be
subjected to auditory
startle..."
Auditory Startle
C - 5

-------
Test device
"Details on the
conduct of this
testing may be
obtained in"
(Adams et al.,
1985a).
No specifics listed
No specifics listed
Study
design
"The mean
response amplitude
on each block of
10 trials (5 blocks
of 10 trials per
session on each
day of testing)
should be made.
While use of
prepulse inhibition
is not a
requirement, it is
highly
recommended
"Motor and sensory function
should be examined in detail at
least once for the adolescent
period and once during the
young adult period (e.g.,
PND60-70).
".. .examples of tests for motor
and sensory function are ...
auditory startle habituation"
Cited refs: (Adams, 1986,
Crofton, 1992, Crofton et al.,
1994, Crofton and Sheets,
1989, Davis and Eaton, 1984,
Ison, 1984)
"Each session consists
of 50 trials. In
performing the auditory
startle test, the mean
response amplitude on
each block of 10 trials
(5 blocks of 10 trials)
should be determined,
with test conditions
optimized to produce
intra-session
habituation. These
procedures should be
consistent with OECD
TG 426 (35).
Data
collection
No specifics listed
No specifics listed
No specifics listed
Data
reporting
"Auditory startle
response amplitude
per session and
intra-session
amplitudes on each
day measured."
"The mean and
standard deviation
for each
continuous
endpoint at each
observation time."
No specifics listed
No specifics listed
4. Auditory Startle Response Testing Systems
4.1 Testing Equipment
Auditory startle testing requires the following: 1) sound dampening chambers in which the
animals are tested, 2) equipment and software used to generate the auditory stimuli, and 3)
equipment and software needed to measure the response.
Auditory Startle
C - 6

-------
Most testing apparatuses include sound dampening test chambers. Sound dampening is important
for startle testing because the level of background noise during startle testing is known to impact the
amplitude of the startle response via a process called sensitization (Davis et al., 1984; Sheets et al.,
1988). The level of background noise impacts the startle response in an inverted-U shape function,
with increasing response levels tracking increasing background up to a point at which higher
backgrounds actually decrease response levels (Ison and Russo, 1990). The critical aspect of this
variable is to maintain background noise at a constant level throughout testing, and quantification of
this background level should be included in the report. Commercially available testing systems do
this by a combination of sound-dampening material and supplemental background white-noise, either
from speakers or venting fans, or both. Note that the level of background noise must be below any
acoustic stimuli used during testing.
The generation, presentation, and quality control of acoustic stimuli is a complex process. There are
a number of important issues that should be kept in mind when reviewing DNT studies, and should
be included in the reports, including:
•	Study reports should clearly document how the sound levels of acoustic stimuli were
calibrated. The appropriate scale for sound level is sound pressure level (SPL) measured in
dB. Note that the use of sound calibration equipment with "A-scale" weighting is not
recommended in most circumstances. This weighting scale was developed to mimic the
human hearing range and it not appropriate for use with rodents that can perceive sounds at
much higher frequencies than humans. A review of species specificity in hearing can be
found in Fay (1988);
•	Selection of speakers for use in presenting the acoustic stimuli should be capable of
responding in the proper frequency range with the proper frequency characteristics (e.g.,
rise and fall times, lack of acoustic transients). The use of a broad band signal is preferable
to pure tones;
•	Test chambers should permit sufficient movement, i.e., minimal restraint, so that animals
can generate startle responses. Note that the appropriate chamber size differs according
to age of the animal. A holder too small will confine the animal and prevent a full startle
response, and one too large could allow the animal to move away from the speaker,
altering the delivered stimuli.
There are two main types of equipment used to record the startle response. The first is a load cell
(or force transducer), and the second is an accelerometer. Because these sensors have different
properties and record different information, it is important to understand the differences between
these types of devices and the limitations of the data collected with them. Typically, both systems
output an analog voltage from the sensor to an analog-to-digital converter that transforms the analog
signal to a digital signal that is stored for analysis in a computer system. It should be noted that
commercial startle equipment has improved over the years, but DNT data that are reviewed may
have utilized either older or newer systems.
A load cell converts mechanical force into electrical signals that are output as voltage. It measures the
force (mass and acceleration) applied to the testing platform by the animal. That force is influenced
by the animal's body weight, thus weight impacts the starting, or resting, point, as well as the
magnitude of startle response. Since the load cell and platform typically have very little movement,
Auditory Startle
C - 7

-------
the response is often presented as a mass measurement (e.g., grams, or force in grams). Most load
cells are based on strain gauges that use minute deformations in electrical wires to alter resistance in,
for example, a Wheatstone bridge circuit. With load cells a maximum voltage output is recorded at
the time of maximum force applied to the platform. An advantage of load cells is the ability to
calibrate output voltage to standardized weights, which allow voltage responses to be converted to
grams. Some load cells are based on piezoelectric sensors. However, a disadvantage of piezoelectric
sensors is that they have no voltage output under static, non-moving conditions, and thus lose the
ability to calibrate with standardized weights.
An accelerometer is a device that measures acceleration forces. Early startle response testing
systems used mono-phonograph cartridges resting on the testing platform. As the animal
responds, the platform moves and this movement is detected by the change in resistance of the
phono cartridge. Other versions of accelerometers are either piezoelectric or capacitance-based.
All of these devices measure the change in motion, i.e., the acceleration, of the testing platform.
With accelerometers, the maximum voltage output is recorded at the time that the platform is
undergoing the largest change in movement. Accelerometers are usually calibrated with pulse
calibrators ("thumpers") that provide a standardized movement to the platform. While some
systems allow for conversion of voltage output to Newtons, the use of only one calibrator may
not be adequate to determine linearity of the system needed for accurate conversion of output
voltages to Newtons at all output voltage levels. Some systems convert the voltage output to
units based on the software, which may not represent units that are comparable across
laboratories.
Regardless of the detection system used, most commercial test systems sample the detection
device at a rate of 0.5 or 1 kHz (sampling usually about every 1 or 2 msec) for up to 300 msec.
This allows for a collection of data points that can be used to calculate the main metrics
commonly reported. It is important to remember that the peak startle response is an extremely
rapid reflex that begins some 10-14 msec after onset of the eliciting stimulus, and peaks within
about 20-40 msec after the onset of the startle stimulus (Davis, 1980; Davis and Eaton, 1984).
Sample trace recordings from load cell and accelerator systems are illustrated in Figure 1. The
figures illustrate the three major metrics typically reported for most startle studies. These are:
response magnitude or amplitude, latency to response onset, and latency to peak of response.
Peak Response
Peak response amplitude
Amplitude
Latency to Peak Response
Latency to peak response
7 \\
1? 150
< 100
Response
Onset
Startle Stimulus
Onset
Startle Stimulus
Onset
20	30
Time (msec)
Force Transducer
20
Time (msec)
Accelerometer
Auditory Startle
C - 8

-------
Figure 1. Trace recordings of startle responses from adult male rats exposed to a 115 dB auditory
startle stimulus. The left panel shows results from a force transducer (load cell measured in grams of
force) while the right panel shows results from an accelerometer. Note that the response onset and
peak response times are comparable in both systems. Note that the right panel shows two peaks and
in some cases the second peak is the maximum peak. Latency to startle onset is the time between the
onset of the startle stimulus and the onset of the startle response.
5.	Test Procedures
Many experimental (e.g., environmental conditions, lighting, noise, test apparatus) and
organismic factors (e.g., age, test history, sex) are known to alter behavior (Claassen, 1994;
Slikker et al., 2005; Tyl et al., 2008). In addition, there are a number of factors that specifically
influence startle responses (Davis and Eaton, 1984; Hoffman and Ison, 1980). Procedures to
control these factors should be described in the protocol or experimental design and reported in the
results. These include:
•	Environmental factors such as sound levels, temperature in test area, lighting conditions,
time of testing relative to light-dark cycle;
•	Testing of treatment groups should be counterbalanced over test device, time of day and
each animal should be tested individually;
•	Standardized procedures (chemicals and drying time) for cleaning chambers,
including procedures used between testing animals and between testing days;
•	Identify how animals are loaded into chambers and how individual test sessions are
started. If animals are stagger-started, then identify precautions used to minimize
disturbance of animals already in the test chambers while other animals are loaded into
test chambers. If animals started simultaneously, identify procedures used to load all
animals quickly in test chambers and then start activity recording in all animals;
•	Testing history of the animals. It should be clearly described whether the animals have
been tested for any other behavioral evaluations prior to startle testing;
•	Identify if a period for habituation to the chamber is included in test procedures (e.g.,
5 min habituation to chambers prior to startle trials).
6.	Data Reporting
In order to permit critical evaluation of startle data reported in DNT studies, reports should
include sufficient detail to ensure that it is clear how data were collected and computed. These
settings affect the raw data generated, as well as the software-computed measures and in turn
will affect comparisons across studies between and within laboratories and comparisons across
ages (e.g., preweaning, adult) within a study. The information that should be provided includes:
•	Details of the recording instrumentation (i.e., whether the system uses a load cell or an
accelerometer) including software;
•	Details on how the recording system was calibrated and how this calibration was used to
calculate response units (e.g., grams, Newtons);
•	Definitions of all calculated variables (e.g., peak response magnitude, latency to peak
Auditory Startle
C - 9

-------
response);
•	Details on the testing chambers, including background noise levels;
•	Details on the generation of acoustic signals (frequency, duration, rise times, inter-trial
intervals), and how these signals were calibrated;
•	Details on procedures used to select response data. The first peak may be considered the
startle response even though it is not always the maximum peak. Indeed, it is not
entirely clear how multiple peaks should be evaluated (Grimsley et al, 2007) (Figure 1
shows examples of multiple peaks);
•	Criteria used to define non-startle responses as general movement (e.g., long latency
values for first peak would not be a startle response; small peaks may reflect movement
rather than startle). Note that animals do not always startle on every trial and this should
be addressed in data analysis. While averaging trials over blocks may help to lessen the
impact of trial-to-trial variability, treatment-related increases in the number of low- or
no-response trials may be detected by examining individual data.
•	Some laboratories include "blank" trials, which record response measurements without
an eliciting stimulus. This could reflect baseline activity, or non-stimulus related startles,
and should be reported against which to judge the response data.
The main results that are of interest from DNT startle testing are:
•	The effects of chemicals on the overall startle amplitude (average of startle response
collapsed over all trials or total session mean);
•	Effects of chemicals on habituation in startle magnitude over the test session;
•	Comparison of effects between the two ages of testing;
•	Sex differences in effects of chemicals on startle response amplitude and/or habituation.
In order to evaluate the effects of chemicals on these parameters, it is essential that DNT reports
provide clear information on study design, precise definitions of the dependent variables, and the
statistical procedures used to evaluate the results.
The terminology used in reporting these measures can vary among laboratories and, therefore, in
DNT reports. Table 2 lists the variety of measurement names that were used in a total of 45 DNT
studies submitted to the US EPA. This represents actual data submissions and is presented for
cross-reference when reviewing startle data. In many cases the amplitude variable is the
maximum amplitude of the response. Note that average response amplitude differs from the
maximum response magnitude in that it incorporates the average over the entire recording period
rather than the peak response magnitude. Response latency may be measured as latency to
response onset or to the peak startle response. Care should be exercised when reading the study
in order to determine exactly what the measurement is intended to represent.
Auditory Startle
C- 10

-------
Table 2. Common reporting terms for startle response data and biological measurement
interpretation

Reported Term
Measurement
amplitude (Newtons)
Amplitude
amplitude (volts)
Amplitude
maximum amplitude (no units)
Amplitude
maximum amplitude of the response (mV)
Amplitude
maximum impulse (arbitrary units)
Amplitude
maximum response (Vmax)
Amplitude
maximum response amplitude (Vmax)
Amplitude
maximum startle (voltage)
Amplitude
mean peak startle (no units)
Amplitude
peak amplitude (grams)
Amplitude
peak response amplitude (v)
Amplitude
reflex peak amplitude (V)
Amplitude
response (g)
Amplitude
latency (msec)
Latency
latency (Tmax)
Latency
latency to maximum response (msec)
Latency
latency to peak (msec)
Latency
latency to peak of the response (msec)
Latency
latency to peak response
Latency
mean time to maximum amplitude
Latency
peak latency (msec)
Latency
reflex latency (msec)
Latency
time of maximum startle (msec)
Latency
time to maximum response (Tmax)
Latency
time to maximum startle (msec)
Latency
time to peak amplitude (msec)
Latency
percent inhibition
Sensory inhibition
prepulse inhibition
Sensory inhibition
average response (grams)
Undefined in report
average response amplitude (mV)
Undefined in report
response duration (msec)
Undefined in report
6.1. Dependent Variables
There are a number of dependent measures that can be used depending on equipment setup and
software parameters, and it is important that the dependent variables recorded and analyzed are
unambiguously described. Based on the biology of the response, the following are the most
common startle response metrics:
• Maximum startle response -maximum amplitude of the response, recorded in grams for
load cells and as equipment-specific units for accelerometers;
Auditory Startle
C- 11

-------
•	Habituation - calculated metric for characterizing the change in startle response
magnitude over repeated exposure to the startle stimulus.
•	Latency to onset or peak - time in msec from the startle stimulus onset to the beginning
or to the peak of the startle response
Averages of the response magnitude and latency values should be calculated from individual trial
data. Both the averages for blocks of 10 trials (average for trial block) as well as the total session
average (average for all 50 trials) should be reported. These averages should include treatment
and sex factors. A measure of habituation must be reported. While habituation can be assessed in
a variety of ways, the main characteristic of habituation is a decrease in a response following
repeated exposure to a stimulus that does not involve sensory adaptation/fatigue or motor fatigue
(Leussis and Bolivar 2006; Grissom and Bhatnagar 2009; Thompson and Spencer 1966;
Thompson 2009). The precise method used to determine habituation should be clearly indicated.
Note that the initial trial typically has the highest startle response, and when averaging across
blocks this high value contributes to the higher values for the first block. Typically, habituation is
inferred from statistical analyses that demonstrate a significant effect of trial block that is due to a
decrease in startle response amplitude between the beginning and end of the test session. Figure 2
illustrates control habituation data from multiple laboratories for males at both young and adult
ages (Raffaele et al., 2008). Habituation may also be inferred from slope measurements (negative
slope for response magnitude from first to last trials) or difference in scores between first and last
trial blocks. Regardless, the specific method used to determine habituation must be clearly
described.
Auditory Startle
C- 12

-------
120
100
SO
60
40
2 20
*
a
uT
s
5
%
o
a.
ss
120
100
SO
CO
40
20
0
Young Female
—r~
2
Adult Female
120
100
so
so
40
20
120
100
so
BO
40
20
0
Young Male
—r~
2

Adult Male
Trial Bins (10 Trials/Bin)
Figure 2: Habituation data for amplitude of young (PND20-24) and adult (PND58-68)
male and female rats. These data were submitted from a number of different DNT
testing laboratories (not identified individually). Data are group mean responses for
controls plotted as a function of 5 10-trial blocks within a test session (data from
Raffaele et al., 2008).
Auditory Startle
C- 13

-------
6.2. Reported Data
In order to evaluate treatment effects, the following level of detail should be reported:
•	Startle measures broken down by dose group, sex and trial blocks at each test age. Note
that the US EPA DNT guidelines and the OECD 433 require that data are reported for 5
blocks of 10 trials/block for a total of 50 startle trials;
•	All dependent variables reported as means ± standard deviations for each treatment cell
(trial block, treatment group and sex combination). Additional statistical measures, such
as the median or mode, may also be presented where appropriate (e.g., unequal variance,
non-normal or skewed distributions);
•	Total session amplitude and latency (i.e., collapsed across all trials) reported as mean ±
standard deviations;
•	Details on calculations or analysis for habituation measures clearly described.
7. Data Analysis
Startle data must be statistically analyzed using methods appropriate for repeated-measures data.
Statistical analyses specific for DNT data are described by Holson and colleagues (2008). As
with all developmental toxicity data, the unit of analysis must be the litter. Where a male and
female are sampled from the same litter and data from both are included in data analysis, the data
must be analyzed with sex nested within the litter, using litter as a random factor, or a
hierarchical analysis with sex as a matching factor. Data analysis should use a repeated-measures
analysis of variance (ANOVA) (or comparable procedure) that incorporates treatment and trial
blocks in the analysis. Separate ANOVAs on results for each sex separately can normally only be
justified after finding a significant sex-by-treatment interaction in an overall ANOVA.
Using ANOVA, demonstrating habituation would usually require the detection of a significant
trials (or trial block) effect that is associated with decreased startle response magnitude over
testing. Other analytical approaches that demonstrate a significant decrease in response
magnitude (e.g., regression approaches) can also be employed. It is not appropriate to perform
separate comparisons among treatment groups at different trial blocks in the absence of a
statistically significant interaction between time and treatment.
While there are a number of statistical models for analysis of startle data, repeated-measures
ANOVA is the most generally accepted approach (Holson et al., 2008). With mixed-effects-
modeling, a covariance matrix that best fits the data can be selected from many available types of
covariance matrices. For analysis of results within a specific age group, factors include dose and
trials as the repeated-measure factor, keeping in mind that litter must be included where there is
more than one animal from each litter. Generally, interactions should be examined first, as
significant interactions indicate that main effects are modified by a second factor. Potential
interactions would include treatment-by-trial blocks, trial block-by-sex, treatment-by-sex, and
treatment-by-trial block-by-sex.
Auditory Startle
C- 14

-------
Depending on the nature of the interactions, main effects may also be examined. More
specifically, crossed interactions usually preclude the use of main effects while for parallel
interactions main effects can still provide valuable information.
The impact of body weight on startle response magnitude has been clearly established. For
example, it has been repeatedly demonstrated in studies of the ontogeny of the response where
body weight gain increased the amplitude of the response (Dean et al., 1990). In addition, where
male body weight exceeds female body weight in mature animals there is a corresponding
increase in the amplitude of the startle response. The use of body weight as a covariate in
analyzing startle response data can be useful to reduce error associated with body weight.
However, the use of body weight as a covariate for startle response amplitude data is appropriate
only when there are no treatment-induced effects on body weight (Csomor, 2008; Holson et al.,
2008; Tabachnick and Fidell, 2006). On the other hand, failing to control for body weight could
result in erroneous conclusions, especially regarding sex differences.
There are many examples from the literature where a simple correlation between a chemical's
effect on body weight and the startle response may be independent of treatment effects on startle
responses. For example, developmental exposure to methimazole results in a small decrease in
body weight in male animals but no corresponding effects on startle response. At the same time
female body weight is unchanged but the startle response magnitude in females is decreased
(Henck et al., 1996). There are also study results where no changes in startle response
magnitude occur at young or old test ages despite decreased developmental weight gain (Buelke-
Sam et al., 1998). Similarly, in young animals, decreases as well as increases in startle response
have been recorded in animals having concomitant treatment-related decreases in weight gain
during development (Goldey et al., 1995; Kobayashi et al., 2005). Some have reported no
effects on startle response after weight loss of up to 25% (Crofton and Knight, 1991). Figure 3
illustrates an example of the independence of weight loss and effects on startle response
magnitude. The graph on the left shows a dose-related decrease in PND22 female body weights
in the absence of any effect on startle response amplitude. The graph on the right shows the
converse situation where a dose-related decrease in startle response magnitude occurs in the
absence of any dose-related change in body weight.
Taken together, even though both age-related changes in body weight and sex differences in
body weight do influence startle response magnitude, the available data indicate that treatment-
induced effects on body weight are not necessarily associated with treatment-induced changes in
startle response magnitude. In the end it is best to examine both body weight and startle data
and determine the most appropriate or valid approach for analysis.
Auditory Startle
C- 15

-------
Females - PND22 - Compound Z	ASR PND20 Males - Coupound X
Increased ASR
1% Increased ASR
100 -
12.6% Decreased BW
90 -
Body Weight
O ASR
0
10
20
30
40
50
3.7% increase
100 -
5.2% decrease
90 -
70 -
Body Weight
ASR
43%
Arbitrary Dose	Arbitrary Dose
Figure 3. Example of independence of treatment-induced changes in body weight and startle
response amplitude (see text for details).
7.1. Statistical Models
Below is a brief description of the types of statistical effects that may be obtained from analyses of
DNT startle data.
• Main effects:
o Treatment: indicates that chemical exposure affects the overall level of startle
amplitude (or latency) collapsed across all other factors (i.e., sex, trial blocks,
litter). This is effectively a direct statistical test of the effect of treatment on the
overall total session startle since trial blocks and sex are ignored in this statistical
test. A significant treatment effect indicates that the exposure affects total session
startle amplitude (or latency);
o Sex: indicates a difference in startle response between males and females
while all other factors (treatment, trial block, litter) are held constant;
o Intersession Trial block: indicates that startle response varies across trial blocks
in the test session. A significant effect of trial block is required to demonstrate
habituation. Note that a significant effect of trial is necessary but not sufficient to
demonstrate habituation. A significant effect of trial may indicate an increase in
startle responses, or some other non-monotonic pattern, over trial blocks, while
habituation requires a decrease in startle responses over trial blocks.
Auditory Startle
C- 16

-------
Figure 4. Example of treatment-related alteration in startle
habituation. Adult males were tested for 50 trials and data are
presented as means of five 10-trial blocks (see text for details).
500'
0)
u
3
Q.
E
<
0)
(A
C
o
Q.
(A
0)
o£
400"
300
200"
100 -
*
\
fsr—^-a
" 
Trial Block
• Interactions:
o Treatment-by-Trial blocks: indicates that the effect of trial blocks varies among
treatment groups (or alternately, the effect of treatment varies between trial
blocks). The precise nature of the interaction must be evaluated to determine
how treatment groups differ over trial blocks since it is possible to have a
significant trial-by-treatment interaction that does not indicate habituation (e.g.,
treatment increases startle responses over trials in a test session but the size of
the increase varies between treatment groups). This is illustrated in Figure 4
where there is a similar decrease in startle amplitude over the five trial blocks in
all groups except the high dose group, which shows much less habituation. This
was evidenced statistically as a trial block-by-treatment interaction. Acceptable
approaches to assess the interaction include simple effects tests or specific
contrasts, at each trial block. Depending on the data, another simpler approach
is to treat trial block as a continuous variable (not as a categorical variable) in the
model, and the habituation is described by the slope (coefficient of trial block).
A significant treatment-by-trial block interaction indicates the slopes were
different among the dose groups;
o Treatment-by-sex: indicates that the effect of treatment differs between sexes.
Note that the nature of the interaction must be evaluated, usually with simple
effects or specific contrasts. It is possible to have a significant treatment-by-sex
interaction that reflects that males and females are both significantly affected by
the treatment in the same direction but the magnitude of the change differs
between males and females;
o Sex-by-Trial blocks: indicates a difference in habituation between males and females;
o Treatment-by-Trial blocks-by-sex: indicates that the interaction between sex and
treatment varies over trial blocks (or, alternately, that the effect of treatment varies
Auditory Startle
C- 17

-------
between males and females over trial blocks). This interaction is normally required to
demonstrate that the effect of treatment on habituation differs between sexes. As with
other interactions, the nature of the interaction must be evaluated using mean contrast
interaction tests, simple effects, or specific contrast. It is necessary to show that the
change in startle response over the test session is influenced by treatment and that this
change in startle response is altered differently between sexes.
In addition to ANOVA results described above, the actual magnitude of change should be
considered since large effects are sometimes not statistically significant due to high variability
but could be of substantial biological significance (also see discussion of biological significance
in section 8.4). This is especially useful where critical significance values (e.g., p-values) are
inconsistent (e.g., results between age groups) due to variability in data, while the magnitude of
effects may be comparable. Lastly, actual significance values of statistical tests should be
reported rather than critical p-values like p < 0.05 since all modern statistical software programs
report actual p-values. Reporting results of statistical testing using only critical p-values such as
p < 0.05 can result in qualitatively different interpretation of test results where, for instance,
p=0.048 and p=0.051 when there is no solid basis for treating these two probability values as
substantially different (Holson et al., 2008).
A number of relatively common issues occur in the data analysis of DNT reports that can limit
the interpretation of study results and may even produce misleading results. These include:
•	Sex is often not included in the overall data analysis, with results for males and females
analyzed separately. This approach not only precludes direct testing of sex effects and
treatment by sex interactions, it also reduces statistical power to detect treatment effects;
•	Trial block is not included in data analysis with results reported only for total session
startle response (i.e., collapsed across all trials). Not only does this approach preclude any
possibility of detecting treatment effects on habituation, it may mask treatment effects
where the effect of treatments varies over time blocks;
•	Using body weight as a covariate in data analysis when there is a treatment-related effect
on body weight (see discussion in section 7). Using body weight as a covariate is
acceptable when there are no treatment effects on body weight.
8. Interpretation of Startle Response Results
8.1. Properties of startle control data
Regardless of the system used to measure the auditory startle response, there are a number of
behavioral properties in the measured data that should be evident and should serve as a reference
point for evaluating reported results. Differences in the detection method (load cell vs
accelerometer) must be understood (see section 4) as they influence the use of control data and
the comparability of results between ages. Properties of startle measurement that should be
present in control animals include:
Auditory Startle
C- 18

-------
•	In adults, there is normally a sex-dependent effect on response amplitude with adult males
exhibiting larger startle response amplitude than adult females due to larger body and
muscle mass;
•	Habituation of startle amplitude should be evident in animals tested at or after weaning;
•	Control animals should demonstrate age-appropriate amplitude and latency measurements.
The reviewer is directed to two papers that contain a wealth of exemplary data on control startle
data from regulatory studies (Raffaele et al., 2008, Tyl et al., 2008). Selected examples from
these two papers and the peer-reviewed literature are presented herein.
Table 3 presents data that illustrate some comparisons that are expected in submitted data with
respect to age, sex, and body weight. Note that systems calibrated to an absolute standard (e.g.,
grams, Newtons) should show increased startle response between young and adult, and between
adult male and female animals; this may not be the case when the output is in units that are not
absolute, or may not be consistent across ages. Assuming that the testing system is calibrated to
an absolute standard (e.g., grams, Newtons), then there should be an apparent increase in the
startle response between young, post-weanling animals compared to adults. This can be seen in
Table 3 where males and females were tested at two ages (PND23) and (PND61) and are due to
the significant weight gain that occurs between these two ages (note that in some strains body
weight can increase 4 or 5-fold). Second, there are no significant differences in response
amplitude of PND23 males and females. This is because there is less impact of sex on body
weight or muscle mass at that age, compared to mature animals. Third, there is smaller response
amplitude in adult females compared to adult males. In this study, females at PND61 weighed
about 20% less than males. Fourth, habituation occurred in all groups. This is clearly evident
by a 30-40% decrease from the first block of trials to the last. The statistics presented in the
report support this position with a main effect on trial block.
Auditory Startle
C- 19

-------
Table 3. Example of auditory startle response amplitude data, from PND23 and PND61 male
and female rats. See text for details, (modified from Tyl et al., 2008)
Testing
Trial
Dose (as multiples of the lowest dose)


Age
Block
0
1
5

25



Males



PND23
1-10
379.6 ±98.8
439.3 ±149.2
372.9 ±99.2
302.5
±122.8

11-20
266.6 + 64.6
297.7 ±63.3
297.4 ±103.2
200.3
±64.4*

21-30
233.3 + 63.1
261.3 ±53.3
265.9 ±91.2
176.7
±48.0

31-40
223.7 ±50.9
235.4 ±47.0
251.0 ±72.6
144.6
±54.1**

41-50
208.0 ±55.4
220.7 ±45.7
221.6 ±62.8
148.0
±51.1**
PND61
1-10
1351 ±370.6
1692 ±563.1
1394 ±525.3
1305;
L393.7

11-20
898.6+.258.7
1053 ±280.6
952.8 ±367.3
871.5
±_228.9

21-30
861.2 + 346.1
960.0 ±336.2
846.6 ±350.7
878.1
±324.9

31-40
769.7 + 279.9
963.1 ±350.3
832.4 ±332.1
758.3
±229.3

41-50
701.9+.307.9
929.6 ±368.5
859.5 ±356.0
752.9
+ 270.6



Females



PND23
1-10
374.9 ±121.3
406.3 ±177.0
353.2 ±108.2
297.4
±62.7

11-20
332.9 ±143.1
306.7+ 130.3
281.0 ±122.5
232.8
±45.9

21-30
267.5 + 67.0
294.3 ±174.7
252.9 ±165.0
222.7
±55.0

31-40
234.7 ±58.3
230.1 ±128.6
190.0 ±59.2
202.4
±53.7

41-50
226.3+.44.1
249.0+ 140.1
199.1 ±72.7
189.9
±55.6
PND61
1-10
1064 + 225.1
974.4 ±222.8
1161 ±442.2
1237;
K321.2

11-20
874.2 ±170.2
913.2 ±277.0
1025 ±441.5
963.5
±156.5

21-30
772.0 + 234.1
831.4 ±305.2
908.7 ±588.7
926.2
±203.1

31-40
677.8 +.270.4
813.2 ±442.1
816.2 ±505.0
832.1
±315.4

41-50
652.6 ±185.4
633.5 ±378.1
716.1 ±380.7
798.6
±220.6
a N = 10-14/group. Mean data + SD. *
Statistically significantly different from control value at p< 0.05.
** Statistically
significantly different from control value atp< 0.01.



Table 4 illustrates control data that have some inconsistencies with the known biology of the
startle response in terms of age and sex. These data were generated using an accelerometer with
output listed as millivolts for the amplitude data (Vmax), milliseconds for the latency data
(Tmax), and average amplitude in millivolts (Vave). There is consistency in the fact that the
amplitude values at young ages have no apparent sex differences, and that amplitude values for
older females are lower than values in older males. In addition, the latency data are mostly
biologically coherent, in that the peak amplitude occurs at 22 to 34 msec. A number of issues in
this dataset, however, are apparent. The response amplitude for PND20 males is slightly higher
than the PND60 males and the amplitude of the PND20 females is more than double the
response for PND60 females. Another issue is that the lowest amplitude responses in PND60
females have longer latencies. Normally larger responses have longer latencies to the peak
response, although there are many reasons this may not occur. Additionally, there is no intra-
session information provided to determine if habituation occurred. Since both the EPA and
OECD 433 guidelines require habituation data, this report could be considered as not meeting
this requirement.
Auditory Startle
C - 20

-------
Table 4. Example of auditory startle response data from PND20 and PND60 male and female
rats (modified from Tyl et al., 2008).


Dose Group


Control
Low
Medium
High
MALES
PND20




Vmax (mv)
214.3 + 94.9
182.7 + 84.6
157.3 + 61.7
123.4 + 57.3
Tmax (msec)
24.9^3.6
24.1 ^3.4
25.4^3.5
24.3 + 3.1
Vave (mv)
45.5 + 18.6
39.0 + 16.9
33.8 + 11.0
26.6 + 11.6*
PND60




Vmax (mv)
209.9 + 129.0
189.2 + 76.0
126.0 + 81.5
98.6 + 66.6*
Tmax (msec)
31.5 ^4 .8
29.3 ^5.5
32.9^6.3
33.6^5.3
Vave (mv)
46.5 +_28.0
40.3 ^16.6
29.0^19.2
21.6^14.4*
FEMALES
PND20




Vmax (mv)
181.2 + 50.8
182.5 + 54.1
160.6 + 47.9
129.0 + 56.6
Tmax (msec)
22.7^2.6
24.0^2.6
24.6^4.6
25.3 +_2.0
Vave (mv)
39.5 + 10.0
37.9 + 10.5
34.4 + 10.9
27.4 + 12.5
PND60




Vmax (mv)
78.2 + 36.1
78.7 + 30.7
79.6 + 40.4
80.1 +37.0
Tmax (msec)
34.3 ^4.4
32.9^4.7
33.9^4.5
30.9^4.7
Vave (mv)
16.7^7.6
17.3 ^7.8
16.2^7.0
16.5^6.7
a N = 9-10/group. Data are mean j^SD


* Statistically significantly different from control group value, p< 0.05.

The issues noted in Table 4 should lead to a more thorough review of the report itself. The first
step should be a careful examination of the testing equipment, calibration of noise stimuli, and
calibration of the testing platform. It is possible that the equipment was properly calibrated and
that the gain (circuit amplification) was increased at the young age. Some equipment allows
this to be done via alterations to the sensitivity settings or gain switches when testing young
(and smaller) animals. If the system was properly calibrated and the sensitivity was increased
in the young animals, then this would explain that the lower startle response on PND60 actually
reflects the change in instrument sensitivity setting rather than an actual decrease in startle
response amplitude. Historical control data from the conducting laboratory would also be
useful in determining whether these data are consistent with previous use of this equipment.
Without such an explanation, it would be appropriate to question the adequacy and reliability of
the control data, thus obviating any need to look at possible treatment-related results.
8.2. Variability in auditory startle data
As with all biological measures, there is variability inherent in the auditory startle response. As
mentioned above, separation of experimental and inherent variance can be difficult (Cohen,
1977). There is an excellent review of the variability in auditory startle data from an ILSI
workshop (Raffaele et al., 2008). The findings of that report indicated that regulatory testing
laboratories had coefficients of variation (CVs; standard deviation as a percent of the mean) that
Auditory Startle
C - 21

-------
varied from 20-110% and that some laboratories are much better than others in controlling
experimental variability. Where high levels of variability are reported, checking raw data can
help determine if treatment effects are masked by odd or unusual individual animal scores that
inflate variability and thus decrease power to detect treatment effects. Another reason for high
variability is that the data may not be normally distributed, and a transformation may be needed
before analyzing the data. Review of the laboratory's historical control data (see section 8.1)
will also help in assessing the variability.
8.3. Positive Controls
The EPA DNT Guideline requires testing laboratories to provide positive control data that
demonstrate the ability of the test equipment to detect both increases and decreases in response
amplitude. Positive control data are crucial for evaluating the relative proficiency of a testing
laboratory in detecting chemical-induced changes in the measured endpoint, and serve as
reference standards for the measured endpoint. Positive control data help demonstrate the
dynamic range of a biological response, verify the proficiency of laboratory personnel in a
realistic environment, and can help to characterize intra-laboratory and inter-laboratory
reliability of a test method. Positive control data are valuable in a weight-of-evidence approach
to help determine the biological significance of results and provide confidence in negative
results. A number of reviews of the use of positive controls in DNT studies are available
(Crofton et al., 2004, 2008; Maurissen and Marable, 2005).
The methods used for positive controls should be completely described, and must be the same as
those used in the study being evaluated. For example, equipment used and strains and numbers of
animals tested should be the same. Sessions should contain the same number of trials and use the
same stimuli. Stimulus test parameters (stimulus intensity, duration, inter-trial intervals, etc.)
should be similar. The testing laboratory should use animals exposed in a manner that mirrors the
DNT protocol, although age-matched animals that are exposed just prior to the time of testing are
also acceptable (Crofton et al., 2008). The positive control data should have been collected
within a reasonable time frame before the data collection for the DNT study being evaluated
(e.g., within the last few years). New positive control data should also be collected when
personnel or other critical laboratory elements change. This new positive control data, while
more important for interpretation of treatment-related effects, can also be used to determine
whether a shift has occurred in the control data collected by the testing laboratory. In addition,
the reviewer should look to see if the same patterns are evident between sexes and ages.
During assessment, care should be taken to ensure that the positive control treatment is effective.
In a review of positive control data submitted to the US EPA, 3 of 16 laboratories never
submitted positive controls for startle testing, 4 laboratories failed to demonstrate proficiency
(i.e., lack of effects of positive controls), and 3 laboratories were deemed to have only partially
established proficiency (e.g., lack of statistical significance, lack of appropriate age animals).
Only 5 of 16 testing laboratories were judged to be proficient based on sufficient positive
control data (Crofton et al., 2004).
The lack of adequate positive control data can result in low confidence in negative data which in
turn increases the possibility that negative results may in fact be a false negative finding. A
simple example illustrates the importance of positive controls for data interpretation. In this
Auditory Startle
C - 22

-------
particular case, the test laboratory submitted startle data from males and females tested on
PND23 and PND62. There were no effects reported for any aspects of startle testing in this
study. There was evidence of habituation, but only in the adult age. No historical control or
positive control data were available for PND23. The positive control data from this laboratory
were available for female adult rats only and the treatment was without effect. Consequently, it
was not clear whether the test laboratory's procedures were effective in detecting any treatment-
induced changes in startle responses. In a case like this the deficiency should be noted in the
positive control data.
8.4.	Statistical versus biological significance
Relying solely on statistical differences or a fixed percent change in response (e.g., 20-25%
change) is not always appropriate because of the inability to differentiate inherent variability
from experimental error (Cohen, 1977). Excessively large variances in data, regardless of the
source, make detection and interpretation of treatment-related findings difficult. In case of
experimental-error induced variability, the use of statistics increases the rate of false negatives.
Large variance due to inherent variability and the use of a fixed change in the response will also
increase false negatives. Although increased variance in treated groups may affect statistical
power to detect significant differences (Muller and Benignus, 1992), it may also be an
important indicator of a treatment effect (Cory-Slechta et al., 2001).
8.5.	Dose-response results
The last 30+ years of studies of effects of chemicals on the auditory startle response have shown
that chemicals are capable of both increasing and decreasing the amplitude response, although
decreases are more commonly found (Crofton, 1992). The classic pharmaceutical principle of
increasing dose and increasing effect is a clear indication of an adverse effect of chemical
exposures. These kinds of effects are easy to interpret as effects of exposure.
More difficult to interpret are non-monotonic effects where there are small changes at the low or
mid dose, with no changes at the high dose. Often these occur at one age or in one sex only, or
occur only with some but not all trial blocks in the session. A careful look at all the startle data
in the study can help determine if effects may be a false positive.
Auditory Startle
C - 23

-------
120
S 100
+i
c
TO
~ £
II
to c
<1> 3
S
O TO
Q. -t
£ ra
80
60
40
20
Male and Female Separate
*
125
450
Dose (mg/kg/day)
900
Q

+l
° s
¦73 TO
11
TO c
<1> 3
s ^
O TO
Q. -t
q: to
120
Sex nested model
100
80
60
40
20
125	450
Dose (mg/kg/day)
Figure 5. Startle data from adult males and
females. The submitted assessment revealed a
significant increase in mean session startle
response amplitude at the mid dose only and in
males only (upper panel). Re-analysis of the data
using a model that contained sex nested within the
litter revealed a lack of statistical significance of
treatment on any variable (see text for details).
Figure 5 illustrates this example from a DNT study that had the following findings: 1) control
data revealed similar amplitudes in PND23 males and females, increased amplitudes in adults
compared to PND23 animals, and smaller amplitudes in adult females compared to males; 2)
habituation was present in both the young and adult animals, and 3) CVs in the control data
ranged from 18-33%. The data fit the biological background for the startle response and the
variance was well within an acceptable range (see Raffaele et al., 2008). There were no
significant effects of treatment on any variables in PND23 males or females. In adults there was
no treatment-related effect on habituation. However, there was a main effect of treatment in the
males, with a significant increase in the group mean of the middle dose group only. Review of
the study methods revealed that one male and one female were tested from each litter for all dose
groups, but the sexes were analyzed separately. Re-analysis of the data using a model that
accounted for sex within the litter revealed a lack of statistical significance of treatment on any
variable. A visual inspection of the data plotted in Figure 5 reveals why. The amplitude in the
mid dose male group was approximately 19% higher than the control group and had a very low
CV of only 8%. Females at this same dose group had slightly smaller amplitude relative to
controls. Including sex in the statistical model to directly test the effect of sex revealed a lack of
any sex-by-treatment interaction and presentation of data from both sexes combined clearly
reveals no effect of treatment. A review of the positive control data from this laboratory
Auditory Startle
C - 24

-------
demonstrated that the test device was capable of detecting a decrease in amplitude (using
clonidine), and an increase in amplitude (using DDT). These data are consistent with other
positive control data for startle tests (Crofton et al., 2008).
8.6. Treatment-related Habituation results
Both the US EPA and OECD 433 guidelines suggest assessing auditory startle response
habituation. For auditory startle testing, habituation is demonstrated by a reduction in startle
amplitude over the course of repeated presentation of the startle stimulus. Like other measures of
habituation, it reflects non-associative learning that is fundamental for interactions with the
environment. Under normal circumstances, habituation of startle response amplitude may be
demonstrated in several ways (see section 6.1). A difference in the rate of habituation between
young and adult animals should be evident in control data. It is also important to note that, unlike
response amplitude, startle response latencies do not normally show habituation.
An example of startle habituation data that present challenges for interpretation is illustrated in
Table 5 and also plotted in Figure 6 (Tyl et al., 2008). Male and female offspring were tested on
both PND22 and PND62. There were no treatment-related effects at PND22 in males or females.
Analyses of data from PND62 revealed an interaction of treatment and trial block for females
only, with significantly decreased amplitudes (up to 45% in the low and high doses, but not in the
mid dose group) and only in the 21-30 and 31-40 trial blocks. While these data appear to indicate
a treatment-related increase in habituation, there are two major challenges with this
interpretation. The first is that the effects are not dose-related since there are no significant
effects in the mid dose. Non-monotonic data of this sort should be identified for closer scrutiny.
The second problem in this data set is the seemingly abnormal habituation in the control females.
Each of the three treatment groups show the expected decrease in amplitude over the 5 trial
blocks (up to that equal to -40-50%), while the female control groups show a decrease only at
the last time block. In the fourth block (31-40) there is only a 6% decrease relative to Block 1-10.
A careful review of positive control and historical control data revealed this laboratory has
repeatedly demonstrated habituation in adult female rats that is in the range of 40-50%. Thus the
data in the low and high dose groups demonstrate a more normal amount of habituation, rather
than a treatment-related response. These data suggest that the effects found in only a limited
number of trial blocks in a non-dose dependent manner, coupled with the lack of habituation in
the control group, likely represent a false positive finding.
Auditory Startle
C - 25

-------
Table 5: Adult male and female rat auditory startle amplitude data from a DNT study with
chemical X. Numbers are group mean ±SD. Numbers in brackets are amplitudes as percent of each
dose group Trial Block 1-10.
Trial
Block
Control
Low dose
Mid dose
High dose
Males
1-10
1338.6 ± 569.5
[100]
1214.8 ±440.3
[100]
1146.0 ±732.8
[100]
1261.0 ± 318.2
[100]
11-20
959.1 ± 363.1
[72]
1015.6 ± 256.6
[84]
907.6 ±646.8
[79]
993.5 ±455.7
[79]
21-30
832.7 ±370.0
[62]
797.4 ± 368.6
[66]
846.0 ±495.4
[74]
738.2 ±248.5
[58]
31-40
854.3 ±444.9
[64]
778.9 ± 303.6
[64]
610.1 ± 383.6
[53]
636.1 ±254.5
[50]
41-50
751.0 ±284.4
[56]
671.6 ±247.5
[55]
613.3 ± 503.1
[53]
620.9 ±232.6
[49]
Females
1-10
860.9 ±224.1
[100]
837.1 ±360.9
[100]
819.5 ± 215.4
[100]
800.2 ±212.1
[100]
11-20
858.2 ±247.3
[99]
730.8 ± 398.0
[87]
715.7 ± 271.9
[87]
654.7 ± 175.3
[82]
21-30
894.4 ± 292.6
[104]
559.4** ±392.2
[67]
661.1 ±215.7
[81]
598.8* ± 175.8
[75]
31-40
810.9 ± 345.7
[94]
443.2** ±276.7
[53]
635.0 ±297.0
[78]
469.0** ± 146.2
[59]
41-50
619.8 ±242.1
[72]
428.2* ± 185.1
[51]
564.9 ±249.7
[67]
435.8 ± 151.9
[54]
* Statistically significantly different from control value at p< 0.05.
**Statistically significantly different from control value at p< 0.01
Auditory Startle
C - 26

-------
Figure 6. Plotted values from Table 5 Adult male and female rat auditory startle amplitude data
from a DNT study with chemical X.
E
<
® 1000 -
2000
500 -
Female
•—• Control
Middle
~—~ High
1500 -
1000 -
100%
100%
Female
Blocks of 10 Trials
Blocks of 10 Trials
Auditory Startle
C - 27

-------
9. General Interpretation
The auditory startle response in rodents is a simple behavior controlled by a simple neural circuit
involving sensory, neural and motor systems (described above). Because of this, treatment-
induced changes in the startle response can usually be ascribed to alterations in one or more of
these systems and this reflects a treatment-induced disruption in nervous system functioning
even if the specific location or nature of the alteration is not clearly identified. For instance,
alterations in sensory thresholds or alterations in neuromuscular functions may both impact
auditory startle responding. Treatment-induced alterations in habituation can usually be
interpreted as treatment-related effects on non-associative learning that is mediated by the central
nervous system. In the interpretation of auditory startle data some issues that should be
considered are described below.
•	In evaluating startle test results the reviewer should verify that the testing procedures and
the data reported meets the requirements of DNT test guidelines. Animals were tested at
appropriate ages, both sexes of animals were tested, testing was counterbalanced among
treatment groups and sexes, animals were selected from all litters, etc. If the study design,
animal selection, or test conduct does not comply with DNT test guidelines, the startle test
data results may be inadequate. If inadequate information is provided to assess this, then
additional information should be obtained from the reporting laboratory;
•	It is important that the study report provides adequate details on the methodology and test
equipment used and on the software parameters used for data collection and analysis. For
example, reports should clearly identify if equipment is reporting the first response, or the
largest response, which are not necessarily the same. Similarly, reports that use the average
response reported by some systems may be the average voltage over the entire recording
interval. These results most likely will differ considerably from results from other
laboratories that use actual startle response peak values;
•	The study report should clearly report the actual measures that were obtained (i.e., peak
response, peak first response, average response, latency to response onset, latency to
response peak). The report should clearly indicate how habituation metrics were calculated
(e.g., compare first block and last block of trials). Reporting only the mean response
collapsed over all trials precludes determining habituation and may also mask any trial-by-
treatment interactions;
•	Appropriate statistical analysis for the reported data is essential. Separate analysis of startle
data for males and females is not appropriate and precludes determining if sex differences in
treatment effects occur. Litter must be taken into account in all data analysis. Comparison
between ages can only be performed if equipment used allows direct comparison between
ages;
•	Assessing positive control data, historical control and concurrent negative control data is
essential to determine if startle response data is within normal and acceptable ranges. Where
control animal data are abnormal, startle study data may not be usable. It is important that
control data reported are reasonable and consistent with reports from the same laboratory in
the past. In addition, both the startle response magnitude and latency values in the study
report should be consistent with historical control data from the same laboratory using the
same equipment. Such historical control data should also include measures of habituation
and demonstrate age- and sex-related startle response. Concurrent control should exhibit
Auditory Startle
C - 28

-------
normal habituation response pattern (this should be compared against historical controls as
well), normal sex differences (males exhibiting larger response than female in adulthood),
and reasonable latency values (>10 msec for onset, approximately 20-40 msec for peak
response). Study reports that have latency values that deviate substantially from these may
be problematic or may reflect inappropriate data selection (wrong software settings) or
analysis.
Table 6. Potential interpretation of some of the most common problems for startle results
found in DNT study reports.

Data Problem
Potential interpretations
1
Lack of age dependent
increase in control
responses (e.g.,
PND21 vs PND60)
If not explained by equipment issues (e.g., sensitivity gain
changes between age groups), then treatment-related effects,
or lack thereof, may not be interpretable
2
Lack of sex dependent
differences in control
responses of adult rats
If not explained by equipment issues (e.g., sensitivity gain
changes), then treatment-related effects, or lack thereof, may
not be interpretable
3
Lack of evidence of
habituation at young
age (around weaning)
Animals may have been tested at too young an age, check
historical and positive control data for habituation at that
specific age. If age is not the cause, see #4 below
4
Lack of evidence of
habituation at adult age
Lack of habituation suggests improper experimental conduct
or data analysis. Verify that data were analyzed so that
habituation was tested directly. If data were analyzed
appropriately then the data should not be used for decisions
based on habituation. Main effects of treatment may still be
useful.
5
Significant effects on
latency with no impact
on amplitude
Since effects on latency without changes in amplitude are
extremely rare and may not be biologically reasonable, efforts
should be taken to ensure that testing equipment and data
algorithms are correctly identifying response peak times.
6
Significant effects on
startle amplitude in
only one sex
Lack of effects in both sexes is not a valid reason for
dismissing treatment-related effects. Chemicals that interfere
with some endocrine systems during development may have
sex-selective effects.
7
Significant effects of
treatment only at the
low or middle dose in
one sex.
If sex is not included in the model, then reanalyze data
including sex. If data re-analyses reveal no sex-by-treatment
interaction and no main effect of treatment, then the reported
effect was likely a false positive.
Auditory Startle
C - 29

-------
Dose-related effects of
increasing magnitude
with a lack of
statistical significance
coupled with
excessive variation
Large effects on the amplitude of the startle response may be
masked by excessive variability. Expert judgment is required
to determine the biological significance in light of the lack of
statistical significance. Check historical and positive control
data. Check to ensure large variability is not due to extreme
values for few test animals.
Auditory Startle
C - 30

-------
10. References
Adams J (1986) Methods in Behavioral Teratology. In: Riley EP, Vorhees CV (eds) Handbook
of Behavioral Teratology Plenum Press, New York, pp 67-97.
Adams J, Buelke-Sam J, Kimmel CA, Nelson CJ, Reiter LW, Sobotka TJ, Tilson HA, and
Nelson BK. 1985a. Collaborative Behavioral Teratology Study: protocol design and testing
procedures. Neurobehav. Toxicol. Teratol. 7, 579-586.
Adams J, Oglesby DM, Ozemek HS, Rath J, Kimmel CA, and Buelke-Sam J (1985b)
Collaborative Behavioral Teratology Study: programmed data entry and automated test
systems. Neurobehav. Toxicol. Teratol. 7, 547-554.
Brunjes PC and Alberts JR (1981) Early auditory and visual function in normal and hyperthyroid
rats. Behav. Neural. Biol. 31, 393-412.
Buelke-Sam J, Cohen IR, Wierda D, Griffey KI, Fisher LF, and Francis PC (1998) The selective
estrogen receptor modulator, raloxifene: a segment II/III delivery study in rats. Reprod.
Toxicol. 12, 271-288.
Claassen V (1994) Neglected factors in pharmacology and neuroscience research:
Biopharmaceutics, animal characteristics, maintenance, testing conditions. Vol 12. El sever,
Amsterdam, Netherlands, 1994.
Cohen J (1977) Statistics for the Behavioral Sciences. Academic Press, New York, NY.
Cory-Slechta DA, Crofton KM, Foran JA, Ross JF, Sheets LP, Weiss B, and Mileson B. (2001)
Methods to identify and characterize developmental neurotoxicity for human health risk
assessment. I: behavioral effects. Environ. Health Perspect. 109 Suppl 1, 79-91.
Crofton KM (1992) Reflex modification and the assessment of sensory dysfunction. In: Tilson
HA, Mitchell CL (eds) Neurotoxicology Raven Press, New York.
Crofton KM, Foss JA, Hass U, Jensen K, Levin ED, and Parker SL (2008) Undertaking positive
control studies as part of developmental neurotoxicity testing. Neurobehav. Toxicol. 30, 266-
287.
Crofton KM and Knight T (1991) Auditory deficits and motor dysfunction following
iminodipropionitrile administration in the rat. Neurotoxicol. Teratol. 13, 575-581.
Crofton KM, Lassiter TL, and Rebert CS (1994) Solvent-induced ototoxicity in rats: an atypical
selective mid-frequency hearing deficit. Hear. Res. 80, 25-30.
Crofton KM, Makris SL, Sette WF, Mendez E, and Raffaele KC (2004) A qualitative
retrospective analysis of positive control data in developmental neurotoxicity studies.
Neurotoxicol. Teratol. 26, 345-352.
Crofton KM and Sheets L (1989) Sensory dysfunction: Assessment using reflex modification of
the startle response. J. Am. Coll. Toxicol. 8, 199-211.
Csomor, PA, Yee, BK, Vollenweider, FX, Feldon, J, Nicolet, T, and Quednow, BB. (2008) On
the influence of baseline startle reactivity on the indexation of prepulse inhibition. Behav.
Neurosci.122, 885-900
Davis M (1980) Neurochemical modulation of sensory-motor reactivity: acoustic and tactile
startle reflexes. Neurosci. Biobehav. Rev. 4, 241-263.
Davis M and Eaton RC (1984) The mammalian startle response. Neural Mechanism of Startle
Behavior Plenum Press, New York, pp 287-351.
Davis M, File SE, Peek HVS, and Petrinovich L (1984) Intrinsic and extrinsic mechanisms of
habituation and sensitization. Habituation, Sensitization, and Behavior Academic Press,
New York.
Auditory Startle
C - 31

-------
Davis M, Gendelman DS, Tischler MD, and Gendelman PM (1982) A primary acoustic startle
circuit: lesion and stimulation studies. J. Neurosci. 2, 791-805.
Dean KF, Sheets LP, Crofton KM, and Reiter LW (1990) The effect of age and experience on
inhibition of the acoustic startle response by gaps in background noise. Psychobiol. 18, 89-
95.
Fay RR (1988) Hearing in Vertebrates: A Psychophysics Databook Hill-Fay Associates.
Fechter LD, Young JS, and Annau Z (1986) Reflexive measures. Neurobehavioral Toxicology
The Johns Hopkins University Press, Baltimore.
Geyer, MA and Swerdlow, NR (1988) Measurement of startle response, prepulse inhibition, and
habituation. Current Protocols in Neuroscience, John Wiley & Sons, Supplement 3, 8.7.1-
8.7.15
Grimsley CA, Longenecker RJ, Rosen MJ, Young JW, Grimsley JM, Galazyuk AV. (2015) An
improved approach to separating startle data from noise. J. Neurosci. Methods 253, 206-217
Grissom N and Bhatnagar S. (2009) Habituation to repeated stress: Get used to it. Neurobiol.
Learn. Mem. 92, 215-224.
Goldey ES, Kehn LS, Rehnberg GL, and Crofton KM (1995) Effects of developmental
hypothyroidism on auditory and motor function in the rat. Toxicol. Appl. Pharmacol. 135,
67-76.
Goldey ES, O'Callaghan JP, Stanton ME, Barone S, Jr., and Crofton KM (1994) Developmental
neurotoxicity: evaluation of testing procedures with methylazoxymethanol and
methylmercury. Fundam. Appl. Toxicol. 23, 447-464.
Henck JW, Frahm DT, and Anderson JA (1996) Validation of automated behavioral test systems.
Neurotoxicol. Teratol. 18, 189-197.
Hoffman HS and Ison JR (1980) Reflex modification in the domain of startle: I. Some empirical
findings and their implications for how the nervous system processes sensory input. Psychol.
Rev. 87, 175-189.
Holson RR, Freshwater L, Maurissen JPJ, Moser VC, and Phang W. (2008) Statistical issues and
techniques appropriate for developmental neurotoxicity testing: A report from the ILSI
Research Foundation/Risk Science Institute expert working group on neurodevelopmental
endpoints. Neurotoxicol. Teratol. 30, 326-348.
Ison JR and Hoffman HS (1983) Reflex modification in the domain of startle: II. The anomalous
history of a robust and ubiquitous phenomenon. Psychol. Bull. 94, 3-17.
Ison JR. (1984) Reflex modification as an objective test for sensory processing following
toxicant exposure. Neiirobehav. Toxicol. Teratol. 6,437-445.
Ison JR and Russo JM. (1990) Enhancement and depression of tactile and acoustic startle
reflexes with variation in background noise level. Psychobiol. 18, 96-100.
Kobayashi K, Tsuji R, Yoshioka T, Kushida M, Yabushita S, Sasaki M, Mino T, and Seki T.
(2005) Effects of hypothyroidism induced by perinatal exposure to PTU on rat behavior and
synaptic gene expression. Toxicol. 212, 135-147.
Koch M (1999) The neurobiology of startle. Prog. Neurobiol. 59, 107-128.
Leussis MP and Bolivar VJ. (2006) Habituation in rodents: A review of behavior, neurobiology,
and genetics. Neurosci. Biobehav. Rev. 30, 1045-1064.
Li L, Du Y, Li N, Wu X, Wu Y. (2009) Top-down modulation of prepulse inhibition of the
startle reflex in humans and rats. Neurosci. Biobehav. Rev. 33, 1157-1167.
Maurissen JP and Marable BR (2005) Neurotoxicity test validation, positive controls and
proficiency: are chemicals necessary? Neurotoxicol. Teratol. 27, 545-551.
Auditory Startle
C - 32

-------
Parisi T and Ison JR (1979) Development of the acoustic startle response in the rat: ontogenetic
changes in the magnitude of inhibition by prepulse stimulation. Dev. Psychobiol. 12, 219-
230.
Raffaele KC, Fisher JE, Jr., Hancock S, Hazelden K, and Sobrian SK (2008) Determining normal
variability in a developmental neurotoxicity test: a report from the ILSI Research
Foundation/Risk Science Institute expert working group on neurodevelopmental endpoints.
Neurotoxicol. Teratol. 30, 288-325.
Muller KE and Benignus VA. (1992) Increasing scientific power with statistical power.
Neurotoxicol. Teratol. 14,211-219.
Sheets LP, Dean KF, and Reiter LW (1988) Ontogeny of the acoustic startle response and
sensitization to background noise in the rat. Behav. Neurosci. 102, 706-713.
Shnerson A and Willott JF (1980) Ontogeny of the acoustic startle response in C57BL/6J mouse
pups. J. Comp. Physiol. Psychol. 94, 36-40.
Slikker, Jr. W, Acuff K, Boyes W, Chelonis J, Crofton KM, Dearlove G, Li A, Moser VC,
Newland C, Rossi J, Schantz S, Sette W, Sheets L, Stanton M, Tyl S, and Sobotka TJ.
(2005). Behavioral test methods workshop: A summary. Neurotoxicol. Teratol. 27, 417-
427.
Swerdlow NR, Geyer MA (1998) Using an animal model of deficient sensorimotor gating to
study the pathophysiology and new treatments of schizophrenia. Schizophr. Bull. 24, 285-
301.
Tabachnick BG and Fidell LS. (2006) Using Multivariate Statistics. (5th edition), Pearson, Boston
MA.
Thompson RF and Spencer WA. (1966) Habituation: a model phenomenon for the study of
neuronal substrates of behavior. Psychol. Rev. 73, 16-43.
Thompson RF. (2009) Habituation: A history. Neurobiol. Learn. Mem. 92, 127-134.
Tyl RW, Crofton K, Moretto A, Moser V, Sheets LP, and Sobotka TJ (2008) Identification and
interpretation of developmental neurotoxicity effects: a report from the ILSI Research
Foundation/Risk Science Institute expert working group on neurodevelopmental endpoints.
Neurotoxicol. Teratol. 30, 349-381.
Williams JM, Hamilton LW, and Carlton PL (1975) Ontogenetic dissociation of two classes of
habituation. J. Comp. Physiol. Psychol. 89, 733-737.
Wise LD, Allen HL, Hoe CM, Verbeke DR, and Gerson RJ (1997) Developmental neurotoxicity
evaluation of the avermectin pesticide, emamectin benzoate, in Sprague-Dawley rats.
Neurotoxicol. Teratol. 19, 315-326.
Young JS and Fechter LD (1983) Reflex inhibition procedures for animal audiometry: a
technique for assessing ototoxicity. J. Acoust. Soc. Am. 73, 1686-1693.
Auditory Startle
C - 33

-------
MODULE-D
EVALUATION OF LEARNING AND MEMORY DATA
Learning and Memory
D - 1

-------
1.	Introduction	4
2.	Guideline Requirements	7
3.	Test Description: Position Discrimination Learning Using Letter Mazes (Y, M, E, T)	9
3.1.	Procedures for Position Discrimination Learning	10
3.2.	Standard Test Conditions for Position Discrimination Learning	11
3.3.	Metrics to Evaluate Learning and Retention in Position Discrimination Learning	12
3.4.	Data Reporting for Position Discrimination Learning	13
3.5.	Interpretation of Position Discrimination Results	14
4.	Test Description: Spatial Learning using the Morris Water Maze Test	15
4.1.	Procedures for Spatial Acquisition	17
4.2.	Standard Test Conditions for the Morris Water Maze	22
4.3.	Metrics to Evaluate Learning and Retention in the Morris Water Maze	22
4.4.	Data Reporting for the Morris Water Maze	23
Reported Data	23
4.5.	Interpretation of Morris Water Maze Results	24
5.	Test Description: Associative Learning using Passive Avoidance	24
5.1.	Procedures for Passive Avoidance	26
5.2.	Metrics to Evaluate Learning and Retention in the Passive Avoidance Task	28
5.3.	Data Reporting for the Passive Avoidance Task	28
5.4.	Interpretation of Passive Avoidance Results	29
6.	Test Description: Sequential Learning using the Biel or Cincinnati Maze	30
6.1.	Procedures for the Biel or Cincinnati Maze	31
6.2.	Metrics to Evaluate Learning and Retention in the Biel or Cincinnati Maze	32
6.3.	Data Reporting for the Biel or Cincinnati Maze	33
6.4.	Interpretation of Biel or Cincinnati Maze Results	34
7.	Standard Conditions for All Tests of Learning and Memory	35
8.	Control Data for Tests of Learning and Memory	36
9.	Statistical Analyses	37
9.1.	Statistical Analysis of Learning Data	37
9.2.	Statistical Analysis of Memory Data	38
9.3.	Data Distributions	38
9.4.	Correlated Measures	40
9.5.	Outcomes of Interest	40
10.	Interpretation of Treatment-Related Effects in Tests of Learning and Memory	41
Learning and Memory	D - 2

-------
11. References
43
Learning and Memory	D - 3

-------
MODULE D - INTERPRETATION OF LEARNING AND MEMORY DATA
1. Introduction
Learning and memory are theoretical constructs that are sometimes difficult to separate
experimentally. Learning is defined as a relatively permanent change in behavior that is the
result of experience. Memory is defined as the retention and use of acquired information about
locations, events, or temporal order to modify subsequent behavior. Learning and memory can
be inferred from changes in behavior. Most complex behaviors require the combinatorial
participation of many neural systems that process different attributes of an experience (Gold,
2003). In the process of investigating the biological basis of learning and memory,
neuroscientists, psychologists, and behavioral pharmacologists have devised a variety of
specialized tests to distinguish different types of learning and memory, multiple definitions have
emerged, and distinct neural systems attributed to specific types of learning and memory have
been identified (Gold, 2003; Kihlstrom et al., 2007; Peele and Vincent, 1989; Squire, 2004;
Vorhees and Williams, 2014a).
The simplest type of learning, non-associative learning, is exemplified by processes such as
habituation or sensitization and has been addressed in the module on interpreting motor activity
data and so will not be considered further here. Associative learning is the process by which an
association between two stimuli (classical conditioning) is learned. Learning and memory are
inextricably intertwined as the capacity for learning presupposes an ability to retain the
knowledge acquired through experience, while memory stores the background knowledge
against which new learning takes place. These processes have been assessed in a variety of ways
including classical conditioning paradigms, instrumental learning, spatial learning, and sequential
learning and are probed to a greater or lesser degree by different test procedures.
Some learning and memory tasks include two components of memory, i.e., reference or working
memory. Reference memory involves remembering the rules of a given task that has been
learned. Reference memory is acquired with repeated training and persists from days to months.
This memory is an integral component to acquisition and is an example of how the lines blur
between assessments of learning and assessments of memory. Reference memory is required for
efficient completion of multiple test trials. In a maze task, reference memory might include
acquiring the knowledge of the location of arms in the maze that are never rewarded, or learning
that an escape platform exists and is in the same location over trials. Distinct from reference
memory, working memory is typically defined as a short term memory for an object, stimulus,
or location that is used within, but not typically between testing sessions. Working memory uses
flexible stimulus-response associations and is sensitive to interference (Green and Stanton, 1989;
Olton, 1979). Working memory in animals can be viewed as having a limited lifetime or being
task-relevant for only a brief period. A common feature of working memory tasks that can be
exploited in variety of simple test systems is that performance decreases as the delay over which
a memory is held increases (Carter et al., 1995). Unfortunately, this feature has only rarely been
exploited in DNT guideline studies.
Learning and memory depend upon the coordinated action of different brain regions and
neurotransmitter systems constituting a functionally integrated neural network (D'Hooge and
DeDeyn, 2001). The main learning areas and pathways are similar in rodents and primates but
Learning and Memory
D -4

-------
specific task requirements preferentially engage some brain regions more than others
(Eichenbaum, 2000; Stanton and Spear, 1990). Very superficially speaking, spatial learning is
believed to be primarily based in hippocampal function, fear-based learning involves the
amygdala, and attention tasks require prefrontal cortex input (Clark et al., 2000; Eichenbaum,
2000; Squire, 2004).
The variety of tests available in the literature to assess learning and memory and their potential
application for guideline testing have been reviewed (Graham et al., 2012; Paul et al., 2009;
Peele and Vincent, 1989; OECD Guidance Document 151, 2013; Vorhees and Williams, 2014a,
2014b). A number of considerations must guide the selection of tests for assessing learning and
memory in a regulatory context, not the least of which are cost effectiveness, sensitivity and
selectivity. Typically, methods used in DNT guideline studies tend to be simpler tests; the ones
that appear most frequently are the M, E, Y, and T mazes, Morris Water Maze (MWM), passive
avoidance (PA), and Cincinnati or Biel Maze (see Table 1). Since these tests have been used in
almost all submitted DNT studies, this document focuses only on those. Less often, spatial
delayed alternation and delayed matching to sample tasks have been employed. The latter two
are potentially more rigorous evaluations of memory, as they can be designed to incorporate time
delays. However, they are more complex and labor intensive to run.
It is important to recognize that various tests are assessing different forms of learning or memory
and it is reasonable to expect that the effects of any one chemical may vary depending on the
specific test. It is also well documented that any given test is unlikely to be sensitive to effects
caused by all different types of chemicals (Peele and Vincent, 1989; Tilson and Mitchell, 1984).
Similarly, it is not anticipated that a performance impairment detected in a learning and memory
test would necessarily be reflected in outcomes from brain pathology or brain morphometry, and
vice versa.
In addition, tests that are appropriate for one age of animal may not be appropriate for a different
age. Certainly the physical and neurological substrates that are probed by any given task must be
fully developed before that task can be applied to evaluate chemical-induced impairment (i.e.,
animals must be able to swim, locomote, hear and see to use tests requiring these abilities).
Certainly all of these abilities are fully developed in animals at the time of weaning, the earliest
time point recommended for learning and memory tests in guideline studies. For some tasks,
weanling animals are not as proficient as adults, and this relative capacity is due to the
immaturity of the brain circuitry that underlies the learning rather than physical ability to
perform (Brown et al., 2005; Ehman and Moser, 2006; Green and Stanton, 1989).
Learning and Memory
D -5

-------
Table 1. Summary of typical tests of learning and memory submitted in guideline studies of DNT
testing.				
Test System
Frequency of Report in
North American
Regulatory Submissions
(as of 2014)
Cognitive Ability
Assessed
Comment
Letter Mazes
M, Y, E
M : 34
Y : 12
E : 3
Position
Discrimination
Simplest form of learning assessed
most common but least sensitive
test procedure
Morris Water
Maze
9
Spatial Learning
Typical acquisition is test of
reference memory with single
probe trial to test retention at 24 hr
Passive
Avoidance
33
Associative
Learning
Single trial learning often followed
by immediate recall (typically < 1
hr) to assess learning and delayed
recall (hrs to days) to test memory.
Multiple acquisition trials are often
but unnecessarily run as most of
learning occurs within the first trial
Biel Water
Maze
6
Sequential
Learning
Multiple T-mazes. Combined
reference and working memory
task. Intramaze cues detract from
its sensitivity
Cincinnati
Maze
2
Sequential/
Egocentric
Learning
Multiple T-mazes, improvement
over Biel maze. Combined
reference and working memory
task.
T-Maze
2
Delayed
Alternation
Typically this procedure involves
position discrimination.
Incorporation of delays can
enhance sensitivity. Reference and
working memory assessed.
Although a more rigorous test than
simple position discrimination, it is
not frequently used. Expert
consultation is advised.
T-Maze
1
Delayed Matching
to Position
Infrequently used. Expert
consultation is advised.
Schedule-
Controlled
Behavior
0
Delayed Matching
to Position
Although suggested by US EPA
Guideline 870.6500, it has not
been used to date. Expert
consultation is advised.
Learning and Memory
D -6

-------
2. Guideline Requirements
As cognitive function is one of the basic processes that can potentially be disrupted by chemical
exposure during development, tests of this parameter are included in many studies that are
designed to evaluate DNT (e.g., USEPA OPPTS 870.6300 and OECD TG426). The U.S. EPA
test guideline for a DNT study (US EPA OPPTS 870.6300) does not specify which test to
perform, but rather states more generally that: 1) associative learning must be assessed as either a
change across several repeated learning trials or sessions, or in tests involving a single trial, with
inclusion of some condition that controls for non-associative effects of the training experience;
and 2) the tests should include some measure of memory (retention) in addition to original
learning (acquisition) where appropriate. The guidelines also specify that the choice of test to
evaluate learning and memory in a given circumstance should consider findings in other toxicity
studies for that chemical. Measures of learning and memory are also influenced by sensory,
motor and motivational variables (Tilson and Mitchell, 1984). The interpretation of data from a
learning/memory task should incorporate consideration of these variables by using the
appropriate control procedures or conducting independent assessments (e.g. ability to swim in a
swim maze and find a visible platform). While regulatory test requirements for OECD and EPA
are comparable for Learning and Memory testing as part of DNT studies, there are some
differences that should be considered. These are summarized in Table 2. Both guidelines
require a minimum of 10 animal s/sex/dose and testing in the early postweaning period and in the
young adult; all litters should be represented. Note that while the OECD guidelines state testing
at PND25, testing may need to be several days later if pups are not weaned before PND25. Task
selection should always consider the developmental state of the animal. Although not specified
in the guidelines, US EPA has also recommended that testing should be performed in different
groups of animals at these two ages if the same test is used. This is consistent with OECD's
recommendation that different tests or different animals be used at the two ages as repeated
testing of the same animals in the same test is likely to decrease sensitivity due to carry over
effects. If different tests are evaluated such that learning on one does not interfere with or
facilitate learning on the other, the same animals can be tested at the weaning and adult ages.
Table 2. Learning and Memory testing requirements in EPA and OECD guidelines. The specific
requirements are essentially the same in both EPA and OECD 426 guidelines. The OECD 443
extended 1-generation reproductive toxicity study guideline does not mention cognitive testing
except as a potential additional measure where needed.


US EPA

OECD 443


OPPTS
OECD
Extended


870.6300
426
1-Gen
Age of testing
OECD: PND25±2 andPND60and
older
US EPA: around weaning and
around PND60
V
V
Not applicable
Minimal Test
Requirements
Test of Task Acquisition- Learning,
a change across several learning
trials or a single trial.
V
V

Learning and Memory
D -7

-------

Test of Retention- Short or long
term memory. Cannot be interpreted
in absence of documentation of
asymptotic acquisition on same task.
V
V
Not applicable
Performance
Controls
If the cognitive test reveals an effect
of the test compound, additional
tests may be conducted to rule out
alternative interpretations based on
deficiencies in sensory,
motivational, and/or motor
capacities.
V
V
Not applicable
Choice of
Procedure
Not specified. When possible,
learning and memory test should be
chosen based on demonstrated
sensitivity to the class of compound
under investigation. Same or
different tests may be used at
different ages.
V
V
Not applicable
Suggested
Learning and
Memory Tests
US EPA: Delayed-matching-to-
position, olfactory conditioning,
acquisition and retention of
schedule-controlled behavior.
OECD: US EPA recommended tests
plus Morris water maze, Biel or
Cincinnati maze, radial arm maze,
T-maze.
V
V
Not applicable
History of
Animal
OECD: Different animals must be
used at each test age.
US EPA: No specific guidance
provided. Generally accepted that
same animal can be tested at both
ages if different tests are used.
V
V
If data
indicate need
for cognitive
testing, it must
be integrated
without
compromising
the integrity of
the other
evaluations in
the study.
Learning and Memory
D - 8

-------
3. Test Description: Position Discrimination Learning Using Letter Mazes (Y, M, E, T)
Position discrimination is a very simple form of maze learning and, perhaps because of its
simplicity, has been the most commonly used test of learning in DNT guideline studies
submitted to date. It is not, however, a widely applied test of learning in the academic literature,
due to its lower sensitivity in detecting behavioral impairments. In this task the animal has to
make a binary choice to turn left or right to escape the water (see Figure 1 for example
diagrams). These mazes are in the shape of a Y, M, E, or T, earning them the moniker 'letter
mazes' and can be either land- or water-based. In the former, animals are generally food
restricted and rewarded for correct turning response by delivery of a food reward. As food
restriction procedures may become more complicated (especially in growing pups, or in cases
where there is already a treatment effect on body weight), water-based maze procedures are
more routinely used in submitted regulatory studies. In the water-based version of this task, the
reinforcer, which motivates the animal to learn the task, takes the form of an escape from the
water, via a platform positioned at the end of the correct choice arm. The following discussion
will be restricted to water-based letter mazes used to test for position discrimination learning.
Figure 1. Various renditions of simple position discrimination mazes, T, Y, and E/M. Animals
are placed in the ' Start' arm and must make either a left or a right turn to swim to the escape
platform. The correct position should be counterbalanced across animals in each treatment
group.
Mazes used in the assessment of position discrimination learning can be configured as a T, Y, E
or M but all have the same basic form of straight alley 'start arm' at the end of which is a choice
point for the animal to turn left or right; only one of which is rewarded by the presence of an
escape ladder or platform. The trials are run as discrete entities, with a relatively brief but
consistent intertrial interval. Only one arm position is rewarded for any given animal during
training, but the correct arm should be counterbalanced across animals (i.e. equal number of
animals per arm position, per dose group). The consequences of an 'error' (e.g., turning left
instead of right) are fairly inconsequential to the animal, since it simply turns around and swims
to the opposite 'correct arm'. If the maze is small, this only minimally increases the time in the
water and time to escape.
Position discrimination is often described as a reference memory task as the same arm is
designated the correct response arm throughout the test (e.g., always turn right). Reference
memory is distinct from working memory which can also be assessed in this type of apparatus.
Learning and Memory
D -9

-------
A working memory task (e.g., spatial delayed alternation) requires the animals to alternate
between the right and left arms for sequential trials (e.g., if you went right last time, go left this
time). Working memory tasks are cognitively more demanding than simple position
discrimination, and they require more extensive training for successful performance. Working
memory tasks are disrupted by treatments that negatively impact function of the hippocampus
(Green and Stanton, 1989; Freeman and Stanton, 1991; Watson et al., 2009) yet leave reference
memory (position discrimination) relatively intact (Chadman et al., 2006; Watson et al., 2006).
Because position discrimination learning develops early in postnatal life, as young as PND15,
(Green and Stanton, 1989; Brown et al., 2005; Pagani et al., 2005; Kenny and Blass, 1977; Smith
and Bogomolny, 1983), both weanling and adult animals can be readily assessed for this simple
form of learning. However, because of potential carry-over learning, the same animals should not
be assessed in the same task at both adult and weanling ages. For younger animals, the size of
the maze should be scaled accordingly to accommodate the smaller body mass of the animal and
to avoid fatigue that could interfere with ability to perform the task.
3.1. Procedures for Position Discrimination Learning
As almost all spatial learning studies submitted for regulatory evaluation are water-based mazes,
the following description of procedures will be limited to tasks of this type. Many of the same
principles, however, are also applicable to land-based mazes. Details completely describing the
maze set-up, size, and procedures should be included in the report.
Straight Alley Test for Swim Speed, Acclimation, Side Bias.
To ensure that the animals are physically able to perform the task and to familiarize them with
the minimal task requirements - swimming and escape from the water — animals are initially
placed in a straight alley. Straight channel swim speed provides information regarding any
potential deficit in motor function. An escape ladder or platform is clearly visible at the other
end of the alley. The animal's swim speed can be calculated, a parameter that should be assessed
if latency measures are used as metrics for learning. If video tracking is used for data collection,
swim speed should also be calculated on every trial. Motor dysfunction could potentially
confound interpretation of learning deficits if a decrease in latency is used as the main measure
of learning for the task, in that an increase in latency could be the result of a decrease in swim
speed, rather than an inability to learn.
In some procedures, the first learning trial may also serve as an acclimation trial, in which
animals are allowed to explore both arms and an escape opportunity is provided in the opposite
arm that is entered. The 'correct' arm for that particular animal for all subsequent trials is the
arm opposite of the one first entered. This procedure may guard against a side preference bias
that some animals may exhibit. Ideally, designation of the correct side should be counterbalanced
within and across treatment groups.
Task Acquisition.
Animals are released in the start box and allowed to swim to the choice point and select a left
or a right turn. If the correct choice is made, animals will find an escape route at the end of the
alley. If an incorrect choice is made, the animal remains in the maze until it swims to the
correct alley or until a set period of time (typically 60-90 seconds) has elapsed. In the case of
Learning and Memory
D - 10

-------
the latter, animals may be guided to the escape platform by the experimenter. Alternatively,
the animal may simply be removed from the incorrect arm and the trial ended (non-correction
procedure). Note that here, and in many other tests, setting a maximum time for testing
produces closed-ended values that can impact statistical analyses (see Section 9). Acquisition
testing typically occurs within one day, with a set maximum number of trials for each animal
and a fixed intertrial interval (ITI). With a short ITI relative to maximal trial duration of 60-90
seconds, it is possible that fatigue may compromise performance in those animals, increasing
variability and confounding interpretation of latency data. However, a review of submitted data
has not identified this as a common occurrence.
Generally, testing for each animal will continue until learning has been demonstrated according
to a pre-defined level of proficiency (i.e., a learning criterion has been met) or until some pre-
defined maximum number of trials has been administered. Criterion performance is typically set
at 5 consecutive correct trials, at which point testing is ended for that animal. Different criteria
may be utilized but must be explicitly defined and represent a valid index of task acquisition.
Taking testing out to the maximum number of trials even if criterion has been achieved in fewer
trials facilitates the evaluation of learning curves and statistical analyses. However, this practice
does run the risk of having 'overlearning' which may decrease sensitivity of retention tests
performed at a later date (see below).
Retention.
For position discrimination learning, retention is usually evaluated one week or so after the
acquisition phase. The protocol for retention testing is generally identical to that for
acquisition, with a specific maximum number of trials for each animal and a learning criterion of
a set number of consecutive correct trials. The correct goal is the same as that designated during
acquisition. If the animal has retained the learned information, criterion is frequently achieved
more quickly than in acquisition. Note, however, that only the first trial during that second week
can be considered "memory", since subsequent testing uses the information from the previous
trial, not trials one week prior. Performance after the first trial can be considered re-learning.
It is not possible to test retention in animals that did not achieve criterion performance of the
task during the acquisition phase. Such animals should be excluded from the retention testing,
and the failure to acquire the task should be noted in the presented results. When this is the case,
it should be noted that there will be differing sample sizes across groups. Furthermore, such a
practice leads to testing only a selected population, which may lead to bias in the outcomes.
Alternatively, the acquisition phase could be extended until all animals achieve criterion
performance prior to testing retention.
3.2. Standard Test Conditions for Position Discrimination Learning.
All standard test conditions described below should be maintained for appropriate conduct of
position discrimination learning:
•	Maze size and water temperature should be specified, within acceptable limits, and
appropriate to the age and size of the animal being tested;
•	The animal should be placed in the start position of the maze facing the back wall of
the maze;
•	The procedure used if an animal does not reach the platform within the maximum test
Learning and Memory
D - 11

-------
time must be specified (e.g., whether it is immediately removed or taken to the
platform);
•	The duration of time animal is left on the escape platform on completion of each trial
must be consistent and specified;
•	The intertrial interval (the delay between removing the animal from the platform and the
beginning of the next trial) and handling and holding conditions between each trial must
be consistent and specified;
•	Criterion performance in the acquisition phase must be achieved in all subjects evaluated
for retention learning.
3.3. Metrics to Evaluate Learning and Retention in Position Discrimination Learning
There are a few basic dependent measures that are used to evaluate position discrimination
learning: escape latency, number of errors, and trials to criterion. These may be the only
measures when data are collected by an observer, but more information (e.g., path length,
direction) may be obtained if video-based tracking systems are used. The basic endpoints are
described in more detail below. Some are redundant and should be highly correlated, but with
slight nuanced differences.
Latency to escape
This measure reflects the time from release of the animal in the maze until it reaches the correct
goal and exits the maze. Latency will typically decrease over trials, reaching some asymptotic
level. Note that latency is to some extent dependent on the size of the maze/length of the alleys,
and thus varies from laboratory to laboratory. Standardization of alley length or minimal
requirements would facilitate interpretation and comparison across studies. Extremely short
latencies may indicate a very small maze, and may complicate interpretation of results,
especially when presenting differences as percent change across trials.
Errors
It is imperative that an explicit operational definition of an 'error' is provided (see Section 7).
Errors will typically decrease over trials, with criterion performance set as a specific number of
consecutive trials with no errors. The number of errors by treatment group for each trial should
be noted, as well as the number of animals in each group making any error for each trial.
Trials to criterion
Criterion performance may be measured as the number of trials before criterion is reached
(trials to criterion) or as the number of total errors each animal makes over the testing session
(errors to criterion). In the latter case, it is important to determine whether criterion was in
fact obtained, i.e., whether the total errors to criterion are inclusive of all animals or only those
that reached criterion. In some cases, the percentage of animals in each group that achieved
criterion performance may be provided. In these cases, the total number of trials or errors is
only a single number and as such cannot provide a learning curve that shows change over
time. For this reason, evaluation of error data over trials provides a better representation of
learning.
The percentage of correct trials (number of correct trials for a given animal divided by the total
number of trials completed by that animal) may be useful when there are differences among
Learning and Memory
D - 12

-------
animals in the number of trials run.
Typically, animals that do not meet criterion are considered non-learners, and are not tested for
retention or re-learning. As mentioned above, this practice may lead to different sample sizes
as well as selection bias across groups. Furthermore, the use of criterion data alone may be
misleading, leading to the classification of non-learner being applied for different situations. In
the example shown in Table 3, the rats show very different patterns of behavior, yet all are
considered non-learners by the definition used in this particular laboratory. Non-learners were
defined as rats that did not perform five consecutive errorless trials for a maximum of 15 trials.
Specifically, rat 1 makes some number of errors on every trial. After the 11th trial, it is not
possible for criterion to be met, so that rat is not tested for the last 4 trials and a maximum
value of 15 is recorded. Rat 2 shows a decreasing number of errors over trials, and reaches
criterion but only on the 15th trial; thus this animal by the laboratory's definition is considered
a non-learner. Rat 3 makes relatively few errors, reaching as many as four but not five
consecutive trials with no errors. As shown in the last column of Table 3, these rats display
different outcomes in terms of other endpoints, namely total and mean errors as well as trials
with no errors, which are considerably more informative than the trials to criterion or number
of rats reaching criterion. Rat 3 in this example had fewer errors and more errorless trials than
the other rats, indicative of a more consistent pattern of learning.
Table 3. Example of individual data (number of errors per trial) for three rats. Total is 15 trials,
criterion performance is 5 consecutive trials with no errors. NT shows that the rat was not tested.
Trial
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Trials to
criterion
Total and
mean
errors
Trials
with no
errors
Rat 1
5
2
4
1
2
3
2
1
5
1
2
NT
NT
NT
NT
15
28
2.5/trial
0
Rat 2
3
2
3
2
0
2
1
2
0
2
0
0
0
0
0
15
17
1.1/trial
7
Rat 3
1
0
0
0
2
0
0
1
0
1
0
0
0
0
1
15
6
0.4/trial
11
3.4. Data Reporting for Position Discrimination Learning.
In order to evaluate the effects of chemicals on these parameters, it is essential that DNT position
discrimination reports provide clear information on study design, the measures (dependent
variables), and the statistical procedures used to evaluate the data. The main results that are of
interest from DNT learning and memory assessments are:
•	The effects of chemicals on acquisition, i.e., initial learning rate;
•	Effects of chemicals on memory, i.e., retention of the learned response;
•	If position discrimination was assessed at both weanling and adult ages, is the deficit seen
at both ages;
•	Sex differences in effects of chemicals on learning and/or retention.
Learning and Memory
D - 13

-------
Dependent Variables
It is important that the dependent variables recorded and analyzed are unambiguously described.
For position discrimination, the dependent measures include the following:
•	Latency to reach the correct goal, and how this changes over trials;
•	Number of errors (incorrect turns within each trial), and how this changes over trials;
•	Cumulative as well as average number of errors across all trials (usually reported as
a maximum of one per trial, although more errors could potentially be committed);
•	Number of trials that were not completed (e.g., animal did not reach goal within
allotted time);
•	Trials to criterion;
•	Numbers (proportions) of each group that did (or did not) achieve criterion;
•	Errors to criterion - similar to cumulative number of errors but eliminates animals that
failed to reach criterion.
Reported Data
In order to evaluate treatment effects, the following details should be reported for each sex at
each age group. In cases where data are non-normally distributed, different measures of central
tendency may be more appropriate (e.g., median, mode).
•	Mean ± standard deviations of latency per treatment group on each trial (not averaged
over all trials);
•	Mean ± standard deviations of total number of errors to criterion performance per
treatment group;
•	Median or mode trials to criterion performance per treatment group (which may be more
appropriate than mean);
•	Median or mode for number of errors per treatment group on each trial (which may be
more appropriate than mean because of the type of data, e.g., count, or the underlying
distribution);
•	Number of animals in each treatment group meeting criterion on each trial as well as
number of animals reaching criterion performance;
•	Individual trial data for each animal should be included in an appendix;
•	Summary data should be presented in tabular and graphic form.
Data should be reported separately for acquisition and retention assessments. All data should be
reported on an individual and group basis. Individual data should include every trial performed
for each animal, including the first (acclimation) trial. Note that some laboratories frequently fail
to include data for the first trial(s). Also, note that the number of errors is usually reported as a
maximum of one per trial, although more errors could potentially be committed. It may be
necessary to request the record of all the errors for all trials, including the first.
3.5. Interpretation of Position Discrimination Results
The following points should be taken into consideration in the interpretation of the data:
•	Control animals should demonstrate a clear and statistically reliable reduction in both
latencies and errors over trials. The absence of this in control animals indicates lack of a
'learning curve' and as such the test does not contain the minimal requirements to
detect a change in behavior over time and does not measure learning;
Learning and Memory
D - 14

-------
•	A treatment-induced increase in the number of errors or trials to criterion is evidence of a
learning impairment;
•	A treatment-induced decline in the number of animals reaching criterion performance is
evidence of a learning impairment;
•	When slight differences in swim speed are present, it is advisable to consider their
relative contribution to the magnitude of the impaired performance. Latency increases in
the presence of slower swim speeds in treated animals may indicate performance or
motoric deficits which can confound the interpretation of cognitive impairment. On the
other hand, faster swim speeds induced by treatments may mask impairments based on
latency measures. In cases where swim speed is impacted by treatment, other metrics
(i.e., errors to criterion, number of animals reaching criterion performance) may be
more reliable to determine presence of a learning impairment;
•	Variability must be considered in interpretation of findings (Raffaele et al, 2008).
Excessive variability can make it impossible to demonstrate that learning has occurred,
and will also make it difficult to detect changes in performance between control and
treatment groups. Variability that is minimal or absent may also indicate a lack of
sensitivity of the task to detect potential cognitive impairments, and should also be of
concern;
•	Positive control data as described below in Section 8 must also be included to
display competency that the laboratory, using the same procedures and protocols,
can detect impairment if one exists.
4. Test Description: Spatial Learning using the Morris Water Maze Test
The Morris water maze (MWM) (D'Hooge and De Deyn, 2001; Morris, 1981, 1984) is widely
used in the field of cognitive neuroscience as well as neurotoxicology, and is supported by a rich
literature regarding the effects of drugs on learning and memory and involved brain regions.
Place training in the Morris maze assesses spatial learning abilities, which are believed to be
primarily hippocampal-based (Clark et al., 2000; Squire, 2004; Eichenbaum, 2000), whereas
memory probes tap the nature, persistence and organization of memory representations within
the cerebral cortex.
In the MWM, animals rely on distal cues in the environment to navigate from various start
locations around the perimeter of an open swimming arena to locate a submerged escape
platform. As with the water-based position discrimination mazes, escape from water provides the
motivation, and no food or drinking water restriction is required. Spatial learning is assessed
across repeated trials and reference memory is determined by preference for the platform area
when the platform is absent. Although working memory is less frequently assessed using this
task, protocols have been developed and a number of papers have been published demonstrating
working memory deficits using the MWM (Dudchenko 2004; Dudchenko et al., 2013; Vorhees
and Williams, 2006). The working memory version of the task has not, however, been used in
guideline toxicological assessments and so will not be discussed further here.
Measures of learning and memory are influenced by sensory, motor and motivational variables
(Tilson and Mitchell, 1984). The interpretation of data from a learning/memory task should
incorporate consideration of these variables by using the appropriate control procedures or
conducting independent assessments (e.g. ability to swim in a swim maze and find a visible
Learning and Memory
D - 15

-------
platform). Recently, the MWM has become popular for the fulfillment of International
Congress for Harmonization (ICH) and Food and Drug Administration (FDA) juvenile studies.
Thus, while not as common in standard DNT studies as the letter mazes, there are a number of
studies that have and will use this test paradigm. The MWM has been successfully used to
detect learning and memory deficits in young and adult rats and mice. As with position
discrimination learning, the tank used in the MWM is scaled to the size of the animal and a
number of optimal parameters have been tested and summarized in recent reviews by Vorhees
and Williams (2006, 2014a, 2014b). A number of variations of the MWM basic protocol can be
applied to probe for different aspects of spatial learning and memory. However, despite
extensive use of the MWM, the task has not always been optimally implemented in guideline
or literature-based studies. Some of this stems from an under-appreciation for specific aspects
of the apparatus and testing procedures that are most salient for obtaining the most reliable and
interpretable results.
The maze itself is a simple circular pool that should be at least three-quarters filled with water
(see Figure 2). It is critical that the interior is made such that it is as close to being featureless as
possible to eliminate as much as possible any intramaze cues to help the animal locate the
platform. It is also important that the escape platform lies just below the surface of the water and
cannot be visually detected by the animal as it swims. Making the water opaque or making the
maze and platform the same color are examples of methods that may be used to camouflage the
platform. The platform should be placed far enough from the tank wall so that the rat cannot find
it by simply swimming close to the wall around the tank (thigmotaxic behavior). The goal is to
have the animal locate the hidden escape platform using spatial cues positioned around (outside
of) the maze. The distal cues, e.g., objects or pictures, should be large and not color-based since
rodents have relatively poor and monochromatic vision. As there are no proximal cues available,
i.e., cues within the maze itself, the use of distal cues provides the most effective strategy to
locate the hidden escape platform. The escape platform is typically 10 cm in diameter for rats
(smaller for mice), made of clear acrylic, and may have a rubberized surface to increase the
animal's traction to climb on once located. Scaling of the size of the pool and the size of the
escape platform can highly influence the difficulty of the task - the larger the pool and the
smaller the platform, the more difficult the task. The search area to escape target ratio is an
important variable to consider and optimal ratios vary between mouse and rat (Vorhees and
Williams, 2006; 2014a; 2014b). The maze is divided into quadrants with cardinal positions
arbitrarily designated as north, south, east, or west (N, S, E, W). While there is no accepted
optimal water temperature, studies have shown that rats learn better with water at 19-25° C but
higher or lower than that will impair learning (Salehi et al., 2010). Details completely describing
the water maze set-up, size, and procedure should be included in the report.
Learning and Memory
D - 16

-------
HkJdofi ptet/wm
Wodng
Figure 2. Morris Water Maze. Animals must navigate a large pool to find a hidden
escape platform submerged below the water surface. Different starting points are
used on each trial but the position of the platform remains fixed over trials. Distal
cues are placed on the walls surrounding the maze and remain invariant over the
course of the training.
The platform is positioned in one of the four quadrants and animals start trials from one of the
four cardinal positions as described below. The escape platform is typically positioned in the
middle of one of the quadrants and remains there throughout testing. The animal is released into
the water facing the wall of the tank and must search for the submerged platform. For video
tracking, a camera is suspended from the ceiling and positioned over the middle of the maze and
attached to a recording device. The animal's position in the pool is recorded, from which the
software computes a number of variables, the most common of which are distance and time
measures. Other measures that may be included are positioning data throughout the trial that
provide a spatial analysis of swim patterns (Blokland et al., 2004); however, such data are rarely
reported in guideline studies.
Two important performance control features should be included to rule out treatment-related
visual and motoric impairments that can confound interpretation of cognitive deficits identified
using this task. These include swim speed and cued-trials and are discussed in more detail below.
4.1. Procedures for Spatial Acquisition
Acquisition.
Place or spatial learning is the most basic MWM procedure. Typically either 4 trials are
administered on each of 5 consecutive days or 2 trials/day are administered for 10 to 12 days.
Most protocols use four start locations, N, S, E and W. Each start location is represented each
day and locations are randomly chosen within a day, without replacement. As such, all start
positions are used each day in 4 trials/day protocols and across 2 days in 2 trials/day protocols;
the same start point is never repeated within a day. The order of start points is also varied from
day to day. These are designed so that the animal is not able to learn a specific order of right or
Learning and Memory
D - 17

-------
left 'headings' to locate the platform, but rather must use the distal spatial cues. Animals are
typically run in squads of 4 such that the interval between trials is ~ 5 minutes if all 4 animals are
run in order on each trial. After locating and mounting the escape platform, the animal is left to
rest there for 15-30 seconds. If an animal fails to find the platform within the allotted time
(typically 60-90 seconds), it is guided there or placed on the platform for the same brief period.
The water should be gently stirred between trials to prevent rats following urine trails from
previous animals (Means et al., 1992). Using a heating pad under or around the cage may keep
the animal warm while drying.
Below, Figure 3 depicts learning in an individual rat using video tracking during early (left)
and late (right) learning trials. Path length drawn in these tracings is automatically
calculated by software and is highly correlated with latency. Figure 4 illustrates some
examples of learning curves based on latency to find the escape platform in a 2 trials/day
protocol (left) and a 4 trials/day protocol (right). In Figure 4, propylthioruacil during early
development impaired the acquisition of spatial learning in the adult. A different pattern
was seen with developmental treatment with tebuconazole where longer latencies were
seen in the high dose group on all days of acquisition.
Figure 3. Individual tracings of the same rat on the first and last day of spatial training.
The video system imposes landmarks (e.g., platform positions): platform 2, darkened circle
in the right bottom quadrant, is the target. On the first day, the rat swims mostly near the
tank wall, and does not find the platform within 60 sec. On the last day the rat quickly
heads in the correct direction and has a latency of 10.7 sec (Moser, personal
communication).
Learning and Memory
D - 18

-------
Males
A.
Acquisition
70
± \J tv v®
* * * A j.	^ ...	
• 0 ppm (n= 11)
O 3 ppm (n=15)
Control
-A- 6 mg/kg/d
—O- 20 mg/kg/d
—0- 60 mg/kg/d
I 1 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Day of Training
Figure 4. Examples of acquisition data (latency to find platform) across training
trials for two different chemicals. On the left are latency data (X±SEM) from
animals with thyroid hormone insufficiency following developmental exposure to
propylthiouracil. Control animals show a decrease in latency over training trials (2
trials/day), with longer latencies in exposed animals. Very little learning is evident
in the high dose group. Significant main effects of dose and day, and a significant
dose-by-day interaction were detected in this study (Gilbert and Sui, 2006). On the
right are data on rats exposed developmentally to tebuconazole, in which there was
a significant effect on learning at the high dose (Moser et al., 2001). There were no
significant group differences on the very first trial, but when analyzed across daily
blocks (Monday through Friday the first week, Monday through Thursday the
second week, 4 trials/day) there was an overall dose effect (no significant dose-by-
day interaction). This indicates that the treated rats were able to learn but the rate of
learning was slower than controls. In both studies, swim speed was not altered by
treatment.
Learning and Memory	D - 19

-------
Retention.
To assess reference memory at the end of learning, a probe trial is given in which the platform is
removed. During this time, the animal searches the area where the platform had been located. A
duration of 30 or 60 sec is adequate to observe bias for the target area, and with longer durations
the animals may begin to search more randomly. The most common method is to administer one
probe trial 24 h after the last acquisition day. With some procedures, the probe trial is
administered immediately following the last learning trial but this is not desirable as such a
procedure cannot differentiate between short- and long-term memory. Performance of the probe
trial on the same day as acquisition may reflect memory for the most recent training session
rather than retention of the spatial location from the previous day. A longer interval between the
last training trial and the probe trial is essential if reference memory is to be determined
independently of the memory of the last training session.
Performance Controls.
Test guidelines specify that when cognitive deficits are detected, performance control procedures
(e.g., swim speed and cue learning) should be conducted to dissociate cognitive deficits from
sensorimotor performance impairments. Two often-used performance controls are assessment of
swim speed and visual function. It is common practice in published literature reports to include
data on both of these parameters as they are often deemed critical for data interpretation.
a)	Swim Speed Performance Control.
As in position discrimination learning, it is necessary to ensure that the animals are physically
able to perform the task - swimming and escaping from the water. This can be conducted in a
separate apparatus such as a straight alley as described for position discrimination, or can be
conducted in the MWM if the system is equipped with tracking from which swim speed can be
calculated using path length and latency measures. Swim speed provides information regarding
whether there is any deficit in motor function, which could potentially confound interpretation of
learning deficits if a decrease in latency is used as the main measure of learning for the task -
i.e., an increase in latency could be the result of a decrease in swim speed rather than an inability
to learn, or, conversely, faster swim speeds may mask learning deficits based on latency
measures. Since test systems that use video tracking can calculate swim speed for all test trials,
this information could be used to assess motor abilities throughout training.
b)	Visual Performance Control.
A cued version of the learning task could be conducted as another sensorimotor performance
control in the MWM in order to dissociate a cognitive impairment from a sensory impairment.
Cue learning differs from place learning since the cues are internal rather than external to the
tank. During this trial the platform is moved to a different quadrant (typically the opposite
quadrant), and is raised above the water surface and made clearly visible (colored disk or
position marked with flag). The animal simply has to swim to the visible platform. When
multiple visual trials are conducted, both the platform position and starting point must be moved
randomly on every trial. Cue learning combined with assessments of swim speed provides
information on the animal's sensory and motoric competence as well as motivation to escape.
Other Learning and Performance Tasks.
Other procedures using MWM include reversal learning and a working memory version of the
task. Both have been reported in the literature and may provide additional, and perhaps more
Learning and Memory
D -20

-------
sensitive, assessments of learning and memory (Gilbert and Sui, 2006; Hoh et al., 1999; Morford
et al., 2002; Vorhees and Williams, 2006; Williams et al., 2002; 2003). Although these are not
typically implemented in DNT studies, a brief description and the potential utility of these tasks
are described below.
a) Reversal Learning.
Once animals have acquired the task to asymptotic and comparable degrees across treatment
groups, reversal learning can be evaluated. In the MWM, reversal learning simply entails placing
the escape platform in the opposite quadrant used during initial training and determines the
number of trials required for the animal to recognize the shift. An identical procedure to that
used during acquisition is used and if these trials are to be run, they may precede cue learning as
the visual performance control. Typically, control animals adjust to the new location within 3-5
trials. Perseveration to the previously correct position is common in brain-damaged animals and
may reflect a failure to recognize or adjust to the new task requirements (e.g., Whishaw and
Tomie, 1997; Castane et al., 2010). Continued responding with the previously correct choice may
stem from an inability to recognize the change in the task contingencies (what the animal needs
to do in order to escape), or an inability to learn a new response pattern, both examples of
cognitive inflexibility. Alternatively, treated animals could exhibit fewer errors than controls on
the reversal challenge, due to poorer retention of the initial task and thus less 'interference' as the
requirements change under reversal conditions.
Reversal Learning
40
• 0ppm(n=ll)
O 3ppm (n=9)
35
30
25
20
15
10
5
12 13 14 15 16 1
?
4 5
Day of Acquisition	Reversal
Figure 5. Reversal learning in a subset of animals that were exposed developmentally to
propylthiouracil. The rats had acquired the task to criterion performance (final days of
acquisition on left), and the platform was then placed in the opposite quadrant to evaluate
reversal learning. Both groups quickly acquired the reversal, but the treated animals exhibited
slightly longer latencies suggestive of a reversal learning impairment.
b) Working Memory.
Working memory refers to short-term retention of information, and may be measured using a
match-to-position version of the MWM (Vorhees and Williams, 2014a). In this procedure, the
platform is placed in a new position each day. The first trial allows the animal to identify the
Learning and Memory
D -21

-------
new position, and subsequent trials on that same day test the rat's memory for that location.
While 2 trial sets/day are adequate to observe improved performance, additional trials may be
used to provide a more robust evaluation of working memory.
4.2.	Standard Test Conditions for the Morris Water Maze
As with simple position discrimination learning, all standard test conditions described below
must be maintained for appropriate application of the MWM. Additional requirements are
specific to the MWM as detailed below. Standardization of maze size is recommended and a
minimal diameter is critical. Small mazes are simpler to solve and the dynamic range for
detecting suboptimal task resolution is limited. Larger mazes provide a more challenging spatial
task, and represent more sensitive means to detect learning and memory impairments (Vorhees
and Williams, 2014a).
•	Maze size and water depth and temperature should be specified, within acceptable limits,
and appropriate to the age and size of the animal being tested;
•	The submerged escape platform is not visible;
•	Positioning of the animal at the beginning of each trial should be done with the head
facing the wall of the tank and the start point chosen randomly over trials;
•	The duration of time that the animal is left on the escape platform after each trial must be
consistent and specified;
•	Intertrial interval (the delay between removing the animal from the platform and the
beginning of the next trial) and handling and holding conditions between trials (where the
animal is placed between trials, if the animal is towel dried, if the animal is warmed)
must be consistent and specified;
•	Stable and comparable performance in the acquisition phase is essential for appropriate
interpretation of learning;
•	Performance control procedures including swim speed and cued learning trials should be
incorporated.
4.3.	Metrics to Evaluate Learning and Retention in the Morris Water Maze
There are a number of dependent measures that are evaluated for MWM learning, but submitted
studies typically include assessment of latency and path length to find the hidden platform, and
search time in the correct quadrant on probe trials. Video tracking provides data on numerous
variables, whereas fewer measures are possible if data are collected by a technician with a
stopwatch. The most common variables are described in more detail below.
Latency to Escape over Trials.
This measure reflects the time from release of the animal in the maze until it reaches the escape
platform on each trial. Note that latency is to some extent dependent on the size of the maze and
thus varies from laboratory to laboratory. A decrease in latency over the course of training
provides evidence of learning.
Path length to Locate Hidden Platform.
Path length is the distance travelled to find the platform on each trial. As with latency, a
Learning and Memory
D -22

-------
decrease is evidence of learning. This measure is highly correlated with escape latency, but note
that there is no upper cut-off for path length as there is for latency (maximum trial duration).
Search Parameters during Retention Probe Trial.
Time spent searching in the correct quadrant as well as the number of passes over the platform
position are typical measures assessed during probe trials.
4.4. Data Reporting for the Morris Water Maze
In order to evaluate the effects of chemicals on these parameters, it is essential that MWM
reports provide clear information on study design, the measures (dependent variables), software
algorithms that define the variables, and the statistical procedures used to evaluate the main
results. The main results that are of interest from DNT learning and memory assessments are:
•	The effects of chemicals on acquisition, i.e., initial learning rate;
•	Effects of chemicals on memory, i.e., retention of the learned response;
•	Visual deficits;
•	Swim Speed;
•	Sex differences in effects of chemicals on learning and/or retention;
•	Preexisting differences across groups, evaluated by data from the first trial or first day of
testing.
Dependent Variables
It is important that the dependent variables recorded and analyzed are unambiguously described.
Averaging data across trials within a single day into daily blocks may be appropriate, especially
in instances where trial and block are included in the statistical analyses and there are no
significant interactions therein. For MWM learning, the dependent measures may include the
following:
•	Latency to reach the escape platform, and how this changes over trials;
•	Path length (distance travelled) to reach the escape platform and how this changes over
trials;
•	Percent of time in each quadrant on probe trial (s);
•	Number of platform crossings on probe trial (s);
•	Latency to reach escape platform that is clearly visible to the animal (to control for
sensory integrity that could influence maze performance independently of spatial
learning);
•	Swim speed (to control for motor impairments that may impact escape latencies
independently of spatial learning).
Reported Data
In order to evaluate treatment effects, the following level of data detail should be reported for
each sex at each age group. In cases where data are not normally distributed, different measures
of central tendency may be more appropriate (e.g., median, mode). Data should be reported
separately for acquisition and retention assessments.
Learning and Memory
D -23

-------
•	Mean ± standard deviations of latency per treatment group on each trial;
•	Mean ± standard deviations of path length per treatment group on each trial;
•	Mean ± standard deviations of percent of time spent in each quadrant on probe trial per
treatment group;
•	Mean ± standard deviations of number of platform crossings per treatment group on
probe trial;
•	Mean ± standard deviations of latency and path length per treatment group on cue
learning trials;
•	Mean ± standard deviations of swim speed per treatment group on probe trial;
•	Individual trial data for each animal should be included in an appendix;
•	Summary (means and standard deviations) should be presented in tabular and in graphic
form.
All data should be reported both on an individual and group basis by trial and by day over
treatment groups. Individual data should include every trial performed for each animal,
including cue learning trials when available.
4.5. Interpretation of Morris Water Maze Results
The following points should be taken into consideration in the interpretation of the data:
•	Control animals should demonstrate a clear and statistically reliable reduction in both
latencies and path length over trials. The absence of these properties in control animals
indicates lack of a learning curve and as such the test does not satisfy the minimal
requirements to detect a change in behavior over time and does not measure learning;
•	Control performance on a single probe trial run 24 hours after final acquisition must show
an increase in percent time in the correct quadrant that exceeds chance levels of 25%;
•	A treatment-related decline in performance on a single probe trial 24 hours after final
acquisition can only be interpreted as a retention deficit if comparable levels of learning
have been achieved in all treatment groups on the final day of acquisition;
•	First trial data should be closely examined to determine the potential presence of a pre-
existing motoric difference among treatment groups that may skew learning curves;
•	Measures of sensory and motoric function should be assessed; for example, slower swim
speed producing longer latencies may be misinterpreted as a learning deficit;
•	Variability should be within a reasonable range as described below in Section 9.
5. Test Description: Associative Learning using Passive Avoidance
Unlike most of the other learning tasks currently used in the guideline DNT study, passive
avoidance (PA) is not a water-based maze. PA was originally designed as a task to investigate
the impact of various interventions on memory consolidation. It was specifically designed as a
type of one-trial learning, so that the relationship between the timing of the intervention (with
respect to the timing of the learning and retention trials) and task performance could be
evaluated. This led to an improved understanding of the process of memory consolidation, in
that interference (e.g., by drug administration) at various times before or after initial learning
would lead to different impacts on whether or not the animal remembered the experience
Learning and Memory
D -24

-------
(Ehman and Moser, 2006; Peele et al., 1990). It is also a very useful way to evaluate whether or
not a drug or chemical has a specific effect on one aspect or another of learning or memory,
again based on evaluating response following different types of chemical administration
paradigms.
The basic procedure involves a device with two connected chambers, separated by a barrier
(usually an opening covered by a sliding door) (Jarvik and Kopp, 1967). During the training or
acquisition phase, the test subject is placed into one compartment, a bright light is turned on
and the barrier removed allowing the animal to enter the other chamber which remains darkened
(see Figure 6). Since the bright light can be aversive to rodents that prefer the dark, the test
subject is motivated to quickly enter the dark chamber to escape from the light. Upon entering
the dark chamber, the door is closed and a mild footshock delivered to the animal. This single
pairing between the dark chamber and the occurrence of footshock is usually sufficient to ensure
that the animal will avoid the dark chamber in the future. On the second trial (accurately defined
as the first retention trial), the animal is placed back in the original chamber (lighted side) and
the time to enter the dark side (previously associated with footshock) is measured. Typically,
footshock is not administered on the retention trial. The latency to enter the shock-associated
side (step-through latency) provides a measure of avoidance learning. This task is referred to as
"passive" avoidance since avoiding the shock-associated dark chamber does not require any
active response on the part of the animal, as it just remains in the original lighted chamber.
Depending on the strength of the aversive stimulus (shock magnitude and duration), rats
typically learn this task following a single trial, avoiding entry to the dark chamber on all
subsequent trials. Subsequent retention or "memory" of the aversiveness of the shock-
associated dark chamber is measured by again placing the animal in the lighted non-shock side
of the same device following a specific interval (varying from 1 hour up to 1 week for most
DNT studies).
Because of its relative ease of use and the large neurotoxicology literature for this task, PA has
become one of the most frequently submitted learning and memory paradigms for guideline DNT
studies. In multiple laboratories, it has been used at one or both time points (weanling and adult),
as the required learning and memory evaluation. However, the specific procedure used in DNT
studies varies widely from laboratory to laboratory, and differences in procedure have the
potential to greatly affect the sensitivity of the task to detect treatment-related memory
impairments.
Learning and Memory
D -25

-------
Figure 6. Passive Avoidance Chamber. Animals are placed in lighted side of the chamber and
receive a mild foot shock when they enter the preferred darkened side of the chamber. The ability
to suppress the motivation to move to the dark side of the chamber is assessed the following day
and taken as a measure of memory.
Passive Avoidance
5.1. Procedures for Passive Avoidance
Initial Acquisition
Typically, animals are placed on one side of a two-chamber box that is brightly illuminated, and
allowed to acclimate for a period of 20-60 seconds. A door is raised to allow access to the
darkened side of the chamber, and animals are allowed to move from the lighted area towards the
darkness. Upon entry to the dark chamber, the door between the light and dark sides is closed
and a mild brief foot shock is delivered. The animal is then removed from the chamber and
returned to the home cage until retention testing. Since this task is designated primarily as a
memory test with one-trial learning, there is no metric for 'acquisition'. Despite this, several
laboratories have adopted a multiple trial 'acquisition' phase, perhaps due to a misinterpretation
of guideline specification of learning defined as "a change in performance over time". Although
this guideline specification was not intended to apply to single-trial learning procedures like PA,
a number of repeated-trials variants of this task have appeared in which the animal is
immediately (within 30-60 seconds) placed back into the lighted side of the chamber following
foot shock delivery, with repetition of this procedure until some criterion is met, e.g., two
consecutive trials in which the animal refrains from entering the darkened chamber for a
specified duration. Since this procedure is not typical in the published literature, its utility and
interpretation are unclear.
One variant of a repeated-trials procedure with merit has been described by Peele et al. (1990) in
which daily trials were administered until a criterion of two consecutive trials (days) of latencies
exceeding the cut-off criterion of > 600 sec was achieved. An increased number of trials to
criterion was demonstrated in animals exposed to iminodipropionitrile (mean of 5 vs 2.7 days to
criterion), in addition to deficits in single trial PA conditioning. In this way, single trial PA
learning was used to assess 'retention' of PA learning, while in a second set of animals, the
repeated trial PA protocol (with trials separated by 24 hour intervals) was used to address
Learning and Memory
D -26

-------
'acquisition'. Unlike the study of Peele et al. (1990), submitted guideline DNT studies have
used relatively brief criterion cut-off times (60-180 seconds versus 600 seconds) for entry into
the darkened chamber.
In addition to a wide range of number of trials, inter-trial interval, and avoidance cut-off criteria,
various shock durations and shock intensities have also been employed. A 1 mA 1-second
scrambled foot shock delivered once all four paws have entered the darkened chamber is typical
in the literature. However, submitted studies have incorporated durations of 1-5 sec, and shock
intensities < 1mA. It should be noted that both intensity and duration of the foot shock will
influence memory retention (Ader et al., 1972). In some laboratories, different shock durations
and intensities have also been used for young versus adult animals. All of these parameters
should be considered when evaluating and interpreting PA data.
Retention testing
As described above, the primary metric in PA testing is the measure of retention (memory).
Following delivery of foot shock in the initial training phase, animals are reintroduced to the
same chamber after a delay under conditions identical to those used during training, except no
shock is applied. The latency to enter the dark side of the chamber (shock-associated side)
provides the main measure of memory. Typically retention trial cut-offs are 60 or 180 sec in
duration, whereas in the study cited above, a criterion of 600 sec was effective in detecting
treatment-induced deficits. Short duration retention trials can reduce the sensitivity of the task to
assess treatment-induced memory impairments. Because PA is a very effective procedure to
invoke avoidance, normally a single acquisition trial is sufficient to produce significant
avoidance of the shock-associated dark side of the chamber. However, shock intensity and
duration can influence the efficacy of this single-trial learning to produce avoidance. Typically,
retention is reported as 1 or 2 short-term trials (1- 6 hour delays following initial training)
administered on the same day of training. Long-term retention is assessed with longer intervals
between training and retention testing, typically 1 or 7 days post-training. Latency to enter the
darkened side of the chamber constitutes the measure of retention. It is compared to the latency
to initial entrance before administration of the first foot shock (i.e., the training trial).
As noted above, some laboratories have employed a multiple training trial approach where more
than one acquisition trial is employed (i.e., shock in dark chamber is applied more than once).
This procedure complicates the interpretation of memory results because most animals will
avoid the shock side of the chamber after only one trial. Because the second acquisition trial is
conducted exactly as the first acquisition trial, the subset of animals that do enter the shock-
associated dark side of the chamber is actually providing a measure of retention. On the other
hand, animals receiving a second shock-dark side pairing will be the subset of animals that have
not learned to avoid the chamber after the first shock acquisition trial. While this second shock
pairing will improve avoidance, animals receiving two shock-dark side pairings will have
different training experience compared to animals that learned to avoid the shock side after only
one acquisition trial. It should also be noted that since any second acquisition trial is actually
also the first retention trial, the interval between the first and second acquisition trials is also the
first retention interval. Because of this, care should be exercised in interpreting retention data
where multiple training trials have been employed.
Learning and Memory
D -27

-------
5.2.	Metrics to Evaluate Learning and Retention in the Passive Avoidance Task
Results from this test are usually reported as latency to re-enter the area where shock was
received on the training trial (s).
Various cut-off criteria have been adopted as described above. Errors have been defined as
failing to inhibit the response to enter the darkened chamber. Trials repeated within a given
retention session until 2 (or more) consecutive 'criterion' trials are reached. As described above,
the specific criteria used vary among laboratories.
Latency
Latency may be reported as step-through latency on the first trial at each retention interval and/or
as mean latency across all trials at a given retention interval. The latencies for the retention trials
are compared to latency to enter the dark side of the chamber on the first trial of the initial
training day.
Trials to Criterion
If a criterion performance is specified (e.g., 2 consecutive trials with >60 sec latency to enter the
darkened chamber), the number of trials to attain this are reported for each retention interval
assessed.
5.3.	Data Reporting for the Passive Avoidance Task
In order to evaluate the effects of chemicals on these parameters, it is essential that study reports
provide clear information on study design, shock parameters, the measures taken (dependent
variables), and the statistical procedures used to evaluate the main results. The main results that
are of interest from PA learning and memory assessments are:
•	Effects of chemicals on memory, i.e., retention of the learned response evidenced
by increased step-through latency;
•	Step-through latency on the first trial, which can show pre-training motoric deficits or
activity differences;
•	Sex differences in effects of chemicals on learning or retention.
Dependent Variables
The dependent measure in PA learning is latency to enter the chamber after the initial training
trial (i.e., on retention trials). This is typically assessed on the same day of training with an
interval ranging from 1-6 hours, and again 24 hours later.
Reported Data
In order to evaluate treatment effects, the following level of data detail should be reported for
each sex at each age group. Results should be provided separately for each trial, and for the
acquisition and retention phases. It is likely that data from PA procedures will not be normally
distributed, with a high proportion of censored trials (i.e., maximal values of 60 or 180 sec
entered for retention trials). In these cases different measures of central tendency may be more
appropriate (e.g., median, mode) and nonparametric statistical procedures should be applied (see
Learning and Memory
D -28

-------
Section 9). Data reported must include:
•	Mean ± standard deviations of latency per treatment group on each trial;
•	Individual trial data for each animal, included in an appendix;
•	Summary (means and standard deviations), presented in tabular and in graphic form.
All data should be reported on an individual and group basis. Individual data should include
every trial performed for each animal. For this procedure, it is not possible to evaluate the study
if data from the first acquisition trial are not reported.
5.4. Interpretation of Passive Avoidance Results
The following points should be taken into consideration in the interpretation of the data:
•	All animals must cross into the darkened chamber and receive a foot shock on the first
training trial in the acquisition phase in order to be included in the analysis. Animals that
fail to cross must be excluded from retention assessments since it is not possible to test
retention in the absence of learning. If a disproportionate number of animals among the
study groups fail to cross on the first trial, the data will be biased and the test results will
not provide a valid assessment of learning. The reviewer needs to examine the reported
data carefully to ensure that both of these conditions are satisfied and appropriately
reflected in results tables;
•	With highly variable data on retention trials, one potentially useful analysis that the
reviewer could perform, or request be performed by the reporting laboratory, is
application of a binning criterion to evaluate latencies, e.g., number of animals in each
group that crossed in 15 seconds or less; 30 seconds or less; 45 seconds or less; >60
seconds. Such an analysis may facilitate the comparison of distributions of latencies
across treatments and provide a more meaningful context of the magnitude of the
differences;
•	If a high proportion of animals from all study groups cross into the darkened chamber on
the first trial, but a difference in latency exists between treated animals and controls, this
may complicate the interpretation of retention deficits;
•	Rats must be able to feel the foot shock, or there will be no aversion. A test of shock
sensitivity (for pain and reactivity) could be performed to rule out sensory deficits;
•	Control animals should demonstrate a clear and statistically reliable increase in latency to
cross on the first retention trial (i.e., the trial that is assessed during the training phase,
Trial 2), relative to the training trial (i.e., Trial 1);
•	In this very robust learning task, latencies in control animals on the subsequent retention
trials typically remain close to that observed in the first retention test, but may diminish
with repeated exposures and very long intervals;
•	Variability should be within a reasonable range as described below in Section 10.
Learning and Memory
D -29

-------
6. Test Description: Sequential Learning using the Biel or Cincinnati Maze
1
Figure 7. Cincinnati Maze. The
animal must make a series of
right or left turns to navigate
V
from the start point to the escape
platform at the end of the
labyrinth. Both forward and
reverse paths may be tested. The
wm

—i
reverse path is not the mirror
opposite of the forward task.
Cincinnati Maze
The Biel and Cincinnati mazes are tests of sequential learning, and by that nature, are complex.
Both have been used to evaluate the cognitive effects of chemicals on the developing nervous
system, and they have been shown to be sensitive to a variety of developmentally neurotoxic
substances (Graham et al., 2012; Paul et al., 2009; Vorhees, 1987; Vorhees and Williams, 2016;
Williams et al., 2002). Although the mazes have been in use for many years, the type of
learning required for the task and the brain systems involved in achieving good performance
have not been as thoroughly investigated as is the case for position discrimination (the letter
mazes) and spatial learning (as tested in the MWM). Similar to position discrimination and the
MWM discussed previously, both the Biel and the Cincinnati mazes are typically run as water
mazes, with motivation being provided by the opportunity to escape from the water once the
final goal is reached.
The Biel and Cincinnati mazes can be appropriately described as multiple-T mazes (figure 8), in
which the animal is placed into the water at a defined start location and is required to remember
the correct sequence of right or left turns as a series of 'choice points' (most recently reviewed in
Vorhees and Williams, 2016). The Cincinnati maze was adapted from the Biel maze, expanding
and increasing the complexity by incorporating a larger number of choice points (9 vs 6), while
at the same time preserving the asymmetry of left vs right turns to reach the goal and escape from
the maze. During training, each incorrect choice can potentially result in multiple errors, both
within the blind alleys and when the animal returns to the main path of the maze. Because of
this, the number of errors made during acquisition frequently exceeds the number of choice
points.
As is the case with other types of mazes, testing procedures for the Biel and Cincinnati mazes
often vary among laboratories. Both mazes have been used wherein the starting point and
goals are moved creating different "paths". Figure 8 shows path A and B for the Cincinnati
maze. In path A, each choice takes place at a blind "T", or dead-end, such that there is a
50:50 chance of turning correctly. However, correct turns in path B are located before each
blind "T", such that turns made only at dead-ends are never successful. For this reason, path
B has been shown to be more difficult, and potentially more sensitive, than path A (Vorhees
Learning and Memory	D - 30

-------
and Williams, 2016). In most instances, there will be two acquisition phases for this type of
maze. After learning on one path, the goal and starting points of the maze will be switched and
the animal will learn the maze in the second path. Thus, in addition to evaluating the animals'
ability to learn a path through the maze, the test will also evaluate an animal's ability to
recognize that the correct choice has changed and to respond by changing their behavior and
learning a new path. This adds an additional level of complexity to the task and thus increases its
sensitivity to detecting cognitive deficits, in that animals not impaired on the initial learning
phase may be impaired in the 'reversal' task. Long term retention can also be evaluated by
retesting the animals on one or both paths after several days with no testing.
6.1. Procedures for the Biel or Cincinnati Maze
Initial Acquisition
Prior to acquisition, the animal must be acclimated and its swimming ability evaluated during
several trials in a straight channel. This also allows opportunity for the rat to learn that there is a
platform for escape. On the following day, the animal is placed in the maze at the start point,
facing toward the wall. Once placed in the maze, the animal navigates the various choice points
until it reaches the goal arm that contains a hidden platform allowing escape from the water.
Figure 8. Path B is not the converse of Path A in the Cincinnati or the Biel Maze. In Path A of
this Cincinnati maze, the animal is forced to go left or right at the cross point of the 'T'. In Path
B, there are always more than a simple left/right decision to be made and more blind alley entries
as described above. Path B is not the direct opposite of Path A and as such, does not constitute
true 'reversal learning'. (Vorhees, personal communication).
Both errors and latency are recorded during the trial. Each incorrect entry should be counted as
an error, and a clear definition of errors should be provided. For each trial, a maximum latency
(trial duration) is defined, typically 3-5 minutes. Correction procedures may vary among
laboratories, but generally the animal will be guided through the correct path to the goal of the
maze, at which point it is removed from the water. The number of trials per day may also vary
(usually 2-4), as may the number of days of training (2-4) and the inter-trial interval (at least 5
minutes, to prevent over-fatigue). Although it is reasonable for these parameters to vary among
laboratories, each laboratory should have a defined procedure that is consistently followed for
Learning and Memory
D -31

-------
all animals tested in that study.
'Reversal' acquisition:
After the initial path has been learned (i.e., met criterion), or after a set number of trials has
elapsed (regardless of performance), the start point and goal of the maze can be reversed, and the
ability of the animal to learn the new path evaluated. This is technically not a 'reversal' trial as
the animal does not simply have to make the opposite choice at each choice point. Rather, the
path for initial learning is distinct from the path for this second learning phase. However,
general procedures for this second phase of acquisition should be identical to those used in the
first phase, with the same number of trials per day, run under similar conditions. In some cases,
the second acquisition phase may include additional days of training, as there is usually a large
increase in errors during the first few acquisition trials in the 'new' path as the animal initially
applies previously learned responses to the changed situation.
Retention testing:
After the second acquisition phase, the animal may be re-tested on the initial path, in order to
determine retention. This re-testing may occur immediately following the second acquisition
phase, or there may be a delay of several days between the acquisition and retention testing. In
some cases, retention for both maze paths may be evaluated.
6.2. Metrics to Evaluate Learning and Retention in the Biel or Cincinnati Maze
Results from this test are usually reported as latency to reach goal and number of errors per trial.
Latency
Latency should be measured from the time the animal is placed in the start area until the animal
enters the designated goal area. Similar concerns regarding latency measures as described above
for position discrimination and the MWM also apply to data from the Cincinnati/Biel maze. In
multiple choice mazes in particular, however, latency data seldom reflect error performance in a
one-to-one fashion. This most likely results from animals deliberating at each choice point before
committing to a specific path as learning progresses, resulting in longer latencies in the absence
of higher errors. As in other less complex water-based mazes, swim speed can also affect latency
measures and potentially confound interpretation.
Errors
Error scores are generally reported as the total number of errors per trial. Unlike the position
discrimination maze (for which only one error per trial is generally reported), the number of
potential errors per trial in the Biel/Cincinnati maze is limited only by the time limit imposed on
each individual trial (the animal must decide among two directional options at each choice point,
and some choices may also lead to back-tracking through the maze). Although the pattern of
errors made by each individual animal within the maze could potentially provide useful
information, DNT study reports generally include only the total number of errors per trial. There
may be an occasional animal that stops searching and swims in a small area for the rest of the
trial, and low error rates would be recorded. Averaging group data with these low rates could
underestimate a real deficit. The incidence of such cases should be reported and evaluated as a
potential treatment effect. It is important that a clear definition be provided of what constitutes
an error and how errors cumulate over choice points within a given trial.
Learning and Memory
D -32

-------
Errors may be averaged within days (which will decrease the variability in results); however,
errors should also be reported for each individual trial, as it is useful to evaluate changes in the
number of errors both within and between days. This analysis can provide information regarding
the animal's memory for different retention intervals. Due to the design of the evaluations using
the Cincinnati/Biel maze in a typical DNT study, the animal is tested for a defined number of
trials. Thus, there is no measure analogous to the trials to criterion or errors to criterion metric
used for position discrimination protocols.
6.3. Data Reporting for the Biel or Cincinnati Maze
In order to evaluate the effects of chemicals on learning, it is essential that study reports provide
clear information on study design, the measures (i.e., clear operational definition of the
dependent variables), and the statistical procedures used to evaluate the results. The main results
that are of interest from Cincinnati/Biel maze learning and memory assessments are:
•	The effects of chemicals on acquisition, i.e., initial learning rate;
•	Effects of chemicals on memory, i.e., retention of the learned response;
•	If more than one path is tested, effects of the chemical on the animal's ability to adapt its
response and learn a new one (i.e., comparison of learning for the initial path and the
second path);
•	Effects of the chemical on swim speed;
•	Sex differences in effects of chemicals on learning and/or retention.
Dependent Variables
The dependent measures include the following:
•	Latency to reach the designated goal, and how this changes over trials;
•	Total number of errors for each trial;
•	Swim speed (to control for motor impairments that may impact escape latencies
independently of spatial learning).
Reported Data
In order to evaluate treatment effects, the following level of data detail should be reported for
each sex of each age group. Results should be provided separately for each maze path tested,
including the order in which the paths were evaluated (and for any repetition of paths tested
again at the end of the study). In cases where data are not normally distributed, different
measures of central tendency may be more appropriate (e.g., median, mode).
•	Mean ± standard deviations of errors per treatment group on each trial;
•	Mean ± standard deviations of latency per treatment group on each trial;
•	Mean ± standard deviations of errors per treatment group by day;
•	Mean ± standard deviations of latency per treatment group by day;
•	Mean ± standard deviations of swim speed per treatment group on straight alley trial;
•	Individual trial data for each animal should be included in an appendix;
•	Summary data (means and standard deviations) should be presented in tabular and in
Learning and Memory
D -33

-------
graphic form.
All data should be reported both on an individual and group basis and by day over treatment
groups. Individual data should include every trial performed for each animal. Data should also
be labeled by maze path, and the trial on which the path changed should be clearly indicated.
6.4. Interpretation of Biel or Cincinnati Maze Results
The following points should be taken into consideration in the interpretation of the data:
• As described for other test procedures, control animals should demonstrate a clear and
statistically reliable reduction in both number of errors and latencies over trials; this
pattern should be demonstrated for both the initial maze path and the reversal path, but the
shape of the learning curve will differ in the two phases. The absence of these properties
in control animals indicates lack of a 'learning curve' and as such the test does not contain
the minimal requirements to detect a change in behavior over time and does not measure
learning;
•	The task requirement for change in behavior during the second phase is a more complex
form of reversal learning among common DNT learning and memory procedures. Control
performance on the initial trials of the second path should show an increase in the number
of errors over both: 1) the number of errors in the last trial of the initial path; and 2) the
number of errors in the first trial of the initial path. The increase in errors is typically
followed by a rapid (perhaps more rapid than during initial learning) decrease in errors
during the later trials (Figure 9);
•	Evaluation of performance by treated animals in the second phase is confounded when
similar levels of performance are not demonstrated on the first phase (i.e., if treated
animals do not learn the first path, they may not be impaired in the early trials of the
second path). If this occurs in a small portion of animals, performing an additional
analysis with their exclusion may be informative;
•	Variability should be within a reasonable range as described below in Section 10.
Learning and Memory
D -34

-------
o—OCoilrej	«
*—* Phtflytoin 100 mg/kq
¦ Phanytoin TOO ntf/kf
90
40
u
O »
20
I
10
O—oControl
*Fh«i)rtofo 100 mg/kq
~Ph««ytofa aw w»rtsi
t	r-—V~-
Trials
7 a i 10 « u
Trials
Figure 9. Mean (± SEM) number errors on Path A (left) in the first 6 trials and Path B (right)
for final 6 trials in animals developmentally exposed to phenytoin. Errors decreased over trials
in both versions of the task, and fewer numbers of errors were committed on Path A (left panel)
than the more difficult Path B (right panel). Developmental exposure to phenytoin impaired
learning in both versions of the task, but effects were greatly exacerbated under conditions of
Path B (Vorhees et al., 1991).
As with the MWM reversal learning described above, treatment-related effects could be
manifest in two different ways, depending on how the animal's behavior is changed by the
chemical: a) the treated animal could exhibit fewer errors on the reversal than the control animal,
due to poorer retention of the initial task (and thus decreased 'interference' as the animal tries to
implement a previously learned response in the newly changed conditions) or b) the treated
animal could exhibit more errors on the reversal than the control animal, due to an inability to
adjust its response in the new situation (i.e., it may continue to respond using the old choice
patterns, for a much longer time, due to its inability to recognize that there has been a change in
the test conditions and/or its inability to alter its response patterns or learn new response patterns
in response to that change in conditions).
The reviewer will need to evaluate data from treated animals in light of these potentially
complex response patterns, realizing that impairment can be demonstrated in multiple types of
performance, and that the demonstrated impairment pattern could potentially vary among
treatment groups.
7. Standard Conditions for All Tests of Learning and Memory
In addition to the multitude of environmental factors (e.g., lighting, noise) that may alter
behavior in general (described in Slikker et al., 2005), a number of additional factors can
specifically influence assessments of learning and memory. Many of these are common across
all behavioral assessments, others are specific to learning and memory, and still others are unique
to specific learning and memory tests. The procedures used by the testing laboratory to control
these factors should be adequately described in the protocol or experimental design or reported in
the results. These include:
Learning and Memory
D -35

-------
•	All testing should be performed in a quiet environment with minimal exposure to outside
personnel traffic;
•	Temperature, humidity, and lighting conditions in the test area, as well as time of testing
relative to light-dark cycle, should be kept constant;
•	Overhead lighting, positioning of experimenter relative to the maze, olfactory stimuli
(i.e., presence of other animals in the test room) should be controlled, and olfactory and
visual cues in/on the maze itself (intramaze cues) should be eliminated. This promotes
reliance on extramaze cues and diminishes animal's reliance on a position or cued
learning strategy for spatially-based (i.e., MWM, position discrimination) or sequential
learning tasks (i.e., Biel and Cincinnati mazes);
•	If several mazes are used to increase throughput of testing, treatment groups should be
counterbalanced over specific test devices and time of day. Care should be taken to
ensure there are no differences in visual perspective among mazes;
•	With repeated testing, the animal should always be tested in the same maze and at the
same time of day;
•	Standardized procedures (chemicals and drying time) for cleaning mazes should be
employed, including procedures used between trials, between test animals, and
between test days;
•	In water mazes, water temperature should be constant across days and neither too warm
so as to reduce motivation to escape, or too cold so as to induce hypothermic stress;
•	Maze size (alley length, width, wall height, and water depth) should be specified, within
acceptable limits, and should be appropriate to the age and size of the animal being
tested;
•	Positioning of animal in the start position, confinement of animal in a start box or simple
release in the start alley should be consistent and specified;
•	Correction procedure in event of an error must be described (e.g., if the animal is guided
to the correct goal, removed and placed on the goal, or if some other procedure is used);
•	Operational definition of correct response vs error should be clearly articulated (e.g., how
far does the animal need to go down the incorrect alley in order for an error to be
recorded [e.g., head only, nose only, head and shoulders, entire body]);
•	Duration of time animal is left in the correct goal box/escape platform on completion of
each trial must be consistent and specified;
•	Intertrial interval (the delay between removing the animal from the goal and returning the
animal to the start box for the next trial) and holding conditions between trials must be
consistent and specified (e.g., where the animal is placed, presence of a heat source);
•	Training of all subjects to the same level of 'criterion' performance must be specified;
•	Wherever data are collected by personnel (as opposed to video tracking, for example),
assurances should be made that the observer is blind as to the animal's treatment. In
such cases of multiple observers, some measure of inter-observer reliability should be
provided.
8. Control Data for Tests of Learning and Memory.
A number of behavioral properties in the measured data must be evident and can serve as a
reference point for evaluating reported results.
Learning and Memory
D -36

-------
•	Control animals should demonstrate a clear and statistically reliable reduction in both
latencies and errors over learning trials in escape-based maze tasks. For PA learning,
latency to enter the darkened chamber should increase and be maintained well above
those recorded in the initial trial;
•	Animals should attain criterion performance at a faster rate for retention trials compared
to the original acquisition learning;
•	Variability should be within a reasonable range, based on data from the test laboratory as
well as the literature;
•	Historical control data may also be useful for interpreting potential treatment effects;
however, such data must represent studies using all the exact same experimental
procedures. While historical control data can be supportive of conclusions, the most
appropriate comparison for interpretation and statistical analyses is the concurrent control
group.
To demonstrate that the task is sensitive to detect learning impairments related to chemical
exposure, positive control data using a chemical (preferable) or other insult (e.g., physical brain
lesion) for which learning deficits have been established should be provided, using exactly the
same apparatus, test procedures, and testing conditions as the reported study. The positive
control study should have been conducted within a reasonable time frame relative to the reported
study, to minimize experimental drift over time. The positive control data presentation should be
sufficiently complete to permit evaluation of the sensitivity of the method, including individual
data and measures of variability. The positive control study should preferably include a dose-
response curve, and exposures should include sufficient internal controls to document that
impaired performance is due to changes in learning and memory, not to other effects of the
positive control treatment (e.g., changes in motor function or nonspecific physiological reactions
to the chemical, which may result from high doses of the same substance).
9. Statistical Analyses
There are many considerations necessary for the appropriate analysis of learning and memory
data. Guidance for analysis of DNT data can be found in Holson et al. (2008). For all
developmental toxicity studies, the unit of analysis is the litter. That is, littermates cannot be
analyzed as if they were independent observations. If the experimental design is such that
multiple pups from a litter are tested, the sample size for statistical analysis is the number of
litters (not number of total pups). Where a male and female from the same litter are tested, sex
must be nested within the litter or used as a random factor. Only in step-down analyses
following a significant main effect or interaction with sex can males and females be analyzed
separately. Typically, interactions should be tested first, e.g., dose-by-time interactions. This
situation applies for all dependent variables. While general linear model ANOVA is historically
used for most analyses, homogeneity of variance across treatment groups and/or test times
should be noted. In many cases, mixed-effects models allows application of the appropriate
covariance matrix that best fits the data.
9.1. Statistical Analysis of Learning Data
Acquisition (i.e., learning) is typically measured over a number of test sessions, with improved
performance over repeated testing (or training) being an indicator that the animal has learned the
task. In cases where measures are made repeatedly (e.g., latency to find a platform on each trial
Learning and Memory
D -37

-------
over days), the within-subject data should be analyzed using repeated-measures analyses. For
such data, the overall ANOVA would include the effects of treatment, sex (within litter if
appropriate), and test session (e.g., days or trials), and potential interactions among these. A
simple approach to establish that learning has occurred would be a significant change over
repeated testing (i.e., significant effect of test session), indicating that the performance has
changed (in the appropriate direction). Examples would include decreased latencies to escape
water in the MWM (see Figure 4) or decreased number of choice errors in the Biel maze (see
Figure 9). Applying such a test to data from the control animals should reveal that the control
group has learned the task. This is especially important because it is not possible to
unambiguously determine the meaning of any effects associated with chemical exposure if
control animals did not learn the task. Assuming that controls do indeed learn the task, an overall
treatment effect or interactions with sex and/or dose then can be interpreted as an effect of the
treatment. Typically, a significant effect of treatment on learning is seen as a significant
treatment-by-time (or day) interaction. Note that simply comparing data from control group
animals with treated animals at each test session separately, in the absence of a treatment-by-test
session interaction, is not appropriate.
Learning may also be evaluated as a specific endpoint instead of repeated measurements of
performance. For example, learning may be measured using trials to criterion or days to criterion
instead of latency values from each test session. Since this approach results in only one measure
of learning (e.g., days to criterion), a repeated measures ANOVA cannot be used for data
analysis and a simple factorial ANOVA would be appropriate using sex and treatment as factors
in data analysis.
9.2.	Statistical Analysis of Memory Data
Memory may be evaluated in a single trial after learning has been established, and in those cases
there are no repeated measures. For example, probe trials typically used in the MWM are
normally one trial, measuring time spent in the platform area. Since this measure is typically
collected only once for memory testing, repeated measures ANOVA cannot be used. Where
there are repeated memory trials (e.g., PA testing on multiple days after shock), the use of
repeated measures is necessary.
9.3.	Data Distributions
It is important to recognize that the properties of the measurement data for learning (or memory)
influence the type of statistical analysis procedure that is appropriate. Parametric statistics, such
as ANOVA, require continuous data as the measure (e.g., latency values, distance, etc.) and
assumes that the data is normally distributed. There are a number of statistical tests available that
evaluate the normality of data sets. Count data (e.g., number of errors) are not continuous and
may not be normally distributed; however, while care should be exercised in how such data are
analyzed, ANOVA may be appropriate nonetheless. It should be noted that ANOVA is
relatively robust for modest departures from normality as well as homogeneity of variance, and
may still be used in many cases. However, inappropriately analyzing distributions with marked
deviations from normality using ANOVA can underestimate treatment effects. In some cases,
transformation (e.g., square root) of the data may improve underlying distributions and variance.
Learning and Memory
D -38

-------
Non-parametric analyses, e.g., Kruskal-Wallis test, may be used in cases of non-continuous data
or undefined distributions. For example, learning measurements may be non-normally
distributed in situations which trials-to-criterion is used as the dependent variable, and
nonparametric statistics should be used. Visual inspection of the data should be used as well as
these tests.
Deviations from normal distribution can be created in some assessments of learning and
memory due to a cut-off criterion which generates closed-ended distributions - these are
considered censored data. This is especially of concern when the cut-off is actually the value
used as the endpoint of learning or memory. For example, maze procedures usually set a
maximum trial length (e.g. 60 sec) and a value of 60 sec is assigned to animals that fail to
reach the escape platform within 60 sec. While convenient, this 60 sec value actually
underestimates the time required for the animal to reach the platform (the actual value is
unknown but is greater than 60 sec) and reduces the estimates of variability of the response.
This results in a closed-ended distribution where data values beyond the maximum trial length
are not measured. One simple but intuitive method for approaching the latency data from a
study using PA is provided by the Kaplan-Meier (KM) method of estimating a survival curve
as depicted in Figure 10. With letter mazes such as the M-maze, there are often a high
number of trials with zero errors, which leads to closed-end data that is left-censored and
zero-centric. Low numbers in datasets such as these are not appropriately analyzed by
ANOVA, and require more specific analyses (see EFSA 2010 for example with Biel maze
data).
Trial 1
100
m
*
wm
High
10C
Figure 10. The KM method estimates the probability that the time-to-crossover is greater
than a specific time t. For those values that are not censored, the KM estimate is simply the
proportion of animals that have not crossed over by time t. For censored values KM is
undefined. The KM estimates can then be compared among groups using either the log-rank
test or generalized version of the Wilcoxon test (for example, SAS PROC LIFETEST). In
these graphs, trial 1 is the cross-over latency before shock. Trial 2, 24 hr later, shows the
latency across treatment groups. Most of the controls did not enter the shock compartment,
whereas a number of treated rats did. In this example, a conventional analysis suggested that
there is no treatment effect, while the Kaplan-Meier approach using log-rank survival tests
showed that in fact there is a significant main effect of treatment (Holson et al., 2008).
Learning and Memory
D -39

-------
9.4.	Correlated Measures
For many learning and memory procedures, multiple measures of performance are collected. For
example, latency, speed, and distance measures are often all reported in water mazes. It is
important to recognize that these measures are usually highly correlated or interdependent. For
example, because swimming speed is typically computed as the distance divided by latency,
latency and speed will be correlated. Even where more sophisticated algorithms in modern
equipment software are employed, these two measures would still be highly correlated. Another
example would be for the Biel maze where latency to reach the escape platform will typically be
greater where animals make more choice point errors simply because errors increase time for
task completion. In such a case, the correlations are inherent in the nature of the animal
behavior and not in the nature of the computation of the metrics. These intercorrelations
between multiple measures of performance have implications for data analysis since performing
separate ANOVAs, for instance, on speed and latency data assumes that these measures are
independent when they are not. Though multiple measures of performance are frequently
analyzed separately, it is important to recognize that they are not independent and in most cases
statistical results should confirm comparable results for separate measures. Note that, like
multiple uncorrected pairwise comparisons (e.g., multiple t-tests), multiple ANOVAs on
correlated learning measures also increase type 1 errors (i.e., finding a significant difference
when it does not actually exist). However, it should be noted that these are not as much of
concern in the interest of being cautious and health-protective. Multivariate statistical
procedures are available that adjust for the intercorrelations, but these are not yet widely used
and some are extremely conservative. Another approach to address assessment of multiple
correlated measures from a given dataset is to decrease or eliminate these redundancies (e.g.,
measures only path length or latency, but not both). Unfortunately, this also precludes
evaluating some of the information in the data set.
Another option would be to assess the effects of treatment on the expected correlations between
multiple measures of learning. This would be especially valuable where treatment may affect one
measure but not another. For example, treatments that alter motor function specifically would be
expected to influence speed more than choice errors and hence the correlations between these
measures would differ between control and treated animals. This may aid in interpretation in
cases where speed is not directly measured on every trial.
9.5.	Outcomes of Interest
In most cases, the types of effects that can be derived from learning and memory tests include:
Main effects:
•	Treatment: the dependent variable differs across groups, irrespective of other factors such
as sex and test-session. This indicates the simple case of a treatment effect that is the
same in both sexes across training;
•	Sex: difference between males and females without consideration of other factors. Since
sex differences are well-known for some cognitive tests, this can be used as an
assessment of the overall success of the task. For example, females typically do not
perform as well as males in spatial tasks (depending on the test parameters), and an
overall effect of sex should be expected (e.g., Blokland et al., 2006; Jonasson, 2005;
Learning and Memory
D -40

-------
Perrot-Sinal et al., 1996; Roof and Stein, 1999; Vorhees et al., 2008);
•	Test session: since learning is defined as a change in behavior as a result of training, there
should be an overall effect of test session indicating significant changes from the first to
the last test session. This is especially true for the control group when analyzed
separately; indeed, evidence of learning in controls is necessary for interpretation of
treatment data as well as satisfying guideline requirements.
Interactions:
Significant multi-way interactions need to be followed using appropriate comparison procedures
to assess the precise nature of the interaction. It is also possible to have significant interactions
but not have a significant effect on follow-up tests.
•	Treatment-by-test session: indicates that the rate of change of the behavior over test
sessions differs across treatment groups. This may occur when behavior on the first trial
is similar across groups (e.g., random searching for a platform), and differences in this
behavior become evident as testing continues. Even if asymptotic performance is
achieved, the time (e.g., number of training sessions, trials to criterion) taken to reach this
could indicate an adverse effect;
•	Treatment-by-sex: indicates that the treatment effect differs between male and females.
This may occur when the behavior is affected by treatments in both sexes but the
magnitude of effects differs between sexes, or the direction of change is different
between sexes;
•	Treatment-by-test session-by-sex: this 3-way interaction can indicate multiple outcomes,
such as the interaction between treatment and test session differs across sex, or the
interaction between test session and sex is altered by treatment. The nature of the
interaction must be followed with contrast interaction tests or step-down analyses of one
or two variables. A close examination of the data, or the specific hypothesis tested, can
influence the choice of subsequent analyses.
In all analyses, the data should be closely evaluated for consistency and expected outcomes,
especially in establishing that control animals learned the task. Group differences should be
considered in light of the magnitude of effect, since biologically significant changes could be
masked by high variability or under-powered studies. Likewise, individual data within the
groups should be examined where outliers may suggest animals that are more or less sensitive, or
else indicative of high variability.
10. Interpretation of Treatment-Related Effects in Tests of Learning and Memory
As noted above, in evaluating the results, care should be taken to account for possible non-
parametric distribution of the data (especially given the frequent use of cut-off criteria for trial
number and trial duration). In such cases, the appropriate measure of central tendency should be
reported and non-parametric statistical approaches applied.
For positional discrimination data, in addition to evaluating changes in the mean or median
values, the reviewer should look at changes in the pattern of errors. Changes in error patterns
can sometimes be manifested as an apparent mismatch between the number of errors and the
number of trials to criterion. This consideration is illustrated in Table 3 above. Careful
examination of individual data may be useful in evaluating differences among means/medians
Learning and Memory
D -41

-------
and variability among treatment groups.
As discussed in the Biel/Cincinnati maze section, it is also possible that longer latencies may
occur without an increase in errors or a decrease in swim speed. Such a pattern can result if an
animal hesitates at the choice point and may still reflect a learning impairment.
For tests in which acquisition occurs over several days and there are multiple trials per day, it
should be possible to evaluate within-day acquisition, and frequently to see some decrement in
learning on the first trial of each day, when compared to the last trial of the previous day. Data
should be presented in such a way that these daily learning curves can be assessed.
It is inappropriate to evaluate retention in the absence of comparable levels of performance
between treatment groups in the acquisition phase. Evaluation is straightforward in studies using
trials to criteria, whereas in studies with a fixed number of training trials, comparable
performance may not actually be reached by the last trial. This highlights the importance of
experimental design in interpretation of results.
Variability should be within a reasonable range. A complete lack of variability in control
animals may be indicative of an insensitive task, which can result from different aspects of task
design. For example, if all animals exhibit perfect performance during retention, then the task
may be too easy, the aversive stimulus too strong, or the initial learning criterion set too high.
Excessive variability in control animals decreases study power, making it very difficult to
demonstrate that learning has occurred, and will also make it difficult to detect changes in
performance between control and treatment groups (see Raffaele et al., 2008).
If impairment (e.g., measured changes in latency in position discrimination or the MWM
learning) is observed in the presence of a reduction in swim speed, it is not possible to
distinguish learning from performance deficits. Changes in swim speed may be indicative of
motor dysfunction and should be evaluated in the context of other behavioral assessments of
motor function (e.g., motor activity, cage side observations).
If impairment is detected in only one sex, it can still be considered evidence of a learning deficit.
If impairment is seen at only one age, it can still be considered evidence of a learning deficit.
Presence of an atypical dose-response relationship warrants closer evaluation of individual data
and consideration of the potential contribution of nonspecific effects that may be treatment-
related. For example, increased activity levels may lead to faster swim speeds but more errors, or
hypothermia may change the aversiveness of the stimulus, altering motivation. Consideration of
other information collected in the context of a guideline study may be informative in evaluation
of observed effects in learning tasks.
With these considerations in mind, it is important to emphasize that in the absence of sufficient
control procedures or other scientifically sound explanations, any change of performance in this
task must be interpreted as an adverse effect on cognitive function.
Learning and Memory
D -42

-------
11. References
Ader R, Weijnen JAWM, Moleman P. (1972) Retention of a passive avoidance response as a
function of the intensity and duration of electric shock. Psychon. Sci. 26, 125-128.
Blokland A, Geraerts E, Been M. (2004) A detailed analysis of rats' spatial memory in a probe
trial of a Morris task. Behav. Brain Res. 154, 71-75.
Blokland A, Rutten K, Prickaerts J. (2006) Analysis of spatial orientation strategies of male and
female Wistar rats in a Morris water escape task. Behav. Brain Res. 171, 216-224.
Brown KL, Pagani JH, Stanton ME. (2005) Spatial conditional discrimination learning in
developing rats. Dev. Psychobiol. 46, 97-110.
Carter CS, Freeman JH Jr, Stanton ME. (1995) Neonatal medial prefrontal lesions and recovery
of spatial delayed alternation in the rat: effects of delay interval. Dev. Psychobiol. 28,
269-279.
Castane A, Theobald DE, Robbins TW. (2010) Selective lesions of the dorsomedial striatum
impair serial spatial reversal learning in rats. Behav. Brain Res. 210, 74-83.
Chadman KK, Watson DJ, Stanton ME. (2006) NMD A receptor antagonism impairs reversal
learning in developing rats. Behav. Neurosci. 120, 1071-1083.
Clark RE, Zola SM, Squire LR. (2000) Impaired recognition memory in rats after damage to the
hippocampus. J. Neurosci. 20, 8853-8860.
D'Hooge R, De Deyn PP. (2001) Applications of the Morris water maze in the study of learning
and memory. Brain Res. Brain Res. Rev. 36, 60-90.
Dudchenko PA, Talpos J, Young J, Baxter MG. (2013) Animal models of working memory: a
review of tasks that might be used in screening drug treatments for the memory
impairments found in schizophrenia. Neurosci. Biobehav. Rev. 37, 2111-2124.
Dudchenko PA. (2004) An overview of the tasks used to test working memory in rodents.
Neurosci. Biobehav. Rev. 28, 699-709.
Ehman KD, Moser VC. (2006) Evaluation of cognitive function in weanling rats: a review of
methods suitable for chemical screening. Neurotoxicol. Teratol. 28, 144-161
Eichenbaum H. (2000) A cortical-hippocampal system for declarative memory. Nat. Rev.
Neurosci. 1, 41-50.
European Food Safety Authority (EFSA). (2010) Statistical re-analysis of the Biel maze data of
the Stump et al (2010) study: "Developmental neurotoxicity study of dietary bisphenol A
in Sprague-Dawley rats" EFSA Journal 8:1836. Doi: 10.2903/j.efsa.2010.1836
Freeman JH Jr, Stanton ME.(1991) Fimbria-fornix transections disrupt the ontogeny of delayed
alternation but not position discrimination in the rat. Behav. Neurosci. 105, 386-395.
Gilbert ME, Sui L. (2006) Dose-dependent reductions in spatial learning and synaptic function in
the dentate gyrus of adult rats following developmental thyroid hormone insufficiency.
Brain Res. 1069, 10-22.
Gold PE. (2003) Acetylcholine modulation of neural systems involved in learning and memory.
Neurobiol. Learn. Mem. 80, 194-210.
Graham DL, Schaefer TL, Vorhees CV. (2012) Neurobehavioral testing for developmental
toxicity. In: Developmental and Reproductive Toxicology: A Practical Approach, Third
Edition. RD Hood (editor), Informa Healthcare, London, UK.
Learning and Memory
D -43

-------
Green RJ, Stanton ME. (1989) Differential ontogeny of working memory and reference memory
in the rat. Behav. Neurosci. 103, 98-105.
Hoh T, Beiko J, Boon F, Weiss S, Cain DP. (1999) Complex behavioral strategy and reversal
learning in the water maze without NMDA receptor-dependent long-term potentiation. J.
Neurosci. 19, RC2 1-5.
Holson RR, Freshwater L, Maurissen JP, Moser VC, Phang W. (2008) Statistical issues and
techniques appropriate for developmental neurotoxicity testing: a report from the ILSI
Research Foundation/Risk Science Institute expert working group on
neurodevelopmental endpoints. Neurotoxicol. Teratol. 30, 326-348.
Kenny JT, Blass EM. (1977) Suckling as incentive to instrumental learning in preweanling rats.
Science 196, 898-899.
Jarvik ME, Kopp R. (1967) An improved one-trial passive avoidance learning situation. Psychol.
Reports 21, 221-224.
Jonasson Z. (2005) Meta-analysis of sex differences in rodent models of learning and memory: a
review of behavioral and biological data. Neurosci. Biobehav. Rev. 28, 811-825.
Kihlstrom, JF, Dorfman J., Park, L. (2007) Implicit and explicit memory and learning in M.
Velmans & S. Schneider (Eds.), The Blackwell Companion to Consciousness.
Blackwell, Oxford, U.K.
Means LW, Alexander SR, O'Neal MR. (1992). Those cheating rats: male and female rats use
odor trails in a water-escape "working memory" task. Behav. Neurol. Biol. 58, 144-151.
Morford LL, Inman-Wood SL, Gudelsky GA, Williams MT, Vorhees CV. (2002) Impaired
spatial and sequential learning in rats treated neonatally with D-fenfluramine. Eur. J.
Neurosci. 16, 491-500.
Morris RGM. (1981) Spatial localization does not require the presence of local cues. Learn.
Motiv. 12, 239-260.
Morris R. (1984). Developments of a water-maze procedure for studying spatial learning in the
rat. J. Neurosci. Methods 11, 47-60.
Moser VC, Barone S, Jr., Smialowicz RJ, Harris MW, Davis BJ, Overstreet D, Mauney M,
Chapin RE. (2001) The effects of perinatal tebuconazole exposure on adult
neurological, immunological, and reproductive function in rats. Tox. Sci. 62, 339-352
OECD/OCDE 426. (2007) OECD Guideline for the Testing of Chemicals, Developmental
Neurotoxicity Study.
OECD. Guidance Document Supporting OECD Test Guideline 443 on the Extended One-
Generation Reproductive Toxicity Test. Series on Testing and Assessment, No. 151.
2004.
Olton DS. (1979) Mazes, maps, and memory. Am. Psychol. 34, 583-596.
Pagani JH, Brown KL, Stanton ME. (2005) Contextual modulation of spatial discrimination
reversal in developing rats. Dev. Psychobiol. 46, 36-46.
Paul CM, Magda G, Abel S. (2009) Spatial memory: Theoretical basis and comparative review
on experimental methods in rodents. Behav. Brain Res. 203, 151-164.
Peele DB, Vincent A. (1989) Strategies for assessing learning and memory, 1978-1987: a
comparison of behavioral toxicology, psychopharmacology, and neurobiology. Neurosci.
Biobehav. Rev. 13, 317-322.
Peele DB, Allison SD, Crofton KM. (1990) Learning and memory deficits in rats following
exposure to 3,3'-iminodipropionitrile. Toxicol. Appl. Pharmacol. 105, 321-332.
Learning and Memory
D -44

-------
Perrot-Sinal TS, Kostenuik MA, Ossenkopp K-P, Kavaliers M. (1996) Sex differences in
performance in the Morris water maze and the effects of initial nonstationary hidden
platform training. Behav. Neurosci. 110, 1309-1320.
Raffaele KC, Fisher JE Jr, Hancock S, Hazelden K, Sobrian SK. (2008) Determining normal
variability in a developmental neurotoxicity test: a report from the ILSI Research
Foundation/Risk Science Institute expert working group on neurodevelopmental
endpoints. Neurotoxicol. Teratol. 30, 288-325.
Roof RL, Stein DG. (1999) Gender differences in Morris water maze performance depend on
task parameters. Physiol. Behav. 68, 81-86.
Salehi B, Cordero MI, Sandi C. (2010). Learning under stress: the inverted U-shape function
revisited. Learn. Mem. 17, 522-530.
Slikker W Jr, Acuff K, Boyes WK, Chelonis J, Crofton KM, Dearlove GE, Li A, Moser VC,
Newland C, Rossi J, Schantz S, Sette W, Sheets L, Stanton M, Tyl S, Sobotka TJ. (2005)
Behavioral test methods workshop. Neurotoxicol. Teratol. 27, 417-427.
Smith GJ, Bogomolny A. (1983) Appetitive instrumental training in preweanling rats: I.
Motivational determinants. Dev. Psychobiol. 16, 119-128.
Squire LR (2004) Memory systems of the brain: A brief history and current perspective.
Neurobiol. Learn. Mem. 82, 171-177.
Stanton ME, Spear LP. (1990) Workshop on the qualitative and quantitative comparability of
human and animal developmental neurotoxicity, Work Group I report: comparability of
measures of developmental neurotoxicity in humans and laboratory animals.
Neurotoxicol. Teratol. 12, 261-267.
Tilson HA, Mitchell CL. (1984) Neurobehavioral techniques to assess the effects of chemicals
on the nervous system. Ann. Rev. Pharmacol. Toxicol. 24, 425-450.
US EPA, Health Effects Guidelines OPPTS 870.6300, Developmental Neurotoxicity Study,
1998.
Vorhees CV. (1987) Reliability, sensitivity and validity of behavioral indices of neurotoxicity.
Neurotoxicol. Teratol. 9, 445-464.
Vorhees CV, Herring NR, Schaefer TL, Grace CE, Skelton MR, Johnson HL, Williams MT.
(2008) Effects of neonatal (+)-methamphetamine on path integration and spatial learning
in rats: effects of dose and rearing conditions. Int. J. Dev. Neurosci. 26, 599-610.
Vorhees CV, Weisenburger WP, Acuff-Smith KD, Minck DR. (1991) An analysis of factors
influencing complex water maze learning in rats: effects of task complexity, path order
and escape assistance on performance following prenatal exposure to phenytoin.
Neurotoxicol. Teratol. 13, 213-222.
Vorhees CV, Williams MT. (2006) Morris water maze: procedures for assessing spatial and
related forms of learning and memory. Nat. Protoc. 1, 848-858.
Vorhees CV, Williams MT. (2014a) Assessing spatial learning and memory in rodents. ILARJ.
55, 310-332.
Vorhees CV, Williams MT. (2014b) Value of water mazes for assessing spatial and egocentric
learning and memory in rodent basic research and regulatory studies. Neurotoxicol.
Teratol. 10, 75-90.
Vorhees CV, Williams MT. (2016) Cincinnati water maze: A review of the development,
methods, and evidence as a test of egocentric learning and memory. Neurotoxicol
Teratol. 57, 1-19.
Watson DJ, Herbert MR, Stanton ME. (2009) NMDA receptor involvement in spatial delayed
alternation in developing rats. Behav. Neurosci. 123, 44-53.
Learning and Memory
D -45

-------
Watson DJ, Sullivan JR, Frank JG, Stanton ME. (2006) Serial reversal learning of position
discrimination in developing rats. Dev. Psychobiol. 48, 79-94.
Whishaw IQ, Tomie J. (1997) Perseveration on place reversals in spatial swimming pool tasks:
further evidence for place learning in hippocampal rats. Hippocampus 7, 361-370.
Williams MT, Morford LL, Wood SL, Wallace TL, Fukumura M, Broening HW, Vorhees CV.
(2003) Developmental D-methamphetamine treatment selectively induces spatial
navigation impairments in reference memory in the Morris water maze while sparing
working memory. Synapse 48, 138-148.
Williams MT, Vorhees CV, Boon F, Saber AJ, Cain DP. (2002) Methamphetamine exposure
from postnatal day 11 to 20 causes impairments in both behavioral strategies and spatial
learning in adult rats. Brain Res. 958, 312-321.
Learning and Memory
D -46

-------
MODULE-E
CONSIDERATIONS IN THE ANALYSIS AND
INTEGRATION OF DNT DATA
Considerations and Integration
E- 1

-------
1.	Introduction	3
2.	Step-Wise Approach to The Integrated Analysis of DNT Study Data	4
3.	Overall Data Synthesis	4
3.1	Integration of offspring data	5
3.1.1	Experimental procedures and controls	5
3.1.2	Dose-response	5
3.1.3	Determining Biological Significance	6
3.1.4	Statistical Significance	7
3.2	Integration of maternal data	8
3.3	Integrating other lines of evidence	9
4.	Additional considerations for interpretation	10
4.1	Exposure	10
5.	Human Relevance	12
6.	Conclusions	13
7.	References	14
Considerations and Integration	E - 2

-------
MODULE E - CONSIDERATIONS IN THE ANALYSIS
AND INTEGRATION OF DNT DATA
1. Introduction
The purpose of this module is to provide further points for consideration in performing the
analysis and overall integration of results from DNT studies. This would include aspects related
to statistics, dose-response and control group considerations. In addition, integration of
individual parameters within the DNT study, including maternal effects, along with
consideration of findings in other toxicity studies is discussed. Additional considerations such as
exposure and human relevance are also addressed.
Several principles for evaluation of neurotoxicological effects have been put forth in guidelines
of the US EPA (1998) and OECD (2007). In general:
•	An agent that produces detectable adverse neurotoxic effects in experimental animal
studies is assumed to pose a potential hazard to humans;
•	Developmental neurotoxic effects in animal studies indicate the potential for altered
neurobehavioral development in humans, although there may be differences in the
specific manifestation of those effects;
•	An alteration in behaviors of the offspring, or in the ontogeny of behaviors, is considered
to indicate a developmentally neurotoxic change. These may occur with or without
neuropathological findings;
•	Changes in neuronal organization, structure, or neurochemistry also indicate an effect,
with or without accompanying functional effects;
•	Data from all potentially relevant studies and effective doses or exposures should be
considered in a weight-of-evidence approach to characterizing the potential for
developmental neurotoxicity;
•	While findings at all dose levels are important, those occurring at doses below levels
associated with maternal or general toxicity are generally considered of increased
concern;
•	While an understanding of toxic mechanisms or pathways can inform an overall
evaluation, the lack of such information does not preclude a determination of risk.
With the current DNT test guidelines, there are a number of behavioral tests and other
assessments conducted in the offspring. To evaluate such a wide range of outcomes, the data
need to be integrated, taking into account consideration of statistical and biological significance.
In addition, residual uncertainties resulting from limitations in the conduct of some parameters
assessed within the DNT study must also be considered, along with findings in the toxicology
database as a whole. The complexities in the nature and the assessment of the endpoints
generated in DNT testing make data evaluation a challenge. Reasonable interpretations of the
data are required to ensure that false positives (type I error) and false negatives (type II error) are
avoided. Approaches to integrating and synthesizing the data are outlined in this module, and
are also described in Tyl et al. (2008).
Considerations and Integration
E - 3

-------
2.	Step-Wise Approach to Integrated Analysis of DNT Study Data
The following outline describes the steps in evaluating DNT data as a whole, and integrating
DNT results with the available toxicity database. Data from the individual test measures should
be assessed before an overall synthesis can be initiated. Historical and positive control data can
be an additional resource for evaluating data but do not substitute for data/findings in the
concurrent controls.
Step 1: Evaluate the patterns and/or consistency of effect and the plausibility of each individual
endpoint. Separate guidance is available here for motor activity, acoustic startle
response, learning and memory, and observations (see modules A-D).
Step 2: Compare the effective dose levels, patterns of change, and the severity of effects across
endpoints within the DNT study. Consider changes in some endpoints in the context of
others where possible (e.g., influence of motor activity dysfunction on some cognitive
tests) (section 3.1).
Step 3: Evaluate general health and growth parameters (e.g., litter size and viability, body
weight, systemic toxicity) within the DNT study. Consider the possible impact of such
changes on the sensitivity or reliability of the behavioral test results (e.g., decreased
ability to detect statistically significant changes due to lower sample size due to pup
loss) (section 3.1).
Step 4: Evaluate the effective doses for, and extent of, maternal toxicity (e.g., decreased body
weight gain, systemic toxicity), as well as the effective doses and outcomes in the
offspring. It is important to take into account the test guideline requirement that the
highest dose level "should induce some overt maternal toxicity" (section 3.2).
Step 5: Integrate these determinations with the available toxicity database to evaluate relative
sensitivity of DNT effects and consistency of DNT effects with other measures (section
3.3). Additional factors can be considered, including route of exposure and kinetic
(absorption, distribution, metabolism, excretion) influences (section 4).
3.	Overall Data Synthesis
The major steps outlined above should be followed for the purpose of properly assessing the
evidence or data on patterns of related effects within and across toxicity studies. General
guidance (OECD, 2004; US EPA, 1998) states that changes in any DNT endpoint can reflect an
action on the development of the nervous system, unless data can be provided to support a
different conclusion. The level of confidence in the overall conclusion that a given chemical
adversely affects the developing nervous system can vary depending upon the weight of
evidence. Higher levels of concern are produced with datasets for which multiple, statistically
and biologically significant changes are evident. A lack of consistency or dose-response in the
data may lower, but not necessarily negate, the concern for potential DNT. The steps and
guidance below should be considered in evaluating data from the DNT study itself within the
context of the entire toxicity database.
Considerations and Integration
E - 4

-------
3.1 Integration of offspring data
To describe DNT outcomes, lines of evidence should be assembled and assessed. There is a
multitude of endpoints collected in the offspring during the course of a DNT study. Evaluation
across these endpoints should consider factors such as experimental procedures and controls,
dose response, treatment effects and statistical and/or biological significance and any
uncertainties. Graphical presentation of data is helpful for examining possible patterns of
behavioral changes (e.g., habituation, learning curves).
3.1.1	Experimental procedures and controls
Experimental and procedural details are necessary to assure that appropriate procedures were put
in place, including counterbalancing the testing across time of day, test days, test chamber, dose
groups, and assuring that most/all procedures, but especially subjective evaluations, are
conducted with testing personnel unaware of the specific group to which an animal has been
assigned. An examination of the control data can provide evidence of a well-conducted study.
Control data should be assessed for the following:
•	Do control animals show the expected response, based on the literature and historical
controls? Observed differences between concurrent and historical controls may be due to
a number of factors, e.g., a change in background sensitivity of the subjects (genetic drift)
over time, or a concurrent control group that is atypical. While such a situation may
complicate interpretation of the data, in general, the concurrent control is the most
relevant group against which to compare the treated groups;
•	Are control data internally consistent for the following factors?
o Evidence of learning over test sessions, e.g., a significant effect of trials,
o Habituation of motor activity and startle response over time, e.g., a
significant effect of trials,
o Maturation of certain behaviors with age, e.g., emergence of clear habituation
patterns in motor activity over PND17 to 21.
o Existence of sex differences in certain behaviors, e.g., male rats show better
performance than females in spatially-based cognitive tasks.
3.1.2	Dose-response
Dose-response is a key indicator of a treatment-related effect; however, there is currently
discussion on the relevance of non-monotonic dose-response curves (e.g., only the lowest dose
has an effect, or the dose-response curve does not show a consistent trend in one direction),
especially for certain classes of chemicals (Vandenberg, 2014). Although an orderly trend of
increasing effect with increasing dose is the clearest indication of a treatment-related finding,
there are many issues to consider. In the face of a weak or absent dose-response (i.e., no
gradation of effect), the pattern of individual animal data should be examined to identify changes
in incidence or severity of an effect that may have been present but not reflected in the group
data. Even if atypical, similar dose-response curves across multiple endpoints or dose groups are
particularly supportive of an effect. Considerations for assessing inconsistent dose-response data
include:
Considerations and Integration
E - 5

-------
•	Inappropriate dose selection may result in unexpected shapes that would be better defined
given more doses. The current guideline requirement for the minimum of three dose
levels makes it difficult to determine or define precise dose-response curve shapes;
•	Influence of kinetic factors (absorption, distribution, metabolism, excretion). Non-linear
responses could be reflective of saturation of metabolic systems or adaptation of
response;
•	Neural systems reflect interplay of both inhibitory and excitatory actions, and the relative
influence of these factors may impact a dose response. These may be observed as U-
shaped or inverted U-shaped curves: the classic example is the excitation followed by
sedation produced by ethanol. If possible, this interplay should be evaluated in
conjunction with any related dose-response measures that are available (i.e., similar
findings in tests measuring the same/similar underlying neurological construct).
3.1.3 Determining Biological Significance
There is strong evidence of a neurotoxic effect when several related measures in a battery of tests
are affected, there is biological consistency in the changes across endpoints, and/or the effects
appear to be dose dependent, especially in the absence of systemic toxicity. There are many
reasons, however, why these criteria may not be met in current DNT studies. Thus, while the
following considerations are important, the lack of such findings does not necessarily negate the
concern for developmental neurotoxicity.
Changes in related measures or endpoints, both within the study and across the database,
strengthen the evidence of a biologically significant effect. Where related measures or endpoints
exist within a database, one could expect similar direction of change as well as dose response.
Patterns to look for could include but are not limited to:
•	Altered motor activity levels are important to consider. Although motor activity is
directly measured in the automated assessment of activity, additional information on
motor activity can be gleaned from cognitive tests that include latency as an endpoint.
For example, an animal with low activity may not cross into the shock side in a passive
avoidance chamber, and thus confound the assessment of learning. Measuring the
latency to cross on the first trial will provide baseline data that can differentiate a learning
deficit from a difference in activity levels;
•	Motor deficits may be evidenced as neuromuscular weakness or inability to meet the
motor demands of a test. Such deficits may be suggested by decreased activity levels,
decreased startle amplitude, observations of sluggishness, or lethargy. In a water maze
cognitive task, poor swimming ability can impact latency; however, it may not actually
impair learning. In this case, swim speed information is useful for looking at motor
function separate from task performance, but the data from other measures may also be
informative;
•	Auditory dysfunction would most likely be detected in the auditory startle test; however,
other sensory deficits may not. Visual function, while necessary for some tests of
learning (especially spatial tasks), is not often evaluated. Likewise, other senses (e.g.,
olfaction) are ignored. Where tests of other sensory function are included, similarities of
effect across the senses would not necessarily be expected;
Considerations and Integration
E - 6

-------
•	Alterations in habituation pattern may be evident in motor activity and/or auditory startle
data. However, since there are subtle differences in the habituation patterns for these
measures, the data may not necessarily be concordant. Thus, the presence or absence of
concordance across these measures may not alter the interpretation of an effect;
•	In adult animals, brain weight is highly conserved in the presence of changes in body
weight, and the convention of using organ to body weight ratios is less useful when
studying brain weight changes (Sellers et al., 2007). Studies of malnutrition during
development suggest that severe reductions in body weight (e.g., <50% of controls) will
result in lower brain weight (Fernandez et al., 1985), but this has been less studied with
regards to moderate or slight (e.g., <10% of control) body weight differences. In general,
effects on brain weight cannot be dismissed even in the presence of body weight
differences, and should be considered treatment-related and adverse;
•	Body weight differences (reduced body weight or lower weight gain) or delayed
maturation in the offspring may affect some of the DNT measures. Data on growth and
body weight are necessary for evaluating neurobehavioral measures. Considerations
include:
o Physical development correlates with body weight, such that developmental
landmarks and neuronal functions may be delayed in smaller pups. Many of
the behavioral tests are influenced by motoric capacity and body size, e.g.,
auditory startle amplitude may be less in smaller pups. For this reason, if
changes are evident in the presence of delayed growth, it may not be possible
to distinguish whether changes in performance are related to delayed growth or
to delayed neurobehavioral development per se; for example, delayed
development (hair growth) results in a hairless pup whose performance may be
compromised in tasks where swimming is required;
o Most often the source of body weight effects is not known, and could be either
maternally mediated (e.g., decreased milk quantity or quality, decreased care)
or the results of a direct adverse effect on the pup (e.g., systemic toxicity,
neurotoxicity that affects feeding ability).
3.1.4 Statistical Significance
For each test, appropriate statistical analyses must be applied to determine significant differences
compared to controls. Necessary factors in statistical analysis include:
•	Litter as the unit of analysis when multiple littermates are tested (including males and
females);
•	Sex as a within-litter or random factor for statistical analyses where males and females
are sampled from the same litters;
•	Treatment, sex, and within-subject repeated measures in an overall design, followed by
step-down analyses and post-hoc group comparisons that are guided by the overall main
effects or interactions.
In some cases the statistical data may be misleading. Possible explanations for this include:
Considerations and Integration
E - 7

-------
•	Inappropriate methods of analyses (e.g., using parametric analyses for binary data);
•	Underpowered statistical measures. Low statistical power in the study can result from
poor experimental control over testing procedures or small sample size. While
coefficients of variation (CV, standard deviation as a percent of average values) can be
useful quantitative measures of variability, it should be noted that: 1) high CVs can be
influenced by numerically low values, and 2) CVs can only be calculated for parametric
data;
•	Individual outliers that affect group means (and increase group variability). This could
reflect overly sensitive or insensitive subsets of animals, but may also be a result of
technician or instrument error. Individual animal data are necessary to assess this;
•	Censored data based on pre-determined levels of significance (e.g., p-values > 0.05 but <
0.1). The reporting of actual significance probabilities is useful in this regard, especially
where treatment effects approach significance criterion values (e.g., p=0.052);
•	The magnitude of change may be too low to exceed the statistical threshold. However, a
small change may still be indicative of developmental neurotoxicity. Indeed, the
magnitude of effect may merely be a function of dose selection.
Tyl et al. (2008) provide case studies that include a discussion of how to balance biological and
statistical significance. Statistical significance of treatment effects does not necessarily mean that
the effect is biologically relevant. Because of the large number of potential endpoints in DNT
studies, care must be exercised in ensuring that experiment-wise error rates are appropriately
controlled to minimize false positives (Holson et al. 2008). However, because DNT studies are
conducted to screen chemicals for possible adverse effects on the developing nervous system
and these studies are often the only study with these biologically important neurological
endpoints, the willingness to accept a higher false positive rate instead of a lower false negative
rate may be a more conservative and protective approach.
Most importantly, biologically significant, treatment-related findings may occur in the absence of
statistical significance. If reported variances in a given behavioral test compared with other
reports from the same laboratory (historical controls) or with other laboratories suggest excessive
variance within a study, the study itself may be flawed, at least for the parameter under
consideration. The utility of these test results should be considered in the context of the other
measures in the study, at which point the particular parameter may not be used, or this could
warrant a repeat of the study with better experimental control.
Thus, the conclusion of whether or not a chemical exposure has resulted in adverse outcomes on
neurodevelopment should not rely solely on evidence of a statistically significant treatment-
related effect. Consideration should be given to factors such as multiplicity of statistical
findings as well as the level of statistical significance in the weight of evidence determination;
however, statistical significance should not override biological significance.
3.2 Integration of maternal data
The DNT test guidelines require that the highest dose level "should induce some overt maternal
toxicity", although the criteria for this determination are not described. The DNT test guidelines
do not specifically require evaluation of maternal behavior, but body weight (and sometimes
food/water consumption) will be available for review. The comparison of toxicity in the dam
Considerations and Integration
E - 8

-------
with that seen in the offspring is important in determining age-related sensitivity for certain
regulatory jurisdictions; issues to consider include:
•	A large decrease in food/water intake could result in decreased body weight that exceeds
an acceptable level, resulting in malnutrition of the dam. While this situation may result
in neurobehavioral and physiological impacts on the offspring, the level of maternal
malnutrition that consistently leads to impairment of neurobehavioral development has
not been determined;
•	Maternal care (e.g., nursing, nesting, grooming) is critical for neuronal development.
This can be monitored by observations of how well the litters are kept together (the dam
usually retrieves pups that go astray), temperature of the pups (coolness to the touch),
and the presence of milk bands. While alterations in maternal behaviors should be
considered in assessing potential neurotoxic effects in the offspring, these simple
assessments are not typically included in regulatory DNT studies;
•	Treatment-related differences in litter size, fetal birth weight, and pup growth can result
from maternal influences, as well as from direct toxic effects of the chemical on the fetus
and/or newborn. Data available from DNT studies are not sufficient to differentiate
between maternal and fetal origin of the effect; such findings should be considered
treatment-related and adverse.
There are several approaches for designing studies to address whether the effects are attributable
to maternal toxicity or direct neurotoxicity on the offspring. The most common of these is a
cross-fostering procedure. These additional studies are not typically conducted, and the reviewer
is advised to consult a neurobehavioral expert if such studies are submitted.
3.3 Integrating other lines of evidence
After conducting a critical analysis of the data from the DNT study, the reviewer should consider
findings from other studies:
•	Reproductive toxicity studies, either single or multigenerational. These exposures cover
the pre-mating, gestational and lactational periods as well as the post-natal periods in the
offspring (Fl, F2 generations). This continuous exposure is in contrast to the discrete
perinatal exposure of the DNT study. Information from such studies can be used to aid in
interpretation of DNT studies; however, differences in the dosing paradigms across
studies could lead to different manifestations of toxicity. For that reason, a lack of
similar findings across different types of studies should not be used to dismiss discordant
effects. Information from these studies that may inform DNT findings include:
o Developmental delay(s) in physical and reproductive parameters. There may be
an expectation of similar effects in reproductive and DNT studies, especially if
the studies include comparable doses. As described in section 3.1.3,
developmental delays may alter behavioral endpoints, especially when testing at
an early age. Pup body weight may be used to infer developmental delay. Many
developmental landmarks (e.g. pinna detachment, incisor eruption, eye opening)
are highly correlated with body weight, and a delay in physical development often
correlates with later appearance of these landmarks. Delayed onset of puberty
Considerations and Integration
E - 9

-------
(delayed preputial separation or vaginal opening) is also often a reflection of
delayed physical growth;
o Marked or obvious changes in the behavior of the offspring noted as part of
routine clinical observations. Since chemical exposure continues after weaning in
reproductive toxicity studies, signs of neurotoxicity may be due to a direct action
of the chemical on neuronal function, which may differ from a permanent change
in the development of the nervous system itself. The reproductive toxicity study
design does not allow differentiation between these types of causality, but either
type of effect would be of concern.
•	Other toxicity studies. For many chemicals, data from other studies, including those
assessing subchronic neurotoxicity and chronic toxicity, may be available. These studies
are conducted in adult animals and thus would reflect direct chemical effects on the adult
animal rather than effects on the developing nervous system. That said, these data can be
compared to maternal data derived from the DNT study, especially where there is
similarity in dose levels and route of administration. Effective dose ranges across studies
can inform sensitivity of different target organs and toxicity manifestations. Care should
be taken to consider the potential impact of other dosing regimens in any comparisons;
•	DNT studies in the literature. The exposures and experimental procedures used in DNT
studies reported in the peer-reviewed literature differ from those used in the DNT test
guidelines in almost all cases. There may be some consistency in DNT effects of some
chemicals across studies, but for many others there may be considerable differences;
these may result from differences in exposure, test selection and conduct, age at
assessment and many other factors. The evaluator should be cognizant of these
differences when making comparisons between the literature data and the study under
review.
4. Additional considerations for interpretation
Additional issues may be considered in the overall interpretation of DNT data. These include
exposure considerations and kinetic factors, both of which impact levels of the test chemical in
the fetus/offspring. In addition, kinetic information may inform extrapolation to humans.
4.1 Exposure
The reviewer should consider details of the exposure scenario, which can influence chemical
levels in the fetus and offspring:
•	In DNT studies, the dam receives the test chemical and fetal exposure is assumed to occur
through placental transfer. For postnatal exposure, the DNT guidelines specify that there
should be assurance of continued (i.e., lactational) exposure, which requires
demonstration of transfer of the test chemical to the pup through the milk. In ideal
situations, data demonstrating the presence of the chemical in milk or in the pup would be
provided. If data indicate a lack of lactational transfer, this would suggest the need for
direct dosing of the pups after birth. While more challenging, direct dosing of pups has
been used in regulatory DNT studies and can be accomplished without adversely
impacting the study (Moser et al., 2005);
Considerations and Integration
E- 10

-------
•	Dietary exposure (food, water) is considered to be a relevant dose route for potential
human exposure, and it is often used in DNT studies. However, food and water intake of
the dam changes over the study, and greatly increases during lactation. Thus, exposure at
fixed concentrations (i.e., ppm in diet or water), will result in much higher chemical
intake by the dam on a mg/kg bw basis during lactation (Yoon and Barton, 2008). This
higher intake could result in additional maternal or pup toxicity during lactation. For this
reason, changes in chemical intake over time should be noted. Since water and/or food
intake is often measured during the study, actual dose should be calculated and reported.
On the other hand, gavage dosing (to the dam or pup) generally provides a consistent
amount of compound on the basis of body weight;
•	It should be noted that by the second to third postnatal week (towards the end of the
lactation period), pups start consuming water and feed in addition to their nursing
activity. If the test chemical is administered in the diet or in drinking water, it is likely
that chemical intake increases for some of this time due to these multiple routes of
exposure (lactational plus direct oral);
•	Other relevant exposure routes may include inhalation or dermal, and review of such
studies should consider the additional complexities of these exposures. For example,
daily inhalation and dermal exposures may be stressful to the pregnant rat, and stress
alone can produce adverse outcomes. For inhalation exposure, there are two options for
continued exposure after birth, both of which have shortcomings: 1) exposure of the dam
only, but this requires separation from the pups for many hours a day, limiting their
maternal care and nourishment, or 2) exposure of the entire litter, but this may lead to
inconsistent uptake due to the pups nesting and burrowing, preventing access to the
vapor. For dermal exposure, direct dosing of the pups should take into account their
immature skin and lack of hair after birth.
Kinetic parameters are important in determining whether there are differences in systemic
exposure in the young animal compared to adults. Some physiological changes underlying
absorption, distribution, metabolism, and elimination (ADME) may take days to weeks to reach
adult capacities. However, there is often a lack of age-specific kinetic information and therefore
one must rely on some generalizations:
•	Metabolic capabilities are often low in the very young, and the ontogeny of different
metabolic pathways occurs at different rates across species (Saghir et al., 2012). The
resulting impact may vary depending on the development of the specific systems through
which the test chemical is metabolized. This can result in higher tissue levels in the pup,
but it may also result in lower levels of the toxic moiety if it is the metabolite, not the
parent compound, that is active;
•	Absorption, distribution, and elimination of the test chemical may differ in the young
compared to the adult, leading to greater or lower blood levels. Furthermore, the blood-
brain barrier and transporter proteins change during development, which may influence
uptake into the brain. The impact of these differences in terms of higher or lower tissue
levels may depend on specific characteristics of the test chemical (e.g., lipophilicity,
molecular size).
Considerations and Integration
E- 11

-------
5. Human Relevance
There are specific processes and/or critical developmental periods in nervous system
development, during which chemical-induced disruptions may result in long-lasting adverse
outcomes. Awareness of differences in brain development and function across species can help
inform extrapolation of rodent DNT data to human infants and children.
Nervous system development across species:
•	Nervous system development begins in early gestation and continues through
adolescence in both rodents and humans (reviewed in Bayer et al., 1993; Rice and
Barone, 2000). As the underlying biological processes are similar across species,
chemical effects targeting those processes may be comparable. Proliferation and
migration of neural and glial cells occur in waves across brain regions. Neural
connections are established through differentiation and synaptogenesis, and are
influenced by a number of molecular signals. Myelination processes are most protracted,
and a period of rapid glial proliferation (known as the brain growth spurt) follows. This
occurs postnatally in rodents (Dobbing and Sands, 1979; Dobbing and Sands, 1973) as
opposed to other mammals (including humans) where this occurs in utero. These basic
developmental processes occur in a specific, tightly controlled sequence that differs
across brain regions and cell types. While the general processes and sequence in brain
development is conserved across species, the rate and timing of specific processes
relative to parturition differ between species. The timing of neurogenesis can be matched
across species (Clancy et al., 2007); a web-based model providing temporal comparisons
across a number of mammals is available at http://translatingtime.net/;
•	Consideration of temporal differences in timing of brain development may provide
insight into identifying potential targets for neurodevelopmental effects. However, it
should be noted that in guideline DNT studies, exposure occurs throughout the pre-
weaning developmental period and effects cannot be attributed to any specific exposure
windows (i.e., gestational versus lactational exposure). Since late gestational
development in humans corresponds to early postnatal life in rodents, relative exposures
of the human fetus and the postnatal pup can become an important issue. For example,
adequate postnatal exposure in the newborn rat is needed to evaluate effects that could
occur from late-gestational exposure in the human fetus.
Nervous system function across species:
•	Nervous system reflexes and responses are highly conserved across species, allowing
direct extrapolation of specific behaviors (e.g., startle reflex); however, higher order
functions (e.g., learning, memory and attention) are less well understood and defined, and
relevant testing procedures may not be entirely homologous between rodents and
humans. Behavioral tests of cognition for rodents and humans may not be specifically
analogous due to species differences in behaviors, motor abilities, and motivation;
•	The issue of comparability of human and rodent cognitive measures in studies of
pharmacology and neurotoxicology has generated a large body of literature, much of
which is relevant to the evaluation of DNT data. A few reviews of possible interest are
found in D'Mello and Steckler (1996), Sarter (2004), and Sharbaugh et al. (2003).
Considerations and Integration
E- 12

-------
6. Conclusions
The DNT study was designed as a screening tool to investigate a wide range of possible
developmental neurotoxic effects, and it should not be considered a diagnostic tool to address
specific human neurological conditions. Data from the rat DNT study are expected to be
predictive of adverse outcomes in human infants and children, given similarities in nervous
system development. Interpretation of DNT data, however, can be challenging and requires
close evaluation and consolidation of the results from testing in offspring and maternal animals
as well as other available studies. With attention to details and understanding of the test
measures, informed decisions can be made regarding the potential for a test chemical to produce
developmental neurotoxicity.
In closing, key points for consideration include:
•	Assessments should take into account biological plausibility, consistency or pattern of
effects, dose response, and systemic toxicity in the offspring. A higher level of
confidence in the conclusions is attained in cases of consistent patterns of effect,
appropriate experimental methodologies and study conduct, and accurate statistical
analyses;
•	Sufficient data are not usually available during the DNT evaluation process to allow full
causal inferences to be made with respect to human disease conditions, which require
availability of a number of experimental and possibly epidemiological studies;
•	It cannot be assumed that toxicity in the offspring is due to maternal toxicity. While clear
or severe maternal toxicity may result in toxicity to the pups, or may influence
neurological outcomes, this is not always the case. Furthermore, the designs of most
standard DNT studies do not provide data to address whether the effects are attributable
to maternal toxicity or direct neurotoxicity on the offspring;
•	Evaluations should take into account available information on kinetic parameters of the
test chemical as well as other available toxicological information, to the extent that this
other information is available.
Considerations and Integration
E- 13

-------
7. References
Bayer SA, Altman J, Russo RJ, Zhang X. (1993). Timetables of neurogenesis in the human brain
based on experimentally determined patterns in the rat. Neurotoxicology 14:83-144.
Clancy B, Finlay BL, Darlington RB, Anand KJS: Extrapolation brain development from
experimental species to humans. Neurotoxicology 28:931-937, 2007.
D'Mello GD and Steckler T. (1996) Animal models in cognitive behavioural pharmacology: an
overview. Cog. Brain Res. 3:345-352.
Dobbing J, Sands J. (1973). Quantitative growth and development of human brain. Archives of
Disease in Childhood, 48, 757-767.
Dobbing J, Sands J. (1979) Comparative aspects of the brain growth spurt. Early Human
Develop. 3:79-83.
Fernandez SF, Menendez MF, Fernandez BM, Patterson AM. (1985). Malnutrition in utero and
during lactation in the rat: Relationship of dams weight gain and development of
suckling. Nutrition Res. 5:413-421.
Holson RR, Freshwater L, Maurissen JP, Moser VC, Phang W. (2008) Statistical issues and
techniques appropriate for developmental neurotoxicity testing: a report from the ILSI
Research Foundation/Risk Science Institute expert working group on
neurodevelopmental endpoints. Neurotoxicol. Teratol. 30, 326-348.
Moser VC, Walls I, Zoetis T. (2005). Principles for direct dosing of pre-weaning laboratory
animals in toxicity testing and research. Int. J. Toxicol. 24:87-94.
Organisation for Economic Co-operation and Development. (2004). Draft Guidance Document
on Reproductive Toxicity testing and Assessment. OECD Series on Testing and
Assessment No. 43.
OECD/OCDE 426. (2007) OECD Guideline for the Testing of Chemicals, Developmental
Neurotoxicity Study.
Rice D, Barone S Jr. (2000) Critical periods of vulnerability for the developing nervous system:
Evidence from humans and animal models. Environ. Health Perspect. 108 (suppl 3),
511-533.
Saghir, SA, Khan, SA, McCoy, AT. (2012) Ontogeny of mammalian metabolizing enzymes in
humans and animals used in toxicological studies. Crit Rev. Toxicol. 42:323-357.
Sarter M. (2004). Animal cognition: defining the issues. Neurosci. Biobehav. Rev. 28:645-650.
Sellers RS, Morton D, Michael B, Roome N, Johnson JK, Yano BL, Perry R, Schafer K.
(2007).
Society of Toxicologic Pathology position paper: Organ weight recommendations for toxicology
studies. Toxicol. Path. 35:751-755.
Sharbaugh C, Viet SM, Fraser A, McMaster SB. (2003). Comparable measures of cognitive
function in human infants and laboratory animals to identify environmental health risks to
children. Environ Health Perspect. Ill: 1630-1639.
Tyl RW, Crofton K, Moretto A, Moser VC, Sheets LP, Sobotka TJ. (2008) Identification and
interpretation of developmental neurotoxicity effects: A report from the ILSI Research
Foundation/Risk Science Institute expert working group on neurodevelopmental endpoints.
Neurotoxicol. Teratol. 30, 349-381.
US Environmental Protection Agency. (1998) Guidelines for Neurotoxicity Risk Assessment.
EPA/630/R-95/001F.
Vandenberg LN. (2014) Low-dose effects of hormones and endocrine disruptors. Vitam. Horm.
94:129-165.
Considerations and Integration
E- 14

-------
Yoon M, Barton HA. (2008) Predicting maternal rat and pup exposures: how different are they?
Toxicol. Sci 102:15-32.
Considerations and Integration
E- 15

-------