OVERVIEW OF METHODS TOR EVALCATIMC EFFECTS OP
PESTICIDES ON REPRODUCTION .TN BIRDS
Richard S.
Environment A! Reeeareh Laboratory
200 Stf 35th Str««t
CorvalJ.i*, Oregon 97333
ar-d
Lisb M-
Mantech Envircn«ental Technolcyy, Inc.
USEPA Environci'srita L Rese&rch Laboratory
noo SW 35th Street
Corval'.is, --regcn 97333
3IVI20NM5NTAL RESEARCH LArCKATORY-CORVALX.IS
OFFICE O? RES3.MvCH WiD C'SVELOPr-IENT
U. S. EtfVIRQlWZITTAL PROT£C'PIOi? AGENCY
S, OREGON 97333
-------
OVERVIEW OF METHODS FOR EVALUATING EFFECTS OF
PESTICIDES ON REPRODUCTION IN BIRDS
by
Richard S. Bennett
USEPA Environmental Research Laboratory
200 SW 35th Street
Corvallis, Oregon 97333
and
Lisa M. Ganio
Mantech Environmental Technology, Inc.
USEPA Environmental Research Laboratory
200 SW 35th Street
Corvallis, Oregon 97333
ENVIRONMENTAL RESEARCH LABORATORY-CORVALLIS
OFFICE OF RESEARCH AND DEVELOPMENT
U. S. ENVIRONMENTAL PROTECTION AGENCY
CORVALLIS, OREGON 97333
-------
FOREWORD
This report provides an overview of methods for laboratory
evaluations of the effects of pesticides on avian reproduction,
including a review of current guidelines for conducting an avian
reproduction test and possible alternative methods for improving
hazard assessment. Field methods for assessing reproductive
effects are not covered in this report. This report does not
establish Agency policy for conducting avian reproduction tests.
The methods discussed in this report that are at variance with
the existing guidelines should not be taken as establishing new
policy or superseding existing policy, however, they are
presented for consideration in future policy decisions.
Guidelines for the avian reproduction test are scheduled for
revision in the near future.
This report is intended to address several issues raised by
persons outside and within the Agency concerning the adequacy and
utility of the current avian reproduction test. The primary
audience for the report is persons involved with conducting and
evaluating avian reproduction tests, but it is hoped that the
discussion of these issues will lead to further research by
persons interested in the effects of chemicals on avian
populations.
ii
-------
ABSTRACT
The standard test methods used by the U. S Environmental
Protection Agency for evaluating the effects of pesticides on
avian reproduction are critiqued. The intent of the report is to
review several concerns that have been raised about the adequacy
of the test methods, discuss ways to improve the regulatory
utility of the test, and present alternative methods that discuss
aspects of avian reproduction that are not adressed by current
test methods. The overview of current test methods includes the
selection of measurement variables, experimental design,
selection of test animals, testing environment and husbandry
methods, egg collection and incubation, observations of progeny,
and data analysis and interpretation. The emphasis is on
reducing variability in test data that is unrelated to pesticide
treatment to improve the ability of the test to detect pesticide
effects if they exist. Alternative test methods are discussed
that determine a dose-response relationship of pesticide effects,
employ a shorter exposure period for less persistent pesticides,
or utilize parental incubation to determine pesticide-induced
behavioral anomalies. Alternative measurement endpoints and data
analysis methods also are presented.
iii
-------
ACKNOWLEDGMENTS
The authors wish to thank all those individuals that
reviewed and commented on various drafts of this report,
including (in alphabetical order) Jewel Bennett, Brian Collins,
David Farrar, Ed Fite, Mary Frankenberry, Gary Heinz, Ron
Kendall, Brad Marden, Dennis McLane, Pierre Mineau, Al Nebeker,
and Robert Ringer. Their excellent comments and recommendations
were extremely helpful in developing this report, although some
of the recommendations will have to be the subject of future
reports. We owe a special thanks to Jewel Bennett, whose earlier
collaboration on several research projects provided the catalyst
for this effort, and Anne Fairbrother, whose continuing support
made this report possible. Also, we owe special thanks to Anne
Barton and James Akerman for providing the senior author with the
opportunity for a rotational assignment in the Office of
Pesticide Programs to experience the regulatory process. The
advice and example set by James Akerman will not be forgotten.
IV
-------
TABLE OF CONTENTS
Section Page
FOREWORD ii
ABSTRACT iii
ACKNOWLEDGMENTS iv
1. INTRODUCTION 1
2. OVERVIEW OF CURRENT AVIAN REPRODUCTION TEST METHODS 6
Criteria for Determining When Test is Required .... 6
Current Criteria 6
Potential Reproductive Effects Not Covered by
Criteria 7
Procedures for Conducting Avian Reproduction Tests 7
Selection of Measurement Endpoints 7
Experimental Design 11
Selection of Test Animals 23
Selection of Testing Environment and Husbandry 25
Egg Collection, Storage and Incubation 27
Observations of Progeny 29
Data Analysis and Interpretation 30
3. ALTERNATIVE APPROACHES FOR EVALUATING REPRODUCTIVE
EFFECTS 40
Alternative Test Protocols 41
Determination of Dose Response Relationship .. 41
Test for Effects from Short Term Exposure .... 44
Parental Incubation 46
Alternative Endpoints 50
Eggshell Strength 50
Plasma Calcium Concentrations in Females 53
Parental Organ Size and Weight 54
Alternative Data Analysis Methods 54
4. CONCLUSIONS 60
5. RECOMMENDATIONS 63
REFERENCES 66
APPENDICES 79
APPENDIX A. Subdivision E Guidelines for Conducting an
Avian Reproduction Test 79
APPENDIX B. Selected Methods for Experimental Design
and Analysis 90
APPENDIX C. Potential Effects of Cracked Eggs on the
Determination of Pesticide-related Effects 101
-------
SECTION 1
INTRODUCTION
The Office of Pesticide Programs (OPP) of the U. S.
Environmental Protection Agency (EPA) is charged under the
Federal Insecticide, Fungicide and Rodenticide Act (FIFRA) with
determining whether a pesticide can be registered for a
particular use. Under FIFRA, the EPA Administrator shall
register a pesticide if it is determined that "when used in
accordance with widespread and commonly recognized practice it
will not generally cause unreasonable adverse effects on the
environment" (P.L.95396, Sec 3 (c)(5)(D)). Additionally, FIFRA
states that "the Administrator may conditionally amend the
registration of such pesticide....if the Administrator determines
that ...amending the registration....would not significantly
increase the risk of any unreasonable adverse effects on the
environment" (ibid. Sec. 3(c)(7)(B)). The determination of
whether adverse effects are "unreasonable" is a risk management
function that requires the integration of the benefits to society
of the use of pesticides with the risks posed to human health and
the environment.
The extent of ecological risk from a proposed use of a
pesticide is estimated by developing an ecological risk
assessment, which takes into account laboratory data, and in some
instances field data, on the toxicity and effects of the
pesticide (hazard) and the potential for exposure to nontarget
organisms in the environment. OPP has developed a Standard
Evaluation Procedure detailing the state-of-the-art for
ecological risk assessment (USEPA 1986). For consistency and
comparability of hazard data generated for the development of an
ecological risk assessment, OPP has established Pesticide
-------
Assessment Guidelines for conducting laboratory and field testing
of pesticides in support of registration. The evaluation of
pesticide hazards to nontarget wildlife and aquatic organisms are
stated in Subdivision E of these guidelines (USEPA 1982a) (See
Appendix A.). This report will focus on the avian reproduction
test, section 71-4 of the Subdivision E guidelines, a laboratory
test in the avian and mammalian testing series.
The avian reproduction test was first developed to evaluate
the reproductive effects of the persistent organochlorine
pesticides. These pesticides exist chronically in the
environment and have been found to produce effects on avian
reproduction, principally by effects on eggshell thickness
(Anderson and Hickey 1972, Ratcliffe 1970) and embryonic
development (Haegele and Hudson 1973, Heath et al. 1969, Longcore
and Samson 1973). Avian reproduction tests were also used to
evaluate potential effects of other persistent chemicals, such as
polychlorinated biphenyls (Dahlgren and Linder 1971, Peakall et
al. 1972) and methyl mercury (Heinz 1974, 1976a, Peakall and
Lincer 1972).
As stated in Subdivision E, the avian reproduction test is
required "when basic data and environmental conditions suggest
possible problems" (USEPA 1982a). The stated purpose for the
test is that "these tests are used principally:
-To estimate the potential for chronic impacts, taking into
account the measured or estimated residues in the
environment; and
-To determine if additional field or laboratory data are
necessary to further evaluate hazards" (USEPA 1982a).
The test simulates a chronic dietary exposure to the
2
-------
pesticide with a goal of determining the highest dietary level
(concentration) that produces no observable adverse effect
(NOAEL) on a suite of reproduction parameters and the lowest
dietary level that produces an observable adverse effect (LOAEL).
The results of this test are then evaluated to determine if
observed adverse effects occur at concentrations expected to
exist in the environment from the proposed uses of the pesticide.
If the ecological risk assessment based on these test results
indicates that a proposed use would pose an unreasonable adverse
effect, the registration may be referred to special review for
risk/benefits analysis or additional laboratory or field testing
may be required to substantiate or rebut the presumption of risk.
Currently, the avian reproduction test is the only
standardized test in Subdivision E that simulates a chronic
dietary exposure, and thus represents the only method for
detecting the long-term effects of chemicals that are either
persistent in the environment or that are applied to the
environment repeatedly or continuously. In addition to measuring
chronic toxicity and reproductive impairment, the chronic dietary
exposure presents a method for evaluating the potential for
pesticide bioaccumulation in organism tissues and deposition into
eggs. Additionally, the avian reproduction test is the only
standardized laboratory test in Subdivision E that focuses on
parameters other than mortality.
Although the avian reproduction test provides information on
pesticide toxicity not found in any other laboratory test, there
are several concerns with the current test and the associated
data analysis. First, the test data generated are often highly
variable so that biologically significant effects on reproduction
may not be statistically detectable. Second, guidance for
carrying out statistically sound analyses of test data is lacking
in the current protocols. Third, chronic dietary exposure
-------
methods may not provide the most representative exposure scenario
for detecting effects on reproduction from newer, less persistent
pesticides. Effects on avian reproduction do not result only
from chronic exposure, but can occur after short-term pesticide
exposures. Fourth, because reproducing birds may come in contact
with pesticides for the first time during any phase of the
reproductive period, including incubation and brood-rearing
periods, many potential effects on reproductive output may not be
evaluated by the current test guidelines. Fifth, standardized
field methods for evaluating pesticide effects on avian
reproduction are limited and their relationship to the laboratory
avian reproduction test are not well understood. Therefore, it
is difficult to conduct field tests that adequately evaluate the
risks predicted by laboratory tests. The purpose of this report
is to provide an overview of current test methods, evaluate the
bases for the first four concerns listed above, and suggest
alternative methods for addressing these concerns. This report
will not address field methods or the comparability of laboratory
and field methods. However, this report will discuss some
aspects of how the avian reproduction test relates to what is
known about pesticide effects on the reproduction of wild birds.
To achieve the goal set for the future of OPP's ecological
risk assessment process (USEPA 1986, p. 1) of broadening
assessments to the population, community and ecosystem levels of
concern, a broader evaluation of direct and indirect effects of
pesticide use is required, including nonlethal effects on
reproductive potential. The dynamics of a defined wildlife
population are governed by the three principle population
parameters—birth, death, and movements in and out of the defined
population. Any effect of the use of a pesticide that changes
the birth rate per female of reproductive age is considered a
reproductive effect. Effects on the birth rate per female may
occur as a change in the number of eggs per clutch, number of
-------
clutches per year, rates of egg fertility, embryo survival, or
hatch success, or as mortality of reproducing adults.
Additionally, mortality of hatchlings before recruitment into the
population is often measured as a reproductive effect. The
deaths of pre-recruitment young may result from direct toxicity
from embryonic or post-hatching exposure to the pesticide or
pesticide-related effects to the parent. Ideally, an ecological
risk assessment evaluates if, and to what extent, the proposed
use of a pesticide would adversely affect the number of young
produced per female of reproductive age and their recruitment to
the population.
The avian reproduction test does not and can not provide all
this information for an ecological risk assessment. It is the
first line screening test for identifying potential reproductive
effects to birds. As such, it utilizes a suite of measurement
endpoints to evaluate the potential effects of pesticides on
several components of avian reproduction and to help identify
cause and effect relationships. In practice, the avian
reproduction test is often the first step in a multiple-step
process of assessing the risk to avian reproduction from the
proposed use of a pesticide. The focus of this report is to
discuss methods for improving the avian reproduction test as a
screening tool for ecological risk assessment.
-------
SECTION 2
OVERVIEW OF CURRENT AVIAN REPRODUCTION TEST METHODS
CRITERIA FOR DETERMINING WHEN TEST IS REQUIRED
Current Criteria
The Subdivision E guidelines (USEPA 1982a) state that "data
on avian reproduction are required by 40 CFR 158.145 to support
the registration of an end-use product which meets one or more of
the following criteria:
1) Its labeling contains directions for using the product
under conditions where birds may be subject to repeated or
continuous exposure to the pesticide or any of its major
metabolites or degradation products, especially preceding or
during the breeding season.
2) The pesticide or any of its major metabolites or
degradation products are stable in the environment to the
extent that potentially toxic amounts may persist in avian
feed.
3) The pesticide or any of its major metabolites or
degradation products is stored or accumulated in plant or
animal tissues, as indicated by the partition coefficient of
lipophilic pesticides (tests 165-3, -4, and -5 of
Subdivision N)(USEPA 1982c), metabolic release and retention
studies (test 83-4 of Subdivision F)(USEPA 1982b), or as
indicated by structural similarity to known bioaccumulative
chemicals.
-------
4) Any other information, such as that derived from
mammalian reproduction studies (test 83-4 of Subdivision
F)(USEPA 1982b), that indicates that reproduction in
terrestrial vertebrates may be adversely affected by the
anticipated use of the pesticide product".
Potential Reproductive Effects Not Covered bv Criteria
The above criteria were developed when the test was
primarily used to determine effects of organochlorine pesticides
and other persistent chemicals and reflect the concern for
pesticides with chronic exposure patterns. The criteria would
not necessarily trigger a test for pesticides that pose risk of
adverse reproductive effects from short-term exposure. Several
pesticides have been shown to reduce egg production within days
after initiation of dietary exposure (Bennett and Bennett 1990,
Bennett et al. 1991). Effects on eggshell quality (Bennett and
Bennett 1990, Haegele and Tucker 1974) and incubation and brood
rearing behavior (Bennett et al. 1991, Brewer et al. 1988, Busby
et al. 1990) have also resulted from short-term pesticide
exposures.
PROCEDURES FOR CONDUCTING AVIAN REPRODUCTION TESTS
Selection of Measurement Endpoints
The Subdivision E guidelines (USEPA 1982a) provide an
extensive list of data requirements to be reported for complying
with the test standards. Much of this information relates to the
general test conditions (see section 70-3, General test
standards), with specific information on test methods and the
test substance and organism required. Additional information on
the test conditions is specified in section 71-4, Avian
-------
Reproduction Test. Information that defines the conditions for
the entire test includes: ambient temperature and humidity,
photoperiod and lighting intensity, description of the test diet
(including composition and proximate analysis), source of food
and water supply, pretest and test history of medical and
chemical administration, dimensions and materials of test pens,
temperature and humidity of egg incubators, and egg turning
frequency. These measurements are critical for evaluating the
appropriateness of test conditions and for comparing results
between tests.
Other test standards specified in section 71-4 represent
measurements made on each experimental unit (i.e., pen) in the
test to be analyzed for pesticide treatment effects. Several of
the measurements are not discussed in the section on statistical
analysis and are often used as qualitative indicators. However,
each of these measurements could, and sometimes are, quantified
and analyzed statistically. These include signs of abnormal
behaviors, signs of intoxication in hatchlings, observed
morphological and physiological responses of adults, observations
on the palatability or repellency of test diets, and postmortem
necropsy findings. These measurements are often critical for
explaining treatment differences in the quantitative
measurements.
Measurements endpoints obtained for each treatment (dietary
concentration group) specifically mentioned for statistical
analysis include:
—number of adult mortalities by sex
—adult body weight by sex
—average adult food consumption per pen per day or season
—average eggshell thickness per pen per day or season
—number of cracked eggs per pen per day or season
8
-------
—proportion of cracked eggs per eggs laid per pen per
day or season
—number of eggs laid per pen per day or season
—number of fertile eggs laid per pen per day or season
—proportion of fertile eggs per eggs set per pen per day or
season
—number of live 3-week embryos per pen per day or season
—proportion of live 3-week embryos per fertile eggs per pen
per day or season
—number of hatchlings per pen per day or season
—proportion of hatchlings per live 3-week embryos per pen
per day or season
—proportion of hatchlings per eggs laid per pen per season
—number of 14-day-old survivors per pen per day or season
—proportion of 14-day-old survivors per number of
hatchlings per pen per day or season
—proportion of 14-day-old survivors per eggs laid per pen
per day or season
—average weight of hatchlings per pen per season
—average weight of 14-day-old survivors per pen per season
Most of the above measurements are based on the entire test
period. Body weights are recommended at the initiation and
termination of the test and at biweekly intervals until egg
production begins. Measuring body weight during egg production
is discouraged because it could have adverse effects on egg
production. Hughes and Black (1976) found that handling laying
hens adversely affected eggshell quality by increasing the
incidence of cracks and equatorial bulges. Also, handling laying
hens can lead to broken eggs in the oviduct. The average
eggshell thickness per pen is calculated by collecting and
measuring all eggs laid on one day every two weeks. The
Subdivision E guidelines recommend that, for consistency, eggs be
collected during weeks 1, 3, 5, 7 and 9 of the egg laying period.
-------
The Subdivision E guidelines recommend that food consumption
be recorded at least biweekly throughout the test. With
mallards, it may be necessary to record consumption and change
feed more frequently because they tend to put water in their
food, which may become moldy. Fungi in the genus Aspercrillus
thrive on moist grain and feeds and can cause respiratory
infections and death (Friend 1987). Also, it is more difficult
to separate food consumption from spillage with mallards if the
spilled food becomes wet so that the weight is difficult to
estimate. Unless measures are taken to quantify spillage, the
food consumption measurements represent a combination of food
consumed and spilled, which compromises the detection of treated-
related changes in consumption if birds vary greatly in the
amount of food spilled.
The Subdivision E guidelines do not discuss the relative
importance of each of the listed endpoints. A variety of
endpoints are measured so that the most sensitive endpoints for
each tested pesticide can be determined. Each endpoint may be an
important indicator of potential effects on avian reproduction.
This provides valuable information for determining the specific
mechanism of action of the pesticide. However, of the above list
of measurement endpoints, the number of 14-day-old survivors per
pen may be the most biologically meaningful and comprehensive at
evaluating overall effects of a pesticide on avian reproduction.
The number of 14-day-old survivors per pen is a function of the
number of eggs laid, the proportion of fertile eggs of those set,
the proportion of live embryos of those fertile, the proportion
of hatchlings of eggs with live embryos, and the proportion of
hatchlings that survive to 14 days of age. Significant effects
on these endpoints also may be observed as a significant effect
on the number of 14-day-old survivors per pen if the power of the
test is sufficient and there is no compensatory effect in another
10
-------
endpoint.
Experimental Design
The goal of an avian reproduction test should be to detect
pesticide-related effects on the reproduction of the test
population at the lowest dietary concentrations that produce
biologically significant effects. This can be accomplished by
choosing the appropriate measurement endpoints and developing a
sufficiently powerful experimental design that will control
variability unrelated to treatment in the test system by reducing
influence of confounding variables. Failure to address these
aspects of the test can lead to a test that fails to identify
effects or can detect only very large effects, thus failing to
detect effects at lower concentrations that may be biologically
significant.
However, circumstances beyond the scientist's current
knowledge or control often make it difficult to ensure that all
avian reproduction studies are sufficiently powerful. Some
general guidelines will be discussed in this report.
An experimental design is defined to be the choice of number
and type of treatments (in this case, diets amended with
pesticide at various concentrations) and experimental units (in
this case, pens of reproductive birds) and the process of
assigning the treatments to experimental units. Note that
because the amended diet is provided to a pen and not to single
birds, the pen is the experimental unit. A treatment group
consists of several replicate experimental units that are
administered the same dietary treatment. In a broad sense, other
techniques necessary for the development and maintenance of the
study, such as animal husbandry and the choice of birds, can also
11
-------
be thought of as aspects of the design of the study.
The Subdivision E guidelines list the minimum number of
treatments (with one serving as a control group), replicates per
treatment, and numbers of males (M) and females (F) per pen for
an acceptable protocol for two avian species, the northern
bobwhite (Colinus viroinianusl, hereafter referred to as
bobwhite, and mallards (Anas platyrhynchos). These are:
Species Treatments No. of birds Replicates
Bobwhite 3 1M & 2F 12
3 1M & IF >12
Mallard 3 2M & 5F 5
3 1M & IF >12
Although pairs are acceptable, these guidelines recommend the use
of the group-pen (more than one male and one female per pen)
design, based on an unreferenced analysis showing greater
sensitivity to detect pesticide effects using the group-pen
design. However, many investigators prefer the use of pairs for
a variety of reasons, including the fact that both species are
monogamous. Given the amount of data collected since the
Agency's original analysis, it would be worthwhile again to
compare the sensitivity of these two approaches for housing test
animals.
The Subdivision E guidelines also provide an example of an
avian reproduction protocol that is considered an acceptable
design for both bobwhites and mallards. While the protocol
states minimum values for the number of treatments and replicate
pens per treatment and guidance on the selection of appropriate
dietary concentrations, it is the responsibility of the
investigators to determine an adequate experimental design and
justification for concentrations chosen for the particular
12
-------
pesticide tested.
Selecting dietary concentrations—
The guidance in Subdivision E for selecting dietary
concentrations states that they "should be based on measured or
calculated residues expected in the diet from the proposed use
pattern(s). The concentrations should include an actual or
expected field residue exposure level and multiple level such as
five" (USEPA 1982a). One of the concerns with the current test
guidelines is that if no significant differences are detected in
a test, it is very difficult to determine if this reflects that
the concentrations tested are truly below those causing
reproductive effects or if the test was inadequately designed to
detect effects that occur. In such a case, the NOAEL is the
highest concentration and the LOAEL is not determined. If a test
fails to detect a reproductive effect, little is learned about
the types of effects potentially caused by the pesticide at
higher concentrations or about how close the tested
concentrations are to those causing effects. Conversely, if
reproductive effects are observed in all treatment groups, it may
not be possible to determine if the same or different effects
would be observed at lower concentrations. If at a later date, a
new use for the pesticide is proposed that would result in higher
or lower environmental concentrations, the test results may be of
little value in assessing the potential risk of the new use, and
the reproduction test would have to be repeated at different
concentrations.
An alternative approach to selecting dietary concentrations
on the basis of estimated environmental concentrations (EEC) is
to conduct the test using concentrations in the range that
produces effects in the more sensitive endpoints, unless the
13
-------
dietary concentrations required would be much greater than the
EEC (e.g., 100 x EEC). American Society for Testing and
Materials (ASTM 1990) discusses this approach as one option for
choosing dietary concentrations. The test would be conducted
with at least one treatment group producing significant effects
on reproductive parameters and one occurring below effect
concentrations. The advantages to this approach are: 1) the
test could more accurately define the NOAEL and LOAEL for making
a regulatory decision; 2) the relative distance between effect
concentrations and the EEC for a proposed use could be
established; and 3) the types of reproductive effects caused by
a pesticide would be identified. The disadvantages of this
approach are: 1) a range finding test or tests would be required
to establish appropriate dietary concentrations for a definitive
test; and 2) the determination of tissue and egg residues may be
less meaningful if the test was conducted at concentrations
considerably different from the EEC.
Length of test—
The Subdivision E guidelines state that birds should receive
the treated diet ad libitum for the duration of the study, which
includes "at least 10 weeks prior to the onset of egg laying" and
an egg laying period of approximately 10 weeks, although the
total duration of the test will be dependent on the onset of egg
production. Most of the measurement endpoints are based on the
calculated value for the entire egg laying season for each pen.
One confounding issue in this approach that may be unrelated
to treatment is that some birds do not consistently lay eggs or
go completely out of egg production before the end of a 10-week
laying period. As the proportion of birds that terminate laying
increases through the test, the variability in parameters, such
as the number of eggs laid, increases among pens within a
14
-------
treatment. For example, all control birds may start out as
consistent egg producers, but late in the laying period, some
control birds will go out of production whereas others will be
consistent producers until the end of the test. As
intratreatment variability increases during the course of the
test, the ability to detect treatment effects will be reduced.
Although the test was originally designed to produce a large
number of eggs per pen so that the estimates for each endpoint
were based on a large sample, the 10 week laying period pushes
the biological limits of some birds for egg production.
ASTM (1990) guidelines suggest an alternative approach that
addresses this situation. It suggests that "it may be
unnecessary to collect more eggs than might be laid in the wild
with two clutches. For the mallard and bobwhite, egg collections
may be terminated when all control pens produce 25 eggs; or 6
weeks after 50% of the control hens have laid one egg". The
advantage to this approach is that setting the egg laying period
at a length that is within the biological limits of the test
species should reduce one source of variability that is unrelated
to the pesticide treatment. However, this approach would reduce
the numbers of eggs upon which the proportional measures are
calculated, which may increase the variability in those
endpoints.
On a related issue, some test substances may delay the onset
of egg production, even though they may not affect the total
number of eggs laid during a laying season. Using the ASTM
criteria for defining a shorter laying period for the test may
increase the probability of detecting treatment differences due
to delayed onset of production. However, greater care must be
exercised in the interpretation of biological significance of
such an effect. Is the delay in onset of laying biologically
significant if reproduction is not otherwise affected?
15
-------
Role of food consumption—
Some pesticides are known to reduce food consumption at the
dietary concentrations that affect reproduction parameters. The
reduction in food consumption alone can affect egg production.
Because energy requirements for egg formation (34 kcal/bird/day)
in bobwhite at 25°C are similar to their basal metabolic
requirements (35 kcal/bird/day)(Case 1972), laying bobwhite
approximately double their food consumption during egg
production. It has been estimated that the daily maximum cost of
mallard egg production is 52 to 70% of daily energy intake at
constant body weight (King 1973). If a pesticide-treated diet
results in reduced food consumption, are observed reductions in
egg production the result of direct pesticide effects on
reproductive mechanisms or a function only of reduced caloric
intake? To address this question, several investigators have
used a parallel "pair-fed" test that pairs a treated pen of birds
with a pen that receives the same amount of untreated food each
day as was consumed by the treated pen. Tests with methamidophos
(Stromborg 1986a), methyl parathion (Bennett and Bennett 1990),
and monocrotophos (Stromborg 1986b) did not produce evidence of/.a
direct pesticide effect other than could be explained by reduced
food consumption alone. Stromborg (1981) found that diazinon
produced an effect on egg production beyond that produced by
reduced food consumption alone. Rattner et al. (1982) found that
parathion reduced bobwhite egg production by directly altering
the secretion of reproductive hormones without a significant
reduction in food consumption.
The mixed results from these tests with organophosphorus
insecticides do not definitively answer the question of the role
of food consumption in explaining reproductive effects, but it is
clear that pesticide-induced reductions in food consumption can
play an important role in affecting egg production. What has not
16
-------
been addressed is how the responses observed in laboratory tests
for pesticides with significant reductions in consumption compare
to bird responses in the field. It is not known to what extent
wild birds may experience the same reductions in food consumption
following an exposure to an agricultural application of pesticide
or respond to the presence of the pesticide by seeking less
contaminated food sources. Some pesticides that are avoided in
laboratory tests can still be acutely poisonous to birds in the
field (Grue et al. 1983). The laboratory test provides a worst
case exposure scenario. Investigators should document the
effects of each pesticide on food consumption to aid in the
interpretation of reproductive effects data.
Randomization—
Once the number of dietary concentrations, number of birds
per pen and number of replicate pens has been decided upon, it is
necessary to assign the individual birds to pens and pens to
treatment groups. The Subdivision E guidelines (section 70-3)
state that "organisms in each test, should as nearly as
practicable, be of uniform weight, size and age. Organisms
should be randomly assigned to test groups." It further states
(section 70-4) that the report should include a "randomization
plan for treatments." However, further guidance for carrying put
a randomization plan is not discussed.
The random allocation of birds to pens and pens (i.e.,
experimental units) to treatments is necessary to remove the
possibility of systematic error in the estimation of treatment
effects (Cox 1958) and will allow for the estimation of
experimental error. Note that because the pens are the
experimental units (and not the birds), pens must be randomly
assigned to treatments. However, it is clear that it is also
important to randomly assign the birds to the pens, and in the
17
-------
process, assure that the pens contain birds of uniform weight,
size and age. It is also desirable to determine the
compatibility of pairs, although this is often not apparent until
after the beginning of the test. The incompatibility of birds
may result in infertile eggs, injuries, or death of one of the
pen mates and is a problem because it increases variability in
measurement endpoints from causes that may be unrelated to
treatment. One possible solution is to have an acclimation
period of sufficient length to identify and remove incompatible
pairs from the test population. This should not be done after
dietary treatment has begun.
One method of randomizing birds is to divide the pool of
test birds into classes of equivalent weight, size and age. Then
randomly assign the birds to pens such that there are equivalent
distributions of birds based on weight, size and age in each pen.
This ensures that a particular treatment group is not dominated
by heavy or older birds.
Note that "randomize" does not mean to choose what appears,
to the researcher, to be a haphazard order. The careful use of a
random order generator such as a table, computer or blind draw is
essential. See Cox (1958) for thorough examples of good and bad
randomization schemes and a discussion of what can go wrong if
randomization is not carried out properly. Further discussion of
randomization can be found in Appendix B.
When the entire pool of pens are randomly allocated to
treatment groups in one generation of random order, the design is
said to be a "completely randomized design (CRD)" (Peterson
1985). This is the type of design that Subdivision E assumes is
being used in its discussion of design and in its prescription of
a one way analysis of variance to analyze the data. There are
however, other randomization plans that may be justified for some
18
-------
circumstances that give rise to a variety of other designs. Two
of the most important are called "randomized block designs (RED)"
and "split plot designs (SPD)". It is important to recognize
when these designs have been used because the design must be
incorporated into the analysis.
Designs other than the CRD can be more efficient and precise
at estimating and detecting treatment effects if they are used in
the proper context. Their use is mandatory in some
circumstances. RBDs are useful when groups of pens are similar
within the group but dissimilar between groups; for example, when
one group of pens is housed in a separate test room from another
group. This design allows the differences between the groups to
be accounted for in the analysis but separated from treatment
effects. Split plot designs are used when a factorial
arrangement of treatment is used; one factor is randomly applied
to the experimental units and then the other factor is randomly
applied within the first factor assignment.
When using alternative experimental designs, the rationale
for why it is more appropriate than the CRD design outlined in
Subdivision E should be stated clearly. Such rationales can be
incorporated into protocols submitted to EPA prior to the study.
Blocking factors and randomization patterns should be well
documented. It is also important to recognize when designs other
than CRDs have been used, because the one-way analysis will not
apply to these cases. The SPD is especially important to note
because it involves the generation of two experimental error
terms and its analysis is non-trivial. Use of a one way analysis
with this design will lead to incorrect inference. Definitions
of RBDs and SPDs and descriptions of their respective analyses
can be found in Peterson (1985).
Power—
19
-------
In designing a reproductive study and in presenting the
results, it is important to know that the study will be able to
detect a meaningful difference between the treatments if a
difference exists. This is the statistical concept of power.
Technically, power is defined to be the probability of detecting
a statistically significant treatment effect in the sample when
such an effect exists in the population. Note that since "power"
is a probability, its calculation requires that the distribution
of the response is known or assumed. Alternatively, given the
power of a study, one can determine the number of pens per sample
necessary to detect treatment differences in the population.
In discussing the major issues involved with the avian
reproduction test, the Subdivision E guidelines state that the
Agency has studied the sensitivity of these avian reproductive
tests and has concluded that with the group design "a
statistically significant reproductive impairment of 20% or more"
can be detected using the recommended number of pens given above.
It goes on to state that "for pairs testing to achieve this
sensitivity, more than 12 replicates are necessary; calculations
have indicated that as many as 25 replicates may be necessary".
Section 70-4 states that the "number of pens needed for a
particular level of sensitivity" be calculated and the reader is
referred to Walpole and Myers (1972) for a method. The ASTM
(1990) standard practice for conducting avian reproduction tests
is that a test should be designed to detect a difference of 25%
with the probability of a Type I error less than 5% in 8 out of
10 such experiments (Power = 0.8). The Office of Pesticide
Programs, whenever possible, follows this guidance when
evaluating test results.
No references are provided to substantiate the claim by the
Agency that 12 replicates are adequate to achieve the desired
power. Differences in variability between studies, treatment
20
-------
groups and endpoints may make the "12 replicate" value an
oversimplification. One method of assessing this would be to
examine power curves and sample sizes from a variety of avian
reproduction studies that have already been conducted. Such a
study is currently being carried out at the National Wildlife
Research Centre of Environment Canada (Pierre Mineau, Wildlife
Toxicology Division, pers. comm.). The results of such a study
would be extremely valuable in helping the Agency update its
expectations of power and detectable differences. It should also
be noted that the required number of replicates are theoretically
adequate only when the one way ANOVA is the appropriate analysis
to use.
Two areas of concern regarding the calculation of power and
number of replicates are: 1) the power depends on the
variability of the response under consideration and many
responses are measured in an avian reproduction study; and 2)
these calculations depend on the assumption that the data are
obtained from a normal distribution of responses.
Since many reproductive endpoints are analyzed using ANOVA
and each endpoint has a different level of variability and the
power calculations depend on endpoint variability, the power that
is attained for one endpoint will be different from the power for
another endpoint. Thus for a single avian reproduction study
some endpoints will be more or less useful, in a statistical
sense, for detecting significant effects. When designing the
study, the most variable endpoint should be considered and the
sample size chosen to obtain the desired power, based on that
endpoint. Then the proposed design will, at least, achieve the
required power for all endpoints. Alternatively, power could be
calculated based on a selected endpoint. In this case the choice
of endpoint should be well documented and it should be recognized
that the power statements will only apply to endpoints that are,
21
-------
at roost, as variable as it.
A more worrisome problem has to do with the assumptions of
normality and homogeneous variance of treatment groups implicit
in the analysis of variance and the assumption, implied in the
calculation of power, that the difference between treatments is a
shift in the mean value of a normally distributed response. When
the response is not normally distributed or when the effect of
the treatment is more complicated than just a shift in the
average of a normal population, power calculations are not
accurate. For example, if eggshell thickness is normally
distributed in the population of control birds, but the effect of
treatment is not to change the mean but to result in a skewed
distribution of eggshell thickness, then the computation of power
will not be accurate. As discussed in the section on data
analysis, it is important to check that there is reasonable
evidence in the data to support the assumption of normality and
homogeneity of variance for all treatment groups and control.
See Appendix B for methods of checking these assumptions. Note
also, that when a transformation of the response is used to
achieve the assumptions of homogeneous variance and normality,
then the power calculations apply to, and must be carried out on,
the transformed scale.
In interpreting the results of the analysis of an avian
reproduction study, attention should be paid to the power and the
value of the difference from control that was detectable for each
endpoint. Although there may be pesticide related effects in a
variety of responses, the power may not have been sufficient for
detecting it in all of them. Statistically significant effects
in some endpoints and not in others may be due to differences in
power. In addition, the biological significance needs to be
considered. For example, if the power was high and a very small
difference from control was detectable and statistically
22
-------
significant in the lowest dietary concentration, it should be
considered whether or not that difference is biologically
significant. Very small differences may not be biologically
meaningful.
Selection of Test Animals
OPP recommends the bobwhite and mallard as the test species
for use in avian reproduction studies. The Subdivision E
guidelines (USEPA 1982a) state the criteria used for selecting
species and the rationale for not including other species at this
time. Briefly, the criteria used were: 1) species should have
demonstrated sensitivity to pesticides as determined from an
extensive data base; 2) species should be ecologically
significant; 3) species should be aesthetically or economically
important; 4) species should be readily available for test
purposes; and 5) species should have characteristics that are
appropriate for this type of test. There is little data
available to demonstrate that laboratory tests with songbirds
would be more useful than tests with mallards and bobwhite for
predicting the reproductive effects of a pesticide to all exposed
bird species. However, given their smaller body size and higher
metabolic rates, songbirds may be more vulnerable to pesticide
exposures during this energetically demanding period. Based on
acute toxicity testing, many songbird species are more sensitive
to pesticide exposures than bobwhite or mallards (Schafer and
Brunton 1979). Much more work is needed to evaluate the relative
sensitivity of birds to pesticide exposures during reproduction
and to develop songbird reproductive test methods, if
appropriate.
The Subdivision E recommendation for test animals used is
"pen-reared birds, previously untreated, approaching their first
23
-------
breeding season, and phenotypically indistinguishable from wild
birds" (USEPA 1982a). Birds may originate from research breeding
colonies or from gamebird vendors. If birds are shipped, they
should be examined for physical injuries, and it is advisable to
have a 2- to 6-week health observation period prior to the start
of the test. The state of health of shipped birds can also be
assessed by necropsy of representative birds at a diagnostic
laboratory at the time of arrival. Bennett and Fairbrother
(unpubl. data) compared the health of mallards from three
commercial game farms and found birds to be generally healthy and
free of parasites and contagions, but all had low levels of
organochlorines and heavy metals in their tissues.
The ASTM (1990) standard practice for avian reproduction
tests makes several additional recommendations that are intended
to reduce sources of variability that are unrelated to treatment
effects. First, birds must come from one source and strain.
Second, the age of birds should be within ±10% of the mean age of
the test population. Third, birds must be rejected for test
purposes if they are deformed, in poor physical condition, or
different in plumage from wild birds.
Although test birds may appear to be "phenotypically
indistinguishable from wild birds," it is extremely difficult to
judge their genotypic comparability to wild birds. One problem
that can arise is that phenotypically indistinguishable adults
may produce young that clearly do not look like wild stock. Gile
and Meyers (1986) found in a study with mallards that 3 out of 24
pairs produced white ducklings, all of which were males. These
white ducklings were considerably larger than normal-color
ducklings. The USEPA Environmental Research Laboratory-Corvallis
has also observed gamefarm-reared bobwhite that produced
nonnormal colored chicks in a reproduction test. This genetic
variation may be a confounding factor that introduces greater
24
-------
uncertainty to the validity of test results, especially if the
atypical young differ in body weight, growth, or survival rates.
Selection of Testing Environment and Husbandry
All birds should be housed in breeding pens of adequate size
for the species and number of birds per pen. Minimum floor space
specifications for quail are 0.023 m2 per bird, and for 0.5 to
1.5 kg chickens are 0.093 m2 (National Institutes of Health
1985). Similar NIH standards do not exist for mallards.
However, all birds must be provided with sufficient head room to
stand erect and wide enough to fully extend the wings. The
Canadian Council on Animal Care (1984) specifies that mallards in
a single pen have a minimum space requirement of 0.33 m2 floor
area with a 35 cm height and small groups should have 0.66 m2
floor space per bird. ASTM (1990) provides a comparative table
of pen sizes and numbers of birds per pen for several species
from the published literature. For other recommendations on
housing and care of laboratory animals, see the "Guide for Care
and Use of Laboratory Animals" (National Institutes of Health
1985). Control of test room temperature and humidity are
desirable and should be recorded throughout the test period. The
recommended environmental conditions are 21°C and 55% relative
humidity (USEPA 1982a). The NIH recommends a room ventilation
rate of 10 to 15 room air exchanges per hour (National Institutes
of Health 1985).
The photoperiod throughout the test is extremely important
to the success of a test. It is recommended in Subdivision E
that birds be maintained under 7 hr light: 17 hr dark each day
during the first eight weeks of the test. The dark period should
not be interrupted during this period. As little as 15 minutes
of light exposure during the dark period can cause increased
25
-------
gonadal development (Kirkpatrick 1955). Bobwhite will become
sexually active if given 60 minutes exposure to light during the
dark period. Consequently, during the first eight weeks of the
test, the dark period of each day must be maintained to keep
birds in nonreproductive condition.
After eight weeks of the short day length, the photoperiod
should be increased to 16 to 17 hours of light per day. This
increase is often accomplished in one step, but sometimes
bobwhite become aggressive after a sudden increase in day length,
causing injuries to pen mates (Bennett et al. 1990b). An
alternative method is to increase day length gradually over a two
week period to reduce aggressiveness. The Subdivision E
guidelines also suggest that the light period can be maintained
either at a constant duration or increased by 15 minutes each
week until the end of the test. Unless there is a documented
benefit to continuously increasing the light period, it may be
better to maintain a constant maximum light period, rather than
adding a potentially confounding variable.
The Subdivision E guidelines state that a lighting intensity
of 6 footcandles at the bird level is adequate. Lighting
intensity seems to be less critical than day length to
reproductive performance. Bobwhite reproductive performance was
similar at 0.1, 1, 10, and 100 footcandles for a 17 hour light
period (Kirkpatrick 1955). The Subdivision E guidelines
recommend that shorter wavelength "cool white" fluorescent lights
that do not emit the daylight spectrum should be avoided.
Food and water should be provided ad libitum throughout the
test. Mallards consume large amounts of water and there are
several ways of providing it. Water can be provided by regularly
filling a static water bowl, using a continuously flowing water
system (Bennett et al. 1991, Heath et al. 1969, Heinz 1974), or
26
-------
using a licking valve (Bennett et al. 1990a). Licking valves
have the advantage of providing clean water on demand while
reducing the flow of water, but birds need to be provided a bowl
of water at least twice a week so they can clear accumulated
matter around the nares.
Egg Collection. Storage and Incubation
It is recommended in the Subdivision E guidelines that eggs
be collected daily, marked with the pen number from which they
were collected, and stored at 16°C and 65% relative humidity.
ASTM (1990) recommends storage at 12 to 16°C. Eggs should be set
in an incubator at weekly intervals. Incubator temperature,
humidity, and egg turning rate are not specified in Subdivision
E. Mallard eggs have been incubated at 37.4 to 37.5°C with
relative humidity ranging from 62 to 80% (Heinz 1976a, 1976b,
Greenwood 1975, Holmes et al. 1978). Prince et al. (1969) found
that the incubation time of mallards eggs decreased as incubation
temperature increased from 35.6 to 39.4°C, with the highest
percent hatch occurring at 37.5°C and relative humidity of 70 to
80%. Flegal and Sheppard (1976) recommend that quail eggs be
incubated at 37.6°C and 60% relative humidity.
The Subdivision E guidelines recommend that eggs laid on one
day every two weeks should be collected for measuring eggshell
thickness. It is recommended that eggs be candled at day 0 of
incubation to check for eggshell cracks, approximately half way
through incubation to determine early embryo mortality, and
approximately three quarters through incubation to measure embryo
survival.
The rate of egg cracking can be sufficiently variable among
tests that it warrants discussion. Eggs can be cracked
27
-------
intentionally or unintentionally by the birds or accidentally by
test personnel or equipment at any point between laying and
hatching. Cracked eggs typically are discarded from further
analysis because of the adverse effect most cracks have on
embryonic development. The Subdivision E guidelines state that
the proportion of eggs laid that are cracked should be
statistically analyzed to determine treatment effects. This
represents one measure of eggshell quality. However, there are
several factors, unrelated to the pesticide treatment, that may
produce confounding influences on the sensitivity of egg cracking
as a measure of eggshell quality. Non-treatment egg cracking can
be-primarily a function of the physical test system, with key
factors being cage materials, slope of cage floor, and access of
birds to eggs. If cracked eggs originate primarily from a small
number of pens within each treatment group, this may be a
reflection of the abnormal reproductive behaviors of some birds
in reaction to the physical test system. Consequently, test
systems with a low rate of background egg cracking may have a
different sensitivity for identifying treatment-related effects
on cracking than test systems with a high rate of background
cracking, especially when the distribution of cracked eggs is
concentrated in a fraction of the pens (See example in Appendix
C). Consequently, the rate of cracked eggs may be a very poor
indicator of treatment effects on eggshell quality.
Pesticide-related changes in eggshell quality are determined
more directly and efficiently by measuring eggshell thickness or
other measures of eggshell quality, rather than using the
indirect measure of cracked eggs. The higher the rate of cracked
eggs, the greater the confounding influence on other measures of
reproductive success. The proportional measures of fertility,
embryonic development, hatchability, and survival will be
affected by reducing the number of eggs on which the proportion
is based. The greatest confounding effect may be in reducing the
28
-------
number of 14-day-old survivors per pen for reasons unrelated to
the test substance. The higher the rate of cracked eggs, the
less meaningful this variable becomes as a measure of treatment
effects. Incidental cracking of eggs is a nuisance variable that
should be minimized to the extent possible. The rate of eggshell
cracking may be more indicative of the quality of the physical
test system than as a indicator of treatment-related eggshell
quality. Rates of egg cracking should be used primarily to
determine if cracking rates in control groups are within
acceptable criteria established by OPP from historical data.
The Subdivision E guidelines recommend that at day 21 of
incubation for bobwhite and day 23 for mallards the eggs are
transferred to a separate nonturning hatcher or hatching
incubator set at 39°C and 70% relative humidity. Flegal and
Sheppard (1976) recommends that hatchers be set at 37.2°C and 85%
relative humidity for bobwhite and pheasant. Stromborg (1986a,
b) used a hatcher temperature of 37.6°C for bobwhite. Heinz
(1974) used a hatcher set at 37.5°C for mallards. The incubator
trays should be partitioned so that hatchlings can be identified
as to parental pen.
Observations of Progeny
The Subdivision E guidelines recommend that hatchlings be
removed from the hatcher at day 24 of incubation for bobwhite and
day 27 for mallards and housed in brooders according to parental
pens for 14 days on a control diet. The rate of hatchability and
hatchling survival to 14 days of age should be recorded for each
pen. Hatchlings are weighed at 14 days of age (or older if
mortality is observed late in the observation period). Since the
analysis of weights of 14-day-old survivors is conducted on the
mean weight for each parental pen, either individual weights for
29
-------
all hatchlings in a season or the average of weighted weekly
means are required to calculate the mean weight properly.
However, individual weights would allow investigators to examine
the distribution of weights for each pen and compare
distributions among treatments.
The method of segregating hatchlings by parental groups as a
means of determining survival rates and body weights for each
parental pen requires several assumptions. First, it has to be
assumed that the rate of survival is equal for hatchlings housed
singly (i.e., only one egg hatched for a pen) compared to those
housed in groups. Second, it must be assumed that there are no
differences in the environmental conditions for housing chicks
among all the parental pens. Third, it must be assumed that the
system will not allow hatchlings from one parental pen to escape
into a section holding hatchlings of another. If these
assumption are not met, the determination of percent survival
will be confounded by variables unrelated to the pesticide
treatment.
An alternative approach is to individually mark hatchlings
(e.g., with numbered wing tags or leg bands), record their
parental pen number, and house hatchlings in common brooders.
This method adds additional requirements to individually mark
hatchlings and assumes that they will not lose their
identification markers, but standardizes group size and
environmental conditions to reduce potentially confounding
variables that could affect hatchling survival.
Data Analysis and Interpretation
• The Subdivision E guidelines (section 71.4) require that
"for continuous variables, experimental groups should be compared
30
-------
to controls by analysis of variance. For most discrete
variables, survival percentages should be computed and arc sine
transformed prior to analysis. Alternatively, a chi square
analysis of survival (contingency table) may be used". In
addition, Section 70.4 states that, in addition to the data at
each treatment level, LD50, LC50, EC50 and 95 percent confidence
intervals be calculated when "sufficient doses and test organisms
are used to establish a dose-response line" and that "no observed
effects levels" and statistical methods used and a reference to
them in published literature be provided for each avian
reproduction test.
To assist in the evaluation of test reports, OPP has
prepared a SAS (1985) program that is currently used to carry out
a one way analysis of variance (ANOVA) on some response variables
and a one way weighted ANOVA on arc sine square root transformed
survival proportions, where the weights are the denominator of
the survival proportions. The current statistical analysis
process involves running this program on data submitted for
review and assessing the statistical significance of the effects
as provided by the program.
The chi-square contingency analysis that the Subdivision E
guidelines also suggest for analyzing discrete response variables
does not seem to be widely used. This may be due to the fact
this analyses can lead to the conclusion that an association
exists but there are no conclusions as to the pattern of that
association. It is not an informative analysis. Generalized
linear models, discussed in Section 3, can provide an analysis
that will describe the patterns of dependency in this type of
data.
31
-------
Use of the Analysis of Variance (ANOVA)—
Because an analysis of variance, or any statistical
analysis, is based on assumptions, the analysis is valid only to
the extent that the assumptions are justified. It is important
that the assumptions be checked because some assumptions may not
be justified for some responses obtained from reproduction tests.
The ANOVA and associated F-tests assume: 1) the data are
sampled from a normal distribution of response such that although
the mean response of the treatment groups may be different; 2)
the variability is assumed to be the same from group to group
(homogeneity of variance); and 3) the effects of the treatment
will be to change the mean response by the addition of a quantity
due to the treatment while the variability will remain unchanged.
This additive quantity is called the treatment effect.
The assumption of additive treatment effects, the third
assumption above, can not always be checked using the data. The
experience of the scientist and the biology of the problem will
often bear on this decision.
The ANOVA is known to be robust to departures from the
assumption of normality (Miller 1986). That is, the p-values
obtained from the ANOVA F-test for treatment effect will provide
the correct conclusion from the data even if the distribution of
the response is somewhat nonnormal. .However, although valid,
these tests may not be the most powerful tests for the nonnormal
distribution. Thus verifying an approximate normal distribution
or using a transformation to help achieve this will ensure that
powerful tests are being employed.
In order to check the first assumption above, it is
important to recognize what types of responses may or may not be
32
-------
normally distributed. First, continuous measurements, which may
take on any positive or negative value can be normally
distributed. Some responses, such as weight which usually cannot
be negative, can be normally distributed if the average value of
the response is not close to 0. Also, some discrete variables
(which can only take on integer values) can be thought of as
having normal distributions if the average count is large, say 25
or more. This is the reason that counts of numbers of eggs laid
per pen can be analyzed using ANOVA, if the number of eggs laid
per pen is large.
Generally, ratios, proportions, small counts of individuals
or responses that differ by orders of magnitudes, will not be
normally distributed. In addition, these types of responses will
have variances associated with them that depend on the mean value
of the response and thus the ANOVA assumption of homogeneous
variance is almost always violated for these types of responses.
However, there can be continuous variables which do not have
normal distributions and proportions, counts or ratios which do.
It is important to evaluate the shape of the distribution and the
extent to which variability is the same from treatment group to
treatment group on a case by case basis.
The distributions of all endpoints to be analyzed should be
examined prior to analysis using histograms, box plots or scatter
plots (as described in Appendix B), in order to decide if the
first ANOVA assumption is justified. The plots should be
obtained for each treatment group including control. The
distributions should be approximately symmetrical, tapering off
evenly in both directions from a single mode. Examination of the
distributions of the data can also provide justification of the
second assumption. If the range over which the data are spread
out is approximately the same for each group then the homogeneous
variance assumption can be justified. However, the authors have
33
-------
seen cases of toxicological data where this is not the case.
If an ANOVA has already been carried out then residuals from
the ANOVA can be plotted to check assumptions. Residuals (datum
minus the treatment group average) from all treatment groups
should be centered around a value of zero and the variability
should be approximately the same in all groups. In SAS (SAS
1985) residuals can be obtained in PROC GLM using an OUTPUT
statement. If the assumption of homogeneous variance is not
justified, then statistical tests may not be carried out at the
intended level and incorrect inference can occur. Wetherill
(1981) and Miller (1986) discuss how the conclusions obtained
from ANOVA can be affected when the variance is not equal between
treatment groups.
It is often suggested that Bartlett's test (Snedecor and
Cochran 1980) for homogeneity of variance be used to check the
second assumption, but it is known that Bartlett's test is very
sensitive to departures from normality (Miller 1986). Levene's
test (Levene 1960) is a simple and robust alternative to
Bartlett's test and it is described in Appendix B. -
When the number of pens per treatment is small, it can be
difficult to decide if the data are normally distributed or if
the variability is the same between groups. In such cases it is
suggested that good scientific judgement, as well as scatter
plots and histograms, be employed. In addition, the advice of a
trained statistician can be very helpful.
Transformations of the Data—
Although some types of responses may not appear to satisfy
the assumptions of an ANOVA, there may be a transformation of the
response which does. That is, on a different scale, the ANOVA
34
-------
assumptions are justified. Some transformations have been
developed for specific types of problematic data. For example,
the arc sine square root transformation of proportions is
typically suggested for adjusting the heterogeneous variance of
proportions. The square root transformation for counts of
individuals is another example developed for the same reason.
The Subdivision E guidelines suggest that the arc sine
square root transformation be applied to all proportions and then
ANOVA be carried out on the transformed variables. Although
these transformations may be helpful in making the distributions
of the data more bell-shaped, transformations will not
necessarily achieve normality, homogeneous variance nor
additivity of effects in one fell swoop. McCullagh and Nelder
(1989) and Atkinson (1985) discuss this point and acknowledge
that frequently a transformation of the data is not the best
solution to the problem. Therefore, once the data have been
transformed, it is imperative to recheck the data in order to
verify that the assumptions are justified on the new scale.
The arc sine square root transformation is suggested for the
proportions obtained in an avian reproduction study, but if the
transformation does not help to achieve normality and homogeneity
of variance, one recourse is to investigate other simple
transformations. Box and Cox (1964) developed a family of
transformations from which a choice of the most appropriate
transformation for achieving normality and homogeneous variance
could be chosen based on the data. Some common transformations
that might be used for counts or proportions, other than the arc
sine square root transformation, include taking the square root
or the logarithm of the response. One problem with using the
logarithmic transformation for positive data is that data values
of zero need to be receded since it is clear that they do not
belong at minus infinity (log(O)). One approach, similar to a
35
-------
method suggested by Tukey et al. (1985) is to choose Y*, the
receded value of log(O), such that:
[Log(Yt) - Y*]/[Log(Y2) - Log(Yl)] = [Y, - (>]/[¥, - Y2]
where Y1 is the smallest datum other than 0 and Y2 is the next
smallest datum.
In some cases, even on a transformed scale, the assumptions
for ANOVA cannot be justified. If these data are analyzed using
ANOVA, the conclusions that are reached may be incorrect and
means and variances may not be well estimated. A practical
approach to an analysis in which there is some question as to the
validity of the assumption is to carry out the analysis but
provide written documentation of the inconsistencies in
assumptions or data. Make it clear that the conclusions reached
from the statistical analysis are valid only to the extent that
the assumptions are justified.
There may be cases when the failure of the ANOVA assumptions
leads a scientist to look toward methods other than ANOVA such as
non-parametric statistics (Daniel 1978) or generalized linear
models (McCullagh and Nelder 1989) such as logit regression or
log linear models or a trend test. These methods and other
motivations for them are discussed in Section 3.
Alternative Designs —
As discussed previously, it is important that the
statistical analysis used reflect the design of the experiment.
It is possible that designs other than a Completely Randomized
Design (CRD) , as described in Section 2, are judged to be the
more appropriate for the test in a given laboratory.
36
-------
For example, Randomized Block Designs (RED) could be used to
reduce variability due to using multiple testing chambers. It is
important that the use of alternative designs be clearly stated
and justified so that test data can be evaluated by the Agency
using the correct statistical analyses. It is quite possible for
the data from a well designed RED study, in which effects are
pronounced, to show no statistically significant effect when
analyzed as if it was a CRD.
As another example, data tabulated from a design that
results in multiple error terms and requires a special analysis
such as a split plot design, may not appear any different from
data tabulated from a CRD. The difference lies within the design
description. Analyzing this type of data with a one way ANOVA
would lead to statistical tests that use the wrong error terms
and hence results would be highly suspect.
Therefore, it is crucial that the statistical analysis be
carried out based on the design that was used to generate the
data. Test protocols should be broadened to include provisions
for basic experimental designs other than the completely
randomized design and scientists should be aware of alternative
analyses required by these designs. There is no reason to
believe that the CRD is the best or most efficient design for all
situations.
Interpretation of Results—
Along with using statistical methods to identify significant
differences between control and treatment groups, the biological
reasons for such differences also should be considered. Finding
a statistically significant effect does not necessarily represent
a biological effect. For example, there may be unmeasured
confounding variables, unrelated to the pesticide effect, that
37
-------
contribute to the apparent significant difference. Or in the
case of a large number of replications, the statistical power nay
be great enough that treatment differences may be identified as
statistically significant even though they may not be large
enough to be of biological significance. On the other hand, the
lack of a significant effect does not mean that there is no
information in the data about pesticide-related effects. Random
variation in biological studies can be quite large and this may
make it difficult to detect statistically significant effects
even when steps have been taken to design a powerful study. When
trends or differences between treatment groups are clearly
noticeable, even if they are not significant, their presence
should be noted and this information added to the "weight of
evidence" for the pesticide in question.
The biological significance of apparent treatment effects
can be explored through a series of questions posed after the
study has been completed. Is the relationship between patterns
of effects and dietary concentrations explainable in the context
of our understanding of dose/response relationships? Are effects
observed in one measurement variable also observed in other
related variables? For example, if there is a treatment-related
decrease in the number of eggs produced, it is possible that
there would also be a treatment-related difference in the number
of 14-day-old juveniles produced, if the test has sufficient
statistical power. If not, is there a reason why the outcomes
for the variables are different? Are there mitigating factors or
does the increase in variability with each successive step during
development mask the effects of treatment? Are the differences
observed in quantitative measurements supported by the
qualitative observations in the test? For example, are
observations of behavioral abnormalities in adults correlated
with measurements such as food consumption or egg production?
38
-------
It is also important to ask questions of data sets where no
statistically significant differences were detected. Are there
trends in response variables that are not statistically
significant, but may represent biologically significant effects?
If so, would effects be significant using a more powerful
experimental design? Are there qualitative observations of
pesticide effects in adults or juveniles?
The benefit to this interpretation is the identification of
the questions that have been answered, questions that have
remained unanswered or unclear, and questions that have been
generated from test results that merit further consideration. If
further testing is required to develop the ecological risk
assessement, it is very important to identify the questions of
concern arising from the avian reproduction test. Identifying
the questions of greatest concern will help to determine the most
effective method for addressing the questions, such as repeating
the laboratory test, using a specialized laboratory test, or
conducting a field test.
39
-------
SECTION 3
ALTERNATIVE APPROACHES FOR EVALUATING REPRODUCTION EFFECTS
There are many approaches that could be taken in evaluating
the effects of pesticides on avian reproduction. However, there
are several important considerations in determining when
alternative methods would be appropriate or acceptable for
supporting a regulatory action. There is a need for consistency
in testing methods for comparability of results. Consequently,
to achieve consistency, it is desirable to have a "standard"
protocol used for testing all chemicals. On the other hand,
there is a need for flexibility in testing methods because some
questions or data sets may not be appropriate for a standard
protocol. In these cases there should be strong scientific
justification for using an alternative approach that is suited to
the question being addressed or the data set obtained. The use
of alternative methods should be clearly explained and justified
as to why they are superior to methods recommended in the
Subdivision E guidelines.
Alternative approaches would be justified and useful in
situations where previous experience (i.e., experience with
related chemical or similar data distribution) indicates that a
new or existing pesticide may pose reproductive effects that are
not adequately tested and evaluated using the current guidelines.
Because avian reproduction tests are used primarily as a toxicity
screening tool, test approaches must be broad enough in scope to
identify potential effects from a variety of pesticides, unless
they are designed for a specific regulatory question or specific
chemical. Especially in the case of existing pesticides where
considerable information may be available concerning both the
laboratory and field effects, alternative test protocols may be
clearly indicated to address specific questions to support a
40
-------
pesticide registration. The following section discusses several
alternative approaches to avian reproduction testing, with
special attention given to the advantages and disadvantages of
the alternatives and guidance on the situations in which the
alternative may be appropriate.
ALTERNATIVE TEST PROTOCOLS
Determination of a Dose-response Relationship
When possible, a statistical analysis of quantitative
responses from an avian reproduction study is desired. Under the
current Subdivision E guidelines, this analysis results in
estimating the mean response for each treatment group, a
description of which treatment groups are statistically different
from control, and the identification of a NOAEL and LOAEL dose
level. Although the concepts of a NOAEL and LOAEL are easily
understood, they are scientifically difficult to define and
measure accurately. What is identified as a LOAEL under one
experimental design may be the NOAEL under a similar design with
a different choice of dietary concentrations or replicates.
More information about the pattern of response over the
range of dietary concentrations and the potential effects of
pesticides could be obtained by using a design and analysis for
the determination of a dose-response relationship. The 'dose1 in
this case, would be the dietary concentration used in the study
and the 'response1 could be any of the quantitative responses
listed in Section 2. The test design would include a range of
several dietary concentrations, including control, that produce
effects on reproductive parameters for calculating the dose-
response relationship. This approach could be comparable to the
dose-response approaches used for measuring relative toxicity
41
-------
(e.g., LD50, LC50).
The advantages to this approach are: 1) the relative
distance between effect concentrations and the expected
environmental concentration for a proposed use could be
established; 2) the types of reproductive effects caused by a
pesticide would be identified; 3) the pattern of toxicity over
the dietary concentrations could be established; 4) a more
accurate definition of the NOAEL and LOAEL; and 5) the ability
to calculate additional statistics that help define the pattern
of toxicity such as ECSO's and the slope of the curve defining
the dose-response.
As was discussed in Section 2, if a test failed to detect
reproductive effects using current methods, nothing would have
been learned about the types of effects that might have been
observed under different exposure scenarios. Conversely, if
reproductive effects are observed in all treatment groups, it may
not be possible to determine if the same or different effects
would be observed at lower concentrations. The alternative
approach would circumvent these problems.
The disadvantages of this dose-response approach are: 1) a
range finding test or tests would be required to establish
appropriate dietary concentrations for a definitive test; and 2)
the determination of tissue and egg residues may be less
meaningful if the test was conducted at concentrations
considerably different from the EEC. Another possible
disadvantage is that, as the number of dietary concentrations
increases and the number of replicate pens per treatment group is
held constant, the size of the test will increase. However a
trade-off may be made by reducing the number of pens per
treatment. The number of replicate pens per treatment may be
different from the currently prescribed number if an analysis
42
-------
other than the one way ANOVA is used (e.g., maximum likelihood
methods).
The appropriate choice of the number of treatment groups and
the number of replicates per treatment group depends on which
statistics are to be obtained and the data analysis method to be
used. For example, if it is desired to estimate the EC50 then the
dose (dietary concentration) groups should be centered around the
likely values of the EC50. If a logit regression analysis is to
be used and the endpoint is the proportion of hatchlings out of
eggs set per pen, then it is important to have a large number of
eggs set over all pens in the treatment group than to have simply
many pens. A statistician should be consulted at this point.
Designing dose-response studies is not always the same as
designing the type of study defined under Subdivision E.
Analysis methods can include simple and multiple regression for
normally-distributed response variables, logit or probit
regression for proportions and log-linear models for data in the
form of integer counts.
The dietary concentrations used in the study should be
spread out over the range of response that is of interest, i.e.,
extrapolating the dose-response beyond the range of the data is
statistically dangerous. Increasing the number of dietary
concentrations used will result in more precise statistics.
However, some statistical analyses that do not rely on
assumptions of normal distributions do not have the same types of
requirements concerning replication as does the analysis of
variance (see the example two paragraphs above). In addition,
power calculations relevant to an ANOVA may not apply in these
cases. It is important not to assume that the methods defined in
Subdivision E carry over exactly to an alternative type of study.
Seek the advice of a trained statistician when'nonstandard
43
-------
methods are used.
Test for Effects from Short-term Exposure
The current avian reproduction test is designed to determine
the effects of pesticides with chronic exposure patterns (i.e.,
continuous or repeated exposure) on reproductive mechanisms.
Since the development of reproductive capacity in birds begins
months before initiation of egg laying (Kirkpatrick 1959), the
test originally was designed to start treated diets well in
advance of laying because of the bioaccumulative properties of
the organochlorine insecticides. However, many newer pesticides
are much less persistent in the environment and their use
patterns are such that the initial contact with these pesticides
may come at any time during reproduction. The current test
protocol using a chronic exposure period may not effectively
identify potential effects'of less persistent pesticides. An
alternative method is to utilize a shorter exposure period
initiated after the test population is in egg production. This
approach is mentioned in the ASTM guidelines (1990), though not
discussed. Bobwhite tests with organophosphorus (OP)
insecticides have shown significant reproductive effects with
treatment periods of 8 days (Bennett and Bennett 1990), 10 days
(Rattner et al. 1982), and 3 weeks (Stromborg 1981, 1986a, I986b,
and Bennett et al. 1990b). A mallard test with an OP insecticide
used a treatment period of 8 days and showed a pesticide-related
response (Bennett et al. 1991). The length of tests using
shorter exposure periods has not been standardized, but one
approach would be to use environmental chemistry data to set the
test length as a function of the environmental degradation rate.
Several of the organophosphorus and carbamate insecticides
have been shown to significantly reduce egg production within
44
-------
days of initiation of dietary exposure (Bennett and Bennett 1990,
Bennett et al. 1991). Changes in eggshell quality also have been
observed within days after treatment (Bennett and Bennett 1990).
Further, because many of the less persistent pesticides will have
shorter residual times in the environment, they may not meet the
current criteria for initiating an avian reproduction test, even
though they present a potential hazard to reproduction from
short-term exposures. Therefore, there is a need to establish
new criteria to initiate a short-term test. Given the
relationship between the amount of food consumption and the rate
of egg production, one possible criterion is testing pesticides
that produce significant reduction in food consumption (e.g.,
>50%) in the higher concentrations of an acceptable avian dietary
toxicity test (i.e. LC50 test). Since the relationship between
pesticide concentrations that affect food consumption of
juveniles and those that affect reproduction in adults is poorly
understood, an adult feeding test may be required to determine
the effect levels in reproducing adults. It is also unclear if
all chemicals that reduce the food consumption of juvenile birds
will result in effects on reproduction.
There are several potential advantages to using a short-term
exposure test. First, the test can be conducted with known
layers of fertile eggs, thus reducing the variability in test
data by not including pairs that produce no eggs or only
infertile eggs. Second, the pretreatment values for measurement
endpoints can be used as controls for each pen (i.e.,
covariates). For example, birds typically lay eggs of consistent
size, shape, and eggshell strength (Thompson et al. 1983), making
the use of pretreatment values from each bird as covariates a
useful method for reducing between-hen variation. Finally, the
test is timed to coincide with the maximum egg production. All
three of these points will reduce the variability from factors
that are unrelated to the pesticide treatment and increase the
45
-------
chances of detecting reproductive effects if they exist. This
approach could also be used in conjunction with the dose-response
alternative described above.
A disadvantage of this approach is that it would not detect
a pesticide-related delay in the onset of laying because
treatment would start during the laying period. Also, effects
may be delayed or the severity of the effects may increase during
the treatment period, so a short treatment period could not
assess delayed effects. Effects that are first expressed near
the end of the treatment period may be overlooked if the data
analysis is conducted only on mean values for the entire
treatment period. The short-term test should not be used with
pesticides with slow-developing or delayed effects, since the
current test guidelines would provide a better approach.
Parental Incubation
Because the current test guidelines (USEPA 1982a) recommend
artificial incubation, parental influences on hatching and
survival of juveniles are eliminated. The test does not provide
information about pesticide effects on parental behavior after
oviposition. However, many pesticides have been observed to
affect parental behaviors that lead to nest or brood
abandonments, abnormal parental care or other behaviors that
adversely affect production of juveniles.
There are several field examples of pesticide-related
behavioral effects during incubation. Forest spraying operations
with fenitrothion resulted in disruption of incubation and nest
abandonment by white-throated sparrows (Zonotrichia albicollisl
(Busby et al. 1990); reproductive success in a sprayed forest
area was one-third of that observed in a control area. Laughing
46
-------
gulls (Larus atricilla) dosed with parathion incubated
significantly less tine during the following three days than
controls (White et al. 1983), although no effects on nest defense
behavior or hatching success were observed (King et al. 1984).
Fyfe et al. (1976) found that merlin (Falco columbarius) nests
with high DDE residues in the eggs were deserted more and
defended less than nests with lower residues, although Fox and
Donald (1980) concluded that the reduced nest defense associated
with high DDE egg concentrations was of minor importance in
explaining nest, failures. Fox et al. (1978) hypothesized that
pollutant-induced endocrine dysfunction in Lake Ontario herring
gulls (Larus argentatus) may have caused reduced nest
attentiveness and defense.
Insecticides have also been observed to adversely affect
interactions of parents with their young. Female starlings dosed
with dicrotophos made significantly fewer sorties to feed their
broods and were absent for longer periods of time than control
females (Grue et al. 1982). Meyers and Gile (1986) found that
mallards breeding on ponds supplemented with food treated with 80
ppm chlorpyrifos did not produce any ducklings surviving to 7
days old, partially because the hens were not attentive to the
broods, while control ponds produced six to eight ducklings per
hen. Brewer et al. (1988) observed brood abandonment by wood
duck (Aix sponsa) and blue-winged teal (Anas discors) hens after
agricultural applications of methyl parathion.
Several studies have been conducted with captive ducks using
parental incubation to more closely simulate field situations
(Bennett et al. 1991, Custer and Heinz 1980, Finley and Stendell
1978, Franson et al. 1983, Haseltine et al. 1980, Longcore and
Samson 1973). Black ducks (Anas rubripes) fed 10 ppm DDE
produced eggshells 22% thinner at the equator than controls, with
a cracking rate of naturally incubated eggs that was fourfold
47
-------
greater than artificially incubated eggs (Longcore and Samson
1973). Breeding black ducks fed 3 ppm mercury appeared
hyperactive, and fewer incubated their clutches (Finley and
Stendell 1978). A range of responses were observed in 23 mallard
hens exposed to dietary methyl parathion for 8 days either early
or late in incubation (Bennett et al. 1991). No response to
treatment was observed in 6 hens that successfully hatched broods
compared to 7 hens that abandoned clutches, 6 hens that exhibited
reduced nest attentiveness by leaving their nests for extended
periods on one or more days, and 4 hens that died while
incubating. An additional treatment group exposed to methyl
parathion for 8 days during the egg laying period was observed to
either begin incubating immediately after the end of the
treatment period or resume egg laying after treatment by either
completing their clutch or abandoning their first clutch and
starting a new nest (Bennett et al. 1991). Even though this
group was treated only during egg laying, these kinds of
reproductive behaviors could only be observed in a test where
birds were allowed to incubate their own eggs.
Parental incubation tests using pesticides and other
environmental contaminants also have been conducted with several
other species, such as ring doves (Stretopelia risoria) (Haegele
and Hudson 1973, McArthur et al. 1983, Peakall et al. 1972,
Peakall and Peakall 1973), Bengalese finch (Lonchura striata)
(Jefferies 1971), screech owls (Otus asio) (McLane and Hughes
1980), barn owls fTvto alba) (Mendenhall et al. 1983), and
kestrels fFalco sparverius) (Porter and Wiemeyer 1969, Wiemeyer
and Porter 1970). Ring doves fed dietary concentrations of 10
ppm polychlorinated biphenyl (Aroclor 1254) were less attentive
of their nests than controls and produced fewer hatchlings than
eggs that were artificially incubated from treated birds (Peakall
and Peakall 1973).
48
-------
The primary advantage to using parental incubation is that
pesticide effects on parental behavior can be observed at the
transition from laying to incubation and during incubation and
brood rearing. Tests using artificial incubation can not
identify pesticides that induce birds to: 1) incubate smaller
than normal clutches; 2) continue laying excessively large
clutches without incubating; 3) abandon nests or broods; or 4)
reduce parental care for the nests or broods. These behaviors
could have as great an effect on the production of young as any
measurements made in the current avian reproduction test.
There are several disadvantages to using parental
incubation. First, variability among birds within treatments may
increase requiring substantially larger numbers of replicate pens
per treatment to achieve the same power in the test as using
artificial incubation. By keeping each hen as the incubator
instead of using one mechanized incubator, incubation behavior of
individual hens becomes a factor that can add significant
variability to reproduction parameters in all treatment groups.
Not all control hens that initiate clutches will begin incubation
(Bennett et al. 1991, Haseltine et al. 1980). Not all control
hens that begin incubation will hatch young (Bennett et al. 1991,
Haseltine et al. 1980, Finley and Stendell 1978). Due to the
variability in incubation behavior among individuals, hatching
percentages can range from 0 to 100% in control groups. Bennett
et al. (1991) found that 35 mallard pairs per treatment would
have been required to detect the 57% decrease in ducklings in one
treatment group as statistically different with a Type I error
rate of less than 5%. Second, different pens may be required to
induce hens to initiate a nest and incubate. Besides providing
nest bowls or boxes, additional space and a greater degree of
isolation may be required for hens to behave normally. This is
especially true for bobwhite, which are more sensitive than
mallards to human contact. Third, fewer eggs would be laid per
49
-------
pen so that proportional values (i.e., fertility, hatchability,
and survival) would be calculated from smaller samples.
Parental incubation tests should be used when the concern
for a pesticide is primarily with effects on adult behavior. It
can be an effective method for identifying potential effects
during incubation and brood rearing prior to attempting a field
study of reproductive effects. However, investigators conducting
tests using parental incubation should justify clearly the
rationale for using this method and the experimental design
chosen.
ALTERNATIVE ENDPOINTS
Are there other endpoints that should be considered when
conducting an avian reproduction test? The value of alternative
endpoints should be judged based on satisfying one or more of the
following criteria: 1) the endpoint detects reproductive
effects that are not detected by current methodology; 2) the
endpoint is more sensitive at detecting effects than existing
endpoints; and 3) the endpoint provides important insight into
exposure patterns or mode of action of the pesticide. There is a
variety of biochemical endpoints (e.g., serum chemistry, enzyme
activity, endocrine function) that may be affected by particular
pesticides or classes of pesticides at the concentrations that
affect other reproductive endpoints, although they do not
represent reproductive effects themselves. These endpoints can
be very useful in understanding individual differences in
sensitivity and identifying causes for other observed effects.
However, it is beyond the scope of this report to address the
broad range of possible endpoints. This section discusses a few
alternative endpoints that can be useful in explaining
reproductive effects and may address the criteria above.
50
-------
Eggshell Strength
Eggshell thickness was the first measurement used to
determine the effects of environmental contaminants on avian
eggshell quality (Hickey and Anderson 1968, Ratcliffe 1967). The
quality of an eggshell is reflected in its ability to maintain a
protective environment for the developing embryo by resisting
cracking or puncture and is a function of its thickness,
ultrastructural characteristics, and the size and shape of the
egg. While eggshell thickness is an extremely valuable measure
of quality, it may not be the most sensitive indicator of adverse
effects on eggshell quality posed by all chemicals. Eggs of
mallards exposed to dietary DDE had reduced shell strength at
lower dietary concentrations than were observed to affect shell
thickness (Carlisle et al. 1986). Bobwhite exposed to
sulfanilamide produced eggs with reduced breaking strength
without reducing shell thickness (Bennett et al. 1988). Scanning
electron microscopy of the eggshells indicated changes in their
ultrastructure, characterized by poorly formed mammillae. In
eggs collected from wild white-faced ibis (Pleaadis chihil,
eggshell strength decreased to a greater extent than shell
thickness with increasing DDE residues in the yolks (Henny and
Bennett 1990). In addition to DDE, many other classes of
pesticides have been shown to reduce eggshell thickness under
laboratory conditions (Haegele and Tucker 1974, Bennett and
Bennett 1990), but few studies of contaminant effects have also
measured eggshell strength. Until more is known about the effect
of chemicals on eggshell strength, it would be a useful
additional endpoint to measure for all pesticides.
Two methods have been used for measuring eggshell strength.
The puncture test uses a punch to penetrate the shell and
represents a measure of shear fracture force. The compression
test uses two parallel flat surfaces and represents a measure of
51
-------
the tensile fracture force. Both methods provide results that
are linearly related to eggshell thickness in uncontaminated
eggs, although the puncture test was more highly correlated with
thickness (Hunt et al. 1977). The puncture method has the
advantage that repeated measures can be made on each egg for
calculation of an average puncture force value, whereas the
compression test can only be performed once per egg. However,
the compression test is a direct measure of eggshell strength
that simulates field conditions (Hunt et al. 1977). Eggshell
cracking is related to tensile fracture properties, which should
provide an index of eggshell resistance to field insults.
Another source of variation to consider is the occurrence of
body-checked eggs or eggs cracked in the shell gland that are
partially repaired by additional calcification (Roland 1982).
Body-checked eggs were thinner than control eggs and required 18%
less force to crack eggs in a compression test (Roland 1982).
Candling can be used to identify body-checked eggs.
The primary advantage to measuring eggshell strength is that
it may detect effects on eggshell quality at lower dietary
concentrations than measuring thickness alone or effects where
thickness is unaffected. The eggshell strength test integrates
the effects on several parameters such as thickness and
ultrastructural integrity into one measure of overall eggshell
quality. Consequently, eggshell strength may be more indicative
of potential field effects than thickness.
One disadvantage to this endpoint is that compression
strength can not be measured on cracked or broken eggs, whereas
eggshell thickness can. If the rate of egg cracking is high in a
test, the number of data gaps due to unmeasured cracked eggs may
be unacceptably high. This is a further reason to take all
precautions to minimize the rate of cracked eggs in a test so
that reproductive parameters can be measured directly. Another
52
-------
disadvantage is that the compression test requires mechanized
testing equipment; models of table top testing instruments exist
for under $10,000. Less expensive instruments that do not
mechanically control compression are available, but their
accuracy and precision need to be rigorously documented before .
they are acceptable for measuring treatment effects on eggshell
quality.
Plasma Calcium Concentrations in Females
Concentrations of calcium in plasma of laying birds can be a
helpful indicator of how pesticides influence eggshell quality
and egg production. Eggshells consist of 98 to 99% calcium
carbonate and 1 to 2% of proteins and polysaccharide material
(Wilbur and Simkiss 1968). In chickens, the formation of the
eggshell takes 16 to 20 hours (Talbot and Tyler 1974). Eggshell
formation takes approximately 20 hours for the ring dove with 60%
of the calcium derived from dietary sources and the remainder
from the marrow of bones (Peakall 1970). During eggshell
formation, only a small fraction (1-2%) of the calcium needed for
the shell is in the blood at any one time. Reductions in the
concentrations of calcium in the blood may lead to abnormally
thin eggshells or birds may cease egg laying rather than
producing thin eggs.
Plasma calcium concentrations in female mallards are
approximately twice as high during egg laying as other times of
the year (Fairbrother et al. 1990). Serum calcium concentrations
of laying bobwhite were 2.3 times higher than in males (Bennett
et al. 1990b). Dietary methyl parathion produced a dose related
decrease in serum calcium concentrations of laying females at
levels that also affected egg production, where calcium
concentrations in males were unaffected by treatment (Bennett et
al. 1990b). This may be related to the pesticide-induced
53
-------
reduction in food consumption, which would reduce overall intake
of calcium. Reduced food consumption alone has been observed to
reduce eggshell thickness and strength (Haegele and Tucker 1974,
Bennett and Bennett 1990).
The eggshell thinning observed from exposure to DDE is not
related to reductions in blood concentration of calcium. Because
the calcium concentration in shell gland cell was higher than in
control birds, Lundholm (1984) concluded that the translocation
of calcium between blood and the shell gland was not impaired.
Eggshell thinning from DDE is a function of reduced secretion
from the shell gland to the egg. Consequently, the value of
measuring plasma calcium concentrations is to use it as a
diagnostic tool for determining the reasons for observations of
reduced eggshell quality or egg production in an avian
reproduction test. It may be especially useful for addressing
specific questions of the effects of a particular pesticide on
eggshell quality prior to attempting to document the presence of
these effects under field conditions.
Parental Organ Size and Weight
The Subdivision E guidelines state that the test report
should contain information from post-mortem necropsies, but does
not mention specifically the reporting of information on the size
and weight of internal organs. Treatment-related changes in
organ weights can be useful for determining causes of other
reproductive effects. For example, the size and weight of
reproductive organs (e.g., testes, hemipenis (mallard), oviduct
and ovary) can provide much additional information to explain
lack or cessation of egg production, infertility of eggs, or lack
of mating behaviors. Changes in other internal organs may
indicate the need for histological examinations to identify
specific cellular changes. Identifying treatment-related changes
54
-------
in organs in conjunction with other reproductive effects can aid
in the development of additional laboratory and/or field testing
to assess pesticide risks to avian reproduction.
ALTERNATIVE DATA ANALYSIS METHODS
The data analysis methods required by Subdivision E are to
provide a mathematical/statistical model for the data obtained in
an avian reproduction study. This model (the one way linear
model for normally distributed errors) has been shown capable of
handling many kinds of data and is flexible to some deviations
from the assumptions implicit in the analysis. But there may be
times when another mathematical model is scientifically more
defensible. This section presents some alternative methods that
may provide better models in some cases. The decision to use an
alternative analysis should be based on sound scientific
reasoning and the reasons for choosing an alternative analysis
should be clearly documented.
The one way ANOVA, recommended in the Subdivision E
guidelines, tests for differences in means of treatment groups
and makes no supposition of increasing effect with dose.
However, in toxicological studies that is very often the case and
interest may lie in testing such a hypothesis. Tukey et al.
(1985) have proposed an alternative trend test for toxicological
studies which can be carried out on data collected from an avian
reproduction study as described by subdivision E. This procedure
identifies the highest dietary concentration at which no
statistically significant trend is detected; a concept similar to
a NOAEL. Tukey et al. (1985) claim that this test has high power
when the response is highly correlated with the dietary
concentration and reasonable power against a wide variety of
response patterns.
55
-------
Alternative data analyses nay be required if the ANOVA
assumptions of homogeneous variance and normality cannot be met
even on a transformed scale or if it can be shown that the data
fit a distribution other than the normal distribution and
analysis is desired on the untransformed scale. Or, if a dose-
response study is carried out it may be desired to fit the
response as a linear function of both continuous covariates (such
as dietary concentration) and classification variables (such as
sex) but the data do not have a normal distribution or it is
desired to use a statistical model that can account for patterns
of variability that are different from the normal model. This
latter alternative can apply when a dose-response type of study
is carried out.
Some alternative methods that can be used in the above
situations include nonparametric methods, generalized linear
models for responses with nonnorroal distributions and linear
models for responses where only the mean/variance relationship
(and not the entire response distribution) is known.
If some ANOVA assumptions do not seem to be justified by the
data, then the choice of an alternative method of analysis may
depend on which assumption is violated. If the variability in
the treatment groups seems to be approximately similar but the
distribution of the response is very skewed even when transformed
then nonparametric methods (Daniel 1978) can be used (see
Appendix B). But if heterogeneous variance between treatment
groups is the problem, then some nonparametric methods that
require homogeneous variance will not be appropriate either.
Monte Carlo randomization tests are nonparametric tests which
deserve mention. Such tests can provide exact statistical test
but they may require the computer generation of all possible
outcomes or the generation of many possible samples. Because of
this, they are not useful as a routine analysis but can provide
56
-------
some insight when other methods cannot.
When the data consist of counts of individuals affected out
of the total in the brooder, and where transformations have not
helped to stabilize variance between groups, it may be possible
to model the data as coming from a binomial distribution and use
logistic or probit regression to decide if there are differences
between treatment groups. If the response is not a proportion
but an integer count, then the data may be able to be modeled as
coming from a Poisson distribution and a log linear model can be
used (McCullagh and Nelder 1989). The methodology is similar to
that used for probit regression.
Toxicology researchers in human health applications have
used the above types of analyses for studies using litters of
mice which are analogous to pens or clutches of birds. There is
extensive literature on the application of these methods to
toxicological studies, especially where correlations between
individuals in the same litter (analogous to a pen or clutch)
exist. Williams (1975) described a maximum likelihood method for
analyzing binary response data from experiments involving such
litter effects. Haseman and Kupper (1979) reviewed various
models that are used to account for litter effects in
toxicological studies with dichotoroous responses such as death or
the presence of malformation. Haseman and Soares (1976) discuss
the impact of litter effects on modeling and estimation and
compare various analyses methods. Kupper et al. (1986) use the
method given by Williams (1975) to model intralitter correlations
and to study the biases and variances of the maximum likelihood
estimators.
Intralitter correlations can lead to variability that is
greater than that predicted by a simple parametric model such as
the binomial distribution. This extra variability has been
57
-------
attributed to differences between groups of subjects that are not
related to treatment effects. This same phenomenon occurs in
avian reproduction studies and examples have been pointed out in
Section 2. For example, the tendency for a hen to have higher
numbers of cracked or infertile eggs due to her own
characteristics can lead to extra variability. Finney (1971)
labeled this extra variation heterogeneity and provided a method
for estimating it in the context of probit regression. Williams
(1975) suggested the use of a beta-binomial model to account for
the extra variation.
Wedderburn (1974) provided a relatively simple and general
method for estimating and accounting for extra variation in the
evaluation of treatment effects in the context of a linear model
such as regression or ANOVA. If the distribution of the response
is not known, but the relationship between the mean response and
the variance of the response can be assumed to be similar to that
of a known distribution such as binomial or Poisson, except for a
constant of proportionality, then Wedderburn has shown that
statistically consistent and unbiased estimates of treatment
effects can be obtained with quasilikelihood methods. Although
these methods are similar to the maximum likelihood methods of
logistic and probit regression, they do not assume the complete
form of the distribution, only the relationship between the mean
and variance. In fact, these methods apply not only to data in
the form of counts or proportions, but to any data for which a
linear model is postulated for some function of the mean.
McCullagh and Nelder (1989) provide a good review and explanation
of this topic. Such models can be fit using existing software
such as SAS (1985). The software package GLIM (Baker and Nelder
1978) was specifically developed for fitting generalized linear
models either by maximum likelihood or quasilikelihood and the
package is widely available as well. In a more recent
application, Chen and Kodell (1989) use the beta-binomial
58
-------
distribution to account for litter effects and a Weibull dose-
response model for teratogenic effects to obtain low-dose risk
estimates for reproductive and developmental toxic effects rather
than a NOEL.
Although these alternative analyses are nonstandard and
require extra effort and understanding by the data analyst and by
the scientist, their use should not be discouraged for these
reasons. When required, they may be the most appropriate methods
to use. That is, they do not depend on assumptions unjustified
by the data. They may also be used by registrants when more
information is sought than is required by the registration
process. It is inefficient to require the registrant to carry
out a less informative experiment or analysis when the
information required by OPP already exists in the alternative
analysis. The expertise of a trained statistician can be
invaluable in interpreting the results of such analyses.
59
-------
SECTION 4
CONCLUSIONS
The avian reproduction test is unlike other standardized
laboratory toxicity tests for birds where the primary endpoint of
the test is death. Effects on avian reproduction can be
expressed in a variety of ways. The current avian reproduction
test measures a suite of parameters, but there are many more that
are not or can not be measured by this test. Alternative methods
can be employed to address parameters not currently addressed,
but the obvious conclusion from the above discussion is that
screening for reproduction effects is not a one-size-fits-all
proposition. No one test or experimental design can adequately
address all aspects of avian reproduction or be adequate for all
chemicals. No one test can have equal sensitivity to detect
treatment effects across all parameters.
The current avian reproduction test provides a strong basis
for screening pesticides for potential effects if conducted in a
manner that achieves sufficient statistical power to detect
effects that exist. There are many steps that must be taken by
investigators to control for sources of variability that are
unrelated to the pesticide treatment. This report has discussed
several means of reducing variability unrelated to the pesticide
treatment. If the evaluation of test data leads to a conclusion
that the use of a pesticide at proposed application rates would
produce unacceptable adverse effects on avian reproduction,
several options exist for rebutting the presumption of risk using
laboratory or field methods. Many of the alternative approaches
discussed above may be helpful in clarifying concerns raised in
the avian reproduction test. Taking these concerns to the field
will provide a more realistic scenario for verifying or rebutting
60
-------
presumed risks, although there is a paucity of standardized field
methods for evaluating avian reproduction. Specialized
laboratory tests may be very helpful in identifying specific
questions to address in the field environment.
If in the evaluation of test data the Agency does not find
that pesticide exposure at the dietary concentrations tested
produced adverse effects on avian reproduction, many questions
may remain unanswered. Was the experimental design and data
analysis adequate to detect existing reproductive effects? Were
the measurement endpoints appropriate for identifying the kinds
of reproductive effects potentially produced by the pesticide in
the field? Were the test species adequate for assessing
potential reproductive effects to other taxonomic groups of
birds? There is often little information available to answer
these questions. Investigators can and should provide an
analysis demonstrating that a test had sufficient power to detect
pesticide-related effects if they existed at the dietary
concentrations used. However, if a test that does not detect
treatment effects, there is no information about the potential
for effects at higher concentrations or the kinds of effects
potentially caused by the pesticide. Consequently, we have
recommended that tests should be conducted using dietary
concentrations that produce effects in the more sensitive
endpoints. This should improve the risk assessments by providing
a measured effect level to compare with the EEC, rather than
making a decision amongst the uncertainty surrounding a test that
failed to detect effects.
There are also situations where an avian reproduction test
may not be triggered for a particular pesticide because its use
pattern will not result in repeated or continuous exposures to
birds. Pesticide effects on avian reproduction are more than a
chronic phenomenon. They can occur rapidly with some chemicals,
61
-------
with potentially significant consequences. Not only are new
approaches needed to properly evaluate the effects of short-term
exposures to pesticides, but new criteria need to be established
to identify when pesticides nay present a hazard to avian
reproduction under any exposure scenario.
62
-------
SECTION 5
RECOMMENDATIONS
This report has reviewed the current Subdivision E
guidelines (USEPA 1982a) for conducting avian reproduction tests
and discussed the advantages and disadvantages of several options
and alternative methods. Many of the options are designed to
strengthen the test described in the Subdivision E guidelines by
reducing variability in the data from sources unrelated to the
pesticide treatment. Many of the alternative methods are
designed to add dimensions to the test that do not currently
exist, but could be ecologically relevant for evaluating
reproductive effects. The intent of the report has been to
provide technical input for future revisions of the Subdivision E
guidelines and for discussions on the future of the avian
reproduction test.
The avian reproduction test is experimentally much more
complex than other standardized avian tests. Consequently, there
is a need for the Agency to clearly define the objectives of the
test and to explain how test results are to be evaluated and used
in the risk assessment process. There is also a need for
investigators to be very clear and explicit in their descriptions
of test methods, analytical methods, and data interpretation, so
that the test reports can be properly reviewed by the Agency.
Several concerns were listed in the Introduction about the
methods used for evaluating the potential effects of pesticides
on avian reproduction that OPP should examine when reviewing
guidelines for conducting avian reproduction tests. Many of
these concerns can be addressed through modifications to existing
guidelines and through alternative test methods, although some of
these alternatives have not been standardized. Attention should
63
-------
be given to identifying and reducing sources of variation that
are unrelated to the pesticide treatment. Examples of these
sources of variation, which are generally related to test
procedures and the design of physical test systems, are given
throughout the text. This could greatly improve the utility of
tests conducted under the existing test guidelines.
The test guidelines could be more specific in defining what
constitutes an acceptable test by establishing acceptability
guidelines for parameters such as rate of egg cracking, eggshell
thickness, and percent fertility, hatching and survival. A
compilation of historical control data would be invaluable for
determining what have been achievable and acceptable values for
these parameters and indicating what may be currently achievable.
This is not to say that these acceptability guidelines for
specific endpoints should be used to judge the acceptability of
the entire test.
To maximize the amount of information gained from an avian
reproduction test, the test should be conducted in the range of
dietary concentrations that produce significant effects on avian
reproduction, unless the concentrations required would be
excessively high (e.g., 100 X EEC). This would better define the
kinds of effects potentially produced by a chemical, better
define the relationship between the EEC for a particular use and
the range of dietary concentrations causing effects, and reduce
the need for repeating tests when new uses are proposed.
It is also very important that the power of the avian
reproduction tests be carefully considered. It is the
responsibility of the investigator to use information on the test
chemical and estimates of response variability observed in
previous tests with a particular test system and species to
design tests to achieve sufficient statistical power when
possible. This is one of the most important issues to confront
64
-------
in future revisions of the guidelines. There is a need for a
detailed analysis of existing reproduction test data to provide
guidance on the adequate numbers of birds per pen, numbers of
replicate pens per concentration, and numbers of concentration
given various degrees of response variability. This analysis
would be very important in determining the most efficient
experimental design necessary to achieve the regulatory
objectives defined by the Agency.
Pesticide effects on avian reproduction are not simply a
function of chronic exposure. This report has discussed several
examples of reproductive effects resulting from relatively brief
exposures to pesticides. New criteria need to be established for
pesticides that do not satisfy existing criteria for initiating
an avian reproduction test, but may affect reproduction from
short-term pesticide exposures. Test methods for measuring
reproductive effects of pesticides with short-term exposures also
need to be standardized.
In summary, future revisions of Subdivision E guidelines can
focus on maximizing the information provided by an avian
reproduction test, identifying and reducing sources of
variability unrelated to the pesticide treatment, reevaluating
questions relating to power and sample size, establishing
baseline guides for endpoints based on past data and establishing
criteria for effects resulting from short-term exposure.
65
-------
REFERENCES
American Society for Testing and Materials. 1990. Standard
practice for conducting reproductive studies with avian species.
American Society for Testing and Materials, Vol. 11.04. E 1062-
86. Philadelphia, PA, pp. 660-670.
Anderson,D. W. and J. J. Hickey. 1972. Eggshell changes in
certain North American birds. In K. H. Voous, (ed.), Proc. XV
Int. Ornithol. Cona. E. J. Brill. Leiden, The Netherlands, pp.
514-540.
Atkinson, A. C. 1985. Plots Transformations and Regression: An
Introduction to Graphical Methods of Diagnostic Regression
Analysis, Oxford Science Publications, Oxford University Press,
NY.
Baker, R. J. and J. A. Nelder. 1978. The GLIM System. Release
3., Numerical Algorithms Group, Oxford.
Bennett, J. K. and R. S. Bennett. 1990. Effects of dietary
methyl parathion on northern bobwhite egg production and eggshell
quality. Environ. Toxicol. Chem. 9:1481-1485.
Bennett, J. K., S. E. Dominquez and W. L. Griffis. 1990a.
Effects of dicofol on mallard eggshell quality. Arch. Environ.
Contam. Toxicol. 19:907-912.
Bennett, J. K., R. K. Ringer, R. S. Bennett, B. A. Williams and
P. E. Humphrey. 1988. Comparison of breaking strength and shell
thickness as evaluators of eggshell quality. Environ. Toxicol.
Chem. 7:351-357.
66
-------
Bennett, R. S., R. Bentley, T. Shiroyama and J. K. Bennett.
1990b. Effects of the duration and timing of dietary methyl
parathion exposure on bobwhite reproduction. Environ. Toxicol.
Chem. 9:1473-1480.
Bennett, R. S., B. A. Williams, D. W. Schmedding and J. K.
Bennett. 1991. Effects of dietary exposure to methyl parathion
on egg laying and incubation in mallards. Environ. Toxicol.
Chem. 10:501-507.
Box, G. E. P. and D. R. Cox. 1964. An Analysis of
Transformations. J. Roval Stat. Soc.. Series B. 26:211.
Box, G. E. P., W. G. Hunter and J. S. Hunter. 1978. Statistics
for Experimenters. An Introduction to Design. Data Analysis and
Model Building. John Wiley and Sons, NY.
Brewer, L. W., C. J. Driver, R. J. Kendall, C. Zenier and T. E.
Lacher, Jr. 1988. Effects of methyl parathion in ducks and duck
broods. Environ. Toxicol. Chem. 7:375-379.
Brown, M. B. and A. B. Forsythe. 1974. Robust tests for the
equality of variances. J. Amer. Stat. Assoc. 69:364-367.
Busby, D. G., L. M. White and P. A. Pearce. 1990. Effects of
aerial spraying of fenitrothion on breeding white-throated
sparrows. J. Appl. Ecol. 27:743-755.
Canadian Council on Animal Care. 1984. Guide to the Care and
Use of Experimental Animals. Canadian Council on Animal Care.
Vol. 2, pp. 39-46.
Carlisle, J. C., D. W. Lamb and P. A. Toll. 1986. Breaking
67
-------
strength: An alternative indicator of toxic effects on avian
eggshell quality. Environ. Toxicol. Chem. 5:887-889.
Case, R. M. 1972. Energetic requirements for egg-laying
bobwhites. Proceedings of the First National Bobwhite Quail
Symposium. Stillwater, Oklahoma, April 23-26. pp. 205-212.
Chen, J. J. and R. L. Kodell. 1989. Quantitative risk
assessment for teratological effects. J. Amer. Stat. Assoc.
84:966-971.
Conover, W. J., M. E. Johnson and M. M. Johnson. 1981. A
comparative sutdy of tests for homogeneity of variances, with
application to the outer continental shelf bidding data.
Technometrics 23:351-361.
Cox, D.R. 1958. Planning Experiments. John Wiley and Sons, NY.
Custer, T. W. and G. H. Heinz. 1980. Reproductive success and
nest attentiveness of mallard ducks fed Aroclor 1254. Environ.
Pollution 21:313-318.
Dahlgren, R. B. and R. L. Linder. 1971. Effects of
polychlorinated biphenyls on pheasant reproduction, behavior and
survival. J. Wildl. Manage. 35:315.
Daniel, W. W. 1978. Applied Nonparametric Statistics. Houghton
Mifflin Company.
Ellman, G. L., K. D. Courtney, V. Andres, Jr. and R. M.
Featherstone. 1961. A new and rapid colorimetric determination
of acetylcholinesterase activity. Biochem. Pharmacol. 7:88-95.
Fairbrother, A., M. A. Craig, K. Walker and D. O'Loughlin. 1990.
68
-------
Changes in mallard (Anas platyrhvnchosl serum chemistry due to
age, sex, and reproductive condition. J. Wildl. Pis. 26:67-77.
Finney, D. J. 1971. Probit Analysis. Third Edition, Cambridge
University Press, London.
Finley, M. T. and R. C. Stendell. 1978. Survival and
reproductive success of black ducks fed methyl mercury. Environ.
Pollution 16:51-64.
Flegal, C. J. and C. C. Sheppard. 1976. Managing Gamebirds.
Extension Bulletin E-692, Cooperative Extension Service, Michigan
State University, East Lansing, Michigan. 15 pp.
Fox, G. A. and T. Donald. 1980. Organochlorine pollutants,
nest-defense behavior and reporductive success in merlins.
Condor 82:81-84.
Fox, G. A., A. P. Gilman, D. B. Peakall and F. W. Anderka. 1978.
Behavioral abnormalities of nesting Lake Ontario herring gulls.
J. Wildl. Manage. 42:477-483.
Franson, J. C., J. W. Spann, G. H. Heinz, C. Bunck and T. Lament.
1983. Effects of dietary ABATE on reproductive success, duckling
survival, behavior, and clinical pathology in game-farm mallards.
Arch. Environ. Contam. Toxicol. 12:529-534.
Friend, M. (ed.) 1987. Field guide to wildlife diseases. U. S.
Fish and Wildl. Serv. Resour. Publ. 167. 225 pp.
Fyfe, R. W., R. W. Risebrough and W. Walker II. 1976. Pollutant
effects on the reproduction of prairie falcons and merlins of the
Canadian prairies. Can. Field-Nat. 90:346-355.
69
-------
Gile, J. D. and S. M. Meyers. 1986. Effect of adult mallard age
on avian reproductive tests. Arch. Environ. Contain. Toxicol. 15:
751-756.
Glaser, R. E. 1982 Encyclopedia of Statistical Sciences,. Volume
4. S. Kotz and N. L. Johnson, Editors-in Chief. John Wiley and
Sons, NY.
Greenwood, R. J. 1975. Reproduction and development of four
mallard lines. Prairie Natural. 7: 9-16.
Grue, C. E., W. J. Fleming, D. G. Busby and E. F. Hill. 1983.
Assessing hazards of organophosphate pesticides to wildlife.
Trans. N. Amer. Wildl. Nat. Res. Conf. 48:200-220.
Grue, C. E., G. V. N. Powell and M. J. McChesney. 1982. Care of
nestlings by wild female starlings exposed to an organophosphate
pesticide. J. APP!. Ecol. 19:327-335.
Haegele, M. A. and R. H. Hudson. 1973. DDE effects on
reproduction of ring doves. Environ. Pollution 4:53-57.
Haegele, M. A. and R. K. Tucker. 1974. Effects of 15 common
environmental pollutants on eggshell thickness in mallards and
coturnix. Bull. Environ. Contain. Toxicol. 11:98-102.
Baseman, J. K. and L. L. Kupper. 1979. Analysis of dichotomous
response data for certain toxicological experiments. Biometrics
35:281-293.
Haseman, J. K. and E. R. Scares. 1976. The distribution of
70
-------
fetal death in control nice and its implication on statisical
tests for dominant lethal effects. Mutation Res. 41:277-288.
Haseltine, S. D., M. T. Finley and E. Cromartie. 1980.
Reproduction and residue accumulation in black ducks fed
toxaphene. Arch. Environ. Contam. Toxicol. 9:461-471.
Heath, R. G., J. W. Spann and J.R. Kreitzer. 1969. Marked DDE
impairment of mallard reproduction in controlled studies. Nature
224:47-48.
Heinz, 6. H. 1974. Effects of low dietary levels of methyl
mercury on mallard reproduction. Bull. Env. Contam. Toxicol. 11:
386-392.
Heinz, G. H. 1976a. Methylinercury: second year feeding effects
on mallard reproduction and duckling behavior. J. Wildl. Manage.
40:82-90.
Heinz, G. H. 1976b. Behavior of mallard ducklings from parents
fed 3 ppm DDE. Bull. Env. Contam. Toxicol. 16:640-645.
Henny, C. J. and J. K. Bennett. 1990. Comparison,of breaking
strength and shell thickness as evaluators of white-faced ibis
eggshell quality. Environ. Toxicol. Chem. 9:797-805.
Hickey, J. J. and D. W. Anderson. 1968. Chlorinated
hydrocarbons and eggshell changes in raptorial and fish-eating
birds. Science 162:271-273.
Holmes, W. N., K. P. Cavanaugh and J. Cronshaw. 1978. The
71
-------
effects of ingested petroleum on oviposition and some aspects of
reproduction in experimental colonies of mallard ducks (Anas
platvrhynchos). J. Reprod. Fertil. 54:335-348.
Hughes, B. O. and A. J. Black. 1976. The influence of handling
on egg production, eggshell quality and avoidance behavior of
hens. Br. Poult. Sci. 17:135-144.
Hunt, J. R., P. W. Voisey and B. K. Thompson. 1977. Physical
properties of eggshells: A comparison of the puncture and
compression tests for estimating shell strength. Can. J. Anim.
Sci. 57:329-338.
Jefferies, D. J. 1971. Some sublethal effects of p,p'-DDT and
its metabolite p,p'-DDE on breeding passerine birds. Meded.
Fakult. Landbouwwetenschappen. Gent. 36:34-42.
King, J. R. 1973. Energetics of reproduction in birds. In D.
S. Earner (ed), Breeding Biology of Birds. National Academy of
Science, Washington, D.C. pp. 78-107.
King, K. A., D. H. White and C. A. Mitchell. 1984. Nest defense
behavior and reproductive success of laughing gulls sublethally
dosed with parathion. Bull. Environ. Contain. Toxicol. 33:499-
504.
Kirkpatrick, C. M. 1955. Factors in photoperiodism of bobwhite
quail. Phvsiol. Zool. 28:255-264.
Kirkpatrick, C. M. 1959. Interrupted dark period: tests for
refractoriness in bobwhite quail hens. In R. B. Withrow (ed),
Photoperiodism. Amer. Assoc. Advan. Sci. Publ. No. 55,
Washington, D.C. pp. 751-758.
72
-------
Kupper, L. L., C. Portier, M. D. Hogan, and E. Yamamoto. 1986.
The impact of litter effects in dose-response modeling in
teratology. Biometrics. 42:85:98.
Levene, H. 1960. Robust tests for equality of Variances. In I.
Okin, S. G. Shurye, W. Hoeffding, W. 6. Madow, and H. B. Mann
(eds.), Contributions to Probability and Statistics. Stanford
University Press, Stanford.
Longcore, J. R. and J. R. Samson. 1973. Eggshell breakage by
incubating black ducks fed DDE. J. Wildl. Manage. 37:390-394.
Lundholm, C. E. 1984. Effect of DDE on the Ca metabolism of the
duck eggshell gland and its subcellular fractions; relations to
the functional stage. Comp. Biochem. Physiol. 78C:5-12.
McArthur, M. L., G. A. Fox, D. B. Peakall and B. J. R. Philogene.
1983. Ecological significance of behavioral and hormonal
abnormalities in breeding ring doves fed an organochlorine
mixture. Arch. Environ. Contain. Toxicol. 12:343-353.
McCullagh, P. and J. A. Nelder. 1989. Generalized Linear
Models. Second Edition. Chapman and Hall, NY.
McLane, M. A. R. and D. L. Hughes. 1980. Reproductive success
of screech owls fed Aroclor 1248. Arch. Environ. Contam.
Toxicol. 9:661-665.
Mendenhall, V. M., E. E. Klass and M. A. R. McLane. 1983.
Breeding success of barn owls fTvto alba) fed low levels of DDE
and dieldrin. Arch. Environ. Contam. Toxicol. 12:235-240.
Meyers, S. M. and J. D. Gile. 1986. Mallard reproductive
73
-------
testing in a pond environment: A preliminary study. Arch.
Environ. Contain. Toxicol. 15:757-761.
Miller, R. G. Jr. 1968. Jackknifing variances. Ann. Mathemat.
Stat. 39:567-582.
Miller, R. G. 1986. Beyond Anova. Basics of Applied Statistics.
John Wiley and Sons, NY.
I
National Institutes of Health. 1985. Guide for the Care and Use
of Laboratory Animals. National Institutes of Health, NIH Publ.
No. 85-23. 83 pp.
Peakall, D. B. 1970. Pesticides and the reproduction of birds.
Sci. Amer. 222:72-78.
Peakall, D. B. 1985. Behavioral responses of birds to
pesticides and other contaminants. Res. Reviews 96:45-77.
Peakall, D. B. and J. L. Lincer. 1972. Methyl mercury: Its
effect on eggshell thickness. Bull. Environ. Contain. Toxicol. 8:
89-90.
Peakall, D. B. and M. L. Peakall. 1973. Effect of a
polychlorinated biphenyl on the reproduction of artificially and
naturally incubated dove eggs. J. Appl. Ecol. 10:863-868.
Peakall, D. B., J. L. Lincer and S. E. Bloom. 1972. Embryonic
mortality and chromosomal alterations caused by Aroclor 1254 in
ring doves. Environ. Health Perspect. 1:103-104.
Peterson, R.G. 1985. Design and Analysis of Experiments. Marcel
Dekker, Inc., NY.
74
-------
Porter, R. D. and S. N. Wiemeyer. 1969. Dieldrin and DDT:
Effects on sprarrow hawk eggshells and reproduction. Science
165:199-200.
Prince, H. H., P. B. Siegel and G. W. Cornwell. 1969.
Incubation environment and the development of mallard embryos.
J. Wildl. Manage. 33:589-595.
Ratcliffe, D. A. 1967. Decrease in eggshell weight in certain
birds of prey. Nature 215:208-210.
Ratcliffe, D. A. 1970. Changes attributable to pesticides in
egg breakage frequency and eggshell thickness in some British
birds. J. APP!. Ecol. 7:67-107.
Rattner, B. A., L. Sileo and C. G. Scanes. 1982. Oviposition
and the plasma concentrations of LH, progesterone and
corticosterone in bobwhite quail (Colinus virqinianus) fed
parathion. J. Reprod. Fert. 66:147-155.
Roland, D. A., Sr. 1982. Relationship of body-checked eggs to
photoperiod and breaking strength. Poult. Sci. 61:2338-2343.
SAS Institute, Inc. 1985. SAS/STAT Guide for Personal
Computers. 6th ed., Gary, NC.
Shafer, E. W., Jr. and R. B. Brunton. 1979. Indicator bird
species for toxicity determinations: Is the technique usable in
test method development? In J. R. Beck, ed., Vertebrate Pest
Control and Management Materials. ASTM STP 680. American
Society for Testing and Materials, Philadelphia, PA, pp. 157-168.
Shorack, G. R. 1969. Testing and estimating ratios fo scale
parameters. J. Amer. Stat. Assoc. 64:999-1013.
75
-------
Snedecor, G.W. and W.G. Cochran. 1980. Statistical Methods,
Seventh Edition, Iowa State University Press.
Stromborg, K. L. 1981. Reproductive tests of diazinon on
bowhite quail. In D. W. Lamb and E. E. Kenaga, ed., Avian and
Mammalian Wildlife Toxicology; Second Conference. ASTM STP757.
American Society for Testing and Materials, Philadelphia, PA,
pp. 19-30.
Stromborg, K. L. 1986a. Reproduction of bobwhites fed different
dietary concentrations of an organophosphate insecticide,
methamidophos. Arch. Environ. Contain. Toxicol. 15:143-147.
Stromborg, K. L. 1986b. Reproductive toxicity of monocrotophos
to bobwhite quail. Poult. Sci. 65:51-57.
Talbot, C. J. and C. Tyler. 1974. A study of the progressive
deposition of shell in the shell gland of the domestic chicken.
Br. Poult. Sci. 15:217-224.
Thompson, B. K., A. A. Grunder, R. M. G. Hamilton and K. G.
Hollands. 1983. Repeatability of egg shell quality measurements
within individual hens. Poult. Sci. 62:2309-2314.
Tukey, J. W., J.L. Ciminera and J. F. Heyse. 1985. Testing the
statistical certainty of a response to increasing doses of a
drug. Biometrics 41, 295-301.
U. S. Environmental Protection Agency. 1982a. Pesticide
assessment guidelines. Subdivision E. Hazard evaluation:
Wildlife and aquatic organisms. EPA-540/9-82-024 (NTIS PB83-
76
-------
153908). Washington, DC.
U. S. Environmental Protection Agency. 1982b. Pesticide
assessment guidelines. Subdivision F. Hazard evaluation: Human
and domestic animals. EPA-540/0-82-025 (NTIS PB83-153916).
Washington, DC.
U. S. Environmental Protection Agency. 1982c. Pesticide
assessment guidelines. Subdivision N. Chemistry: Environmental
fate. EPA-540/9-82-021 (NTIS PB83-153-973). Washington, DC.
U. S. Environmental Protection Agency. 1986. Hazard Evaluation
Division Standard Evaluation Procedure: Ecological Risk
Assessment. EPA-540/9-85-001. Washington, DC.
Walpole, R.E. and R.H. Myers. 1972. Probability and Statistics
for Engineers and Scientists. The MacMillan Company, NY. pp. 387-
392.
Wedderburn, R. W. M. 1974. Quasi-likelihood Function,
generalized linear models and the Gauss-Newton method.
Biometrika 61:439-447.
Wetherill, G.B. 1981. Intermediate Statistical Methods. Chapman
and Hall, London.
White, D. H., C. A. Mitchell and E. F. Hill. 1983. Parathion
alters incubation behavior of laughing gulls. Bull. Environ.
Contam. Toxicol. 31:93-97.
Wiemeyer, S. N. and R. D. Porter. 1970. DDE thins eggshells of
American kestrels. Nature 227:737-738.
77
-------
Wilbur, K. M. and K. Simkiss. 1968. Calcified shells. In M.
Flockin, (ed.), Comprehensive Biochemistry,, Elsevier, Amsterdam,
Vol. 26, pp. 229-295.
Williams, D. A. 1975. The analysis of binary responses from
toxicological experiments involving reproduction and
teratogenicity. Biometrics 31:949-952.
78
-------
APPENDIX A
SUBDIVISION E GUIDELINES FOR CONDUCTING AN
AVIAN REPRODUCTION TEST
This appendix includes the test guidelines for conducting an
avian reproduction test (section 71-4) from the Subdivision E
guidelines (USEPA 1982a) on Hazard evaluation: Wildlife and
aquatic organisms. Other relevant information, not included in
the appendix, concerning the conduct of all tests in the guidance
document can be found in the introductory sections on General
information (section 70-1), Definitions (section 70-2), General
test standards (section 70-3), and Reporting and evaluation of
data (section 70-4).
79
-------
71-4 Avian reproduction test.
(a) When required. (1) Data on avian reproductive effects
are required by 40 CFR § 158.145 to support the registration of an
end-use product which meets one or more of the following criteriai
(1) Its labeling contains directions for using the product
under conditions where birds may be subject to repeated or continuous
exposure -to the pesticide or any -of its major metabolites or
degradation products, especially preceding or during the breeding
season.
(11) The pesticide or any of its major metabolites or
degradation products are stable in the environment to the extent
that potentially toxic amounts may persist in avian feed.
(ill) The pesticide or any of its major metabolites or
degradation products is stored or accumulated in plant or animal
tissues, as indicated by the partition coefficient of lipophilic
pesticides (${ 165-3, -4, and -5 of Subdivision N) metabolic release
and retention studies (f 85-1 of Subdivision P), or a* indicated by
structural similarity to known ttoaccumulative chemicals*
(iv) Any other information, such as that derived frcm mammalian
reproduction studies (| 83-4 of Subdivision F), that indicates the
reproduction in terrestrial vertebrates may be adversely affected
by the anticipated use of the pesticide product.
(2) Applicants for registration of avicides should consult •
with the Agency prior to conducting this test.
80
-------
(3) See 40 CFR J 156.50, •Poraulators' exemption," to determine
whether these data must be submitted. Section XX-A of this Sub-
division provides an additional discussion on this subject.
(b) Test standards. Data sufficient to satisfy the require-
ments in 40 CFR J 156.145 should be derived from tests which comply
with the general test standard* la J 70-3 and all of the following
test standards:
(1) Test substance. Data shall be derived from testing
conducted with the technical grade of each active ingredient la the
product*
(2) Species. Testing should be performed on the bobwhite
quail and mallard.
(3) Pose levels. At least two treatment level groups and a
vehicle control group should be used.
(4) Number of test animals, when other test data reveal
bioaccumulative potential, the number of test animals in the test
group should be increased sufficiently to partly offset animal
deaths or data-gathering problems associated with morbidity or with
tissue residue determinations.
(5) Age. Birds approaching their first breeding season should
be used.
(6) Duration of administration. Birds should be exposed to
treated diets beginning not less than 10 weeks before egg laying is
expected, and extending throughout the laying season.
(c) Reporting and evaluation of data. Xn addition to the
Information provided in § 70-4, the test report should contain:
(1) Test results. The following information, reported for
all test groups:
(i) All observed abnormal behavior;
(ii) All observed morphological and physiological responses)
(iii) Post mortem autopsy.
(2) Test conditions. The following information, reported
for each treated and untreated test group:
(i) Speciesi
(ii) Straini
(iii) Agei
(iv) Body weight!
81
-------
(v) Number of birds par test (include sex ratio)»
(vi) Individual identification of bird*;
(vii) Diet}
(viii) Storage}
(ix) Peed consumption (grama per day)i
(x) Observation on palatability or repellancyi
(xi) Bousing eonditiona of teat birda, including!
(A) Space allocatlona for mating and nesting;
(B) Protection from weather and injuries} and
(C) Lighting program, including hours per day and wattage
or foot candles at bird leveli
(xii) Diagram of test layout}
(xiii) Temperature}
(xiv) Hater supply}
(xv) pretest and test hiatory or medical and chemical
administration} and
(xvi) Length of treatment period and observation period.
(3) Egg and hatching data. The following information, reported
for each treated and untreated teat group:
(i) Egg shell thickness}
(ii) Number and percent of cracked eggs;
(iii) Eggs laid (number eggs per bird per day and per season)i
(iv) Batching egg storage datat
(*) Temperature}
(B) Buaidityt
(C) Incubation data}
(D) Eggs set} and
(E) Egg turning frequency}
82
-------
(V) Fertility (viable embryos);
(vi) Live 3-week embryosj
(vii) Babryos that nature, embryos that pip shell* and embryo*
that liberate themselves, end a determination of hatchabilityi
(viii) Dead embryos!
(ix) Fourteen-day-old survivors*
(x) Crippled survivorst
(xi) Post-hatching sort ability}
(xii) Heights of fourteen-day-old survivorsi and
(xiii) Any signs of intoxication in post-hatching survivors.
(4) Feed analysis data. Levels of concentration of
pesticide in the feed used in each test, and the rationale for
choice of such levels.
(d) Acceptable protocol. Except where noted, the following
example of avian reproduction protocol is acceptable for the testing
of both bobwhites and mallards. This study is a modification of a
study that appears on pages 23 to 50 in an unpublished draft report
to EPA from the American institute of Biological Sciences (AXES),
titled analysis of Specialized Pesticide Programs* Volume VI,
Wildlife Toxicology Study. The report is dated October, 1974, and
was funded under DA Contract Mo. 68-01-2457.
Test animals. Pen-reared birds, previously .
untreated, approaching their first breeding season,
and phenotypically indistinguishable from wild birds,
should be used as test animals. Zf shipped, all birds
should be examined following shipment for possible
physical injury that may have been encountered in
transit. Zf deemed necessary, several birds may be
randomly selected for pretreatment necropsy at a diag-
nostic laboratory to assess the state of health upon
arrival. Zt is desirable to have a 2- to 6-week health
observation period prior to selection of birds for
treatment.
83
-------
A history of rearing practice for the bird* to
be tested should be obtained if possible. This history
should include lighting practices during rearing,
disease record, drug and any other medication adminis-
tered, and exact age.
Test groups - Bobwhite. * minimum of 3 test
groups of bobwhite should be used. One group should
serve as a control and 2 groups as treated birds. By
random distribution, 1 male and 2 females per pen,
replicated by a minimum of 12 pens, should be used per
group. If individual pairs (1 male and 1 female) are
to be used per pen, more pens (greater than 12) per
teat group should be used to proide similar sensitivity
to the group testing design. To determine the Bomber
of pens needed for a particular level of sensitivity,
see Waipole and Myers (1972). Control and treated
birds should be kept under the same experimental con-
ditions.
Test groups - Mallards. A minimum of 3 test
groups of mallards should be used. One group should
serve as a control and 2 groups as treated birdas. By
random distribution, 2 males and 5 females per pen,
replicated by 5 or more pens, should be used per group.
Zf individual pairs (1 male and 1 female) are to be
used per pen, considerably more pens (greater than 12
per test group should be used to provide similar sen-
sitivity to the group testing design. To determine
the number of pens needed for a particular level of
sensitivity, see Wai pole and Myers (1972). Control
and treated birds should be kept under the same experi-
mental conditions.
Diet preparation. Concentrations for the test
substance should be based on measured or calculated
residues expected in the diet from the proposed use
pattern(s). The concentrations should include an
actual or expected field residue exposure level and a
multiple level such as five. The highest nonlethal
level may be estimated from data developed from the
avian dietary USD (f 71-2).
The test material should be added to table grade
corn oil or other appropriate vehicle and premixed
with an aliquot of basal diet, utilising a mortar and
peetle or mechanical blender. Zt is recommended that
the aliquot of basal diet used for the premix be
screened to remove large particles of diet before
blending in the corn oil and test material. The final
diet should be a uniformly mixed composition consisting
of 98 to 99 parts by weight of basal diet and 1 or 2
84
-------
parti by weight of corn oil. The basal diet Bhould be
• commercial game bird breeder ration (or its equivalent)
that is treated with an equivalent amount of vehicle.
The premix should be stored under conditions which
maintain stability. Test diets should be analysed for
pesticide concentrations at intervals during the tests.
If other long-term animal tests have demonstrated a
propensity for the test chemical to persist or bioac-
cumulate, the degree .of bioaccumulation in birds should
be determined by measurement of tissue residues in the
birds from an extra pen group put through the reproduc-
tion test. Two or three tissues should be selected
for residue analysis at the end of the exposure period,
based on tissues known from other studies to hold
highest residues.
Testing phase - test environment. The birds
should be housed in breeding pens of adequate size
conforming to good husbandry practices. The mallard
pens should be screen-bottomed or kept clean of
spilled food and excrement. Zt is desirable to offer
mallards water in which to bathe.
Since light is extremely important, both during
rearing and during the egg laying period, all birds
should be maintained for the first 8 weeks under a
regime of 7 hours of light per day for maximum egg
production.
The photoperiod should then be increased to
16-17 hours of light per day and either maintained
at this level or increased by 15 minutes per week
for the following 12 weeks. (The 12-week period
may vary depending upon the time required for the
onset of egg production.) An illumination intensity
of 6 footcandles at the bird level during the lighting
phase of the reproductive study is adequate. Avoid
the use of shorter wavelength 'cool white" fluorescent
lights which do not emit the daylight spectrum.
Temperature and relative humidity control through-
out the reproductive test is desirable end should be
recorded. Recommended levels are 21*C and 55 percent
relative humidity. Ventilation is necessary.
Feeding and husbandry. All birds should receive
the appropriate diet ad libitum for the duration of
the study. Hater is to be provided ad libitum. The
test chemical should be administered for at least 10
weeks prior to the onset of egg laying.
85
-------
Body weights should be recorded at test Initiation
prior to onset of laying, and at termination. Daring
egg laying, body weight recording is discouraged because
of the adverse effects that handling Bay have on egg
production*
Pood consumption should be recorded at least
•t biweekly intervals throughout the study.
Mortality should be recorded by date and morbidity
(noted together with clinical signs) throughout the
test phase. Cross pathology data should be obtained
for birds that die during the course of the test phase
and for some survivors.
Egg collection, storage, and incubation. All egg*
should be collected daily, marked according to pen
from which collected, and stored at 16*C and 65 percent
relative humidity. Eggs should be set at weekly
intervals for incubation in a commercial incubator.
All eggs should be candled on day 0 for eggshell
cracksi on approximately day 11 for bobwhites and
day 14 for mallards to measure fertility and early
death.pf.embryos) and on day 18 for bobwhite and
day 21 for mallards to measure embryo survival. For
hatching, transfer of the eggs to a separate
commercial incubator or hatcher should be made on
day 21 for bobuhites and Day 23 for mallards.
Recommended temperatures and relative humidity
during hatching phase are 39*C and 70 percent,
respectively.
Bobwhite chick observations. On Day 24 of
incubation, the hatched bobwhite chicks should be
removed, hatchability recorded, chicks housed according
to the appropriate parental grouping, and maintained
on control diet for 14 days. The time period should
be extended if mortality occurs appreciably late.
The diet should be a commercial.bobwhite starter
diet or its equivalent.
Duckling observation. On Day 27 of incubation,
the hatched mallard ducklings should be removed,
hatchability recorded, ducklings housed according
to the appropriate parental grouping, and maintained
on control diet for 14 days. The time period should be
extened if mortality occurs appreciably late. The
diet should be commercial mallard starter diet or its
equivalent.
86
-------
Eggshell thickness. One day every two weeks newly
laid eggs should be collected and neasured for eggshell
thickness. For consistency, the eggs used for thickness
determinations should be collected during weeks 1, 3, 5,
7 and 9 of the egg-lay ing period. An accepted
procedure is to crack open the eggs at the widest
portion (girth or waist), wash out all egg contents,
air-dry the shell* for at least 48 hours, and then
measure the thickness of the dried shell plus the
membranes st 3 or 4 points sxound the girth using •
micrometer calibrated to 0*01 SOB units.
Analysis. Reproductive data consists of continuous
variables (e.g., shell thickness, and body weight data)
and discrete variables (e.g., number of eggs laid or
14-day-old survivors). For continuous variables,
experimental groups should be compared to controls by
analysis of variance. For most discrete variables,
survival percentages should be computed (e.g., 14-day-
old survivors of eggs laid) and arcsine transformed
prior to analysis of variance. Alternately, a chi
square analysis of survival (contingency tables) may
be used for discrete variables. Analyses should include:
body weight, food consumption, eggs laid, eggshell thick-
ness, eggs cracked, viable eggs, fertility, live 3-week
embryos, hatchability, number of normal chicks or
ducklings, 14-day-old survivors (per number of eggs
hatched, per hen, and per number of eggs laid). Sample
units are generally the pens within each group.
Withdrawal. Zf the test substance is toxic (re-
duced reproduction evident), then a withdrawal study
period should be added to the test phase. The withdrawal
period need not exceed 3 weeks. Continued observations
should be made on egg production, fertility, hatchability,
and hatching survival.
Definitions;
1• Eggs laid. The total egg production during a
breeding season (which is approximately 10 weeks)*
2* Eggs cracked. Xggs determined to have cracked
shells when inspected with a candling lamp) fine cracks
cannot be detected without utilising a candling lamp
and if undetected will bias data by adversely affecting
embryo development.
3. Bggs set. All eggs placed under insubation, i.e.,
total eggs laid minus cracked eggs and those selected for
eggshell thickness analysis.
87
-------
4. Viable embryos (fertility). Eggs in which
fertilization has occurred and embryonic development has
begun. This is determined by candling the eggs 6 to
14 days after Incubation has begun. It is difficult
to distinguish between the absence of fertilisation
and early embryonic death. This distinction can be
made by breaking out eggs that appear infertile and
examining further. This is especially important when
a test compound induces early embryo mortality.
5. Live 3-week embryo. Embryo that is developing
normally after 3 weeks of incubation. This is determined
by candling the *gg.
6. Hatehability. The percentage of embryos that
mature, pip the shell, and liberate themselves from their
•ggs as computed from the number of fertile eggs. For
quail this generally occurs on day 23 or 24 of incubation,
and for mallard on day 25, 26, or 27.
7. 14-dav-eld-survivers. Birds that survive for
2 weeks following hatch.
8. Eggshell thickness. The thickness of the
shell and the membrane of the egg at the girth after the
•gg has been opened and washed out, then the shell with
membrane dried for at least 4B hours at room temperature.
(e) References. The following references can provide useful
background information in developing acceptable protocols) some
outline useful statistical procedures for handling data.
(1) Cochran, W.C. 1943. Analysis of variance for percentages
based on unequal numbers. Am. Stat. Assoc. 36*287-301.
(2) Davidson, JC.L., and J.L. Sell. 1974. DDT thins shells
of eggs frcn mallard ducks maintained on ad libitum or controlled-
feeding regimes. Arch, environ. Contain. Toxieol. 2(3)i222-232.
(3) Duncan, D.B. 1955. Multiple range and multiple P tests.
Biometrics 11i1-42.
(4) Heath, R.G., J.W. Spann, and J.r. Ereitzer. 1969.
Marked DDE impairment of mallard reproduction in controlled studies.
Hature 224(5215)t47-48.
(5) Beath, R.C., J.W. Spann, J.R. Xreitzer, and C. Vance.
1970. Effects of polychlorinated biphenyls on birds* Presented at
the XV Internal. Ornith. Congress, The Bague, 30 Aug - 5 Sept.,
1970. Pp. 475-485 in Proceedings of XV Interaat. Ornith. Congress.
K.H. Voous, ed. S.J. Brill, (pub.) Leiden.
-------
(6) Heinz, C. 1974. Effect! of dietary levels of Methyl
mercury on mallard reproduction. Bull. Inv. Cent. Toxieol. 11«386-
392.
(7) tongcore, J.R., P.B. Sampson, and T.W. Mhittendale, Jr.
1971. DOE thin* eggshells and lowers reproductive success of
captive black ducks. Bull. Env. Cont. Toxieol. 6t465-490.
(8) Prince, B.H., P.B. Seigel, and C.W. Cornwall. 1969.
Incubation environment and the development of mallard embryos.
J. Wildlife Manage. 33:589-595.
(9) Stronborg, X.L., 1981. Reproductive test of diazinon on
bobwhite quail, Avian and Mammalian Vildlife Toxicologyi Second
conference, ASTM BTf 757, D.V. Laab and X.Z. Xanaga, Bds., American
Society for Testing and Materials, pp. 19-30.
(10) Halpole, R.E. and R.H. Myers. 1972. Probability and
statistics for engineers and scientists. The MacMillan Company,
Mew York. Pp. 387-392.
89
-------
APPENDIX B
SELECTED METHODS FOR EXPERIMENTAL DESIGN AND ANALYSIS
RANDOMIZATION
The proper use of randomization in a designed experiment
removes the potential for systematic biases in the estimation of
treatment effects. Randomization usually means assigning
individuals to groups such that every individual is equally
likely to fall into any group. Note however, that other
randomization schemes that assign individuals to groups with
unequal probabilities are also possible. The alternative to
randomization is the assignment of individuals to groups based on
an ordering that appears to be haphazard to the scientist. Cox
(example 5.6, 1958) discussed this method and how its use can
invalidate the results of designed experiments. Correct
randomization procedures involve the use of an algorithm that has
been scientifically shown to choose individuals with known,
usually equal, probability.
One method of randomly assigning individuals to groups is to
number the individuals from 1 to n (where n is the total number
of individuals) and then, using a random number table, pick off a
sequence of the numbers from 1 to n in the order given by the
table. For example, given 5 individuals numbered 1 through 5,
obtain from the random number table the random order 3, 2, 5, 1
and 4. Then individual 3 goes in the first pen, individual 2
goes in the second pen, individual 5 goes in the third pen and so
on. This allocation plan can be used to assign birds in age
classes to pens or pens to treatment groups (diets). Another
method of obtaining random order would be to use a computer
program with a good random number generator to generate
orderings.
90
-------
INVESTIGATING DISTRIBUTIONS OF RESPONSE VARIABLES
The results of a single avian reproduction test are assumed
to be representative of results that would be obtained if all
birds ever exposed to the pesticide were examined and the true
distribution of the responses could be examined. However, since
it is not possible to examine all birds, the distributions of the
responses from only the birds in the test are used. A
statistical analysis will attempt to describe the distributions
of the response and the extent to which the results of the test
might vary if different birds or pens had been used.
The determination of significant treatment differences is
made by examining the distributions of responses obtained from
different pens for different treatment groups and asking if the
distributions appear to be "different" from each other in the
face of the variability in response that is observed. For
example, an analysis of variance will decide whether two or more
distributions are centered at the same value assuming that the
distributions have the same variance.
Usually, distributions are summarized with an average (or a
median) to describe the most likely response and a standard
deviation (or range or interquartile range) to describe the
amount of variability in the response. The average and standard
deviation are the theoretical best summary statistics for the
normal or bell-shaped distribution and since normal distributions
are quite common in biological studies, the use of these
statistics has become almost routine. However, it is important
to remember that some responses may have distributions that are
not bell-shaped and in these cases other types of summary
statistics, such as the median or interquartile range, can
91
-------
provide better summaries of the distribution of the response.
In any study, it is important to visually examine the
distribution of the responses for each treatment group (Box,
Hunter and Hunter, p 193, 1978). In addition, summary statistics
can be calculated, but only by looking at the distributions of
the observed responses can it be known what the means, medians or
standard deviations are attempting to describe. Histograms,
frequency charts, stem and leaf plots, box plots and scatter
plots of the data are suggested methods for looking at
distributions of data. These plots can be generated by SAS's
PROC UNIVARIATE. Box plots, generated by PROC UNIVARIATE and the
plot option, with means superimposed on them, are especially
useful for evaluating whether or not unusual points (outliers)
are present. In addition, these plots can be used to decide
whether or not variability is similar between groups and whether
it is likely that the data come from an approximately normal
distribution or whether they come from a distribution that is
quite un-normal.
The analysis of variance is relatively robust to deviations
from normality and so small amounts of skewness can be tolerated
(Miller, 1986). However, ANOVA is not robust to deviations from
homogeneity of variance and it is important to be aware of this
type of deviation in the data. Visual examination of the data
allows the scientist to be aware of specific strengths and
weaknesses in the data.
A hypothetical dataset, given in Table 1, will be used to
illustrate how exploratory data analysis techniques can be used
to validate the use of particular statistical tools. This
dataset is based on a hypothesized avian reproduction study
involving 3 dietary concentrations (Ctl, Lo and Hi), 1 hen per
pen, 15 pens per conentration group and it supposes that the
92
-------
study is carried out for 10 weeks (70 days).
Figure B-l is a frequency chart for each dietary
concentration of the number of fertile eggs with the average,
median and standard deviation superimposed on it. It is clear
from the diagram that even if the responses from each dietary
concentration came from normal distributions, they appear to have
very different variances. Levene's test for homogeneity of
variance (described below) was conducted using the absolute value
of the residuals, rejected the hypothesis of equal variance with
a p-value of 0.0063. Hence, using ANOVA on this data may not be
appropriate since the assumption of equal variance does not seem
to be justified. In addition, there is some evidence from a
visual inspection of the plot that the distributions may not be
normal. The test for normality, from PROC UNIVARIATE, gave p-
values of 0.0001, 0.0419 and 0.1486 for the Ctl, Lo and Hi groups
respectively. However, the issue of non-constant variance is of
more importance since ANOVA is relatively robust to this
departure from the assumptions.
Figure B-2 is a similar chart for the eggshell thickness
data. The variability within each dietary concentration group
appears, from the plot, to be about the same. Levene's test
resulted in a p-value of 0.8705 indicating that there is no
evidence to reject the hypothesis of equal variances. The p-
values for the tests of normality were 0.5832, 0.4754 and 0.4044
for the Cnt, Lo and Hi groups respectively. Visual examination
of Figure 2 shows that the responses appear to come from normal-
like distributions. Thus there is no evidence to believe that an
ANOVA should not be used.
This first step in any data analysis is relatively easy to
do and yet it can provide more insight than many other
sophisticated analyses. The generation of histograms, or similar
93
-------
plots, to check assumptions for future statistical analysis
should be incorporated into the statistical protocol.
LEVENE'S TEST
Since Bartlett's test for homogeneity of variance is very
sensitive to deviations from normality (Miller, 1986), Levene's
test (Levene, 1960) is a robust alternative. It is easily
programed in SAS as well. Levene's test is a statistical test of
the hypothesis that the variance is constant in all of the
treatment groups.
To carry out the test in an avian reproduction study where
the treatment groups are different dietary concentrations of
pesticide, run the appropriate ANOVA on the data and obtain the
residuals from this analysis. Then square the residuals and
treat these squared residuals as if they are independently,
identically normally distributed (under the null hypothesis of
constant variance) and apply the usual ANOVA F test to them. -
That is, carry out an ANOVA (using a model similar to the one
used on the original data) on the squared residuals and test the
hypothesis that the means of the residuals are all equal.
As discussed in Miller (1986), these squared residuals do
not satisfy the assumptions imposed upon them by the test.
However, as demonstrated by Levene (1960), Miller (1968), Shorack
(1969) and Brown and Forsythe (1974) the test preforms well and
is robust to departures from normality. A simple alternative,
less sensitive to heavy tailed distributions of the original
data, is to use the absolute value of the residual, rather than
the squared residual.
A simulation study of the performance of more than 50
proposed homogeneity of variance tests as well as a bibliography
94
-------
of these tests is given by Conover et al. (1981).
A sample SAS program that carries out Levene's test using
the absolute value of the residuals, for a one way model, is
given below.
DATA ONE;INFILE 'TEST.DAT1;
INPUT EGGS CHICKS TRT;
RUN;
PROC GLM; CLASS TRT;
MODEL CHICKS=TRT;
OUTPUT OUT=LEVENE R=RESID;
RUN;
DATA TWO;SET LEVENE;
ABSRES=ABS(RESID);
RUN;
PROC GLM DATA=LEVENE;CLASS TRT;
MODEL ABSRES=TRT;
RUN;
NON-PARAMETRIC STATISTICS
When the ANOVA assumptions of normality and constant
variance cannot be justified by the data nor by a transformation
t
of the data, then one alternative is to use non-parametric
statistics. Daniel (1978) is a good reference for such
techniques since assumptions inherent in these analyses and
examples using real data are provided. SAS, in PROC NPAR1WAY,
provides non-parametric analyses for one-way classifications of
data.
It is important to know when nonparametric techniques are
appropriate. First of all, because ANOVA is relatively robust to
i
departures from normality, it is best to use ANOVA when the
variance seems to be constant between treatment groups and there
is only slight evidence that the distribution of the response is
95
-------
not symmetric (Miller, 1986). But if it is evident that the
response distribution is exceedingly non-symmetric then non-
parametric statistics may be used. Secondly, some non-parametric
tests assume that the distribution of the responses are
identical, except for a difference in mean or median. Thus,
responses for which the variability is different for different
treatment groups should not be analyzed by these methods. The
Kruskal-Wallis one way analysis (given by PROC NPAR1WAY) is one
example. However, the k-sample median test in PROC NPAR1WAY only
assumes that the probability of being larger than the overall
median is the same for all treatment groups (Daniel, 1978). For
some cases of heterogeneous variance, this may not be a problem.
Thus, although non-parametric methods make less assumptions
about the distribution of the response, they still require that
the response behave in certain prescribed ways. As for ANOVA, it
is best to be certain that the data satisfy whatever assumptions
ts
are inherent in the analysis.
96
-------
TABLE B-l. SELECTED RESULTS FROM A HYPOTHETICAL AVIAN
REPRODUCTION STUDY
Dietary
CONC.
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Hi
Hi
Hi
Hi
Hi
Hi
Hi
Hi
Hi
Hi
Hi
# Eggs
FERTILE
53
54
54
54
55
55
56
56
56
56
56
57
57
57
58
47
48
49
49
50
50
51
52
52
53
54
55
56
57
57
37
38
40
41
41
41
41
42
42
44
45
(continued)
Eggshell
THICKness
0.24486
0.24048
0.23870
0.22711
0.24954
0.25704
0.21630
0.23669
0.23876
0.23382
0.23434
0.23987
0.24373
0.25232
0.25177
0.22984
0.25984
0.23661
0.23736
0.23780
0.25032
0.24532
0.24840
0.24282
0.24862
0.24724
0.23867
0.23219
0.21183
0.25195
0.19182
0.20519
0.19119
0.21250
0.19215
0.20661
0.19116
0.20605
0.21265
0.19921
0.20144
97
-------
Table B-l. (continued)
Dietary i Eggs Eggshell
CONC. FERTILE THICKness
Hi 46 0.1978
Hi 46 0.19135
Hi 47 0.19420
Hi 48 0.17650
98
-------
\o
10
fi
-------
HIGH
xs.20
s = .0097
c
0>
3
cr
o>
o
o
.18 .20 .22 .24
Eggshell Thickness (mm)
n .
n
.26
LOW
xs.24
s «.0115
CONTROL
x = .24
s».0104
Rg. B-2 Frequency chart, summary statistics and 95% confidence intervals
for eggshell thickness, by treatment group.
-------
APPENDIX C
POTENTIAL EFFECTS OF CRACKED EGGS ON THE DETERMINATION
OF PESTICIDE-RELATED EFFECTS
The number of cracked eggs in an avian reproduction study is
often obtained and analyzed as an indicator or pesticide related
effects or as a measure of the quality of the study. There have
been many discussions concerning its usefulness and applicability
as a valid endpoint. It should be recognized that eggshell
cracking may be due to a variety of causes, including technical
aspects such as pen construction, animal husbandry techniques
such as handling of the birds, pesticide related effects or to
characteristics of individual birds. It is this latter cause
which has been observed to lead to problems in the analysis of
other endpoints such as the number of 14 day old chicks.
It has been observed by one author that some hens are more
likely, perhaps due to behavioral characteristics, to have
clutches of eggs that contain cracked eggs than other hens. Thus
in a study one may find hens with clutches that contain a large
proportion of cracked eggs and other hens whose clutches contain
no cracked eggs. Even if this pattern of cracked eggs is not
related to the effects of the pesticide it can lead to incorrect
conclusions concerning pesticide effects.
A hypothetical example, given in Table 2, was devised to
illustrate the problem that cracked eggs may have on the
detection of pesticide related effects on the number of 14 day
old chicks. The example was designed to represent a hypothetical
avian reproduction study involving bobwhite quail over a period
of 10 weeks (70 days). In this example, it is theorized that
cracking is not due to the pesticide treatment. The incidence of
cracked eggs is the same in all of the treatment groups.
101
-------
However, the presence and pattern of cracked eggs will make
detecting the true effect of the pesticide more difficult. In
order to provide a relatively simple example, endpoints other
than the number of eggs laid, the number of viable embryos and
the number of live 14 day old chicks are not included.
The hypothetical study from which these data arose had one
hen per pen, 15 pens per dietary concentration group and 3
different dietary concentrations; a control (Ctl), a low
concentration (Lo), and a high concentration (Hi). For each
concentration group, the number of eggs laid was generated as a
binomial random variable with a binomial sample size of 70. For
the control group, the probability of laying an egg on any given
day was set equal to 0.80. For the low group, the probability
was also 0.80, but for the high concentration group, the
probability was 0.60. Thus on average, the control and low
concentration groups would lay about 56 eggs over the course of
the study and the high concentration group would lay about 42
eggs.
It was decided to have a study-wide cracking rate of about
10%. However the pattern of cracking is crucial. This pattern
reflects a typical pattern observed by one of the authors. For
each dietary concentration group, 20% of the pens in each group
have a pen-wide cracking rate of about 50%. The remaining pens
have no cracked eggs.
The number of viable embryos in the test is calculated as
the number of eggs laid minus the cracked eggs. The number of 14
day old chicks was generated as a binomial random variable with
the binomial sample size equal to the number of viable embryos.
For the control group, the probability that an embryo develops
into a live chick is 0.65. For the low concentration group, the
probability of an embryo developing into a live chick is 0.55,
102
-------
and for the high concentration group the probability is 0.40.
Thus, the dataset is generated to reflect a pesticide
induced effect in the number of eggs laid in only the high
concentration group, and an effect in the number of 14 day old
chicks in both the high and low dietary concentration groups.
The cracking rate is not affected by the pesticide, that is, the
same cracking rate is seen in all groups.
When the number of eggs laid (LAID), viable embryos
(EMBRYOS) and 14 day old chicks (CHICKS) are analyzed using a
one-way ANOVA, a significant difference is detected between the
high concentration group and control but not between the low
concentration group and control That is, the analysis detected
the true pesticide effect in the number of eggs laid, but failed
to detect the difference between the control and low
concentration groups for the number of 14 day old chicks.
A hypothetical scenario in which no cracking occurs is also
simulated. The variable CHICK2 is generated as a binomial random
variable with the binomial sample size equal to the number of
eggs laid (and not the number of viable embryos). When CHICK2 is
analyzed using the one-way ANOVA, a significant difference is
detected between the low concentration and the control group and
the high concentration and the control group.
That is, in the absence of the cracked eggs, the true effect
of the pesticide is detected. However, when cracked eggs are
present, the effects is masked. It should be noted that this
phenomenon is due to the distribution of cracking, i.e., 50% in a
few pens and 0% in other pens. If cracking occurs at the same
rate in all pens this effect would not occur. In addition, the
cracking in this example was hypothesized to be independent of
treatment. There may be cases where cracking is due to the
103
-------
pesticide treatment or to a combination of husbandry/technical
and pesticide effects. In these cases it is not easy to predict
the effect on the analysis.
104
-------
TABLE C-l. SELECTED RESULTS FROM A HYPOTHETICAL AVIAN
REPRODUCTION STUDY
CONC.
group
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Ctl
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Lo
Hi .
Hi
Hi
Hi
Hi
Hi
Hi
Hi
Hi
Hi
Hi
# eggs
LAID
54
53
55
58
49
54
53
55
58
53
57
57
57
52
48
51
58
54
54
54
48
55
58
53
62
56
56
60
56
60
42
41
41
47
36
42
42
44
43
40
41
i
CR
0
0
27
29
0
0
0
0
0
26
0
0
0
0
0
26
29
0
0
0
0
0
0
0
0
0
0
30
0
0
0
0
0
23
0
21
0
0
0
0
0
eggs # viable
ACKED EMBRYOS
54
53
28
29
49
54
53
55
58
26
57
57
57
52
48
25
29
54
54
54
48
55
58
53
62
56
56
30
56
60
42
41
41
24
36
21
42
44
43
40
41
(continued)
# 14 day
CHICKS
39
35
19
20
33
32
31
40
37
16
35
37
36
29
32
12
18
28
34
26
33
30
37
27
34
23
28
17
30
31
17
18
13
9
14
9
23
22
17
14
15
# 14 day
CHICKS 2
34
37
40
36
29
37
25
35
40
30
33
34
42
34
31
23
30
30
29
34
23
32
33
30
27
35
30
30
22
32
21
15
10
18
16
14
21
18
15
13
26
105
-------
Table C-l. (continued)
CONC. # eggs # eggs # viable # 14 day # 14 day
group LAID CRACKED EMBRYOS CHICKS CHICKS2
Hi 45 22 23 4 19
Hi 40 0 40 15 20
Hi 42 0 42 18 16
Hi 47 0 47 15 17
106
------- |