Retrospective Theses and Dissertations
2005
Mapping quantitative trait loci for economic traits
in chickens
Joseph Paul McElroy
Iowa State University
Follow this and additional works at: http://lib.dr.iastate.edu/rtd
Part of the Agricultural Science Commons, Agriculture Commons, Agronomy and Crop
Sciences Commons, Genetics Commons, and the Veterinary Pathology and Pathobiology
Commons
Recommended Citation
McElroy, Joseph Paul, "Mapping quantitative trait loci for economic traits in chickens " (2005). Retrospective Theses and Dissertations.
Paper 1583.
This Dissertation is brought to you for free and open access by Digital Repository @ Iowa State University. It has been accepted for inclusion in
Retrospective Theses and Dissertations by an authorized administrator of Digital Repository @ Iowa State University. For more information, please
contact [email protected].
Mapping quantitative trait loci for economic traits in chickens
by
Joseph Paul McElroy
A dissertation submitted to the graduate faculty
in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
Major: Genetics
Program of Study Committee:
Jack C. M. Dekkers, Co-major Professor
Susan J. Lament, Co-major Professor
Rohan L. Fernando
Michael Lee
Max F. Rothschild
Iowa State University
Ames, Iowa
2005
Copyright © Joseph Paul McElroy, 2005. All rights reserved.
UMI Number: 3184641
Copyright 2005 by
McElroy, Joseph Paul
All rights reserved.
INFORMATION TO USERS
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
UMI
UMI Microform 3184641
Copyright 2005 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
ii
Graduate College
Iowa State University
This is to certify that the doctoral dissertation of
Joseph P. McElroy
has met the dissertation requirements of Iowa State University
Signature was redacted for privacy.
Co-major Professor
Signature was redacted for privacy.
Co-major Professor
'
Signature was redacted for privacy.
For the Major Program
iii
TABLE OF CONTENTS
LIST OF FIGURES
iv
LIST OF TABLES
v
ABSTRACT
vi
CHAPTER 1
GENERAL INTRODUCTION
CHAPTER 2A MOLECULAR MARKERS ASSOCIATED WITH
1
53
GROWTH AND CARCASS TRAITS IN MEAT-TYPE
CHICKENS
CHAPTER 2B TRAIT LOCI AFFECTING WHITE MEAT PERCENT
60
AND OTHER GROWTH AND CARCASS TRAITS IN
COMMERCIAL BROILER CHICKENS
CHAPTER 3
MICROSATELLITE MARKERS ASSOCIATED WITH
94
RESISTANCE TO MAREK'S DISEASE IN LAYER
CHICKENS
CHAPTER 4
COMPARISON OF METHODS FOR ANALYSIS OF
122
SELECTIVE GENOTYPING SURVIVAL DATA
CHAPTER 5
GENERAL CONCLUSIONS AND DISCUSSION
142
CHAPTER 6
ACKNOWLEDGEMENTS
160
iv
LIST OF FIGURES
CHAPTER 1
Figure 1 F2 and BC generation of linkage disequilibrium
9
between marker and QTL (no recombination)
CHAPTER 2B
Figure 1 QTL detected on Gga 3 with line-cross, half-sib,
90
or combined models
Figure 2 QTL detected on Gga 5 with line-cross, half-sib,
91
or combined models
Figure 3 QTL detected on Gga 3 with parent of origin models
92
Figure 4 QTL detected on Gga 5 with parent of origin models
93
CHAPTER 3
Figure 1 Population design used to generate each of the five
119
Grand-parental backcross families
Figure 2 Distribution of length of survival on the backcross
120
Population
Figure 3 Markers associated with Marek's disease resistance
121
On chromosomes 2, 5, and Z
CHAPTER 4
Figure 1 Distribution of survival times from the real population
of chickens
141
#
V
LIST OF TABLES
CHAPTER 2A
Table 1
Significant QTL detected at the 5% chromosome-wise
58
Level by interval mapping
Table 2
Markers suggestively significant at the 10% comparison-
59
wise level by association analysis on chromosomes with
single markers
CHAPTER 2B
Table 1
Estimated map positions of microsatellite markers used
84
for analysis and their corresponding consensus map
positions
Table 2
Phenytyping correlations between traits and means and
86
standard deviations of traits
Table 3
Estimates of position, effect, and mode of expression and
87
inheritance of QTL
CHAPTER 3
Table 1
False positive rates for the linear regression and Cox's
117
proportional hazards models
Table 2
Markers associated with Marek's disease survival
118
False positive rates for Cox proportional hazards, linear
137
CHAPTER 4
Table I
regression, and Weibull models
Table II
Power of different genotyping scenarios
Table III Power of Cox proportional hazards, linear regression,
138
139
and Weibull models
Table IV
Average estimates of effects in days for Cox proportional
hazards, linear regression, and Weibull models
140
vi
ABSTRACT
Identification of genomic regions harboring genes affecting economic traits is the primary
step for the improvement of agricultural species through marker-assisted selection. Many
important traits are difficult and expensive to measure, and, thus, selection can be improved
by selecting directly upon genomic regions affecting these traits (QTL).
White meat
percentage (WM%) and Marek's disease (MD) resistance are examples of important traits in
commercial chickens. One objective of the research presented herein was to identify QTL
affecting WM%, other growth-related traits, and MD resistance. Another objective was to
compare statistical models for identifying these QTL.
Data (phenotypes and genetic
markers) on an F2 broiler (meat-type chicken) cross were analyzed using half-sib, line cross,
combined, and parent of origin models to identify QTL affecting WM% and other growth
related traits. Sixty-eight QTL were identified at the 5% chromosome-wise level, including
six QTL affecting WM% and 20 putative imprinted QTL. The use of multiple segregation
and expression models proved to be beneficial for identifying QTL. A commercial egg-layer
backcross was used to identify QTL affecting MD resistance based on marker genotypes of
long and short survivors (selective genotyping). Seventeen markers associated with MD
survival were identified at P < 0.10 using linear regression (LR) and Cox proportional
hazards (CPH) models. Using simulated data reflecting the MD virus-challenged population,
analyses using LR, CPH, and Weibull models were compared. Little difference in power
was found between the CPH and the LR model when few individuals survived to the end of
the experimental period (low censoring) and when all or selected individuals were
genotyped. The simulated data did not follow a Weibull distribution, and thus the Weibull
model generally resulted in less power than the other two models. The LR model was
recommended for analyzing survival data when the amount of censoring is low because of
the ease of implementation of the model and interpretation of estimates. Including nongenotyped individuals in the selective genotyping analysis increased power, but resulted in
LR having an inflated false positive rate. The QTL identified in this research can be an
integral step for the improvement of commercial chickens through marker assisted selection
programs or identification of candidate genes.
1
CHAPTER 1
GENERAL INTRODUCTION
1.1 INTRODUCTION
One of the main concerns of the livestock industry is to improve economically important
traits in animals. Typically, the traits of interest are quantitative traits; traits that change from
animal to animal or generation to generation incrementally and are determined by a
combination of the effects of many genetic factors and the environment in which the trait is
expressed (Dekkers and Hospital, 2002).
Improvement in these traits can come from
selecting the best animals from one generation for breeding to create the next generation.
The improvement of a trait from selection from one generation to the next directly reflects
the effectiveness of the selection criteria in determining an animal's genetic superiority for
that trait, i.e. the heritability of the selection criteria (Falconer and Mackay, 1996). The
amount of genetic gain in a trait per unit time is determined by genetic variation, generation
interval, accuracy of selection, and selection intensity (Falconer and Mackay, 1996).
Improving lowly heritable and hard to measure traits is difficult to accomplish by phenotypic
selection (Dekkers and Hospital, 2002). A trait is lowly heritable when the phenotypic
variation due to genetic variation is relatively small (Falconer and Mackay, 1996). When a
trait is lowly heritable, the accuracy of selecting the genetically best animals based on
phenotype will be low, so phenotypic selection may result in selecting animals that have a
high trait value because they have been subjected to a favorable individual environment, not
because the selected animals have superior genetics. Because only genetic differences are
passed to offspring, the improvement from generation to generation through phenotypic
selection on a lowly heritable trait will be minimal.
Animals have to be removed from the breeding population to measure many traits that are
important in the livestock industry, such as carcass yield, meat quality, and pathogen
response traits, unless gametes or embryos from the animals can be preserved for future use.
Information from relatives can be used to estimate an animal's genetic merit for the trait
2
without sacrificing that animal. When selecting an animal based on information of relatives,
however, accuracy and effectiveness of selection is determined by the number of relatives
evaluated and their relationship to the animal under selection, as well as the heritability of the
trait (Falconer and Mackay, 1996).
Some traits cannot be measured until late in the life of an animal, such as longevity and
lifetime production, requiring the breeder to maintain all animals for an extended period of
time before the unselected animals can be culled. Making selection decisions early in the life
of an animal reduces the number of animals that have to be maintained until adulthood, or
until the trait is measured, and can reduce generation intervals. Accurate selection early in
life for these types of traits reduces the cost of the breeding program, and allows for
maintenance of selection intensity without housing the entire population from which the
animals are selected.
Other traits, such as milk yield, egg production, and scrotal
circumference, can only be measured in one sex, requiring the breeder to wait until the
progeny of an animal can be evaluated before a selection decision on that animal can be
made or to evaluate the animal based on existing relative's records.
An approach to selection, if the phenotypic value is not reliable for an accurate assessment of
genotypic value or is expensive or difficult to measure, is to select directly upon the genetics
of an animal (Dekkers and Hospital, 2002). This type of selection can be accomplished by
selecting animals that have genomic regions containing favorable alleles of a genetic
sequence affecting a trait. These regions can be identified using polymorphic DNA markers,
which are DNA sequences that may not have an effect on the trait of interest, but are linked
to genetic sequences that do. The genetic sequences that directly affect a quantitative trait
are commonly referred to as Quantitative Trait Loci (QTL) (Geldermann, 1975). Markers
that are linked to genetic sequences affecting the trait can be used to identify animals having
the favorable alleles at the QTL. Adding marker genotypes to a traditional selection program
is commonly referred to as Marker Assisted Selection, or MAS (Lande and Thompson,
1990). By using MAS, selection pressure is placed directly upon the genetic content of an
individual, which partially circumvents the need for making direct phenotypic measurements.
3
MAS is most useful for traits that are difficult or expensive to measure, require animal
harvest to measure, are only displayed in a single sex, are expressed late in life, are
controlled by genes involved in environmental or genetic interactions, or that are lowly
heritable (Dekkers and Hospital, 2002).
MAS is also useful when the frequency of a
favorable allele of a gene is low or zero in a population, such as in marker-assisted
introgression programs (Hospital et al., 1992).
1.2 RATIONALE AND OBJECTIVES FOR DISSERTATION RESEARCH
Two of the main segments of the commercial chicken industry are egg-layers and broilers
(meat chickens). Some traits, such as disease resistance, are important to both segments,
whereas other traits are of more interest to either layer or broiler breeders, such as egg quality
in layers or white meat weight in broilers. Growth rate is also an important trait in broiler
production. Intense selection of meat-type chickens began in the 1950s (Rose, 1997). Since
then, response to selection for increased growth rate has been effective and consistent
(Koerhuis, 1996). Estimates of heritability of growth rate average 0.50, although they vary
widely among studies (Chambers, 1990). Selection for live weight results in an increase in
the weight of carcass and parts. Genetic correlations of live weight with weights of the
carcass and its parts are high (0.49 - 0.91), except with abdominal fat, which is moderate
(0.35) (Cahaner and Nitsan, 1985; Marks, 1995). However, selection for growth has a
negative effect on reproductive traits (Chambers, 1990; Koerhuis, 1996). Therefore, female
lines used in industry are also selected for reproduction (Stevens, 1991), which reduces the
intensity with which they can be selected for growth.
White meat, or the breast meat, is the most economically valuable part of the chicken
(Stevens, 1991). Because of the high genetic correlation between white meat weight and
body weight (0.76 (Le Bihan-Duval et al., 1998)), selection for growth rate has resulted in
increased white meat weight. The genetic correlation between body weight and other, less
valuable, parts of the chicken, however, is also high (Marks, 1995). Because of this high
correlation, selection for growth rate increases the weight of less valuable parts of the
chicken as well as white meat weight. Broilers that are large because of high white meat
4
weight are more economically valuable birds than broilers that are large because of the size
of other parts, such as the giblets or the legs. Birds with large breasts relative to their
bodyweights can be selected for by applying selection pressure on white meat weight as a
percentage of live weight, which will be referred to as white meat yield, as opposed to
selecting only on white meat weight, which would be less effective in increasing the white
meat yield. Direct phenotypic selection on white meat yield is not possible because birds
have to be slaughtered to measure white meat yield. The genetic correlation between white
meat yield and live weight is low (0.13 in males and 0.16 in females (Le Bihan-Duval et al.,
1998)), which indicates that selection for live weight results in little progress in increasing
white meat yield, even though the genetic correlation between live weight and white meat
weight is high. Therefore, including information on genomic regions affecting white meat
yield in a selection program through MAS could increase the improvement made for breast
meat yield in a population.
One trait that has been useful in improving white meat yield is conformation score.
Conformation score is a subjective measurement on live birds, made by a skilled technician,
which is indicative of the overall build of a bird.
Conformation score has a reported
heritability of 0.36 and a genetic correlation of 0.6 with white meat yield (Vereijken, 1992).
Therefore, selection on conformation score will improve white meat yield. However, also
including information on genomic regions affecting white meat yield in selection schemes
can improve progress in increasing white meat yield.
Marek's disease (MD) is one of the major diseases affecting the chicken industry. The loss
from MD has been estimated at approximately 1 to 2 billion dollars per year worldwide
(Purchase, 1985; Morrow and Fehler, 2004). Several vaccines [Rispens (Rispens et al.,
1972), Mdll/75C (Liu and Lee, 1983), SB1 (Schat and Calnek, 1978), and HVT (Okazaki et
al., 1970)] have been used to effectively combat MD since the late 1960s, but the virus has
become resistant to many of them. Outbreaks in vaccinated chickens resulted in the need to
use new vaccines (Biggs, 2001). As MD becomes resistant to more vaccines, it will be
difficult to find effective vaccines to prevent outbreaks. Partial genetic control of Marek's
5
disease resistance has been shown in several studies (Hanson et al., 1967; Lament, 1998; Liu
et al., 2001b; Wakenell et al., 1996). Ameli et al. (1992) estimated the heritability of Marek's
disease incidence to be 0.06 and 0.24, and of Marek's disease mortality to be 0.34 and 0.4 in
two generations of three unselected Leghorn strains. Studies in the 1970s estimated the
heritability of Marek's disease resistance to range from 0.06 to 0.67 (reviewed by Gavora,
1990). Therefore, there are opportunities to increase chickens' genetic resistance to the
disease. While pathogen challenge studies are successful for determining which birds are
genetically superior in resistance to MD, they take considerable time and money to perform
(Weigend et al., 2001). Direct selection on genomic regions affecting MD resistance through
MAS may reduce the time and cost of improving MD resistance in a population.
Identification of genes affecting Marek's disease resistance will enable selection upon them
in an MAS program.
To use MAS, information about marker-gene linkage is needed. Markers are polymorphic
sequences of DNA that do not nessessarily directly affect the trait, but may be linked to QTL
that do. Although it might not be possible to directly observe segregation of alleles of a
QTL, observing the segregation of marker alleles that are linked to the causative QTL allows
inferences to be made about segregation at the QTL. Experiments to obtain this information
may involve specific mating designs, such as an F2 or a backcross between genetically
diverse lines (Andersson-Eklund et al., 1998; Malek et al., 2001; Rohrer and Keele, 1998a, b;
van Kaam et al., 1999b; van Kaam et al., 1999a; van Kaam et al., 1998). Several types of
statistical analyses can then be performed on the mapping resource population to determine
associations of (segregation of) alleles in specific genomic regions with phenotypic
differences. Markers can be used to identify these regions for future genetic selection. The
objectives of the research reported herein is to identify QTL affecting economic traits in
chickens, including white meat yield and other growth and carcass traits in broiler chickens,
and MD resistance in layer chickens, and to determine the proper statistical models for
analyzing selectively genotyped survival data.
6
1.3 ORGANIZATION OF DISSERTATION
The remainder of this chapter (Section 1.4) is a review of literature relevant to the research
presented herein. This thesis is presented in the alternate format, in which manuscripts
prepared for scientific journal publication (Chapters 2, 3, and 4) follow a review of literature
relevant to those manuscripts (Section 1.2). The overall theme of the research presented
herein is the identification of genomic regions (QTL) associated with economic traits in
commercial chickens, and a comparison of statistical methods to detect such QTL.
Chapter 2 is comprised of two sections: a manuscript (McElroy et al., 2002) that was
published in the Proceedings of the 7th World Congress on Genetics Applied to Livestock
Production, presenting the identification of QTL affecting white meat percentage and other
growth and carcass traits in a single half-sib sire family generated from a commercial broiler
chicken cross, and a manuscript (in preparation for submission to Poultry Science) describing
the identification of QTL affecting white meat percentage and other growth and carcass traits
in a commercial broiler chicken cross, using an extension of the population that was utilized
in the study reported in McElroy et al. (2002). Chapter 3 (submitted to Poultry Science)
describes the identification of QTL affecting Marek's disease resistance in a commercial
layer cross. Chapter 4 (in preparation for submission to Genetics, Selection, and Evolution)
describes a comparison, by simulation, of the statistical models (linear regression, Cox
proportional hazards, and Weibull models) used to identify QTL in Chapter 3. Because the
manuscripts were prepared for publication, the organization and format of Chapters 2 to 4 are
in accordance with requirements of the respective journal format requirements. Coauthors of
the manuscripts are:
Chapter 2A: J. P. McElroy1, D. E. Harry2'4, J. C. M. Dekkers1, and S. J. Lamont1
'Department of Animal Science, Iowa State University; 2Aviagen Intl.
Chapter 2B: J. P. McElroy1, J.-J. Kim1'3, D. E. Harry2'4, S. Brown2, J. C. M. Dekkers1, and S.
J. Lamont1
7
'Department of Animal Science, Iowa State University; 2Aviagen Intl.;
^Present Address: School of Bitechnology, Yeungnam University,
Gyeongsan, Gyeongbuk, 712749 South Korea; ^Present Address:
Genetic Foundations, P.O. Box 3897, Napa, CA 94558.
Chapter 3: J. P. McElroy1, J. C. M. Dekkers1, J. E. Fulton2, N. P. O'Sullivan2, M. Soller3, E.
Lipkin3, W. Zhang4, K. J. Koehler4, S. J. Lamont1, and H. H. Cheng5
'Department of Animal Science, Iowa State University; 2Hy-Line Intl.;
3The
State
Hebrew University of Jerusalem; department of Statistics, Iowa
University;
^United
States
Agricultural Research Service,
Department
of
Agriculture,
Avian Disease and
Oncology
Laboratory.
Chapter 4: J. P. McElroy', W. Zhang2, K. J. Koehler2, S. J. Lamont1, and J. C. M.
Dekkers'
'Department of Animal Science, Iowa State University; 2Hy-Line Intl.;
3The
Hebrew University of Jerusalem; department of Statistics, Iowa
State University; 5USDA-ARS-ADOL.
For Chapter 2, the first author (McElroy) performed some of the laboratory work (DNA
isolation, PCR amplification), all of the genotype scoring and statistical analyses, and was
primarily responsibly for drafting the manuscripts. For Chapters 3 and 4, the first author
(McElroy) was the primary person responsible for the statistical analyses and drafting the
manuscripts.
1.4 REVIEW OF RELEVANT LITERATURE
Approaches For Mapping QTL
To perform MAS, marker/QTL linkage information is needed. Two types of experimental
approaches are commonly employed to obtain this linkage information: anonymous marker
approaches and candidate gene studies. These approaches will be described in the following
8
sections. Both approaches exploit gametic phase or linkage disequilibrium (LD), which
refers to the state where the probability of a combination of alleles at different loci occurs
more or less frequently than the product of their individual frequencies in the population
(Falconer and Mackay, 1996). Markers in LD with genes affecting a trait should show an
allelic association with the trait.
Therefore, by analyzing markers or genes for allelic
associations with phenotype of a trait, the regions in the genome that contain genes affecting
the trait (QTL) can be determined.
Anonymous Marker Approaches
The use of molecular markers to identify trait-associated regions of the genome has been the
focus of many studies (e.g., Malek et al., 2001; van Kaam et al., 1999b; Zhou et al., 2003).
The concept behind this type of analysis is that QTL that segregate in a population can be
detected by identifying markers that are linked to the QTL in the population.
Population Structures. Two types of crosses are primarily used in a QTL detection analysis
in animals: intercrosses (usually an F2 cross) and backcrosses (Falconer and Mackay, 1996)
(Figure 1). An F2 population is created by making a cross between two lines to form an F1
generation, and then crossing the F1 generation within itself to form the F2 generation. A
backcross population is created by crossing two lines to form an F1 generation, and then
crossing the F1 individuals to one of the parental lines. Both types of crosses are used
because they create considerable between-line LD (Figure 1).
An F2 cross permits an
analysis that contrasts three genotypic classes of offspring for a given locus: homozygotes for
each of the parental line alleles and hétérozygotes, which have one allele from each line
(Soller et al., 1976). Dominance can be determined from an F2 cross because the mean trait
value of the two homozygous genotype classes can be determined, and the deviation of the
heterozygous individuals from this mean value can be calculated (Soller et al., 1976). A
backcross only creates two genotypic classes at a given locus: homozygotes for the allele
from the parental line used to create the backcross generation and hétérozygotes having an
allele from each of the parental lines (Soller et al., 1976) (Figure 1). Because individuals
homozygous for the allele from the parental line not used to create the backcross generation
9
BC
F2
L1XL2
L1XL2
X
M1
M2 q
M1 Q
M1 Q
7
M1 Q
F1X F1
F2
M1_Q ^
M1 Q
T'
M1_ Q
M1 Q
M1
M2 q
QTL +2
effect
+1
+ dominance
F1XL2
BC
QTL
effect
M2 q
X
M2 q
Ml
Q
M2 q
M2 q
M2 q
M2 q
M2 q
M2 q
+1
0
+ dominance
QTL allelic effects
Q-» +1
q-*"0
Figure 1. F2 and BC Generation of Linkage Disequilibrium Between Marker and QTL
(No Recombination)
10
are not produced in a backcross mating design, the trait mean of the two parental line
homozygous classes cannot be calculated, and therefore the dominance and additive effects
cannot be determined separately. The effect estimated from a backcross is the additive effect
+/- the dominance effect (Soller et al., 1976).
Three other types of populations used for QTL detection are the half-sib design (Knott et al.,
1996), the grandprogeny design (Weller et al., 1990), and the three generation full-sib design
(van Kaam et al., 1998). In general, the half-sib and full-sib designs can be utilized within a
breed or line to detect QTL segregating within that breed or line, whereas the F2 and
backcross designs utilize crosses between different breeds or lines to detect QTL segregating
differentially between the breeds or lines. However, a half-sib or full-sib analysis can also be
performed within a backcross or F2 population. In half-sib designs, mating sires to multiple
dams generates half-sib families and the offspring from these matings are the subjects for
genetic analysis. Only the sire allelic effects, averaged across all dam alleles at a locus, i.e.
the allele substitution effects (Falconer and Mackay, 1996), are analyzed. The sire must be
heterozygous at a given marker locus for the analysis. Offspring classified by sire allele
received are contrasted in the half-sib analysis. A grandprogeny population design (Weller et
al., 1990) is similar to a half-sib design. The difference between the two analyses lies in the
origin of the phenotypes. In a grandprogeny design, the offspring of the half-sibs (i.e.
grandprogeny of the original sires) are used to estimate a breeding value for the half-sibs,
which is used as their "phenotype" in the analysis. The advantage of the granddaughter
design over a half-sib design is that most of the environmental variance is removed from the
"phenotypes" of the half-sib individuals (Weller, 2001). With most of the environmental
variance removed, the differences in the "phenotypes" of the individuals is to a greater
degree determined by genetics, which results in more power in the analysis.
A three
generation full-sib design (van Kaam et al., 1998) is analyzed similarly to a half-sib design.
However, in full-sib analysis, marker alleles from both parents are analyzed, as implied by
"full-sib," and the full-sib animals' trait values are based on the performance of their
offspring in the F3 generation, as implied by "three generation."
11
Statistical Analyses. Single Marker Analysis. Originally, markers or haplotypes (Thoday,
1961) were analyzed individually to detect trait-associated regions in the genome (Edwards
et al., 1987; Sax, 1923; Soller et al., 1976). Marker-trait association tests are based on
comparisons of mean trait values of animals grouped by inherited allele or genotype for a
marker or haplotype using linear regression or one-way ANOVA (Lander and Botstein,
1989). Although the "apparent" marker effects of single marker analysis can identify regions
of a genome containing QTL affecting a trait, the position of the QTL and the effect of the
QTL are confounded (Lander and Botstein, 1989). The term "apparent" is used here because
the effect determined for each locus is really an effect of a linked QTL, not a direct effect of
the marker. A marker tightly linked to a small effect QTL and a marker loosely linked to a
large effect QTL can have similar estimated associations with the trait.
Interval Mapping. By analyzing pairs of markers, the position and effect of a QTL can be
separated. Conceptually, two fully informative markers (A and B) with a QTL midway
between them should show nearly identical apparent effects. In comparison, a QTL closer to
marker A, which is further from marker B, would result in marker A showing a greater
association with the trait than marker B. Comparing the apparent effect of marker A and
marker B allows us to estimate the position of the QTL. As the position of the QTL is more
accurately estimated, the QTL effect can be more accurately estimated (Haley and Knott,
1992).
Lander and Botstein (1989) proposed using a maximum likelihood method to estimate QTL
effects and location based on multi-marker data. This method is performed by placing a
putative QTL at fixed cM steps, e.g., 1 cM, between two markers, and determining the
location of the QTL that best explains the distribution of phenotypes in the population given
the marker data. The best location for a QTL is determined by the value of a LOD score for
each QTL position, which is the logio of a likelihood ratio test, where the numerator is the
likelihood of observing the data given a QTL at the position, and the denominator is the
likelihood of observing the data in the absence of a QTL (Lander and Botstein, 1989).
12
Haley and Knott (1992) proposed using least squares regression for interval mapping in
inbred line F2 crosses. Least squares interval mapping involves the regression of phenotypic
data onto the probability of inheriting a specific allele at a given position between two
markers. Using flanking marker data, the probability of inheriting an allele at a putative QTL
location between the markers can be determined by calculating the expected amount of
recombination between that point and the markers based on map distances.
Genotypic
probabilities at a locus can then be calculated by multiplying the probabilities of the two
alleles that make up a genotype at the locus. Phenotypic values are then regressed on these
genotypic probabilities using the model:
_y(i) = Xb + aPm + JPd(i) + 6(0,
where
is the phenotype of F2 individual i, X is the design matrix, b is the vector of
coefficients for fixed effects, a is the additive effect of the QTL, d is the dominance
deviation, Pa(i) the probability of one homozygous type at the putative QTL locus given the
marker information minus the probability of the other homozygous type at the locus given
the marker information for animal i, Pd(i) is the probability of being heterozygous at the
putative QTL locus given marker genotypes for animal i, and % is the residual error for
animal i. The model is fitted at fixed positions, e.g., 1 cM intervals in a chromosome or
chromosomal region with multiple markers and the position with the highest significant F
ratio test value is considered the most likely location of a QTL. Least squares interval
mapping is much less computationally expensive than maximum likelihood, and yields
similar results in most instances (Haley and Knott, 1992).
Haley et. al (1994) extended least squares regression interval mapping to outbred crosses
where the QTL are considered fixed for alternate alleles between the two parental lines. The
alternate allelic fixation need not be the case for the method to be useful; only a difference in
allele frequencies between the two lines is required. The QTL effects will, however, be
underestimated when the two parental lines are not fixed for alternate alleles (Haley and
Knott, 1992).
13
Knott et al. (1996) extended least squares regression interval mapping to half-sib analysis.
When analyzing half-sib populations, the model for a single half-sib family with the sire as a
common parent is:
J/(i) = Xb + CtHS-Ps(i) + 6(j),
where y^ is the phenotypic value for animal i, X is the incidence matrix relating fixed effects
to observations, b is the vector of fixed effects, ans is the substitution effect (Falconer and
Mackay, 1996) for the QTL, Ps(i) is the probability that the progeny i inherited one versus the
other QTL allele from the sire based on marker genotypes, and % is the residual error for
progeny i. Dominance cannot be determined in the half-sib analysis (Soller et al., 1976). For
multiple half sib families, the regression would need to be nested within sires because the
linkage phase between the QTL and markers may differ between sires, and marker alleles in
one sire may not be identical to the alleles in other sires.
Other Multiple Marker QTL Mapping Methods.
Several other multiple marker QTL
mapping methods have been proposed after the methods suggested by Lander and Botstein
(1989) and Haley and Knott (1992). Two methods that use other markers in the genome as
cofactors to reduce the background genetic noise when examining an interval for the
presence of a QTL were proposed by Zeng (1993) using multiple regression and Jansen and
Stam (1994) using maximum likelihood. Kao et al. (1999) presented a method for QTL
mapping in which multiple marker intervals are analyzed simultaneously for QTL. Kruglyak
and Lander (1995a) proposed a non-parametric method that utilizes the Wilcoxon rank-sum
test to map QTL when the phenotypes under study are not normally distributed, and Uimari
et al. (1996) introduced a Bayesian method for multiple marker mapping of QTL using
Markov chain Monte Carlo algorithms.
Selective Genotyping and DNA Pooling.
Because of the large number of genotypings
required to perform a typical QTL mapping experiment, methods have been proposed and
14
utilized that reduce the number of genotypings with a minimal loss in statistical power to
detect the QTL. Selective genotyping (Lander and Botstein, 1989) is a marker/QTL linkage
analysis method in which only individuals in the extremes of the phenotypic distribution are
genotyped.
Most of the power in detecting marker QTL linkage comes from individuals
from the extremes of the phenotypic distribution (Lander and Botstein, 1989).
Analyzing selective genotyping data by normal regression analysis is not appropriate because
the effect estimates will be biased (Henshall and Goddard, 1999). Maximum likelihood
analysis can be used to get unbiased effect estimates by including the phenotypic data of
ungenotyped animals in the analysis (Lander and Botstein, 1989). Muranty and Goffinet
(1997) applied an approximation of maximum likelihood to selective genotyping data to map
QTL. The approximation, a second order Taylor expansion of the likelihood function, was
shown to accurately estimate QTL effect and position for QTL with small to medium effects
(< 25% of the trait mean). Henshall and Goddard (1999) proposed a logistic regression
method for analyzing selective genotyping data. In this method, marker genotype is treated
as the dependent variable and phenotype as the independent variable. Reversing the roles of
the genotypic and the phenotypic data in the model results in unbiased estimates of the QTL
effect. This method was also applied to interval mapping of selective genotyping data.
The use of selective DNA pooling Darvasi and Soller (1992) can further reduce the number
of genotypings considerably. The design of a selective DNA pooling experiment is the same
as selective genotyping, but instead of genotyping each individual from the phenotypic
extremes, equal amounts of DNA from all of these individuals are pooled into one sample.
Then, the frequency of alleles at a marker in the pool is determined by quantification of the
signal strength from each allele divided by the sum of the signal strengths of both alleles.
Alleles of a marker that are not linked to a QTL are expected to be evenly distributed across
the phenotypic distribution, and therefore to have a frequency of 0.5 in animals from each
extreme "tail" of the distribution. By either examining the deviation of an allele's frequency
in a tail from .5, or by comparing the allele's frequency between the two tails for a deviation
from equality, linkage of that marker with a QTL affecting the trait can be determined. If
15
just two alleles are present for a marker in the population, only a single allele need be
analyzed because the frequency of one allele is completely dependent upon the frequency of
the other allele. One drawback of using selective genotyping or pooling is that only one or
several highly correlated traits can be examined for each selective genotyping or pooling
analysis, because the animals genotyped are generally chosen based on their phenotypic
values for a single trait.
Survival Models to Detect QTL in Livestock.
Many agriculturally important traits in
livestock follow a survival distribution (e.g. survival in a disease state and length of
productive life). Survival traits typically follow a non-normal distribution and it is common
for animals to leave the study before the trait of interest can be recorded on them (censoring).
To analyze survival traits, the Weibull model or the Cox proportional hazards model (Cox,
1972) is often employed (Benard et al., 1999; Hirst et al., 2002; Kuurman et al., 2003;
Maizon et al., 2004; Roxstrom et al., 2003). The Weibull model, a generalization of the
exponential model, is parametric and is therefore appropriate for only specific distributions,
whereas the Cox model is semi parametric and rank based, and therefore should be
appropriate for all distributions (Smith, 2002).
Both models can appropriately analyze
censored data. Weigend et al. (2001) is the only study using survival models to identify QTL
in livestock that the author located in the literature.
Significance Testing. When testing for an association between a genomic region and a
phenotype, criteria are needed to determine whether or not the results are indicative of the
presence of a QTL or if they are the result of random chance. Witte (1989) defines a
significance level as an indicator of "the degree of rarity among random outcomes required to
reject the null hypothesis." For example, a significance level of 5% means that five percent
of the tests are expected to reject the null hypothesis due to random chance if the null
hypothesis is true. For QTL mapping, the null hypothesis is that there is no QTL affecting
the trait of interest in LD with the markers under analysis. The alternate hypothesis is that
there is a QTL affecting the trait of interest in LD with the markers under analysis.
16
There are several types of significance levels, differing in their stringencies and
interpretation. Comparison-wise significance is the chance that any test will yield a false
positive result. For example, a 5% comparison-wise significance level would result in any
individual test having a 5% chance of being a positive under the null.
With multiple
independent tests at the comparison-wise 5% significance level, the chance of at least one of
the tests being a false positive is l-(.95)n, where n is the number of tests.
Thus, the
probability of getting at least one false positive will be greater than .05 when using the
comparison-wise significance test on multiple tests. Chromosome-wise significance (Lander
and Kruglyak, 1995) controls the expected number of false positives on a chromosome.
Using a chromo some-wise significance level of 5% would result in a 5% chance of getting at
least one false positive per chromosome.
When analyzing multiple chromosomes, the
probability of getting at least one false positive on any chromosome would then be l-(.95)n ,
where n is the number of chromosomes. This number is always greater than .05 when
analyzing multiple chromosomes. Genome-wise significance (Lander and Kruglyak, 1995)
controls the number of false positives across a genome. A genome-wise significance level of
5% indicates a 5% probability of at least one false positive in an analysis of a whole genome.
Another term that is commonly used with these stringencies of significance is "suggestive"
significance.
Lander and Kruglyak (1995) defined suggestively significant linkage as
"statistical evidence that would be expected to occur one time at random in a genome scan."
However, many studies do not follow Lander and Krugliyak's definition.
Typically,
significance and suggestive significance just indicate false positive probability levels that are
lower than specific levels chosen by the particular authors. If the authors make clear their
definitions of significance and suggestive to the reader, then the readers can come to their
own conclusions about the results of a study.
A significance threshold is the boundary indicating the minimum value of a test statistic
required for a specific significance level. Determining significance thresholds is not a trivial
matter when performing anonymous marker analyses because of the large number of
correlated tests involved.
Several methods have been proposed to determine proper
17
significance thresholds for the multiple comparison situations that arise during anonymous
marker analyses.
For instances in which marker intervals are far enough apart to be considered independent
tests, Lander and Botstein (1989) suggested using a/n, the Bonferroni correction, as the
significance threshold, where a is the significance level desired, and n is the number of
intervals tested. When the marker spacing approaches zero, Lander and Botstein (1989)
found that the expected number of false positive test results when no QTL is present could be
determined by a formula from physics and engineering, describing a specific case of
Brownian motion. The typical marker spacing that will be encountered in interval mapping
studies, however, is an intermediate of these two extremes. Lander and Botstein (1989)
found through extensive simulation that a LOD score of 2 to 3 will usually give a genomewise false positive rate of 5% when the markers are spaced intermediately (i.e. not so far
apart that each test is independent, and not so close together that the recombination fraction
approaches zero between consecutive markers).
Churchill and Doerge (1994) argued that assumptions have to be made that are unlikely to
occur in a real world situations when using the methods presented in Lander and Botstein
(1989) to determine threshold values, and that the large differences in parameters from
experiment to experiment created the need for experiment-specific methods to derive
significance thresholds. They suggested use of a permutation test, first proposed by Fisher
(1935), for determining significance thresholds. The permutation test consists of randomly
assigning the phenotypic values in a population to the marker genotypes observed for that
population, which breaks any association that might exist between genotype and phenotype.
The data are then considered to be following the null hypothesis, i.e. no QTL are present that
are in LD with the markers under analysis. Next, the specific test that is being used to detect
QTL, e.g., least-squares interval mapping, is performed on the data. This process is repeated
many times, usually 1000-10,000 times, and the value for which the desired percentage of
permutations had a greater value of the test statistic (e.g. F or LOD) is used as the
significance threshold for the experiment. For example, to obtain a significance level of a =
18
0.05, the F-value at which 5% of the permutation generated F-values is higher would be used
as the threshold. The permutation test is very robust to different data structures because it
always uses experiment specific data structures. Difficulties in using the permutation test
arise when the type of mapping method utilized requires a large amount of computing power.
Candidate Gene Analysis
The approach of a candidate gene study is to choose genes for analysis based on either a
hypothesized or known functional relationship to the trait of interest (biological candidate
genes), to choose genes for analysis based on their genomic location (positional candidate
genes), or a combination of the two (Rothschild and Soller, 1997). Once the gene has been
verified to exist in the experimental population, individuals are screened to identify
polymorphisms in the gene segregating in the population. The gene can then be tested in the
population for an allelic association with the trait of interest by a simple comparison of
phenotypic means of candidate gene allelic groups while taking appropriate fixed
and
random effects into account...
Biological candidate genes can be selected for their functions in the species of interest or
other species. Sequence data and comparative mapping can be used to locate a biological
candidate gene in the experimental population. Kim et al. (2000), Rothschild et al. (2000),
and Zhou et al. (2001) are examples of successful biological candidate gene studies. The
positional candidate gene approach begins by identifying genomic regions that are associated
with a trait of interest. These regions can be identified through an anonymous marker study.
Positional candidate genes are then selected based on the knowledge of genes that are known
to be located in these trait-associated regions, either through previous mapping studies in the
species of interest or comparative mapping with other species. The positional candidate gene
approach has been used successfully in several studies (Nielsen et al., 2000; Smith et al.,
2000b; Yokoi et al., 2000).
The candidate gene approach can be carried out in almost any population design. In crosses
such as an F2 or backcross, however, LD can span considerable distances on a chromosome,
19
so associations between the candidate gene mutation and phenotype do not necessarily
indicate tight linkage between the candidate gene mutation and the causative mutation (Zhao
et al., 2003). Therefore, outbred populations are ideal for determining tight linkage between
the candidate gene mutation and the causative mutation.
Candidate gene studies have appeal because they use prior information about the effect or
position of a gene that might affect a trait of interest. However, the candidate gene approach
only reveals markers (biological) or additional markers (positional) linked to the causative
gene. To prove that a mutation in a candidate gene is causative, extensive protein analysis, a
transgenic experiment, or a knock-out/recovery experiment must be performed.
It is
interesting to note that the hypothesis under which a candidate gene is chosen, i.e. that the
gene selected affects the trait, is never actually tested in a candidate gene association
analysis. Instead, the hypothesis tested is that the genes chosen are in LD with a gene
affecting the trait. This information, however, is sufficient for using the candidate gene in a
MAS program.
Candidate Gene vs. Anonymous Markers
Although both the candidate gene approach and the anonymous marker approach have been
used successfully, both approaches have limitations. The candidate gene approach is limited
by the amount of prior knowledge that is available about genes affecting a trait or genes in a
particular area of the genome. Also, this method is only used to investigate the specific areas
of the genome where the candidate genes are located. In an anonymous marker approach, all
regions of the genome can be analyzed provided that there are markers in those regions, and
the analysis is not limited to having prior information. However, the extensive LD needed in
an anonymous marker approach prevents the differentiation between moderate linkage and
very tight linkage between a marker and a QTL. In a random mating or advanced intercross
population, the candidate gene approach can identify very tight linkage between a mutation
and a QTL because it is quite possible that the candidate mutation is in the causative gene.
Therefore, narrowing the QTL region using a genome scan approach requires the production
of more recombination events (more offspring or more generations) than created in the F2,
20
half-sib, or backcross populations used in the initial genome scan study, or the addition of
more markers within a QTL region for use in population-wide linkage disequilibria mapping
(Meuwissen and Goddard, 2000). A genome scan requires considerably more cost and time
than a single candidate gene study because of the extensive amount of genotyping involved
in performing a genome scan.
Identifying Imprinted Genes/QTL
Imprinting has been observed in humans, mice, and sheep [see Tycko and Morison (2002) for
a review], as well .as swine (de Koning et al., 2001a; de Koning et al., 2001b; de Koning et
al., 2000; Thomsen et al., 2004). Autosomal imprinting is defined by Tilghman (1999) as
"the differential expression of the two parental alleles of a gene." Some believe that the
mechanism for imprinting evolved for the purpose of gene silencing due to differing optimal
strategies for each parent (mother and father) for resource allocation of the mother's
resources to the offspring in viviparous animals, i.e. the conflict hypothesis (Moore and Haig,
1991), and that this mechanism evolved after the evolutionary split between birds and
mammals (Yokomine et al., 2005). Therefore, following this theory, imprinting should be
absent in chickens.
However, Pardo-Manuel de Villena et al. (2000) hypothesized that
imprinting may have evolved because of the need to distinguish between homologous
chromosomes (one from each parent) and sister chromosomes (exact copies) during meiosis
and mitosis, and that gene silencing is just a side effect of the mechanism of imprinting.
Following this hypothesis, all sexually reproducing animals would need the imprinting
mechanism, or some equivalent mechanism, to distinguish between homologous and sister
chromosomes. Identification of parent of origin QTL in the chicken has been reported by
Tuiskula-Haavisto et al. (2004). Koski et al. (2000) found the insulin-like growth factor II
gene (IGF2) to have parent of origin expression in chickens. O'Neill et al. (2000), however,
also studied IGF2 in chickens and found no evidence of imprinting.
Statistical methods for detecting imprinted QTL in F2 crosses have been developed by Knott
et al. (1998) and were modified by de Koning et al. (2000). The methods of de Koning et al.
(2000) were used by de Koning et al. (2000), de Koning et al. (2001a), de Koning et al.
21
(2001b), and Thomsen et al. (2004) to identify imprinted QTL in swine. The models, fitted
at each 1 cM position along the genome, were:
mendelian expression model (Mend):
= Xb + aPa(j) + dPà(i) + e(i),
full imprinting model (Full):
y ( l ) = Xb + ap-dXPpiltil) + amatPm3im + dPd(i) + e(i),
paternal expression model (Pat):
jK(i) = Xb + apatPpat(i) + e(i),
maternal expression model (Mat):
^ = Xb +
null model:
><i) = Xb + e(i),
timatPmat(i)
+
where y^) is the phenotype of F2 individual i, X is a design matrix, b is the vector of
coefficients for fixed effects, % is the residual error for individual i, and apat, amat, and d are
the paternally inherited, maternally inherited, and dominance QTL effects, respectively.
Coefficients Pa(j) is the line-origin coefficient for animal i at a given position conditional on
flanking marker genotypes, Ppat(i) is probability of inheriting a Line 1 allele vs. a Line 2 allele
from the sire of animal i, Pmat(j) is probability of inheriting a Line 1 allele vs. a Line 2 allele
from the dam of animal i, and P^i) is the probability of animal i being heterozygous.
The following decision tree was used by Thomsen et al. (2004) to determine the
expression pattern of the QTL:
If the Mend model vs. the null model was significant:
1) The Full model was tested against the Mend model at each lcM position in that
genomic region. If this test was not significant, then the QTL was classified as a
Mend QTL.
2) If the Full model vs. the Mend model was significant, then the Full model was tested
against the Pat and Mat models at each lcM position in that genomic region.
a. If the Full model vs. the Pat model was not significant and the Full model vs.
the Mat model was significant, then the QTL was classified as a paternally
expressed QTL.
22
b. If the Full model vs. the Pat model was significant and the Full model vs. the
Mat model was not significant, then the QTL was classified as a maternally
expressed QTL.
c. If the Full model vs. the Pat model and the Full model vs. the Mat model were
both significant or both not significant, then the QTL was classified as a
partially expressed QTL.
If the Mend model vs. the null model was not significant, the Full model was tested against
the null model. If this test was significant, then the Full model was tested against the Mat
model and Pat model as described in step 2 above.
A paternally (maternally) expressed QTL is one that shows a significant allelic effect when
inherited from F1 sires (dams) without showing a significant allelic effect when inherited
from F1 dams (sires). A partially expressed QTL is one that shows an allelic effect when
inherited from F1 sires and F1 dams, but the effect is different depending on the sex of the F1
parent from which it was inherited.
Molecular Genetic Markers
Several types of genetic markers are commonly used in QTL detection: restriction fragment
length polymorphisms (RFLPs), randomly amplified polymorphic DNAs (RAPDs), single
nucleotide polymorphisms (SNPs), microsatellite markers, amplified restriction fragment
length polymorphisms (AFLPs), and single-strand conformational polymorphisms (SSCPs).
These will be described in further detail below.
Restriction Fragment Length Polymorphisms
Restriction enzymes are proteins that recognize specific DNA sequences and cut the DNA
within these sequences (Becker et al., 1996). The enzyme will not cut the DNA if the
restriction sequence differs by one base from the required sequence. Therefore, different
alleles at a restriction site result in different lengths and numbers of fragments of DNA
(corresponding to whether or not the enzyme cut the DNA), which can be visualized by gel
23
electrophoresis of the DNA after digestion. RFLPs are codominant markers, meaning that
both alleles can be visualized in a heterozygous individual. The consensus linkage map of
the chicken genome, which is an integration of maps created from the reference populations
of Compton, East Lansing, and Wageningen, contains over 200 RFLPs (Groenen et al., 2000;
Schmid et al., 2000).
Randomly Amplified Polymorphic DNA
Amplifying random regions of genomic DNA by the polymerase chain reaction (PCR) using
a short sequence primer generates RAPDs (Williams et al., 1990). Only genomic regions
that have complimentary sequence to the primer sequence in both the forward and reverse
directions and have a distance between the forward and reverse sequences that permits
amplification will be amplified. Usually, multiple sites in the genome will be amplified by a
given primer and can be differentiated by their amplified product size. A mutation in the
primer sequence can prevent amplification, and therefore alleles at a locus can be determined
by the presence or absence of a PCR product.
RAPDs are dominant markers, since a
heterozygous individual will still result in an amplified product, meaning that the
hétérozygote cannot be distinguished from the amplified homozygote. More than 60 RAPDs
have been placed on the consensus linkage map of the chicken genome (Groenen et al., 2000;
Schmid et al., 2000).
Single Nucleotide Polymorphisms
SNPs are the most common type of mutation in the genome (Weller, 2001). SNPs are single
base mutations at a locus, and can be identified through the direct sequencing of regions of
the genome. Utilizing the newly available complete sequence of the chicken genome (Hillier
et al., 2004), 2.8 million SNPs have been mapped in the chicken (Wong et al., 2004).
Microsatellite Markers
Microsatellite markers (Litt and Luty, 1989) consist of short tandem DNA base sequence
repeats. Microsatellites are highly polymorphic, which is likely due to unequal crossing over
during meiosis (Jeffreys et al, 1985) or slippage during replication (Romberg, 1980). This
24
highly polymorphic nature makes microsatellites useful for mapping, because markers have
to be polymorphic for use in identification of chromosomal regional inheritance. Alleles of a
microsatellite are distinguished by their repetitive region lengths, which usually differ by a
multiple of the number of bases in the repeated sequence. Microsatellites are codominant
markers, meaning that heterozygous individuals can be distinguished from either of the
corresponding homozygote genotypes. The consensus linkage map of the chicken genome,
contains over 800 microsatellite markers (Groenen et al., 2000; Schmid et al., 2000).
Microsatellite markers are ideal for mapping QTL in the chicken genome because many have
been mapped in chickens, they are highly polymorphic, they are codominant, and primer sets
for them are readily available.
Amplified Restriction Fragment Length Polymorphisms
AFLPs are generated by first cutting genomic DNA with two restriction enzymes, ligating
adapters to the cut ends of the DNA, and then amplifying this DNA with labeled primers
which have complementary sequence to the adapter specific to one of the restriction enzyme
cut sites (Masiga et al., 2000). The number of labeled amplified products is often large, but
can be reduced by putting bases at the 3 ' end of the primers so that they will only amplify
sequences with these specific bases directly after the cut sites. AFLPs are dominant markers.
The consensus linkage map of the chicken genome contains 552 AFLPs (Groenen et al.,
2000).
Single-Strand Conformational Polymorphisms
SSCPs are detected by polyacrylamide gel electrophoresis of single stranded DNA. Gel
mobility shifts between differing alleles of homologous stretches of DNA are a result of a
single stranded conformational change caused by a mutation (Orita et al., 1989). Alleles
differing by as little one base pair can be detected by SSCP analysis (Orita et al., 1989).
SSCPs are codominant markers. The consensus map of the chicken genome contains 59
SSCP markers (Groenen et al., 2000).
25
Chicken Genomic Sequence
A draft of the complete chicken genome is now available (Millier et al., 2004). This draft has
already facilitated the identification of 2.8 million SNPs (Wong et al., 2004), and will be a
valuable tool for physically mapping existing markers and identifying both positional and
biological candidate genes.
Traits Analyzed in This Dissertation Research
Growth and Breast Meat Yield
Increasing amounts of food are needed to sustain a growing human population. From 1995
to 2020, the world population is expected to increase by 32 percent (Pinstrup-Andersen et al.,
1999) and agricultural production will also have to rise to meet the growing demand. More
efficient methods of food production will aid in meeting this demand.
High-quality protein is needed for normal growth and repair of the human body. A diet
deficient in the required amounts of good quality protein can result in mental and physical
abnormalities (Lasley, 1987). Generally, protein from animals is superior in quality to plant
proteins (Lasley, 1987), and therefore, meat is a very good source for the required proteins.
The demand for meat in the developing world is expected to double from 1995 to 2020
(Pinstrup-Andersen et al., 1999). Much of this expected increase is due to increased income
in the developing countries because, generally, as per capita income grows, meat
consumption grows (Gehlhar and Coyle, 2001). Per capita meat consumption in the U.S. has
also grown, increasing more than 11 percent from 1970 to 2000 (Haley, 2001).
Roenigk (1999) reported that poultry is second only to pork in world meat consumption. In
1970 in the U.S., beef and veal was the preferred meat for consumption, with pork second,
and poultry third. In 2000, poultry was the leading type of meat by pounds consumed, and
accounted for most of the increase in meat consumption in the U.S. (Haley, 2001). The
success of poultry can be attributed to population and economic growth (which has resulted
in an increase in meat consumption as a whole), the relatively low cost of poultry as
26
compared to other meats, the nutritional value of poultry, the relatively fast cooking time of
poultry, product development of poultry meat industry relative to consumer demands, and
technological development in the poultry industry (Haley, 2001; Roenigk, 1999). These
technological developments encompass breeding, feeding, production, processing, and
marketing (Roenigk, 1999).
Commercial broiler breeding companies perform selection based on pure-line selection for
growth rate or bodyweight at a fixed age (Emmerson, 1997). In general, the pure-lines they
use fall into two categories: male lines and female lines. Selection in both lines is for growth
rate, edible meat yield, and, sometimes, feed conversion ratio. The female line is also
selected for egg production (Pollock, 1999). Commercial birds are three or four-way crosses
of the male and female lines. Because of the amount of multiplication that occurs between
the pedigreed lines and the market birds, the genetic superiority of a single pedigree-level
sire is realized in many thousands of commercial broilers.
Genetic selection in the broiler industry has been very successful. Since the mid 1900s, time
to reach slaughter weight has halved (Goldspink and Yang, 1999) and feed conversion rate
(kg feed/kg growth) has decreased (Emmerson, 1997). These high genetic gains are partially
due to the relatively short generation interval of poultry as compared to other livestock. The
high heritability of growth rate [approximately 0.5 (Chambers, 1990)] also made rapid gains
possible.
Overall weight is a result of growth of individual parts of a bird, including muscle, bone, fat,
and viscera. Because meat is the central product from broilers, breeders are interested in
improving meat growth while maintaining or reducing the growth of other parts.
By
increasing meat yield relative to other parts, breeders can generate birds that are more
efficient in the utilization of resources to produce the most valuable marketable product.
Breast meat, which makes up about 33.5% of the carcass (Rose, 1997), is the most valuable
part of the broiler. This meat is usually referred to as white meat, reflecting its pale color as
27
compared to other muscles. The color of breast meat is due to a relatively low concentration
of myoglobin compared to other muscle types. The breast muscles are adapted for short
bursts to pull the wings down to escape predators, and tire easily because of the low blood
supply and low numbers of mitochondria (Rose, 1997).
Marek's Disease
MD is one of the major diseases affecting the chicken industry. The loss from MD has been
estimated at approximately 1 to 2 billion dollars per year worldwide (Purchase, 1985;
Morrow and Fehler, 2004). Several vaccines [Rispens (Rispens et al., 1972), Mdll/75C (Liu
and Lee, 1983), SB1 (Schat and Calnek, 1978), and HVT (Okazaki et al., 1970)] have been
used to effectively combat MD since the late 1960s, but the virus has become resistant to
many of them.. Outbreaks in vaccinated chickens resulted in the need to use new vaccines
(Biggs, 2001).
As MD becomes resistant to more vaccines, it will be difficult to find
effective vaccines to prevent outbreaks. Therefore, there is a need to increase chickens'
genetic resistance to the disease.
MD is a lymphoproliferative disease caused by a herpes virus (Biggs, 2001). Initial infection
occurs when a chicken inhales cell free Marek's disease virus (MDV), which can occur as a
result of being in close approximation to an infected bird. Upon infection, the disease can
cause lesions in peripheral nerve cells, gonads, irises, viscera, muscles, and skin (Calnek and
Witter, 1984). Peripheral nerve cell infection results in inflammatory lesions in the nerves,
paralysis, and death.
The progression of the MD infection can be divided into four sequential phases: early
cytolytic, latent, late cytolytic, and transforming (Calnek, 2001). In the early cytolytic phase,
the disease is carried from the respiratory tract to lymphoid organs by phagocytic cells,
resulting in inflammation of these organs. Mostly B cells are infected in this phase. The
latent phase, which begins at about 7 days post infection, is characterized by infection of
peripheral blood lymphocytes and a transient immunosuppression. Most of the cells infected
in this phase are T cells. At approximately 14 days post infection, the late cytolytic phase
28
begins. This phase is characterized by infection of non-lymphoid cells throughout the body.
Cell-free infectious MD is generated in feather follicle epithelium cells in the late cytolytic
phase. This is the only tissue that generates virus that has the ability to be spread from
animal to animal (Calnek et al., 1970). In the transforming phase, which can begin anytime
after the late cytolytic phase, lesions can form in many tissues, resulting in impairment of
function and death.
QTL Detection in Chickens
QTL detected for Growth and Carcass Traits in Chickens
Several studies have been published identifying QTL for growth and/or carcass traits in
experimental populations of chickens. One population, designed as a three generation full-sib
population, consisted of 20 F1 individuals derived from an outbred cross of two White
Plymouth Rock lines, resulting in approximately 451 F2 individuals and approximately 2000
F3 individuals (van Kaam et al., 1999b; van Kaam et al., 1999a; van Kaam et al., 1998). The
Fis and F2s were genotyped, and the phenotypes of the F3s were recorded to assign progeny
trait averages to the F2 individuals. The number of markers used ranged from 368 to 437 per
analysis. All markers were microsatellites. The method of least squares regression (Haley
and Knott, 1992; Haley et al., 1994; Knott et al., 1996) extended to full sib families (van
Kaam et al., 1998) was used for analysis. One genome-wise significant (p < 0.05) QTL for
feed intake was found on chromosome 1, which was suggestive for growth and body weight.
Suggestive QTL were also found on chromosomes 4 and 23 for feed intake, on chromosome
2 for feed intake adjusted for body weight, chromosome 1 for carcass percentage, and on
chromosome 2 for meat color. Jennen et al. (2004) also used this population to identify two
significant (5% genome-wise) QTL: on chromosome 1 for percentage abdominal fat and on
chromosome 13 for body weight.
Tatsuda et al. (2000) and Tatsuda and Fujinaka (2001) searched for QTL for body weight in
an F2 cross between Satsumadori and White Leghorn chicken lines. In these studies, 72 and
78 markers were genotyped and analyzed in 246 and 241 F2 birds, respectively.
The
MAPMAKER/QTL program (Kruglyak and Lander, 1995b), which utilizes a maximum
29
likelihood algorithm, was used for QTL analysis. They found QTL affecting body weight on
chromosomes 1 and 2 with LOD scores > 3.
Tuiskula-Haavisto et al. (2002) used a population of 307 F2 hens generated from a cross
between a Rhode Island Red line and a White Leghorn line. Using 99 markers for the
analysis, least squares regression interval mapping (Haley et al., 1994) was employed to
discover a QTL on chromosome 4 for body weight at 40 weeks of age and feed intake
significant at the 5% genome-wise threshold. Tuiskula-Haavisto et al. (2004) used this
population to identify parent of origin QTL for egg and growth traits. QTL with significant
(10% genome-wise) maternal parent of origin effects were identified on Gga 1 for
bodyweight and on E36 for feed intake.
Sewalem et al. (2002) crossed a white leghorn egg laying line with a commercial broiler sire
line to generate a mapping population of 546 F2 offspring. The least squares regression
interval mapping method of Haley et al. (1994) was used for 101 microsatellite markers to
detect significant (5% genome-wise) QTL for body weight at 3 weeks on chromosomes 1, 7,
13, and Z; for body weight at 6 weeks on chromosomes 1, 2, 4, 7, 8, and 13; and for body
weight at 9 weeks on chromosomes 1, 2, 4, 8, 13, and 27. Ikeobi et al. (2002) also used this
population to map QTL affecting abdominal fat weight, abdominal fat weight adjusted for
bodyweight, skin fat weight, skin fat weight adjusted for body weight, and abdominal fat
weight adjusted for skin fat weight. Least squares regression interval mapping (Haley et al.,
1994) was used to analyze 102 microsatellite markers. Significant (5% genome-wise) QTL
affecting one or more of the fat traits were found on chromosomes 1, 3, 5, 7, 13, 15, and 28.
Carlborg et al. (2004) used the same population genotyped for 101 microsatellite markers to
identify more QTL for growth traits with an epistatic model. Significant (5% genome-wise)
marginal or epistatic QTL for body weight and/or growth rate were identified on
chromosomes 1, 2, 3, 4, 6, 7, 8, 13, 18, and 27.
Deeb and Lamont (2003) used F1 individuals (n=600) from a cross between two broiler sires
and dams from two highly inbred lines to locate QTL affecting body weight at eight weeks of
30
age. The experiment was performed in two stages, starting with a selective DNA pooling
analysis of 136 microsatellite markers, followed by selective genotyping for markers
significant in the pooling analysis (n = 10). The selective genotyping analysis identified
genomic regions significantly (5% comparison-wise) associated with body weight at eight
weeks of age on chromosomes 1 and 2, and on linkage group E46C08W18.
Kerje et al. (2003) utilized an F2 cross between White Leghorn and Red Junglefowl chickens
to identify QTL for growth and body weight. The F2 generation consisted of 851 individuals
that were genotyped for 105 microsatellite markers. Body weight and growth were measured
at intervals from day 1 to day 200 of age. An outbred line least-squares regression method
(Haley et al., 1994) was used to identify genomic regions significantly (5% genome-wise)
associated with growth and/or bodyweight on chromosomes 1, 2, 3, 5, 7, 8, 11, 12, 27, and Z.
Carlborg et al. (2003) used an epistatic model with the same population and markers to
identify significant (5% genome-wise) QTL affecting growth and/or growth rate on
chromosomes 1, 2, 3,4, 5, 7, 8, 11,12, 13, 14, 18, 27, and linkage group E47W24.
De Koning et al. (2003) confirmed a QTL on chromosome 4 affecting bodyweight and feed
intake found in previous studies of experimental crosses (Sewalem et al., 2002; TuiskulaHaavisto et al., 2002; van Kaam et al., 1999a; van Kaam et al., 1998) in a commercial broiler
line.
A three-generation half-sib population design (~ 500 half-sib individuals) was
employed to detect the QTL and the half-sib module of QTL Express software (Seaton et al.,
2002), which utilizes statistical methods described by Knott et al. (1996), was used for the
analysis. Multiple QTL affecting the traits of interest were confirmed (comparison-wise p <
0.05) to be segregating in this population in the region of chromosome 4 under analysis. The
same population was used by de Koning et al. (2004) to verify other QTL for meat
production, including body weight and muscle weights, that had been previously identified in
experimental crosses. The HS module of QTL Express (Seaton et al., 2002) and a multiple
QTL model (de Koning et al., 2001c) were used to identify associations of phenotype with
the segregation of chromosomal regions. QTL identified in various experimental crosses
31
were confirmed (comparison-wise p < 0.01) in this population on chromosomes 1, 3, 4, 5, 7,
8, 9, 11, and 13.
Sasaki et al. (2004) used an F2 (265 F2 individuals) cross between a White Leghorn line and
a Rhode Island Red line that were genotyped for 123 microsatellite markers to identify QTL
for body weight. Map Manager QTX bl8 (Manly et al., 2001), which employs least-squares
regression (Haley and Knott, 1992), was utilized for the analyses. Significant (5% genomewise) QTL affecting body weight were identified on chromosomes 4 and 27.
Siwek et al. (2004) used an F2 population (n = 672) created by crossing two medium-heavy
layer lines that were divergently selected for primary antibody response to sheep red blood
cells. The population was genotyped for 174 microsatellite markers.
A half-sib linear
regression interval-mapping model (de Koning et al., 1999; Knott et al., 1996) and a linear
regression line cross model (Haley et al., 1994) were used to identify genomic regions
associated with body weight at 6, 8, 12, and 18 weeks of age. Significant (genome-wise p <
0.05) QTL were detected on chromosomes 2, 3, and Z. Currently, no reports of studies to
identify QTL associated with breast meat yield in chickens could be identified.
Several genes associated with growth traits in chickens have been identified using the
candidate gene approach. Jiang et al. (2002) identified an association of variation in the
melanocortin-3 receptor gene with body weight and abdominal fat in chickens. Johnson et al.
(1995) found an association between Ornithine Decarboxylase transcription levels and
growth, and Parsanejad et al. (2004) found and association of allelic variation of the
Ornithine Decarboxylase gene with bodyweight at sexual maturity. Allelic variation in the
myostatin gene [chromosome 7: (Sazanov et al., 1999)] was found to be associated with
abdominal fat weight, abdominal fat percentage, birth weight, breast muscle weight, and
breast muscle percentage (Zhiliang et al., 2004), and Guernec et al. (2003) found an
association between mystatin mRNA levels and growth rate. Guernec et al. (2003) also
found that the insulin-like growth factor-I (IGF1) gene (chromosome 1) had an allelic
association with growth rate. Zhou et al. (2005) found the IGF1 gene to be associated with
32
bodyweight, average daily gain, breast meat weight, drumstick weight, breast meat percent,
abdominal fat weight percent, liver weight percent, heart weight percent, drumstick weight
percent, and various metabolic and skeletal traits. Li et al. (2003) identified associations
between transforming growth factor /? genes [TGFB2 (Gga 3), TGFB3 (Gga 5), and TGFB4
(unmapped)] with bodyweight, average daily gain, breast meat weight, abdominal fat weight,
spleen weight, abdominal fat percent, spleen weight percent, and various skeletal
measurements.
QTL Mapping Studies for Disease Resistance in Chickens
Marek's Disease QTL. Vallejo et al. (1998) and Yonash et al. (1999) studied 272 F2
individuals generated from a cross between two White Leghorn lines that differed in
susceptibility to MD to map QTL for MD resistance. Both studies used a multiple stage
approach.
First, extreme animals (based on MD virus concentration two weeks after
challenge with MD virus, number of different tissues showing tumors, and a tumor index)
were genotyped for markers distributed throughout the genome for initial screening, and
second, all animals in the population were genotyped for markers in regions found to have an
association in the first stage. The regions were then analyzed for associations with various
tumor traits, viral load, and survival, which were used to identify susceptibility to MD. Map
Manager QT (Manly and Olson, 1999), which employs least squares regression analysis
(Haley and Knott, 1992), and MAPMAKER/QTL (Kruglyak and Lander, 1995b), which uses
maximum likelihood, were used to locate significant (5% chromosome-wise) QTL on
chromosomes 1, 2, 4, 7, and 8 affecting susceptibility to MD. Bumstead (1998) performed
an interval mapping study on a backcross of the two parental lines used by Vallejo et al.
(1998) and Yonash et al. (1999). A significant (5% genome-wise) QTL for MD resistance
was identified on chromosome 1.
A genetic region of special note when performing genome scans in chickens for disease traits
is the Major Histocompatability Complex (MHC) on chromosome 16. The B blood group
locus is linked to the MHC (Schierman and Nordskog, 1961), and is commonly used as a
marker to identify MHC haplotypes. The MHC has been shown to have an association with
33
resistance to many diseases in poultry, including MD (Bacon, 1987; Bacon and Witter, 1994;
Bacon et al., 1981; Lakshmanan et al., 1997; Lamont, 1989, 1998; Steadham et al., 1987).
Interactions of the MHC with QTL and background genetics affecting disease resistance have
also been found in chickens (Dunnington et al., 1989; Kaiser et al., 2002). Because of the
MHC's known effect on disease resistance and possible interactions with other disease
resistance genes, MHC genotypes must be considered when mapping QTL for disease
resistance in chickens.
In addition to the MHC, several other genes have been identified as having an association
with MD resistance in chickens using the candidate gene approach. The Rfp-Y region of
chromosome 16 has been shown to have an association with Marek's disease resistance
(Lakshmanan and Lamont, 1998; Wakenell et al., 1996). The growth hormone gene (GH1)
on chromosome 1 also has an allelic association with MD resistance (Kuhnlein et al., 1997;
Liu et al., 2001b). Microarray analysis has shown that GH1 expression is associated with
differences in MD resistance (Liu et al., 2001a) and the GH1 protein has been shown to
interact with the SORF2 protein, a protein only found in virulent MD virus strains (Liu et al.,
2001b). Stem lymphocyte antigen 6 complex locus E (LY6E) on chromosome 2 has also
been identified as an MD resistance gene through genetic, RNA, and protein analysis (Liu et
al., 2003).
Other Disease Resistance QTL. Several mapping studies have been successful in identifying
QTL in chickens associated with resistance to diseases other than MD. Yonash et al. (2001)
reported an experiment to locate genomic regions associated with antibody response to
Newcastle disease virus, Escherichia coli bacteria, and sheep red blood cells (SRBC), and
survival rate in meat type chickens.
The population studied was created by crossing
individuals from two broiler lines that had been divergently selected for high and low
antibody response to E. coli. One male offspring from this cross was then mated with
females from the cross and females from each of the parental lines to generate 160 half-sib
progeny. Significant associations were found on chromosome 2 for antibody response to
34
SRBC and the Newcastle disease virus, on chromosome 5 for survival and antibody response
to E. coli, and on chromosome 18 for antibody response to the Newcastle disease virus.
Kaiser et al. (2002) used a multiple stage approach to locate genomic regions associated with
antibody response to Salmonella enteritidis bacteria vaccine. The population under study
was created by crossing four broiler males with dams from highly inbred Fayoumi, Spanish,
and Leghorn lines to generate 388 offspring. The population was first analyzed with an
across-family pooling analysis of 79 microsatellite markers followed by within family
pooling analysis of 15 markers that showed allelic frequency differences in the acrosspopulation pools. From these 15 markers, four were chosen for individual genotyping in the
entire population. Markers on chromosomes 1,5, and 6 were found to be associated response
to the vaccine.
Yunis et al. (2002) used multiple crosses of an F1 population generated from 2 divergently
selected lines for antibody response to E. coli to locate QTL for antibody response to E. coli
and S. enteritidis vaccines.
Markers with significant associations with E. coli and/or S.
enteritidis vaccine antibody response were found on chromosomes 1, 2, 3, 5, 7, 8, 17, and 28.
Mariani et al. (2001) used a population consisting of 321 individuals generated from a
backcross of two lines that differed in salmonella resistance to map QTL for Salmonella
typhimurium resistance, as defined by the number of bacteria in the spleen five days post
infection. Backcross individuals were intravenously infected with the bacteria at two weeks
of age. A significant QTL for S. typhimurium resistance was located on chromosome 5.
Summary and General Conclusions
In conclusion, growth, particularly breast meat yield, and resistance to Marek's disease are
very important traits to the chicken industry. Because of the extensive resources needed to
measure and perform phenotypic selection for these traits, MAS would be very useful in
improving them. To perform MAS, linkage between markers and QTL affecting a trait must
35
be determined. Therefore, the objective of the research that follows is to identify QTL
affecting economic traits in chickens, including white meat yield and other growth and
carcass traits in broiler chickens, and MD resistance in layer chickens. To accomplish this
objective, genome scans using microsatellite markers were performed.
Microsatellite
markers are used in studies presented in this thesis because they are highly polymorphic, they
are codominant, and many have been mapped across the chicken genome. The anonymous
marker least-squares QTL mapping approach was utilized in the studies presented in this
thesis for the following reasons:
1. Anonymous marker vs. candidate gene approach
A. No assumptions of function or position of genes need to be made prior to the
analysis with the anonymous marker approach.
B. Large genomic regions can be analyzed, leading to a greater chance of
discovering multiple QTL with the anonymous marker approach.
C. Regions with genes that would be analyzed in the candidate gene approach
can still be analyzed in the anonymous marker approach.
D. New QTL can be identified that may not have been identified previously in
the chicken or other species with the anonymous marker approach.
2. Least-Squares vs. Maximum Likelihood
A. The least squares method of interval mapping yields very similar results to the
maximum
likelihood
method
for
designed
QTL
mapping
resource
populations, but is much less computationally demanding (Haley and Knott,
1992).
Interval mapping was used in the growth and composition studies in this thesis (Sections 4.1
and 4.2) instead of single marker analysis because interval mapping analysis has more power
to detect QTL than single marker analysis (Lander and Botstein, 1989), and the effect and
position of a QTL can be determined in an interval mapping analysis, whereas they are
confounded in single marker analysis.
36
For the MD resistance study (Section 4.3), marker density and the available genotypes were
not sufficient to allow interval mapping, so single marker analysis was used. Because of the
distribution of the phenotype (i.e. selectively genotyped survival data with censoring) in this
study, survival models (Cox proportion hazards and Weibull models) were used, in addition
to least-squares regression, to analyze the data (and are compared in Section 4.4 through
simulation) because:
1.
The data violated the assumption of normality required for least-squares
regression.
2.
Some of the data were censored, which cannot be appropriately handled by leastsquares regression.
3.
The Cox proportional hazards model is semiparametric, and therefore expected
to be less affected by the missing data in the center of the phenotypic distribution
resulting from selective genotyping.
Bibliography
Ameli, H., J. S. Gavora, J. L. Spencer, and R. W. Fairfull. 1992. Genetic resistance to two
Marek's disease viruses and its relationship to production traits in chickens. Can. J.
Anim. Sci. 72: 213-225.
Andersson-Eklund, L., L. Marklund, K. Lundstrom, C. S. Haley, K. Andersson, I. Hansson,
M. Moller, and L. Andersson. 1998. Mapping quantitative trait loci for carcass and
meat quality traits in a wild boar x large white intercross. J. Anim. Sci. 76: 694-700.
Bacon, L. D. 1987. Influence of the major histocompatibility complex on disease resistance
and productivity. Poult. Sci. 66: 802-811.
Bacon, L. D., and R. L. Witter. 1994. B haplotype influence on the relative efficacy of
Marek's disease vaccines in commercial chickens. Poult. Sci. 73: 481-487.
Bacon, L. D., R. L. Witter, L. B. Crittenden, A. Fadly, and J. Motta. 1981. B-haplotype
influence on Marek's disease, Rous sarcoma, and lymphoid leukosis virus-induced
tumors in chickens. Poult. Sci. 60: 1132-1139.
37
Becker, W. M., J. B. Reece, and M. F. Poenie. 1996. The world of the cell. 3 ed. The
Benjamin/Cummings Publishing Company, Inc., Menlo Park, CA.
Benard, H. J., K. D. Stark, R. S. Morris, D. U. Pfeiffer, and H. Moser. 1999. The 1997-1998
classical swine fever epidemic in the Netherlands—a survival analysis. Prev. Vet.
Med. 42: 235-248.
Biggs, P. M. 2001. The history and biology of Marek's disease virus. In: K. Hirai (ed.)
Marek's disease, p 1-24. Springer, Berlin, Germany.
Bumstead, N. 1998. Genomic mapping of resistance to Marek's disease. Avian Pathol. 27:
S78-S81.
Cahaner, A., and Z. Nitsan. 1995. Evaluation of simultaneous selection for live body weight
and against abdominal fat in broilers. Poult. Sci. 64: 1257-1263.
Calnek, B. W. 2001. Pathogenesis of Marek's disease virus infection. In: K. Hirai (ed.)
Marek's disease, p 25-56. Springer, Berlin, Germany.
Calnek, B. W., H. K. Adldinger, and D. E. Kahn. 1970. Feather follicle epithelium: A source
of enveloped and infectious cell-free herpesvirus from Marek's disease. Avian Dis.
14: 219-233.
Calnek, B. W., and R. L. Witter. 1984. Marek's disease. In: M. S. Hofstad (ed.) Diseases of
poultry, p 325-359. Iowa State University Press, Ames, Iowa.
Carlborg, O., P. M. Hocking, D. W. Burt, and C. S. Haley. 2004. Simultaneous mapping of
epistatic QTL in chickens reveals clusters of QTL pairs with similar genetic effects
on growth. Genet. Res. 83: 197-209.
Carlborg, O., S. Kerje, K. Schutz, L. Jacobsson, P. Jensen, and L. Andersson. 2003. A global
search reveals epistatic interaction between QTL for early growth in the chicken.
Genome Res. 13: 413-421.
Chambers, J. R. 1990. Genetics of growth and meat production in chickens. In: R. D.
Crawford (ed.) Poultry breeding and genetics, p 599-644. Elsevier Science Publishing
Company Inc., New York, New York.
Churchill, G. A., and R. W. Doerge. 1994. Empirical threshold values for quantitative trait
mapping. Genetics 138: 963-971.
38
Cox, D. R. 1972. Regression models in life tables (with discussion). J. Roy. Stat. Soc. B 34:
187-220.
Darvasi, A., and M. Soller. 1992. Selective genotyping for determination of linkage between
a marker locus and a quantitative trait locus. Theor. Appl. Genet. 85: 353-359.
de Koning, D. J., C. S. Haley, D. Windsor, P. M. Hocking, H. Griffin, A. Morris, J. Vincent,
and D. W. Burt. 2004. Segregation of QTL for production traits in commercial meattype chickens. Genet Res. 83: 211-220.
de Koning, D. J., B. Harlizius, A. P. Rattink, M. A. Groenen, E. W. Brascamp, and J. A. van
Arendonk. 2001a. Detection and characterization of quantitative trait loci for meat
quality traits in pigs. J. Anim. Sci. 79: 2812-2819.
de Koning, D. J., L. L. Janss, A. P. Rattink, P. A. van Gers, B. J. de Vries, M. A. Groenen, J.
J. van der Poel, P. N. de Groot, E. W. Brascamp, and J. A. van Arendonk. 1999.
Detection of quantitative trait loci for backfat thickness and intramuscular fat content
in pigs (Sus scrofa). Genetics 152: 1679-1690.
de Koning, D. J., A. P. Rattink, B. Harlizius, M. A. Groenen, E. W. Brascamp, and J. A. van
Arendonk. 2001b. Detection and characterization of quantitative trait loci for growth
and reproduction traits in pigs. Livest. Prod. Sci. 72: 185-198.
de Koning, D. J., A. P. Rattink, B. Harlizius, J. A. van Arendonk, E. W. Brascamp, and M.
A. Groenen. 2000. Genome-wide scan for body composition in pigs reveals important
role of imprinting. Proc. Natl. Acad. Sci. USA 97: 7947-7950.
de Koning, D. J., N. F. Schulmant, K. Elo, S. Moisio, R. Kinos, J. Vilkki, and A. MakiTanila. 2001c. Mapping of multiple quantitative trait loci by simple regression in
half-sib designs. J. Anim. Sci. 79: 616-622.
de Koning, D. J., D. Windsor, P. M. Hocking, D. W. Burt, A. Law, C. S. Haley, A. Morris, J.
Vincent, and H. Griffin. 2003. Quantitative trait locus detection in commercial broiler
lines using candidate regions. J. Anim. Sci. 81: 1158-1165.
Deeb, N., and S. J. Lamont. 2003. Use of a novel outbred by inbred F1 cross to detect genetic
markers for growth. Anim. Genet. 34: 205-212.
Dekkers, J. C., and F. Hospital. 2002. The use of molecular genetics in the improvement of
agricultural populations. Nat. Rev. Genet. 3: 22-32.
39
Dunnington, E. A., A. Martin, R. W. Briles, W. E. Briles, W. B. Gross, and P. B. Siegel.
1989. Antibody responses to sheep erythrocytes for White Leghorn chickens differing
in haplotypes of the major histocompatibility complex (B). Anim. Genet. 20: 213216.
Edwards, M. D., C. W. Stuber, and J. F. Wendel. 1987. Molecular-marker-facilitated
investigations of quantitative-trait loci in maize. I. Numbers, genomic distribution and
types of gene action. Genetics 116: 113-125.
Emmerson, D. A. 1997. Commercial approaches to genetic selection for growth and feed
conversion in domestic poultry. Poult. Sci. 76: 1121-1125.
Falconer, D. S., and T. F. C. Mackay. 1996. Introduction to quantitative genetics. 4th ed.
Longman Group Ltd., Essex, England.
Fisher, R. A. 1935. The design of experiments. 3rd ed. Oliver and Boyd Ltd., London.
Gavora, J. S. 1990.
Disease genetics.
In: R. D. Crawford (ed.) Poultry breeding and
genetics, p 805-846. Elsevier Science Publishing Company Inc., New York, New
York.
Gehlhar, M., and W. Coyle. 2001. Changing structure of global food consumption and trade:
Global food consumption and impacts on trade patterns. WRS-01-1, Economic
Research Service/USDA, Washington, D. C.
Geldermann, H. 1975. Investigations on inheritance of quantitative characters in animals by
gene markers I. Methods. Theor. Appl. Genet. 46: 319-330.
Goldspink, G., and S. Y. Yang. 1999. Muscle structure, development and growth. In: R. I.
Richardson and G. C. Mead (eds.) Poultry meat science No. 25. p 3-18. CAB
International, New York, New York.
Groenen, M. A., H. H. Cheng, N. Bumstead, B. F. Benkel, W. E. Briles, T. Burke, D. W.
Burt, L. B. Crittenden, J. Dodgson, J. Hillel, S. Lamont, A. P. de Leon, M. Soller, H.
Takahashi, and A. Vignal. 2000. A consensus linkage map of the chicken genome.
Genome Res. 10: 137-147.
Guernec, A., C. Berri, B. Chevalier, N. Wacrenier-Cere, E. Le Bihan-Duval, and M. J.
Duclos. 2003. Muscle development, insulin-like growth factor-I and myostatin
40
mRNA levels in chickens selected for increased breast muscle yield. Growth Horm.
IGF Res. 13: 8-18.
Haley, C. S., and S. A. Knott. 1992. A simple regression method for maping quantitative trait
loci in line crosses using flanking markers. Heredity 69: 315-324.
Haley, C. S., S. A. Knott, and J. M. Elsen. 1994. Mapping quantitative trait loci in crosses
between outbred lines using least squares. Genetics 136: 1195-1207.
Haley, M. M. 2001. Changing consumer demand for meat: The U.S. Example, 1970 - 2000.
From USDA: Changing Structure of Global Food Consumption and Trade. WRS-011.
Hanson, M. P., G. R. J. Law, and J. N. Van Zandt. 1967. Differences in susceptibility to
Marek's disease in chickens carrying two different B locus blood group alleles. Poult.
Sci. 46: 1268 (Abstr.).
Henshall, J. M., and M. E. Goddard. 1999. Multiple-trait mapping of quantitative trait loci
after selective genotyping using logistic regression. Genetics 151: 885-894.
Hillier, L. W., W. Miller, E. Bimey, W. Warren, R. C. Hardison, C. P. Ponting, P. Bork, D.
W. Burt, M. A. Groenen, M. E. Delany, J. B. Dodgson, A. T. Chinwalla, P. F. Cliften,
S. W. Clifton, K. D. Delehaunty, C. Fronick, R. S. Fulton, T. A. Graves, C.
Kremitzki, D. Layman, V. Magrini, J. D. McPherson, T. L. Miner, P. Minx, W. E.
Nash, M. N. Nhan, J. O. Nelson, L. G. Oddy, C. S. Pohl, J. Randall-Maher, S. M.
Smith, J. W. Wallis, S. P. Yang, M. N. Romanov, C. M. Rondelli, B. Paton, J. Smith,
D. Morrice, L. Daniels, H. G. Tempest, L. Robertson, J. S. Masabanda, D. K. Griffin,
A. Vignal, V. Fillon, L. Jacobbson, S. Kerje, L. Andersson, R. P. Crooijmans, J.
Aerts, J. J. van der Poel, H. Ellegren, R. B. Caldwell, S. J. Hubbard, D. V. Grafham,
A. M. Kierzek, S. R. McLaren, I. M. Overton, H. Arakawa, K. J. Beattie, Y.
Bezzubov, P. E. Boardman, J. K. Bonfield, M. D. Croning, R. M. Davies, M. D.
Francis, S. J. Humphray, C. E. Scott, R. G. Taylor, C. Tickle, W. R. Brown, J.
Rogers, J. M. Buerstedde, S. A. Wilson, L. Stubbs, I. Ovcharenko, L. Gordon, S.
Lucas, M. M. Miller, H. Inoko, T. Shiina, J. Kaufman, J. Salomonsen, K. Skjoedt, G.
K. Wong, J. Wang, B. Liu, J. Yu, H. Yang, M. Nefedov, M. Koriabine, P. J. Dejong,
L. Goodstadt, C. Webber, N. J. Dickens, I. Letunic, M. Suyama, D. Torrents, C. von
41
Mering, E. M. Zdobnov, K. Makova, A. Nekrutenko, L. Elnitski, P. Eswara, D. C.
King, S. Yang, S. Tyekucheva, A. Radakrishnan, R. S. Harris, F. Chiaromonte, J.
Taylor, J. He, M. Rijnkels, S. Griffiths-Jones, A. Ureta-Vidal, M. M. Hoffman, J.
Severin, S. M. Searle, A. S. Law, D. Speed, D. Waddington, Z. Cheng, E. Tuzun, E.
Eichler, Z. Bao, P. Flicek, D. D. Shteynberg, M. R. Brent, J. M. Bye, E. J. Huckle, S.
Chatteiji, C. Dewey, L. Pachter, A. Kouranov, Z. Mourelatos, A. G. Hatzigeorgiou,
A. H. Paterson, R. Ivarie, M. Brandstrom, E. Axelsson, N. Backstrom, S. Berlin, M.
T. Webster, O. Pourquie, A. Reymond, C. Ucla, S. E. Antonarakis, M. Long, J. J.
Emerson, E. Betran, I. Dupanloup, H. Kaessmann, A. S. Hinrichs, G. Bejerano, T. S.
Furey, R. A. Harte, B. Raney, A. Siepel, W. J. Kent, D. Haussier, E. Eyras, R.
Castelo, J. F. Abril, S. Castellano, F. Camara, G. Parra, R. Guigo, G. Bourque, G.
Tesler, P. A. Pevzner, A. Smit, L. A. Fulton, E. R. Mardis, and R. K. Wilson. 2004.
Séquence and comparative analysis of the chicken genome provide unique
perspectives on vertebrate evolution. Nature 432: 695-716.
Hirst, W. M., R. D. Murray, W. R. Ward, and N. P. French. 2002. A mixed-effects time-toevent analysis of the relationship between first-lactation lameness and subsequent
lameness in dairy cows in the UK. Prev. Vet. Med. 54: 191-201.
Hospital, F., C. Chevalet, and P. Mulsant. 1992. Using markers in gene introgression
breeding programs. Genetics 132: 1199-1210.
Ikeobi, C. O., J. A. Woolliams, D. R. Morrice, A. Law, D. Windsor, D. W. Burt, and P. M.
Hocking. 2002. Quantitative trait loci affecting fatness in the chicken. Anim. Genet.
33:428-435.
Jansen, R. C., and P. Stam. 1994. High resolution of quantitative traits into multiple loci via
interval mapping. Genetics 136: 1447-1455.
Jeffreys, A. J., V. Wilson, and S. L. Thein. 1985. Hypervariable 'minisatellite' regions in
human DNA. Nature 314: 67-73.
Jennen, D. G., A. L. Vereijken, H. Bovenhuis, R. P. Crooijmans, A. Veenendaal, J. J. van der
Poel, and M. A. Groenen. 2004. Detection and localization of quantitative trait loci
affecting fatness in broilers. Poult. Sci. 83: 295-301.
42
Jiang, S. W., L. Jacobsson, S. Kerje, L. Andersson, and Y. Z. Xiong. 2002. Abst. [Studies of
relationship between the melanocortin-3 receptor gene and body weight in chicken
for high and low weight lines' intercross]. Yi Chuan Xue Bao 29: 322-325.
Johnson, R., G. Bulfield, A. Tait, and C. Goddard. 1995. Increased expression and activity of
ornithine decarboxylase in chicks genetically selected for rapid growth rate. Comp.
Biochem. Physiol. B Biochem. Mol. Biol. 110: 531-537.
Kaiser, M. G., N. Deeb, and S. J. Lamont. 2002. Microsatellite markers linked to Salmonella
enterica serovar enteritidis vaccine response in young F1 broiler-cross chicks. Poult.
Sci. 81: 193-201.
Kao, C. H., Z. B. Zeng, and R. D. Teasdale. 1999. Multiple interval mapping for quantitative
trait loci. Genetics 152: 1203-1216.
Kerje, S., O. Carlborg, L. Jacobsson, K. Schutz, C. Hartmann, P. Jensen, and L. Andersson.
2003. The twofold difference in adult size between the Red Junglefowl and White
Leghorn chickens is largely explained by a limited number of QTLs. Anim. Genet.
34: 264-274.
Kim, K. S., N. Larsen, T. Short, G. Plastow, and M. F. Rothschild. 2000. A missense variant
of the porcine melanocortin-4 receptor (MC4R) gene is associated with fatness,
growth, and feed intake traits. Mamm. Genome 11: 131-135.
Knott, S. A., J. M. Elsen, and C. S. Haley. 1996. Methods for multiple-marker mapping of
quantitative trait loci in half-sib populations. Theor. Appl. Genet. 93: 71-80.
Knott, S. A., L. Marklund, C. S. Haley, K. Andersson, W. Davies, H. Ellegren, M. Fredholm,
I. Hansson, B. Hoyheim, K. Lundstrom, M. Moller, and L. Andersson. 1998. Multiple
marker mapping of quantitative trait loci in a cross between outbred Wild Boar and
Large White pigs. Genetics 149: 1069-1080.
Koerhuis, A. N. M. 1996. Quantitative genetic studies on growth and reproductive traits in
broiler chickens, University of Edinburgh, Edinburgh.
Komberg, A. 1980. DNA Replication. Freeman, San Francisco.
Koski, L. B., E. Sasaki, R. D. Roberts, J. Gibson, and R. J. Etches. 2000. Monoalleleic
transcription of the insulin-like growth factor-ii gene (ig/2) in chick embryos. Mol.
Reprod. Dev. 56: 345-352.
43
Kruglyak, L., and E. S. Lander. 1995a. Complete multipoint sib-pair analysis of qualitative
and quantitative traits. Am. J. Hum. Genet. 57: 439-454.
Kruglyak, L., and E. S. Lander. 1995b. A nonparametric approach for mapping quantitative
trait loci. Genetics 139: 1421-1428.
Kuhnlein, U., L. Ni, S. Weigend, J. S. Gavora, W. Fairfull, and D. Zadworny. 1997. DNA
polymorphisms in the chicken growth hormone gene: Response to selection for
disease resistance and association with egg production. Anim. Genet. 28: 116-123.
Kuurman, W. W., B. A. Bailey, W. J. Koops, and M. Grossman. 2003. A model for failure of
a chicken embryo to survive incubation. Poult. Sci. 82: 214-222.
Lakshmanan, N., J. S. Gavora, and S. J. Lamont. 1997. Major histocompatibility complex
class II DNA polymorphisms in chicken strains selected for Marek's disease
resistance and egg production or for egg production alone. Poult. Sci. 76: 1517-1523.
Lakshmanan, N., and S. J. Lamont. 1998. Rfp-Y region polymorphism and Marek's disease
resistance in multitrait immunocompetence-selected chicken lines. Poult. Sci. 77:
538-541.
Lamont, S. J. 1989. The chicken major histocompatibility complex in disease resistance and
poultry breeding. J. Dairy Sci. 72: 1328-1333.
Lamont, S. J. 1998. Impact of genetics on disease resistance. Poult. Sci. 77: 1111-1118.
Lande, R., and R. Thompson. 1990. Efficiency of marker-assisted selection in the
improvement of quantitative traits. Genetics 124: 743-756.
Lander, E., and L. Kruglyak. 1995. Genetic dissection of complex traits: Guidelines for
interpreting and reporting linkage results. Nat Genet 11: 241-247.
Lander, E. S., and D. Botstein. 1989. Mapping mendelian factors underlying quantitative
traits using RFLP linkage maps. Genetics 121: 185-199.
Lasley, J. F. 1987. Genetics of livestock improvement. Prentice-Hall, Inc., Englewood Cliffs,
New Jersey.
Le Bihan-Duval, E., S. Mignon-Grasteau, N. Millet, and C. Beaumont. 1998. Genetic
analysis of a selection experiment on increased body weight and breast muscle weight
as well as on limited abdominal fat weight. Br. Poult. Sci. 39: 346-353.
44
Li, H., N. Deeb, H. Zhou, A. D. Mitchell, C. M. Ashwell, and S. J. Lamont. 2003. Chicken
quantitative trait loci for growth and body composition associated with transforming
growth factor-/? genes. Poult. Sci. 82: 347-356.
Litt, M., and J. A. Luty. 1989. A hypervariable microsatellite revealed by in vitro
amplification of a dinucleotide repeat within the cardiac muscle actin gene. Am. J.
Hum. Genet. 44: 397-401.
Liu, H. C., H. H. Cheng, V. Tirunagaru, L. Sofer, and J. Bumside. 2001a. A strategy to
identify positional candidate genes conferring Marek's disease resistance by
integrating DNA microarrays and genetic mapping. Anim. Genet. 32: 351-359.
Liu, H. C., H. J. Kung, J. E. Fulton, R. W. Morgan, and H. H. Cheng. 2001b. Growth
hormone interacts with the Marek's disease virus SORF2 protein and is associated
with disease resistance in chicken. Proc. Natl. Acad. Sci. USA 98: 9203-9208.
Liu, H. C., M. Niikura, J. E. Fulton, and H. H. Cheng. 2003. Identification of chicken
lymphocyte antigen 6 complex, locus e (ly6e, alias seal) as a putative Marek's disease
resistance gene via a virus-host protein interaction screen. Cytogenet. Genome Res.
102: 304-308.
Liu, X. F., and L. F. Lee. 1983. Kinetics of phytohemagglutinin response in chickens
infected with various strains of Marek's disease virus. Avian Dis. 27 (3): 660-666.
Maizon, D. O., P. A. Oltenacu, Y. T. Grohn, R. L. Strawderman, and U. Emanuelson. 2004.
Effects of diseases on reproductive performance in Swedish Red and White dairy
cattle. Prev. Vet. Med. 66: 113-126.
Malek, M., J. C. Dekkers, H. K. Lee, T. J. Baas, and M. F. Rothschild. 2001. A molecular
genome scan analysis to identify chromosomal regions influencing economic traits in
the pig. I. Growth and body composition. Mamm. Genome 12: 630-636.
Manly, K. F., and J. M. Olson. 1999. Overview of QTL mapping software and introduction
to Map Manager QT. Mamm. Genome 10: 327-334.
Manly, K. F., R. H. Cudmore, Jr., and J. M. Meer. 2001. Map Manager QTX, crossplatform software for genetic mapping. Mamm. Genome 12: 930-932.
45
Mariani, P., P. A. Barrow, H. H. Cheng, M. M. Groenen, R. Negrini, and N. Bumstead. 2001.
Localization to chicken chromosome 5 of a novel locus determining salmonellosis
resistance. Immunogenetics 53: 786-791.
Marks, H. L. 1995. Genetics of growth and development. In: P. Hunton (ed.) Poultry
production, p 157-183. Elsevier Science B. V., Amsterdam, The Netherlands.
Masiga, D. K., A. Tait, and C. M. Turner. 2000. Amplified restriction fragment length
polymorphism in parasite genetics. Parasitai. Today 16: 350-353.
Meuwissen, T. H. E., and M. E. Goddard. 2000. Fine mapping of quantitative trait loci using
linkage disequilibria with closely linked marker loci. Genetics 155: 421-430.
Moore, T., and D. Haig. 1991. Genomic imprinting in mammalian development: A parental
tug-of-war. Trends Genet. 7: 45-49.
Morrow, C., and F. Fehler. 2004. Marek's disease: a worldwide problem. Pages 49-61 in
Marek's Disease, An Evolving Problem. F. Davison and V. Nair, ed. Elsevier Ltd.,
Oxford, U.K.
Muranty, H., and B. Goffmet. 1997. Selective genotyping for location and estimation of the
effect of a quantitative trait locus. Biometrics 53: 629-643.
Nielsen, V. H., C. Bendixen, J. Ambjerg, C. M. Sorensen, H. E. Jensen, N. M. Shukri, and B.
Thomsen. 2000. Abnormal growth plate function in pigs carrying a dominant
mutation in type X collagen. Mamm. Genome 11: 1087-1092.
Okazaki, W., H. G. Purchase, and B. R. Burmester. 1970. Protection against Marek's
disease by vaccination with a herpesvirus of turkey (HVT). Avian Res. 14: 413-429.
O'Neill, M. J., R. S. Ingram, P. B. Vrana, and S. M. Tilghman. 2000. Allelic expression of
ig/2 in marsupials and birds. Dev. Genes Evol. 210: 18-20.
Orita, M., H. Iwahana, H. Kanazawa, K. Hayashi, and T. Sekiya. 1989. Detection of
polymorphisms of human DNA by gel electrophoresis as single-strand conformation
polymorphisms. Proc. Natl. Acad. Sci. USA 86: 2766-2770.
Pardo-Manuel de Villena, F., E. de la Casa-Esperon, and C. Sapienza. 2000. Natural
selection and the function of genome imprinting: Beyond the silenced minority.
Trends Genet. 16: 573-579.
46
Parsanejad, R., D. Praslickova, D. Zadworny, and U. Kuhnlein. 2004. Ornithine
decarboxylase: Haplotype structure and trait associations in White Leghorn chickens.
Poult. Sci. 83: 1518-1523.
Pinstrup-Andersen, P., R. Pandya-Lorch, and M. W. Rosegrant. 1999. World food prospects:
Critical issures for the early twenty-first century, International Food Policy Research
Institute, Washington, D. C.
Pollock, D. L. 1999. A geneticist's perspective from within a broiler primary breeder
company. Poult. Sci. 78: 414-418.
Purchase, H. G. 1985. Clinical disease and its economic impact. In: L. N. Payne (ed.)
Marek's disease, scientific basis and methods of control, p 17-42. Martinus Nkjhoff
Publishing, Boston, Massachusetts.
Rispens, B. H., H. van Vloten, N. Mastenbroek, H. J. Maas, and K. A. Schat. 1972. Control
of Marek's disease in the Netherlands. I. Isolation of an avirulent Marek's disease
virus (strain CVI 988) and its use in laboratory vaccination trials. Avian Dis. 16 (1):
108-125.
Roenigk, W. P. 1999. Muscle growth and development. Keynote address: World poultry
consumption. Poult. Sci. 78: 722-728.
Rohrer, G. A., and J. W. Keele. 1998a. Identification of quantitative trait loci affecting
carcass composition in swine: I. Fat deposition traits. J. Anim. Sci. 76: 2247-2254.
Rohrer, G. A., and J. W. Keele. 1998b. Identification of quantitative trait loci affecting
carcass composition in swine: II. Muscling and wholesale product yield traits. J.
Anim. Sci. 76: 2255-2262.
Rose, S. P. 1997. Principles of poultry science. Cab International, New York, New York.
Rothschild, M. F., L. Messer, A. Day, R. Wales, T. Short, O. Southwood, and G. Plastow.
2000. Investigation of the retinol-binding protein 4 (RBP4) gene as a candidate gene
for increased litter size in pigs. Mamm. Genome 11: 75-77.
Rothschild, M. F., and M. Soller. 1997. Candidate gene analysis to detect genes controlling
traits of economic importance in domestic livestock. Probe 8: 13-20.
Roxstrom, A., V. Ducrocq, and E. Strandberg. 2003. Survival analysis of longevity in dairy
cattle on a lactation basis. Genet. Sel. Evol. 35: 305-318.
47
Sasaki, O., S. Odawara, H. Takahashi, K. Nirasawa, Y. Oyamada, R. Yamamoto, K. Ishii, Y.
Nagamine, H. Takeda, E. Kobayashi, and T. Furukawa. 2004. Genetic mapping of
quantitative trait loci affecting body weight, egg character and egg production in F2
intercross chickens. Anim. Genet. 35: 188-194.
Sax, K. 1923. The association of size differences with seed-coat pattern and pigmentation in
Phaseolus vulgaris. Genetics 8: 552-560.
Sazanov, A., D. Ewald, J. Buitkamp, and R. Fries. 1999. A molecular marker for the chicken
myostatin gene (GDF8) maps to 7pll. Anim. Genet. 30: 388-389.
Schat, K. A., and B. W. Calnek. 1978. Characterization of an apparently nononcogenic
Marek's disease virus. J. Natl. Cancer Inst. 60 (5): 1075-1082.
Schierman, L. W., and A. W. Nordskog. 1961. Relationship of blood type to
histocompatibility in chickens. Science 134: 1008-1009.
Schmid, M., I. Nanda, M. Guttenbach, C. Steinlein, M. Hoehn, M. Schartl, T. Haaf, S.
Weigend, R. Fries, J. M. Buerstedde, K. Wimmers, D. W. Burt, J. Smith, S. A'Hara,
A. Law, D. K. Griffin, N. Bumstead, J. Kaufman, P. A. Thomson, T. Burke, M. A.
Groenen, R. P. Crooijmans, A. Vignal, V. Fillon, M. Morisson, F. Pitel, M. TixierBoichard, K. Ladjali-Mohammedi, J. Hillel, A. Maki-Tanila, H. H. Cheng, M. E.
Delany, J. Bumside, and S. Mizuno. 2000. First report on chicken genes and
chromosomes 2000. Cytogenet. Cell Genet. 90: 169-218.
Seaton, G., C. S. Haley, S. A. Knott, M. Kearsey, and P. M. Visscher. 2002. QTL Express:
Mapping quantitative trait loci in simple and complex pedigrees. Bioinformatics 18:
339-340.
Sewalem, A., D. M. Morrice, A. Law, D. Windsor, C. S. Haley, C. O. Ikeobi, D. W. Burt,
and P. M. Hocking. 2002. Mapping of quantitative trait loci for body weight at three,
six, and nine weeks of age in a broiler layer cross. Poult. Sci. 81: 1775-1781.
Siwek, M., S. J. Comelissen, A. J. Buitenhuis, M. G. Nieuwland, H. Bovenhuis, R. P.
Crooijmans, M. A. Groenen, H. K. Parmentier, and J. J. van der Poel. 2004.
Quantitative trait loci for body weight in layers differ from quantitative trait loci
specific for antibody responses to sheep red blood cells. Poult. Sci. 83: 853-859.
48
Smith, E., L. Shi, P. Drummond, L. Rodriguez, R. Hamilton, E. Powell, S. Nahashon, S.
Ramlal, G. Smith, and J. Foster. 2000a. Development and characterization of
expressed sequence tags for the turkey (Meleagris gallopavo) genome and
comparative sequence analysis with other birds. Anim. Genet. 31: 62-67.
Smith, E. J., L. Shi, P. Drummond, L. Rodriguez, R. Hamilton, S. Ramlal, G. Smith, K.
Pierce, and J. Foster. 2001. Expressed sequence tags for the chicken genome from a
normalized 10-day- old White Leghorn whole embryo cDNA library: 1. DNA
sequence characterization and linkage analysis. J. Hered. 92: 1-8.
Smith, P. J. 2002. Analysis of failure and survival data. Chapman & Hall/CRC, Boca Raton,
FL.
Smith, T. P., E. Casas, C. E. Rexroad, 3rd, S. M. Kappes, and J. W. Keele. 2000b. Bovine
capnl maps to a region of Bta29 containing a quantitative trait locus for meat
tenderness. J. Anim. Sci. 78: 2589-2594.
Soller, M., T. Brody, and A. Genizi. 1976. On the power of experimental designs for the
detection of linkage between marker loci and quantitative loci in crosses between
inbred lines. Theor. Appl. Genet. 47: 35-39.
Steadham, E. M., S. J. Lamont, I. Kujdych, and A. W. Nordskog. 1987. Association of
Marek's disease with Ea-B and immune response genes in subline and F2 populations
of the Iowa State SI Leghorn line. Poult. Sci. 66: 571-575.
Stevens, L. 1991. Genetics and evolution of the domestic fowl. Cambridge University Press,
Cambridge.
Tatsuda, K., and K. Fujinaka. 2001. Genetic mapping of the QTL affecting body weight in
chickens using a F2 family. Br. Poult. Sci. 42: 333-337.
Tatsuda, K., K. Fujinaka, and T. Yamasaki. 2000. Genetic mapping of a body weight trait in
chicken. Anim. Sci. J. 71: 130-136.
Thoday, J. M. 1961. Location of polygenes. Nature 191: 368-370.
Thomsen, H., H. K. Lee, M. F. Rothschild, M. Malek, and J. C. Dekkers. 2004.
Characterization of quantitative trait loci for growth and meat quality in a cross
between commercial breeds of swine. J. Anim. Sci. 82: 2213-2228.
49
Tilghman, S. M. 1999. The sins of the fathers and mothers: Genomic imprinting in
mammalian development. Cell 96: 185-193.
Tuiskula-Haavisto, M., M. Honkatukia, J. Vilkki, D. J. de Koning, N. F. Schulman, and A.
Maki-Tanila. 2002. Mapping of quantitative trait loci affecting quality and production
traits in egg layers. Poult. Sci. 81: 919-927.
Tuiskula-Haavisto, M., D.-J. de Koning, M. Honkatukia, N. F. Schulman, A. Maki-Tanila,
and J. Vilkki. 2004. Quantitative trait loci with parent-of-origin effects in chicken.
Genet. Res., Camb. 84: 57-66.
Tycko, B., and I. M. Morison. 2002. Physiological functions of imprinted genes. J. Cell.
Physiol. 192: 245-258.
Uimari, P., G. Thaller, and I. Hoeschele. 1996. The use of multiple markers in a bayesian
method for mapping quantitative trait loci. Genetics 143: 1831-1842.
Vallejo, R. L., L. D. Bacon, H. C. Liu, R. L. Witter, M. A. Groenen, J. Hillel, and H. H.
Cheng. 1998. Genetic mapping of quantitative trait loci affecting susceptibility to
Marek's disease virus induced tumors in F2 intercross chickens. Genetics 148: 349360.
van Kaam, J. B., M. A. Groenen, H. Bovenhuis, A. Veenendaal, A. L. Vereijken, and J. A.
van Arendonk. 1999a. Whole genome scan in chickens for quantitative trait loci
affecting growth and feed efficiency. Poult. Sci. 78: 15-23.
van Kaam, J. B., M. A. Groenen, H. Bovenhuis, A. Veenendaal, A. L. Vereijken, and J. A.
Van Arendonk. 1999b. Whole genome scan in chickens for quantitative trait loci
affecting carcass traits. Poult. Sci. 78: 1091-1099.
van Kaam, J. B. C. H. M., J. A. M. van Arendonk, M. A. M. Groenen, H. Bovenhuis, A. L. J.
Vereijken, R. P. M. A. Crooijmans, J. J. van der Poel, and A. Veenendaal. 1998.
Whole genome scan for quantitative trait loci affecting body weight in chickens using
a three generation design. Livest. Prod. Sci. 54: 133-150.
Vereijken, A. L. J. 1992. Genetics of body conformation and breast meat yield in broilers.
Pages 98-100 in The Proceedings of the 19th World's Poultry Congress. Amsterdam,
The Netherlands.
50
Wakenell, P. S., M. M. Miller, R. M. Goto, and W. E. Briles. 1996. Association between the
Rfp-y haplotype and the incidence of Marek's disease in chickens. Immunogenetics
44: 242-245.
Weigend, S., S. Matthes, J. Solkner, and S. J. Lamont. 2001. Resistance to Marek's disease
virus in White Leghorn chickens: Effects of avian leukosis virus infection genotype,
reciprocal mating, and major histocompatibility complex. Poult. Sci. 80: 1064-1072.
Weller, J. I. 2001. Quantitative trait loci analysis in animals. CABI Publishing, New York,
New York.
Weller, J. I., Y. Kashi, and M. Soller. 1990. Power of daughter and granddaughter designs for
determining linkage between marker loci and quantitative trait loci in dairy cattle. J.
Dairy Sci. 73: 2525-2537.
Williams, J. G., A. R. Kubelik, K. J. Livak, J. A. Rafalski, and S. V. Tingey. 1990. DNA
polymorphisms amplified by arbitrary primers are useful as genetic markers. Nucleic
Acids Res. 18: 6531-6535.
Witte, R. S. 1989. Statistics. Third ed. Holt, Rinehart, and Winston, Inc., New York, New
York.
Wong, G. K., B. Liu, J. Wang, Y. Zhang, X. Yang, Z. Zhang, Q. Meng, J. Zhou, D. Li, J.
Zhang, P. Ni, S. Li, L. Ran, H. Li, R. Li, H. Zheng, W. Lin, G. Li, X. Wang, W.
Zhao, J. Li, C. Ye, M. Dai, J. Ruan, Y. Zhou, Y. Li, X. He, X. Huang, W. Tong, J.
Chen, J. Ye, C. Chen, N. Wei, L. Dong, F. Lan, Y. Sun, Z. Yang, Y. Yu, Y. Huang,
D. He, Y. Xi, D. Wei, Q. Qi, W. Li, J. Shi, M. Wang, F. Xie, X. Zhang, P. Wang, Y.
Zhao, N. Li, N. Yang, W. Dong, S. Hu, C. Zeng, W. Zheng, B. Hao, L. W. Hillier, S.
P. Yang, W. C. Warren, R. K. Wilson, M. Brandstrom, H. Ellegren, R. P. Crooijmans,
J. J. van der Poel, H. Bovenhuis, M. A. Groenen, I. Ovcharenko, L. Gordon, L.
Stubbs, S. Lucas, T. Glavina, A. Aerts, P. Kaiser, L. Rothwell, J. R. Young, S.
Rogers, B. A. Walker, A. van Hateren, J. Kaufman, N. Bumstead, S. J. Lamont, H.
Zhou, P. M. Hocking, D. Morrice, D. J. de Koning, A. Law, N. Bartley, D. W. Burt,
H. Hunt, H. H. Cheng, U. Gunnarsson, P. Wahlberg, L. Andersson, E. Kindlund, M.
T. Tammi, B. Andersson, C. Webber, C. P. Ponting, I. M. Overton, P. E. Boardman,
H. Tang, S. J. Hubbard, S. A. Wilson, J. Yu, and H. Yang. 2004. A genetic variation
51
map for chicken with 2.8 million single-nucleotide polymorphisms. Nature 432: 717722.
Yokoi, N., S. Shimizu, K. Ishibashi, K. Kitada, H. Iwama, M. Namae, M. Sugawara, T.
Serikawa, and K. Komeda. 2000. Genetic mapping of the rat mutation creeping and
evaluation of its positional candidate gene reelin. Mamm. Genome 11: 111-114.
Yokomine, T., H. Shirohzu, W. Purbowasito, A. Toyoda, H. Iwama, K. Ekeo, T. Hori, S.
Mizuno, M. Tsudzuki, Y. Matsuda, M. Hattori, Y. Sakaki, and H. Sasaki. 2005.
Structural and functional analysis of a 0.5-mb chicken region orthologous to the
imprinted mammalian Ascl2/Mash2-Igf2-hl9 region. Genome Res. 15: 154-165.
Yonash, N., L. D. Bacon, R. L. Witter, and H. H. Cheng. 1999. High resolution mapping and
identification of new quantitative trait loci (QTL) affecting susceptibility to Marek's
disease. Anim. Genet. 30: 126-135.
Yonash, N., H. H. Cheng, J. Hillel, D. E. Heller, and A. Cahaner. 2001. DNA microsatellites
linked to quantitative trait loci affecting antibody response and survival rate in meattype chickens. Poult. Sci. 80: 22-28.
Yunis, R., E. D. Heller, J. Hillel, and A. Cahaner. 2002. Microsatellite markers associated
with quantitative trait loci controlling antibody response to Escherichia coli and
Salmonella enteritidis in young broilers. Anim. Genet. 33: 407-414.
Zeng, Z. B. 1993. Theoretical basis for separation of multiple linked gene effects in mapping
quantitative trait loci. Proc. Natl. Acad. Sci. USA 90: 10972-10976.
Zhao, H., M. F. Rothschild, R. L. Fernando, and J. C. Dekkers. 2003. Tests of candidate
genes in breed cross populations for QTL mapping in livestock. Mamm. Genome 14:
472-482.
Zhiliang, G., Z. Dahai, L. Ning, L. Hui, D. Xuemei, and W. Changxin. 2004. The single
nucleotide polymorphisms of the chicken myostatin gene are associated with skeletal
muscle and adipose growth. Sci. China C. Life Sci. 47: 25-30.
Zhou, H., A. J. Buitenhuis, S. Weigend, and S. J. Lamont. 2001. Candidate gene promoter
polymorphisms and antibody response kinetics in chickens: Interferon-gamma,
interleukin-2, and immunoglobulin light chain. Poult. Sci. 80: 1679-1689.
52
Zhou, H., H. Li, and S. J. Lamont. 2003. Genetic markers associated with antibody response
kinetics in adult chickens. Poult. Sci. 82: 699-708.
Zhou, H., A. D. Mitchell, J. P. McMurtry, C. M. Ashwell, and S. J. Lamont. 2005. Insulin­
like growth factor-I gene polymorphism associations with growth, body composition,
skeleton integrity, and metabolic traits in chickens. Poult. Sci. 84: 212-219.
53
CHAPTER 2A
MOLECULAR MARKERS ASSOCIATED WITH GROWTH AND CARCASS
TRAITS IN MEAT-TYPE CHICKENS
McElroy, J. P., D. E. Harry, J. C. M. Dekkers, and S. J. Lamont. 2002. Molecular markers
associated with growth and carcass traits in meat-type chickens. Communication 0404 in Proceedings of the 7th World Congress on Genetics Applied to Livestock
Production, Montpellier, France. INRA, Cedex, France.
A paper published in the proceedings of the 7th World Congress on Genetics Applied to
Livestock Production.
Joseph P. McElroy ', David E. Harry 2, Jack C. M. Dekkers 1 and Susan J. Lamont1
1
2
Iowa State University, Ames, IA, 50011
Nicholas Turkey Breeding Farms, Sonoma, CA, 95476 and Aviagen Ltd., Newbridge,
Midlothian, EH28 8SZ, Scotland, UK
ABSTRACT
A half-sib family with 201 progeny from an F2 cross between two commercial broiler
breeder lines that differed in breast meat yield was used to identify QTL for growth and
carcass traits using 50 microsatellite markers across the genome. Ten chromosomes with
multiple markers were analyzed by interval mapping. Significant (p<0.05 chromosomewise) QTL for one or more traits were found in nine regions across five chromosomes.
Single markers on nine chromosomes were analyzed by association study.
Four of the nine
markers tested had a suggestive (p<0.10 comparison-wise) association with one or more
traits, of which one showed significant associations at p<0.05 with multiple traits.
Significant QTL for breast meat yield were found on Chromosomes 2 and 5.
54
INTRODUCTION
Breast meat is the most economically valuable part of the chicken and makes up about 50%
of total muscle weight (Stevens, 1991). Improvement in breast meat yield, defined as breast
meat weight relative to body weight, would result in more efficient allocation of resources
consumed by the bird. Direct selection on breast meat yield is, however, difficult. Live
weight has a high genetic correlation with breast meat weight (0.76, Le Bihan-Duval et al.,
1998), but a low genetic correlation with breast meat yield (0.13 in males and 0.16 in
females, Le Bihan-Duval et al, 1998). Genetically superior birds for breast meat yield could
be identified if molecular markers linked to genes affecting the trait were available.
Several studies have identified quantitative trait locus (QTL) regions in the poultry genome
affecting growth and carcass traits; on Chromosome 1 for body weight, feed intake, and
carcass percentage, on Chromosome 2 for body weight, feed intake, and meat color, and on
Chromosomes 4 and 23 for feed intake (Groenen et al., 1997; van Kaam et al., 1998, 1999a,
and 1999b; Tatsuda et al., 2000; Tatsuda and Fujinaka, 2001). No QTL for breast meat yield
have been reported. Thus, the objective of this study was to map QTL for breast meat yield
and other growth and carcass traits in an F2 cross between two commercial broiler breeder
lines.
MATERIALS AND METHODS
Experimental Population. Individuals from male (M) and female (F) commercial broiler
breeder chicken lines were used in a reciprocal cross design to generate five F1 half-sib/fullsib families for each reciprocal type (MxF and FxM). One male from each of ten F1 half-sib
families was crossed with, on average, three females from each of the other four F1 families
within his reciprocal type to produce ten F2 half-sib families, for a total of 1123 F2
individuals. All birds were raised using standard feed, housing, and biosecurity procedures
for evaluating and selecting offspring in commercially relevant foundation lines. F2
55
individuals were slaughtered at six weeks and measured for body weight prior to (PrBW) and
post (PoBW) transport to the processing facility, fat weight (Fat), weight of carcass without
giblets (WOG), weight of the front half of carcass (FrontH), tender weight (Tender), fillet
weight (Fillet), conformation score (Conf), and white meat weight (WM). Breast meat yield
(BMY) was computed as WM/PoBW. The results presented here are based on one F2 halfsib family of 201 F2 progeny from 12 dams.
Markers. F2 progeny were genotyped for 50 microsatellite markers across 19 chromosomes,
of which 10 (1, 2, 3, 4, 7, 9, 11, 13, 19, and Z) had multiple markers. Marker intervals
averaged 56 cM and ranged from 4 to 100 cM. The F1 sire was heterozygous for all 50
markers selected for analysis.
Analyses. Chromosomes with multiple markers were analyzed by interval mapping using the
half-sib option of QTL Express for a single QTL with one cM steps (Seaton et al, 2002.)
using the following model: Trait = Sex + Hatch + Dam + a g + error, where a is the
substitution effect for the QTL, and g is the probability that the progeny inherited one versus
the other QTL allele from the sire, based on marker genotypes (Knott et al., 1996). Marker
location was determined from the consensus linkage map (Groenen et al., 2000).
Chromosome-wise significance thresholds were obtained from 1000 data permutations using
QTL Express (Seaton et al., 2002). Chromosomes with single markers (5, 8, 12, 23, 24, 26,
27, 28, and linkage group E47W24) were analyzed for marker associations. Both sire and
dam allelic effects were tested by the following model: Trait = Hatch + Dam + Sex + Sire
Allele + Dam Allele(Dam) + error.
RESULTS AND DISCUSSION
The results from the interval mapping (Table 1) indicated several regions that were
significantly associated with one or more traits. Chromosome 2 showed three significant
QTL for different traits at substantially different positions, indicating presence of multiple
56
QTL. A QTL affecting breast meat yield was found at 125 cM on Chromosome 2. Tatsuda
and Fujinaka (2001) found a QTL for body weight at 60 cM and van Kaam et al. (1999a) a
QTL for feed intake adjusted for body weight at 41 cM on Chromosome 2, which is close to
the fat QTL in Table 1. The literature has not reported QTL for the other chromosomes that
were identified to harbor QTL in this study. Chromosomes 3 and 13 were found to contain
QTL for several traits at similar positions, which may represent single pleiotropic QTL. No
QTL were found on chromosome 1, which has been reported to contain QTL for body weight
in several studies (Tatsuda and Fujinaka, 2001; Tatsuda et al., 2000; Groenen et al., 1997).
Four markers were suggestive for linkage to a QTL in the association study of chromosomes
with single markers (Table 2). MCW165 on Chromosome 23 was associated with fat and is
in the region of a suggestive QTL found for feed intake by van Kaam et al. (1999a).
MCW193 on Chromosome 5 was significant for many traits, including breast meat yield.
This effect was present for both sire and dam alleles. No reports of QTL on Chromosome 5
have been reported for growth and carcass traits. Markers MCW262 and MCW233 on
Chromosomes 26 and 27 also showed suggestive significance but are not near other reported
QTL in the literature.
The results reported in this paper, though preliminary, add to the increasing knowledge of
QTL in poultry for growth and carcass traits (Groenen et al., 1997; van Kaam et al. 1999a, b;
Tatsuda et al., 2000; Tatsuda and Fujinaka, 2001). Currently, a more extensive study of the
experimental population described in this paper is underway, in which multiple F2 halfsib
families will be genotyped and for additional markers on the 11 largest chromosomes. The
extensive study will identify QTL that remained undetected in the present study due to lack
of chromosomal coverage or lack of power from small numbers of animals, and indicate
which QTL have effects across multiple families
57
ACKNOWLEDGMENTS
The authors thank Dr. Harris Wright, Dr. Jerry Smith, Edward Landers, and Jennifer Green
of Arbor Acres Farms.
Joseph McElroy is a USDA National Needs Fellow in Animal
Biotechnology.
REFERENCES
Groenen, M. A. M., Crooijmans, R. P. M. A., Veenendaal, T., Van Kaam, J. B. C. H. M.,
Vereijken, A. L. J., Van Arendonk, J. A. M. and Van Der Poel, J. J. (1997) Animal.
Biotech., 8: 41-46.
Groenen, M. A., Cheng, H. H., Bumstead, N., Benkel, B. F., Briles, W. E., Burke, T., Burt,
D. W., Crittenden, L. B., Dodgson, J., Hillel, J., Lamont, S., de Leon, A. P., Soller, M.,
Takahashi, H. and Vignal, A. (2000) Genome Res., 10: 137-47.
Knott, S. A., Elsen, J. M. and Haley, C. S. (1996) Theor. Appl. Genet., 93: 71-80.
Le Bihan-Duval, E., Mignon-Grasteau, S., Millet, N. and Beaumont, C. (1998) Br. Poultry
&%., 39: 346-353.
Seaton, G., Haley, C., Knott, S., Kearsey, M., Visscher, P. (2002) QTL Express:
http://qtl.cap.ed.ac.uk (1/13/2002).
Stevens, L. (1991) Genetics and evolution of the domestic fowl. Cambridge University
Press, Cambridge, UK.
Tatsuda, K., Fujinaka, K., and Yamasaki, T. (2000) Anim. Sci. J., 71: 130-136.
Tatsuda, K., and Fujinaka, K. (2001) Br. Poultry Sci., 42(3): 333-7.
van Kaam, J. B. C. H. M., van Arendonk, J. A. M., et al. (1998) Livest. Prod. Sci., 54: 133150.
van Kaam, J. B., Groenen, M. A., Bovenhuis, H., Veenendaal, A., Vereijken, A. L., and van
Arendonk, J. A. M. (1999a) Poultry Sci., 78(1): 15-23.
van Kaam, J. B., Groenen, M. A., Bovenhuis, H., Veenendaal, A., Vereijken, A. L., and Van
Arendonk, J. A. M. (1999b) Poultry Sci., 78(8): 1091-9.
58
Table 1. Significant QTL detected at the 5% chromosome-wise level by interval
mapping.
Chromo­
some
(Gga)
2
2
2
3
3
3
3
3
3
A Significant
B Analysis
Posi­
tion
(cM)
247
69
125
154
31
154
154a
154*
188
Trait
PrBW
Fat
BMY
PoBW
PoBW
FrontH
Fillet
WM
Conf
Allele
substitution
effect
117.2 g
7.0g
0.006
73.0 g
B
31.0 g
21.4 g
24.0 g
0.73 units
Chromo­ Posi­ Trait
Allele
some
tion
substitution
(Gga) (cM)
effect
7
146 Tender
4.0 g
13
15
PrBW
88.3 g
13
15 PoBW
73.8 g
13
15
WOG
52.5 g
13
15 FrontH
32.9 g
13
15
Fillet
18.7 g
13
15
WM
21.2 g
13
15
Conf 0.35 units
155 WOG
Z
52.8 g
at the 1% chromosome-wise level.
only gives the effect for the most significant QTL for a trait on a chromosome.
59
Table 2. Markers suggestively significant at the 10% comparison-wise level by
association analysis on chromosomes with single markers.
Marker
MCW193
MCW193
MCW193
MCW193
MCW193
MCW193
MCW193
MCW193
MCW193
MCW193
MCW193
MCW193
MCW165
MCW262
MCW262
MCW233
Chromosome Position
(Gga)
5
50
50
5
50
5
5
50
5
50
5
50
5
50
5
50
5
50
5
50
5
50
50
5
23
5
26
22
26
22
27
19
Trait
P-value
PrBW
Fat
Fillet
Fillet
FrontH
PoBW
Tender
WM
WM
BMY
BMY
WOG
Fat
PrBW
PoBW
Fat
0.04
0.01
0.09
0.02
0.01
0.04
0.02
0.07
0.04
0.07
0.01
0.01
0.08
0.07
0.07
0.06
Test of sire/
dam allele
Dam
Sire
Dam
Sire
Dam
Dam
Dam
Dam
Sire
Dam
Sire
Dam
Dam
Sire
Sire
Sire
60
CHAPTER 2B
TRAIT LOCI AFFECTING WHITE MEAT PERCENT AND OTHER GROWTH
AND CARCASS TRAITS IN COMMERCIAL BROILER CHICKENS
A manuscript in preparation for submission to Poultry Science.
J. P. McElroy*, J.-J. Kim2*, D. E. Harry3*, S. R. Brown*, J. C. M. Dekkers*, and S. J.
Lamont1*
t
*Iowa State University, Ames, Iowa 50011; Aviagen Group, Huntsville, Alabama 35805
Abbreviation Key: BWT = body weight; HW = harvest weight; CS = conformation score;
FAT = weight of abdominal fat pad; WOG = weight of the carcass without giblets; FNTH =
weight of the front half of the carcass; FIL = weight of the fillet; TEN - weight of the tender;
TWM = total white meat weight; WM% = white meat percent; LC = line cross; HS = halfsib; CB = combined; Mend = mendelian; Pat-e = paternal expression; Mat-e = maternal
expression; Part-e - partial expression
2 Present
Address: School of Bitechnology, Yeungnam University, Gyeongsan, Gyeongbuk, 712749 South
Korea.
Present Address: Genetic Foundations, P.O. Box 3897, Napa, CA 94558.
61
ABSTRACT White meat is the most economically valuable part of a broiler chicken.
Increasing white meat relative to overall body size (white meat percent: WM%) makes a
broiler, gram for gram, a more valuable animal. However, accurately measuring WM%
percent requires removing the bird from the breeding flock. Identification of markers for
genomic regions associated with WM% would allow direct genetic selection on breeders.
The objective of this study was to identify genomic regions affecting WM%, and other
growth and carcass traits in an F2 cross between two commercial broiler lines that differed in
WM%. Two commercial lines were crossed to generate five F1 half-sib families of each
reciprocal cross type. One male from each family was crossed with 3 females from each of
the other families within each reciprocal cross type. Seven F2 half-sib families, consisting of
430 F2 individuals, were analyzed.
Microsatellite markers (n = 73) on the 11 largest
chromosomes were analyzed for associations with various growth and carcass traits by least
squares interval mapping using line-cross, half-sib, combined, and parent of origin models.
Sixty-eight QTL were identified at the 5% chromosome-wise level, including six QTL
affecting WM%. Ten QTL reached 5% genome-wise significance, including one WM%
QTL on Gga 2. This study identified genomic regions harboring QTL affecting WM% and
other carcass and growth traits, which may be useful for direct genetic selection, and also
identified putative imprinted QTL in the chicken. The advantage of using multiple statistical
models was evident, because QTL were identified with the combined and parent of origin
models that were not identified with the line-cross or half-sib models.
Key Words: QTL, chicken, white meat percent, growth, carcass, parent of origin, imprinting
62
INTRODUCTION
White meat, or the breast meat, is the most economically valuable part of the chicken
(Stevens, 1991). Because of the high genetic correlation (0.76) between white meat weight
(TWM) and body weight (BWT) (Le Bihan-Duval et al., 1998), selection for growth rate has
resulted in increased TWM. Genetic correlations between BWT and other, less valuable,
parts of the chicken, however, are also high (Marks, 1995).
Because of these high
correlations, selection for growth rate increases the weight of less valuable parts of the
chicken as well as TWM. Birds with large breasts relative to bodyweights can be more
effectively selected by applying pressure on TWM as a percentage of BWT (WM%) than by
selecting only on TWM. Direct phenotypic selection on TWM and WM% is not possible
because birds must be slaughtered to measure TWM. The genetic correlation between WM%
and BWT is low (0.13 in males and 0.16 in females: Le Bihan-Duval et al. (1998)), so that
selection for BWT results in little progress in increasing WM%, even though the genetic
correlation between BWT and TWM is high. A common field alternative to measuring
TWM or WM% is visual scoring by a trained observer on the live bird (conformation score:
CS). However, the correlation between CS and WM% depends on the scorer and the criteria
for scoring.
Therefore, including information on genomic regions affecting WM% in a
selection program through MAS could increase the improvement made for WM% in a
population.
Several studies have identified QTL for growth and carcass traits in the chicken.
Many of the studies utilized F2 crosses between various lines of chickens.
Tatsuda et al.
(2000) and Tatsuda and Fujinaka (2001) identified QTL for BWT in a cross between
Satsumadori and White Leghorn chicken lines. Tuiskula-Haavisto et al. (2002) used hens
generated from a cross between a Rhode Island Red line and a White Leghorn line to find a
QTL for BWT and feed intake. Sewalem et al. (2002), Ikeobi et al. (2002), and Carlborg et
al. (2004) crossed a White Leghorn line with a commercial broiler sire line to generate a
mapping population to identify many QTL for growth and carcass traits. Kerje et al. (2003)
and Carlborg et al. (2003) utilized a cross between White Leghorn and Red Junglefowl
chickens to identify QTL for growth and BWT. Sasaki et al. (2004) used a cross between a
63
White Leghorn line and a Rhode Island Red line to identify QTL for BWT. Siwek et al.
(2004) used a population created by crossing two medium-heavy layer lines that were
divergently selected for primary antibody response to sheep red blood cells to identify QTL
for BWT. Ikeobi et al. (2004) used a cross between broiler and layer lines to identify QTL
for growth and carcass traits. Several other population designs have been utilized to identify
QTL for growth and carcass traits in the chicken. A three-generation full-sib population
derived from a cross of White Plymouth Rock lines was utilized by van Kaam et al. (1999b),
van Kaam et al. (1999a), van Kaam et al. (1998), and Jennen et al. (2004) to map QTL for
growth and carcass traits.
Deeb and Lamont (2003) used F1 individuals from a cross
between two broiler sires and dams from two highly inbred lines to locate QTL affecting
BWT at eight weeks of age. De Koning et al. (2003) and de Koning et al. (2004) used a
commercial broiler three-generation half-sib population design to confirm QTL that were
previously identified on experimental crosses in commercial lines. An early stage (McElroy
et al., 2002) of the present study utilized a half-sib model to analyze one F2 half-sib family
generated from a cross between a male and a female parental lines (described below) to
identify QTL for various carcass and growth traits. The McElroy et al. (2002) study is the
only one to date reporting QTL associated with WM%.
Many different statistical methods have been employed to identify QTL. One of the
most common methods used in animal agriculture is least squares regression interval
mapping (Haley and Knott, 1992; Haley et al., 1994) because of its simplicity, low
computational requirements, and wide availability of programs that employ this method for
QTL detection (see Seaton et al. (2002) for a description of QTL Express, a user friendly web
based program for QTL analysis using least squares regression).
Within the statistical
framework of least squares regression, many different models have been used to detect QTL,
including half-sib (HS) (Knott et al., 1996), line cross (LC) (Haley and Knott, 1992; Haley et
al. 1994), backcross (Knott et al., 2002), parent of origin (de Koning et al., 2002; Knott et al.,
1998; Thomsen et al., 2004), and epistatic models (Carlborg and Andersson, 2002; Carlborg
et al., 2000). The validity of some models, such as half-sib, LC, and backcross models, is at
least partially determined by population structure. Other models, such as parent of origin or
epistatic models, can be used within the framework of various population structures, and can
64
be used to identify genes that have expression patterns that do not follow mendelian
expression.
The objective of the current study was to identify and characterize QTL
affecting white meat percent and other carcass and growth traits in an F2 cross between two
commercial broiler lines that differed in WM%,
MATERIALS AND METHODS
Experimental Population
The population was created by reciprocally crossing individuals from a founder
broiler line used in generating male parents of commercial birds (Line 1) with individuals
from a female broiler line used in generating female parents of commercial birds (Line 2) to
generate 5 full-sib/half-sib F1 families for each reciprocal type. Line 1 had 2.7% higher
white meat percentage than Line 2. One male from each of these F1 families was then mated
to, on average, three females from each of the other F1 families within his reciprocal cross
type to produce 10 F2 half-sib families. Shipment problems led to loss of DNA samples
from multiple individuals from this cross, including some granddams, F1 dams, and F2
individuals. The remaining F2 population for which DNA samples were available consisted
of 204 males and 226 females from seven F2 half-sib families with 199, 42, 40, 39, 39, 37,
and 34 F2 individuals, for a total of 430 F2 birds. All but two of the individuals in the F2
half-sib family analyzed by McElroy et al. (2002) were retained.
All birds were raised using standard feed, housing, and biosecurity procedures for
evaluating and selecting offspring in commercially relevant foundation lines. Treatment of
animals met or exceeded accepted guidelines presented in Guidelines for the Care and Use of
Agricultural Animals in Agricultural Research and Teaching, 1st revised edition (1999).
Males and females from each hatch (n = 3) were raised in separate pens, thus F2 individuals
resided in one of six pens.
65
Phenotypes
Traits on all birds within a hatch were measured on the same day and birds within a
hatch were slaughtered on the same day. All birds were within 2 d of age of each other when
traits were measured and when they were slaughtered (i.e. 40 d, 41 d, and 42 d).
Phenotypes collected on the F2 individuals at six weeks of age were live weight prior
to transport to the processing facility (BWT); live weight after transport to the processing
facility (harvest weight: HW); conformation score (CS), a subjective measurement of body
structure and fleshing made by a skilled technician; weight of the abdominal fat pad (FAT);
weight of the carcass without giblets (WOG); weight of the front half of the carcass (FNTH);
weight of the fillet (pectoralis major; FIL); and weight of the tender (pectoralis minor; TEN).
Eight additional traits were derived from these eight primary traits: total white meat (TWM:
FIL + TEN); white meat percent [WM%: (TWM/HW) X 100]; fat percent [FAT%:
(FAT/HW) X 100]; WOG% [(WOG/HW) X 100]; FNTH% [(FNTH/HW) X 100]; FIL%
[(FIL/HW) X 100]; TEN% [(TEN/HW) X 100]; and transport loss (TPL: BWT-HW).
Microsatellite Markers
Microsatellite markers on the 11 largest chromosomes (Gga 1-10 and Z) were
selected (Table 1). This strategy allowed each marker to be linked to a larger portion of the
genome than if some markers were on smaller chromosomes, thereby maximizing the
average per marker information.
In addition to the 30 markers for the 11 largest
chromosomes that were used by McElroy et al. (2002), 43 new markers were selected, with
the goal to create a 30 cM average interval between markers on the 11 largest chromosomes
and high heterozygosity of F1 sires. The final set of markers consisted of 13 on chromosome
1; 15 on chromosome 2; 8 on chromosomes 3 and 4; 6 on chromosome Z; 5 on chromosome
5; 4 on chromosomes 6, 7, 9 and 10; and 2 on chromosome 8. The average marker interval
on the 11 largest chromosomes, based on the linkage map generated from marker data on this
population, was approximately 37 cM and ranged from 5.3 cM to 100 cM. The number of
alleles per marker ranged from 2 to 11.
66
For amplification of each microsatellite for each individual, the PGR reagents used
were 2mM MgCl%, 0.2mM dNTP, 0.05(iM of forward primer, reverse primer, and labeled
M13 primer (Getting et al., 1995), 0.02 units of Taq polymerase per |jl of solution, and 2(il of
genomic DNA. For reactions that did not yield product using the M13 primer, 0.2pM of
forward and reverse primers were used for direct PGR. Genomic DNA was diluted 10 to 100
fold (approximately 1 to 10 ng/p.1) before adding to the reaction, based on degree of
amplificability of a particular microsatellite or DNA sample. The total PGR reaction volume
was 12 pi. Electrophoresis was performed on the LI-COR system (LI-COR DNA Sequencer
Model 40004) with manual scoring of genotypes.
Statistical Analyses
Generation of the Linkage Map. Cri-map (Green et al., 1994) was used to derive
linkage maps for all chromosomes. All options were used to order markers and to obtain
distances between markers.
QTL Analyses. Two general groups of least squares interval mapping models were
used for QTL detection on the autosomal chromosomes. For all models, fixed effects were
F1 sire (n = 7), sex (n = 2), and hatch (n = 3). Group 1 models consisted of three models, as
described by Kim et al. (2005a) and Kim et al. (2005b): a line-cross (LC) model, a half-sib
(HS) model, and a combined (CB) model that included coefficients from both the LC and HS
models. These models were based on the least-squares regression models of Haley et al.
(1994) and Knott et al. (1996). Following Kim et al. (2005a) and Kim et al. (2005b), the
models, fitted at each 1 cM position along the 11 largest chromosomes, were:
LC Model
y,j — Xyb + Sj + <zPa(ij) + dPd^ + eis
HS model
y\j = Xyb + Sj + anso^s® + e,j
CB model
yij = Xyb + s,+ aPa(ij) + dP^]} + acB(i>Ps(ij) + £ij
where y^ is the phenotype of F2 individual j from F1 sire i, Xy is the design matrix and b is
the vector of coefficients for fixed effects, Sj is the effect of the ith F1 sire, and ey is the
residual error.
4
In the LC and CB models, a and d are estimates of the additive and
LI-CORR Biotechnology, 4308 Progressive Avenue, P.O. Box 4000, Lincoln, NE 68504.
67
dominance effects contrasting the male line and female line QTL alleles. Coefficients P^)
and Pd(ij) are the line-origin coefficients for each animal at a given position conditional on
flanking marker genotypes. Following Haley et al. (1994), Pa(ij) for an individual is the
difference between the probabilities of being homozygous for the QTL allele that originated
from Line 1 versus Line 2, and fa# is the probability of an individual being heterozygous for
line origin QTL alleles. Following Knott et al. (1996), ans and ace in the HS and CB models,
respectively, are the allele substitution effects for a given sire, and Ps(ij) is the probability that
an F2 offspring received one QTL allele vs. the other QTL allele from its F1 sire.
Each model was fitted across the chromosomes of interest. When at least one of the
models was significant (chromosome-wise P = 0.05), a decision tree was used to define the
segregation type of the QTL, following Kim et al. (2005a) and Kim et al. (2005b): the QTL is
fixed for alternate alleles in the two parental lines (LC QTL); the QTL is not fixed in the
parental lines but has similar allele frequencies in the two parental lines (HS QTL); or the
QTL is not fixed in the parental lines, but has different allele frequencies between the
parental lines (CB QTL). The decision tree was as follows:
1) An LC QTL was declared if the QTL was significant (chromosome-wise P = 0.05) in
the line-cross model, but the lack of fit test between the LC and CB model (at the
most likely LC QTL position) was not significant (comparison-wise P = 0.05).
2) An HS QTL was declared if the QTL was not significant (chromosome-wise P =
0.05) in the LC model, was significant in the HS model (chromosome-wise P = 0.05),
and the lack of fit test between the HS and CB model (at the most likely HS QTL
position) was not significant (comparison-wise P = 0.05).
3) A CB QTL was declared if the QTL was significant in the CB model (chromosomewise P = 0.05) and was not defined as an LC or HS QTL in the first two steps.
Significance thresholds used to detect QTL were determined empirically, as described below.
The overall significance level of a QTL was determined using the model that corresponded to
the classification of the QTL, i.e. LC, HS, or CB.
The second group of QTL models (Group 2) consisted of models to detect parent of
origin QTL, following Thomsen et al. (2004) and Kim et al. (2005b). The base model was
the LC model with mendelian expression from the Group 1 models (above; referred to as
68
mendelian (Mend) here). The second model in this group was the full (partial) expression
model (Full):
Full model:
y ij
—
Xijb +
Sj +
Clp3{P
pat(ij)
QmaxP
mat(ij)
dP
f /(ij)
+ 6jj,
where _yy, Xy, b, Sj, and gy are as defined previously, and apat, amat, and d are the paternally
inherited, maternally inherited, and dominance QTL coefficients, respectively. Coefficient
Ppat(ij)
is the probability of animal j inheriting a Line 1 allele vs. a Line 2 allele from its sire i,
Pmat(ij)
is probability of animal j inheriting a Line 1 allele vs. a Line 2 allele from its dam, and
Pd(ij)
is the probability of animal j being heterozygous. The next models in this group are the
paternal (Pat) and maternal (Mat) expression models, and the null model:
paternal expression model:
y,j = Xyb + s, + a
pat Ppat
(ij) + ey,
maternal expression model: yy = Xyb + Sj + amatPmat(ij)+ ey,
null model:
_yy = Xyb + S; + el},
where all terms are as defined previously. All models were tested at each 1 cM position
along the chromosomes.
To define a QTL as a mendelian, partial, paternal, or maternal expression QTL, the
following decision tree, which was based on the tree used by Thomsen et al. (2004) and Kim
et al. (2005b), was used:
If the Mend model vs. the null model was significant:
3) The Full model was tested against the Mend model at each lcM position in that
genomic region. If this test was not significant, then the QTL was classified as a
Mend QTL.
4) If the Full model vs. the Mend model was significant, then the Full model was tested
against the Pat and Mat models at each lcM position in that genomic region.
a. If the Full model vs. the Pat model was not significant and the Full model vs.
the Mat model was significant, then the QTL was classified as a paternally
expressed QTL.
b. If the Full model vs. the Pat model was significant and the Full model vs. the
Mat model was not significant, then the QTL was classified as a maternally
expressed QTL.
69
c. If the Full model vs. the Pat model and the Full model vs. the Mat model were
both significant or both not significant, then the QTL was classified as a
partially expressed QTL.
If the Mend model vs. the null model was not significant:
1) The Full model was tested against the null model. If this test was significant, then the
Full model was tested against the Mat model and Pat model as described in step 2
above.
2) If the Full model vs. the null model was not significant, then the Pat model and Mat
model was tested against the null model. If the Pat model vs. the null model was
significant, then the QTL is classified as a paternally expressed QTL. If the Mat
model vs. the null model was significant, then the QTL is classified as a maternally
expressed QTL. It is unlikely that both the Pat and Mat models vs. the null model
will both be significant if the Full model vs. the null model is not significant.
A paternally (maternally) expressed QTL is one that shows a significant allelic effect when
inherited from F1 sires (dams) without showing a significant allelic effect when inherited
from F1 dams (sires). A partially expressed QTL is one that shows an allelic effect when
inherited from F1 sires and F1 dams, but the effect is different depending on the sex of the F1
parent from which it was inherited.
This decision tree differs from that used by Thomsen et al. (2004) and Kim et al. (2005b)
in that in the current study, if no QTL were detected with the Mend or Full models, the Pat
and Mat models are each tested against the null model, and a Pat or Mat QTL was declared if
the Pat or Mat test was significant.
All significance thresholds used for these tests to
determine presence and type of QTL were empirically determined chromosome-wise P =
0.05, as described below. The overall significance level reached by a QTL was determined
using the model that corresponded to the classification of the QTL, i.e. Mend, Full, Pat, or
Mat.
For all models, the estimated proportion of phenotypic variance explained by a detected
QTL was calculated, following Kim et al. (2005a), by comparing the reduction of the
residual sums of squares with and without fitting the QTL in the model. The Z chromosome
70
was analyzed only with the half-sib model using QTL Express (Seaton et al., 2002), and
followed only the effects of the alleles inherited from the F1 sires averaged over the Z or W
alleles inherited from the dams. Identification of two QTL was declared for a trait when
peak F-values were 40 or more cM apart.
Significance Thresholds. Empirically derived significance thresholds for all 16 traits
from 1000 permutations were found to be quite similar (data not shown) for a given
chromosome. Therefore, the same threshold was used for all traits on a given chromosome.
To derive these thresholds, F-values from all 1000 permutations from each of the traits were
combined, by chromosome, totaling 16000 permutations per chromosome.
As suggested by
Thomsen et al. (2004) for the parent of origin models, significance tests with the same
degrees of freedom had similar significance thresholds, so empirically derived thresholds for
Pat vs. null were used for Full vs. Mend, and empirically derived thresholds for Mend vs.
null were used for Full vs. Pat and Full vs. Mat.
In
addition
to
chromosome-wise
significance
thresholds,
experiment-wise
significance thresholds were also computed for each trait. Following de Koning et al. (2001):
Pexperiment-wise
—
1
—
(1
—
Pchromosome-wise)
,
where
r = (distance between first
and last markers on a chromosome)/(total genomic
coverage on all 11 chromosomes).
This formula was solved for f experiment-wise = 0.05 and 0.01 to get the equivalent fchromosome-wise
thresholds and F values corresponding to those thresholds.
RESULTS
Phenotypic Analyses
In Table 2 is shown the phenotypic correlations between the 16 traits measured and
the means and standard deviations of the traits from the 430 F2 individuals in the genotypic
analysis. WM% had a low phenotypic correlation with HW, BWT, TPL, and WOG (0.06,
71
0.07, 0.07, and 0.13, respectively), a negative correlation with FAT and FAT% (-0.12 and 0.13, respectively), a correlation of 0.47 with CS, and moderate to high correlations with the
other carcass and percentage traits (0.35 to 0.97).
Linkage Map
The estimated linkage map positions of the markers based on the experimental
population is shown in Table 1, along with approximate positions obtained from the
consensus linkage map (Groenen et al., 2000; Schmid et al., 2000). Estimated distances
between the first to the last marker on a chromosome were shorter than consensus map
distances for all chromosomes except Gga 3.
QTL Analyses
Results from all models are presented in Table 3. For the 16 traits, 68 QTL were
identified at the 5% chromosome-wise level, ranging from one to seven QTL per trait. Ten
QTL reached 5% experiment-wise significance and one QTL, on Gga 5 for FAT%, reached
the 1% experiment-wise significance. In Table 3, a positive effect indicates that the allele
conferring the larger trait value was inherited from Line 1.
LC, HS, and CB Models
For the QTL detected using the LC, HS, and CB models, five were classified as
segregating both between and within lines (CB QTL), 18 were classified as segregating only
within lines (HS QTL), and 25 were classified as segregating only between lines (LC QTL)
(Table 3).
With these models, WM% QTL were identified on Gga 3 and 6 at the 5%
chromosome-wise significance level, on Gga 5 at the 1% chromosome-wise significance
level, and on Gga 2 at the 5% genome-wise significance level. Shown in Figures 1 and 2 are
QTL graphs for traits for which significant QTL were found on Gga 3 and 5, respectively,
using the LC, HS, and CB models. In these graphs, significance and thresholds are expressed
as -LOG]o(comparison-wise P-value), which makes thresholds comparable across models
(Kim et al. 2005b). Gga 3 and 5 were the two chromosomes with the most QTL. Gga 3
appears to have two QTL regions, one near 0 cM and one near position 225 cM.
72
Parent of Origin Models
Twenty QTL showed parent of origin expression effects. Of these 20 QTL, two
showed partial expression, three were expressed maternally, and 15 were expressed
paternally (Table 3). WM% QTL were identified with these models on Gga 2 and 3 at the
5% chromosome-wise significance level.
On Gga 3, there were three regions at
approximately 50 cM, 150 cM, and 225 cM (Figure 3) with parent of origin effects. On Gga
5, a single region at approximately 50 cM (Figure 4) with parent of origin effects was
identified. Other regions with parent of origin effects were found on Gga 2 and 7 (Table 3).
DISCUSSION
Identified QTL
For the 16 traits, 68 QTL were identified in this study at the 5% chromosome-wise
level. Of those, 42 affected body weight or carcass components of body weight, 24 affected
carcass composition percentage traits, one affected CS, and one affected TPL. A QTL was
defined in this study as being significant at the 5% chromosome-wise for a given trait.
Because many of the traits are highly correlated (Table 2), it is likely that QTL identified for
multiple traits in the same region represent a single pleiotropic QTL. Further investigation of
these regions, by fine mapping and identification and further analysis of candidate genes,
may help to determine which of these regions contain a single pleiotropic QTL and which
contain multiple QTL.
Only QTL significant at the 5% experiment-wise level (n = 10) will be further
discussed here. On Gga 1 and Gga 5, QTL for FAT and FAT% were identified in the same
regions. Because of the high phenotypic correlation between these two traits (0.94), it is
expected that these represent pleiotropic QTL. This high correlation also means that FAT is
relatively independent of HW, which is also indicated by the fairly low correlation between
the two traits (0.2). Although the correlation between FAT and FAT% is very close to what
Zerehdaran et al. (2004) found (0.93), the correlation in the present study between FAT and
73
HW is much less than was found by Zerehdaran et al. (2004) (0.45). Near the region for the
FAT and FAT% QTL on Gga 5, a QTL affecting FNTH% was also identified.
The
correlation of FNTH% with FAT and FAT% is very low (-0.04 and -0.03, respectively),
FAT is not included in the FNTH, and there is approximately 20 cM between the QTL, so
two QTL are likely segregating in this region.
Several QTL affecting the economically valuable trait WM% were identified in the
present study. One QTL on Gga2 was detected with the CB model and reached the 5%
experiment-wise threshold. This QTL had a negative effect, meaning that the favorable
allele (higher WM%) had a higher frequency in the Line 2 (dam line) grandparents than in
the Line 1 (sire line) grandparents, which is unexpected since the female line had lower
WM% (cryptic QTL). Another QTL affecting WM% detected near this region was classified
as a being maternally expressed and was also cryptic. Four other regions were identified that
had less significant associations with WM%: two on Gga 3 (5% chromosome-wise level),
one on Gga 5 (1% chromosome-wise level), and one on Gga 6 (5% chromosome-wise level)
(Table 3). The favorable alleles of both of the QTL on Gga 3 originated from Line 1, and the
QTL on Gga 5 was cryptic. Since the QTL on Gga 6 was classified as a HS QTL, a separate
QTL effect was estimated for every F1 sire. For four of the F1 sires, the favorable allele was
inherited from Line 1 (effect estimates were 0.11%, 0.66%, 1.19% and 1.43%), and for the
other three F1 sires the favorable allele was inherited from Line 2 (effect estimates were 0.03%, -0.37%, and-1.69%).
Comparison to Other Studies and Candidate Genes for QTL
Only QTL significant at the 5% experiment-wise level of the current study (n = 10)
will be discussed here. Comparisons will be limited to reported QTL in literature that
reached a 5% genome/experiment-wise significance threshold. On Gga 1, QTL affecting
FAT (168 cM) and FAT% (155 cM) were identified in the current study. Ikeobi et al. (2002)
identified a QTL at nearly the same position (150 cM) affecting FAT adjusted for body
weight. Two likely candidate genes in this region are sterol regulatory element binding
protein-2 (Assaf et al., 2004; Brown and Goldstein, 1997) and insulin-like growth factor-I
(Zhou et al., 2005) because of their effects on fat metabolism. On Gga 2, a QTL affecting
74
WM% (86 cM) was identified in the present study. To date, no other QTL in this region
affecting WM% have been reported. Interleukin-6 is a good candidate gene in this region
because it has been shown to affect skeletal muscle atrophy (Haddad et al., 2005). On Gga 2,
a QTL affecting WOG% (163 cM) was identified, which has not been reported in other
studies. The insulin-like growth factor binding protein-1 gene is in this region and is a likely
candidate gene because the insulin-like growth factor system has been shown to be
associated with various growth traits (eg. Yun et al. (2005)). Two QTL were identified on
Gga 3 for TEN (9 cM and 51 cM). Carlborg et al. (2003), Kerje et al. (2003), and Siwek et
al. (2004) found QTL in the same regions for BWT and growth rate. The calpain gene is in
this region, and calpain activity has been shown to be associated with the differences in
breast muscle between broilers and layers (Schreurs et al., 1995).
The current study
identified a QTL on Gga 3 affecting FNTH (234 cM). Carlborg et al. (2003), de Koning et
al. (2004) and Kerje et al. (2003) identified a QTL in this region affecting BWT. The Mprotein gene is in this region, and M-protein is a structural constituent of skeletal muscle
(Noguchi et al., 1992). The present study identified 3 QTL on Gga 5 affecting FNTH% (48
cM), FAT% (64 cM), and FAT (67 cM). Ikeobi et al. (2002) found QTL in this region for
FAT and FAT adjusted for body weight. The insulin like growth factor-2 (Yokomine et al.,
2001) and proinsulin (Perler et al., 1980) genes are both good candidates in this region
because of their effects on fat metabolism.
The current study identified a paternally
expressed QTL on Gga 3 affecting several growth and carcass traits. Tuiskula-Haavisto et al.
(2004) also identified a QTL in this region showing paternal expression affecting egg weight,
providing additional evidence that this is truly an imprinted QTL. Examples of methods and
successes of identifying positional candidate genes based on previously identified QTL
locations are reviewed by Mackay (2001), Abiola et al. (2003), Mackay (2004), and
Rothschild (2004).
The present study verified several QTL in commercial lines, which were identified in
previous studies of experimental or commercial lines. It is important to verify putative QTL
in multiple populations to provide confidence that the QTL are real, and to verify segregation
of the QTL in commercial lines (de Koning et al., 2004; de Koning et al., 2003). However,
additional new QTL were identified in the current study, including QTL affecting WM%.
75
Although it is encouraging that identified QTL regions continue to be associated with the
traits across studies, it is expected that many QTL will not be consistently identified across
studies.
Inconsistent results between studies can occur for many reasons, including
differences in power, differences in markers used, choice of statistical models, and, most
importantly, the use of different populations.
Use of Multiple Models
The use of multiple models was advantageous for identifying QTL. When using an
F2 cross between non-inbred lines, it is likely that some QTL will be differentially fixed
between the parental lines, and some QTL will not. The use of the LC, HS, and CB models
was successful in identifying QTL that were segregating in different patterns within the
cross. The QTL affecting WM% detected with the HS model on Gga 6 was a good example
of the value of using multiple models. The estimated effects of this QTL ranged from 1.69% to 1.43% within the F1 sires and the QTL was not detected using the LC model.
Using the parent of origin models allowed the detection of some QTL that were not identified
with the mendelian models. The results of the current analyses also provide evidence
supporting the existence of imprinting in chickens.
Conclusions
Several new QTL, as well as ones verifying previous studies, were identified in the
present study. The new QTL identify genome regions harboring genes or closely linked
markers that can be included in selection programs to enhance the genetic improvement of
broiler chickens.
The most useful QTL for selection are the ones affecting traits that are
difficult or expensive to measure, or traits for which birds need to be slaughtered (i.e. carcass
traits). The identified QTL that affect WM% are particularly promising for MAS, because of
the economic importance of this trait in the broiler industry. The complete draft of the
chicken genome will greatly aid in identification of the causative genes affecting these traits
(Hillier et al., 2004; Wong et al., 2004).
76
ACKNOWLEDGMENTS
The authors thank Jerry Smith and other Arbor Acres Farms staff for assistance in the
field; Edward Landers, Jennifer Green, Jeff Goodenbour, and Jenny Lanning for laboratory
assistance; and Robert Benson for computing assistance. The State of Iowa funds, the USDA
National Needs Fellowship program (98-38420-5833), Aviagen Int., and the Midwest Poultry
Research Program provided financial support for the project.
REFERENCES
Abiola, O., J. M. Angel, P. Avner, A. A. Bachmanov, J. K. Belknap, B. Bennett, E. P.
Blankenhom, D. A. Blizard, V. Bolivar, G. A. Brockmann, K. J. Buck, J. F. Bureau,
W. L. Casley, E. J. Chester, J. M. Cheverad, G. A. Churchill, M. Cook, J. C. Crabbe,
W. E. Crusio, A. Darvasi, G. de Haan, P. Dermant, R. W. Doerge, R. W. Elliot, C. R.
Farber, L. Flaherty, J. Flint, H. Gershenfeld, J. P. Gibson, J. Gu, W. Gu, H.
Himmelbauer, R. Hitzemann, H. C. Hsu, K. Hunter, F. F. Iraqi, R. C. Jansen, T. E.
Johnson, B. C. Jones, G. Kempermann, F. Lammert, L. Lu, K. F. Manly, D. B.
Matthews, J. F. Medrano, M. Mehrabian, G. Mittlemann, B. A. Mock, J. S. Mogil, X.
Montagutelli, G. Morahan, J. D. Mountz, H. Nagase, R. S. Nowakowski, B. F.
O'Hara, A. V. Osadchuk, B. Paigen, A. A. Palmer, J. L. Peirce, D. Pomp, M.
Rosemann, G. D. Rosen, L. C. Schalkwyk, Z. Seltzer, S. Settle, K. Shimomura, S.
Shou, J. M. Sikela, L. D. Siracusa, J. L. Spearow, C. Teuscher, D. W. Threadgill, L.
A. Toth, A. A. Toye, C. Vadasz, G. Van Zant, E. Wakeland, R. W. Williams, H. G.
Zhang, and F. Zou. 2003. The nature and identification of quantitative trait loci: A
community's view. Nat. Rev. Genet. 4: 911-916.
Assaf, S., S. Lagarrigue, S. Daval, M. Sansom, B. Leclercq, J. Michel, F. Pitel, M. Alizadeh,
A. Vignal, and M. Douaire. 2004. Genetic linkage and expression analysis of SREBP
and lipogenic genes in fat and lean chicken. Comp. Biochem. Physiol. B Biochem.
Mol. Biol. 137:433-441.
77
Brown, M. S., and J. L. Goldstein. 1997. The SREBP pathway: Regulation of cholesterol
metabolism by proteolysis of a membrane-bound transcription factor. Cell 89: 331340.
Carlborg, O., and L. Andersson. 2002. Use of randomization testing to detect multiple
epistatic QTLs. Genet. Res. 79: 175-184.
Carlborg, O., L. Andersson, and B. Kinghorn. 2000. The use of a genetic algorithm for
simultaneous mapping of multiple interacting quantitative trait loci. Genetics 155:
2003-2010.
Carlborg, O., P. M. Hocking, D. W. Burt, and C. S. Haley. 2004. Simultaneous mapping of
epistatic QTL in chickens reveals clusters of QTL pairs with similar genetic effects
on growth. Genet. Res. 83: 197-209.
Carlborg, O., S. Kerje, K. Schutz, L. Jacobsson, P. Jensen, and L. Andersson. 2003. A global
search reveals epistatic interaction between QTL for early growth in the chicken.
Genome Res. 13: 413-421.
de Koning, D. J., H. Bovenhuis, and J. A. van Arendonk. 2002. On the detection of imprinted
quantitative trait loci in experimental crosses of outbred species. Genetics 161: 931938.
de Koning, D. J., C. S. Haley, D. Windsor, P. M. Hocking, H. Griffin, A. Morris, J. Vincent,
and D. W. Burt. 2004. Segregation of QTL for production traits in commercial meattype chickens. Genet. Res. 83: 211-220.
de Koning, D. J., B. Harlizius, A. P. Rattink, M. A. Groenen, E. W. Brascamp, and J. A. van
Arendonk. 2001. Detection and characterization of quantitative trait loci for meat
quality traits in pigs. J. Anim. Sci. 79: 2812-2819.
de Koning, D. J., D. Windsor, P. M. Hocking, D. W. Burt, A. Law, C. S. Haley, A. Morris, J.
Vincent, and H. Griffin. 2003. Quantitative trait locus detection in commercial broiler
lines using candidate regions. J. Anim. Sci. 81: 1158-1165.
Deeb, N., and S. J. Lamont. 2003. Use of a novel outbred by inbred F1 cross to detect genetic
markers for growth. Anim. Genet. 34: 205-212.
Green, P., K. Fallis, and S. Crooks. 1994. Documentation for cri-map version 2.4,
Washington Univ. School Med., St. Louis, MO.
78
Groenen, M. A., H. H. Cheng, N. Bumstead, B. F. Benkel, W. E. Briles, T. Burke, D. W.
Burt, L. B. Crittenden, J. Dodgson, J. Hillel, S. Lamont, A. P. de Leon, M. Soller, H.
Takahashi, and A. Vignal. 2000. A consensus linkage map of the chicken genome.
Genome. Res. 10: 137-147.
Haddad, F., F. Zaldivar, D. M. Cooper, and G. R. Adams. 2005. IL-6-induced skeletal muscle
atrophy. J. Appl. Physiol. 98: 911-917.
Haley, C. S., and S. A. Knott. 1992. A simple regression method for mapping quantitative
trait loci in line crosses using flanking markers. Heredity 69: 315-324.
Haley, C. S., S. A. Knott, and J. M. Elsen. 1994. Mapping quantitative trait loci in crosses
between outbred lines using least squares. Genetics 136: 1195-1207.
Hillier, L. W., W. Miller, E. Birney, W. Warren, R. C. Hardison, C. P. Ponting, P. Bork, D.
W. Burt, M. A. Groenen, M. E. Delany, J. B. Dodgson, A. T. Chinwalla, P. F. Cliften,
S. W. Clifton, K. D. Delehaunty, C. Fronick, R. S. Fulton, T. A. Graves, C.
Kremitzki, D. Layman, V. Magrini, J. D. McPherson, T. L. Miner, P. Minx, W. E.
Nash, M. N. Nhan, J. O. Nelson, L. G. Oddy, C. S. Pohl, J. Randall-Maher, S. M.
Smith, J. W. Wallis, S. P. Yang, M. N. Romanov, C. M. Rondelli, B. Paton, J. Smith,
D. Morrice, L. Daniels, H. G. Tempest, L. Robertson, J. S. Masabanda, D. K. Griffin,
A. Vignal, V. Fillon, L. Jacobbson, S. Keije, L. Andersson, R. P. Crooijmans, J.
Aerts, J. J. van der Poel, H. Ellegren, R. B. Caldwell, S. J. Hubbard, D. V. Grafham,
A. M. Kierzek, S. R. McLaren, I. M. Overton, H. Arakawa, K. J. Beattie, Y.
Bezzubov, P. E. Boardman, J. K. Bonfield, M. D. Croning, R. M. Davies, M. D.
Francis, S. J. Humphray, C. E. Scott, R. G. Taylor, C. Tickle, W. R. Brown, J.
Rogers, J. M. Buerstedde, S. A. Wilson, L. Stubbs, I. Ovcharenko, L. Gordon, S.
Lucas, M. M. Miller, H. Inoko, T. Shiina, J. Kaufman, J. Salomonsen, K. Skjoedt, G.
K. Wong, J. Wang, B. Liu, J. Yu, H. Yang, M. Nefedov, M. Koriabine, P. J. Dejong,
L. Goodstadt, C. Webber, N. J. Dickens, I. Letunic, M. Suyama, D. Torrents, C. von
Mering, E. M. Zdobnov, K. Makova, A. Nekrutenko, L. Elnitski, P. Eswara, D. C.
King, S. Yang, S. Tyekucheva, A. Radakrishnan, R. S. Harris, F. Chiaromonte, J.
Taylor, J. He, M. Rijnkels, S. Griffiths-Jones, A. Ureta-Vidal, M. M. Hoffman, J.
Severin, S. M. Searle, A. S. Law, D. Speed, D. Waddington, Z. Cheng, E. Tuzun, E.
79
Eichler, Z. Bao, P. Flicek, D. D. Shteynberg, M. R. Brent, J. M. Bye, E. J. Huckle, S.
Chatterji, C. Dewey, L. Pachter, A. Kouranov, Z. Mourelatos, A. G. Hatzigeorgiou,
A. H. Paterson, R. Ivarie, M. Brandstrom, E. Axelsson, N. Backstrom, S. Berlin, M.
T. Webster, O. Pourquie, A. Reymond, C. Ucla, S. E. Antonarakis, M. Long, J. J.
Emerson, E. Betran, I. Dupanloup, H. Kaessmann, A. S. Hinrichs, G. Bejerano, T. S.
Furey, R. A. Harte, B. Raney, A. Siepel, W. J. Kent, D. Haussier, E. Eyras, R.
Castelo, J. F. Abril, S. Castellano, F. Camara, G. Parra, R. Guigo, G. Bourque, G.
Tes1er, P. A. Pevzner, A. Smit, L. A. Fulton, E. R. Mardis, and R. K. Wilson. 2004.
Sequence and comparative analysis of the chicken genome provide unique
perspectives on vertebrate evolution. Nature 432: 695-716.
Ikeobi, C. O., J. A. Woolliams, D. R. Morrice, A. Law, D. Windsor, D. W. Burt, and P. M.
Hocking. 2002. Quantitative trait loci affecting fatness in the chicken. Anim. Genet.
33: 428-435.
Ikeobi, C. O. N., J. A. Woolliams, D. R. Morrice, A. Law, D. Windsor, D. W. Burt, and P.
M. Hocking. 2004. Quantitative trait loci for meat yield and muscle distribution in a
broiler layer cross. Livestock Production Science 87: 143-151.
Jennen, D. G., A. L. Vereijken, H. Bovenhuis, R. P. Crooijmans, A. Veenendaal, J. J. van der
Poel, and M. A. Groenen. 2004. Detection and localization of quantitative trait loci
affecting fatness in broilers. Poult. Sci. 83: 295-301.
Keije, S., O. Carlborg, L. Jacobsson, K. Schutz, C. Hartmann, P. Jensen, and L. Andersson.
2003. The twofold difference in adult size between the Red Junglefowl and White
Leghorn chickens is largely explained by a limited number of QTLs. Anim. Genet.
34: 264-274.
Kim, J. -J., H. Zhao, H. Thomsen, M. F. Rothschild, and J. C. M. Dekkers. 2005a: Accepted
for publication. Combined line-cross and half-sib QTL analysis of crosses between
outbred lines. Genet. Res.
Kim, J. J., M. F. Rothschild, J. Beever, S. Rodriguez-Zas, and J. C. Dekkers. 2005b. Joint
analysis of two breed cross populations in pigs to improve detection and
characterization of quantitative trait loci. J. Anim. Sci. 83:1229-1240.
80
Knott, S. A., J. M. Elsen, and C. S. Haley. 1996. Methods for multiple-marker mapping of
quantitative trait loci in half-sib populations. Theoretical Applied Genetics 93: 71-80.
Knott, S. A., L. Marklund, C. S. Haley, K. Andersson, W. Davies, H. Ellegren, M. Fredholm,
I. Hansson, B. Hoyheim, K. Lundstrom, M. Moller, and L. Andersson. 1998. Multiple
marker mapping of quantitative trait loci in a cross between outbred wild boar and
large white pigs. Genetics 149: 1069-1080.
Knott, S. A., P. E. Nystrom, L. Andersson-Eklund, S. Stem, L. Marklund, L. Andersson, and
C. S. Haley. 2002. Approaches to interval mapping of QTL in a multigeneration
pedigree: The example of porcine chromosome 4. Anim. Genet. 33: 26-32.
Le Bihan-Duval, E., S. Mignon-Grasteau, N. Millet, and C. Beaumont. 1998. Genetic
analysis of a selection experiment on increased body weight and breast muscle weight
as well as on limited abdominal fat weight. Br. Poult. Sci. 39: 346-353.
Mackay, T. F. C. 2001. The genetic architecture of quantitative traits. Annu. Rev. Genet.
35: 303-339.
Mackay, T. F. C. 2004.
The genetic architecture of quantitative traits: lessons from
Drosophila. Curr. Opin. Genet. Dev. 14: 253-257.
Marks, H. L. 1995. Genetics of growth and development. In: P. Hunton (ed.) Poultry
production, p 157-183. Elsevier Science B. V., Amsterdam, The Netherlands.
McElroy, J. P., D. E. Harry, J. C. M. Dekkers, and S. J. Lamont. 2002. Molecular markers
associated with growth and carcass traits in meat-type chickens. Communication 0404 in Proceedings of the 7th World Congress on Genetics Applied to Livestock
Production, Montpellier, France. INRA, Cedex, France.
Noguchi, J., M. Yanagisawa, M. Imamura, Y. Kasuya, T. Sakurai, T. Tanaka, and T. Masaki.
1992. Complete primary structure and tissue expression of chicken pectoralis mprotein. J. Biol. Chem. 267: 20302-20310.
Getting, W. S., H. K. Lee, D. J. Flanders, G. L. Wiesner, T. A. Sellers, and R. A. King.
1995. Linkage analysis with multiplexed short tandem repeat polymorphisms using
infrared fluorescence and Ml3 Tailed Primers. Genomics 30, 450-458.
Perler, F., A. Efstratiadis, P. Lomedico, W. Gilbert, R. Kolodner, and J. Dodgson. 1980. The
evolution of genes: The chicken preproinsulin gene. Cell 20: 555-566.
81
Rothschild, M. F. 2004. Porcine genomics delivers new tools and results: This little piggy
did more than just go to market. Genet. Res., Camb. 83: 1-6.
Sasaki, O., S. Odawara, H. Takahashi, K. Nirasawa, Y. Oyamada, R. Yamamoto, K. Ishii, Y.
Nagamine, H. Takeda, E. Kobayashi, and T. Furukawa. 2004. Genetic mapping of
quantitative trait loci affecting body weight, egg character and egg production in F2
intercross chickens. Anim. Genet. 35: 188-194.
Schmid, M., I. Nanda, M. Guttenbach, C. Steinlein, M. Hoehn, M. Schartl, T. Haaf, S.
Weigend, R. Fries, J. M. Buerstedde, K. Wimmers, D. W. Burt, J. Smith, S. A'Hara,
A. Law, D. K. Griffm, N. Bumstead, J. Kaufman, P. A. Thomson, T. Burke, M. A.
Groenen, R. P. Crooijmans, A. Vignal, V. Fillon, M. Morisson, F. Pitel, M. TixierBoichard, K. Ladjali-Mohammedi, J. Hillel, A. Maki-Tanila, H. H. Cheng, M. E.
Delany, J. Bumside, and S. Mizuno. 2000. First report on chicken genes and
chromosomes 2000. Cytogenet. Cell Genet. 90: 169-218.
Schreurs, F. J., D. van der Heide, F. R. Leenstra, and W. de Wit. 1995. Endogenous
proteolytic enzymes in chicken muscles. Differences among strains with different
growth rates and protein efficiencies. Poult. Sci. 74: 523-537.
Seaton, G., C. S. Haley, S. A. Knott, M. Kearsey, and P. M. Visscher. 2002. QTL Express:
Mapping quantitative trait loci in simple and complex pedigrees. Bioinformatics 18:
339-340.
Sewalem, A., D. M. Morrice, A. Law, D. Windsor, C. S. Haley, C. O. Ikeobi, D. W. Burt,
and P. M. Hocking. 2002. Mapping of quantitative trait loci for body weight at three,
six, and nine weeks of age in a broiler layer cross. Poult. Sci. 81: 1775-1781.
Siwek, M., S. J. Comelissen, A. J. Buitenhuis, M. G. Nieuwland, H. Bovenhuis, R. P.
Crooijmans, M. A. Groenen, H. K. Parmentier, and J. J. van der Poel. 2004.
Quantitative trait loci for body weight in layers differ from quantitative trait loci
specific for antibody responses to sheep red blood cells. Poult. Sci. 83: 853-859.
Stevens, L. 1991. Genetics and evolution of the domestic fowl. Cambridge University Press,
Cambridge.
82
Tuiskula-Haavisto, M., D. J. De Koning, M. Honkatukia, N. F. Schulman, A. Maki-Tanila,
and J. Vilkki. 2004. Quantitative trait loci with parent-of-origin effects in chicken.
Genet. Res. 84:57-66.
Tatsuda, K., and K. Fujinaka. 2001. Genetic mapping of the QTL affecting body weight in
chickens using a F2 family. Br. Poult. Sci. 42: 333-337.
Tatsuda, K., K. Fujinaka, and T. Yamasaki. 2000. Genetic mapping of a body weight trait in
chicken. Anim. Sci. J. 71: 130-136.
Thomsen, H., H. K. Lee, M. F. Rothschild, M. Malek, and J. C. Dekkers. 2004.
Characterization of quantitative trait loci for growth and meat quality in a cross
between commercial breeds of swine. J. Anim. Sci. 82: 2213-2228.
Tuiskula-Haavisto, M., M. Honkatukia, J. Vilkki, D. J. de Koning, N. F. Schulman, and A.
Maki-Tanila. 2002. Mapping of quantitative trait loci affecting quality and production
traits in egg layers. Poult. Sci. 81: 919-927.
van Kaam, J. B., M. A. Groenen, H. Bovenhuis, A. Veenendaal, A. L. Vereijken, and J. A.
van Arendonk. 1999a. Whole genome scan in chickens for quantitative trait loci
affecting growth and feed efficiency. Poult. Sci. 78: 15-23.
van Kaam, J. B., M. A. Groenen, H. Bovenhuis, A. Veenendaal, A. L. Vereijken, and J. A.
van Arendonk. 1999b. Whole genome scan in chickens for quantitative trait loci
affecting carcass traits. Poult. Sci. 78: 1091-1099.
van Kaam, J. B. C. H. M., J. A. M. van Arendonk, M. A. M. Groenen, H. Bovenhuis, A. L. J.
Vereijken, R. P. M. A. Crooijmans, J. J. van der Poel, and A. Veenendaal. 1998.
Whole genome scan for quantitative trait loci affecting body weight in chickens using
a three generation design. Livestock Production Science 54: 133-150.
Wong, G. K., B. Liu, J. Wang, Y. Zhang, X. Yang, Z. Zhang, Q. Meng, J. Zhou, D. Li, J.
Zhang, P. Ni, S. Li, L. Ran, H. Li, R. Li, H. Zheng, W. Lin, G. Li, X. Wang, W.
Zhao, J. Li, C. Ye, M. Dai, J. Ruan, Y. Zhou, Y. Li, X. He, X. Huang, W. Tong, J.
Chen, J. Ye, C. Chen, N. Wei, L. Dong, F. Lan, Y. Sun, Z. Yang, Y. Yu, Y. Huang,
D. He, Y. Xi, D. Wei, Q. Qi, W. Li, J. Shi, M. Wang, F. Xie, X. Zhang, P. Wang, Y.
Zhao, N. Li, N. Yang, W. Dong, S. Hu, C. Zeng, W. Zheng, B. Hao, L. W. Hillier, S.
P. Yang, W. C. Warren, R. K. Wilson, M. Brandstrom, H. Ellegren, R. P. Crooijmans,
83
J. J. van der Poel, H. Bovenhuis, M. A. Groenen, I. Ovcharenko, L. Gordon, L.
Stubbs, S. Lucas, T. Glavina, A. Aerts, P. Kaiser, L. Rothwell, J. R. Young, S.
Rogers, B. A. Walker, A. van Hateren, J. Kaufman, N. Bumstead, S. J. Lamont, H.
Zhou, P. M. Hocking, D. Morrice, D. J. de Koning, A. Law, N. Bartley, D. W. Burt,
H. Hunt, H. H. Cheng, U. Gunnarsson, P. Wahlberg, L. Andersson, E. Kindlund, M.
T. Tammi, B. Andersson, C. Webber, C. P. Ponting, I. M. Overton, P. E. Boardman,
H. Tang, S. J. Hubbard, S. A. Wilson, J. Yu, and H. Yang. 2004. A genetic variation
map for chicken with 2.8 million single-nucleotide polymorphisms. Nature 432: 717722.
Yokomine, T., A. Kuroiwa, K. Tanaka, M. Tsudzuki, Y. Matsuda, and H. Sasaki. 2001.
Sequence polymorphisms, allelic expression status and chromosome locations of the
chicken IGF2 and MPR1 genes. Cytogenet. Cell Genet. 93: 109-113.
Yun, J. S., D. S. Seo, W. K. Kim, and Y. Ko. 2005. Expression and relationship of the
insulin-like growth factor system with posthatch growth in the Korean native Ogol
chicken. Poult. Sci. 84: 83-90.
Zerehdaran, S., A. L. Vereijken, J. A. van Arendonk, and E. H. van der Waaijt. 2004.
Estimation of genetic parameters for fat deposition and carcass traits in broilers.
Poult. Sci. 83: 521-525.
Zhou, H., A. D. Mitchell, J. P. McMurtry, C. M. Ashwell, and S. J. Lamont. 2005. Insulin­
like growth factor-I gene polymorphism associations with growth, body composition,
skeleton integrity, and metabolic traits in chickens. Poult. Sci. 84: 212-219.
84
TABLE 1. Estimated map positions of microsatellite markers used for
analysis and their corresponding consensus map1 positions
Microsatellite
Chromo- Estimated
some
Position2
Approximate
Consensus
Map Position
GCT0050
MCW0106
ADL0307
MCW0297
LEI0101
MCW0068
MCW0195
MCW0200
ADL0148
MCW0283
ADL0183
LEI0134
MCW0107
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
Gga 1
32
97
122
146
246
268
291
307
321
383
409 ,
509
547
32
94
128
162
259
283
302
330
360
414
443
527
565
MCW0205
MCW0082
ADL0190
ADL0309
ADL0212
MCW0042
MCW0173
MCW0088
MCW0185
MCW0264
MCW0051
MCW0166
LEI0031
ADL0146
MCW0143
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
Gga 2
0
26
46
90
136
216
231
258
286
304
329
334
386
391
448
0
30
63
92
152
228
243
273
302
320
358
358
400
403
460
LEI0043
ADL0370
LEI0161
MCW0212
MCW0277
MCW0207
LEI0166
MCW0037
Gga 3
Gga 3
Gga 3
Gga 3
Gga 3
Gga 3
Gga 3
Gga 3
9
109
140
187
213
289
328
349
9
79
115
154
190
250
290
317
Estimated
Position2
Approximate
Consensus
Map Position
Microsatellite
Chromosome
ADL0413
MCW0114
MCW0005
LEI0094
LEI0076
MCW0240
MCW0122
MCW0174
Gga 4
Gga 4
Gga 4
Gga 4
Gga 4
Gga 4
Gga 4
Gga 4
0
69
86
138
160
169
180
208
0
82
101
152
182
200
210
240
LEI0082
MCW0193
ADL0292
MCW0113
ADL0233
Gga 5
Gga 5
Gga 5
Gga 5
Gga 5
32
48
81
113
124
32
50
83
133
150
LEI0192
ADL0230
ADL0377
ADL0142
Gga 6
Gga 6
Gga 6
Gga 6
31
60
88
95
31
63
99
115
MCW0030
MCW0120
MCW0201
MCW0236
Gga 7
Gga 7
Gga 7
Gga 7
5
43
71
101
5
44
79
109
ADL0258
ADL0345
Gga 8
Gga 8
23
51
23
56
ADL0191
MCW0017
ADL0259
MCW0134
Gga 9
Gga 9
Gga 9
Gga 9
44
64
97
107
44
72
122
132
MCW0228
ADL0272
ADL0158
ADL0112
Gga 10
Gga 10
Gga 10
Gga 10
0
51
89
103
0
47
101
120
ADL0022
MCW0055
Gga Z
Gga Z
0
28
0
15
85
TABLE 1 Continued.
LEI0171
MCW0154
MCW0128
LEI0075
1
Gga Z
Gga Z
Gga Z
Gga Z
63
83
134
146
Groenen et al. (2000) and Schmid et al (2000)
positions of the markers on a chromosome were adjusted for the consensus
map position of the first marker on that chromosome.
2 Estimated
78
95
160
165
TABLE 2. Phenotypic correlations between traits and means (on diagonal) and standard deviations of traits (in
parentheses).
FNTH
FAT
WOG
FIL
TEN TWM WM% CS FAT% WOG% FNTH%
HW
BWT
Trait1
0.95
0.06 0.56 -0.13
0.97
0.20
0.98
0.85
0.68
0.86
0.27
0.00
2124(253)
HW
0.07 0.55 -0.10
0.96
0.93
0.84
0.84
0.66
0.02
2336(291) 0.22
0.29
BWT
0.18
0.18
0.09
0.18
0.10 -0.12 0.10 0.94
30(10)
-0.01
-0.04
FAT
0.87
0.70
1534(200) 0.96
0.88
0.13 0.56 -0.14
0.47
0.11
WOG
867(108) 0.92
0.78
0.93
0.29 0.64 -0.13
0.41
FNTH
0.31
359(55) 0.74
0.99
0.55 0.70 -0.19
0.39
0.35
FIL
0.57
81(10) 0.81
0.48
-0.04
0.36
0.43
TEN
440(63) 0.56 0.71 -0.17
0.40
TWM
0.38
WM%
21(2) 0.47 -0.13
0.35
0.75
CS
4(0.9) -0.08
0.21
0.34
FAT%
1(0.5) -0.09
-0.03
WOG%
72(2)
0.49
FNTH%
41(2)
FIL%
TEN%
TPL
1HW
= live weight (after transport to processing facility), g; BWT = body weight (prior to transport to
processing facility), g; FAT = weight of abdominal fat pad, g; WOG = weight of carcass without giblets, g;
FNTH = weight of the front half of the carcass, g; FIL = weight of the fillet, g; TEN = weight of the tender, g;
TWM = weight of the white meat, g; WM% = TWM percent of LWT; CS = conformation score; FAT% =
FAT percent of LWT; WOG% = WOG percent of LWT; FNTH% = FNTH percent of LWT; FIL% = FIL
percent of LWT; TEN% = TEN percent of LWT; and TPL = transport loss, g.
FIL% TEN% TPL
-0.32 0.39
0.15
-0.31 0.60
0.16
-0.13 -0.01 0.15
-0.26 0.41
0.22
-0.13 0.41
0.36
-0.07 0.36
0.64
0.28
0.48
0.41
0.36
0.02
0.63
0.07
0.56
0.97
0.07
0.24
0.51
-0.17
0.09
0.03
0.15
0.23
0.36
0.10
0.69
0.56
17(1) 0.35
0.11
4(0.4) -0.12
212(76)
lBLE
3. Estimates of position, effect, and mode of expression and inheritance of QTL identified at the 5%
chromosome-wise level.
Paternal Effect
'h'omo- Position1
some
Trait2
-log P3
Var(%)4
.
. 5 Estimate
Classification
S.E.
Additive Effect
Estimate
S.E.
Maternal Effect
Dominance Effect
Estimate
Estimate
S.E.
S.E.
1
155
FAT%
3.41**
3.68
LC
0.1
0.0
-0.1
0.1
1
168
FAT
3.56**
3.85
LC
4.1
1.1
-4.2
2.8
1
226
HW
2.66
6.04
CB
61.4
23.8
12.7
29.6
1
231
WOG
2.46
5.75
CB
38.1
17.2
8.9
19.3
-
1
379
FIL%
2.82
5.45
HS6
2
10
FAT%
2.77
3.01
LC
0.1
0.0
-0.1
0.1
2
13
FAT
3.12
3.38
LC
3.3
0.9
-1.5
1.6
2
67
FIL%
2.79
2.30
Mat-e
-0.6
0.2
2
69
WM%
2.78
2.30
Mat-e
-0.6
0.2
2
86
WM%
3.25**
6.84
CB
-0.8
0.2
0.0
0.3
2
87
FIL%
2.92
6.40
CB
-0.7
0.2
0.0
0.3
2
163
WOG%
3.78**
4.08
LC
-0.4
0.3
2.6
0.6
3
9
TEN
3.28**
6.88
CB
-4.6
1.9
4.0
2.2
3
49
BWT
2.85
2.40
Pat-e
82.4
25.6
3
49
HW
2.90
2.50
Pat-e
74.2
22.9
3
51
FNTH
2.49
2.00
Pat-e
32.9
11.1
3
51
TEN
3.86**
4.80
Partial
4.6
1.2
14.5
1.2
3
53
WOG
2.32
2.00
Pat-e
53.7
18.9
3
62
TWM
2.57
2.50
Pat-e
21.9
7.3
3
129
WM%
2.49
2.10
Pat-e
0.3
0.1
3
210
WM%
2.99
3.24
LC
0.4
0.1
-0.4
0.2
3
213
FIL
2.78
3.02
LC
13.9
3.8
-1.9
6.6
-
-
-2.4
1.2
TABLE 3 Continued
3
213
FNTH%
2.41
2.62
LC
3
213
HW
2.45
2.00
Pat-e
3
213
TWM
2.86
3.10
LC
3
214
FNTH
2.45
2.00
Pat-e
14.4
4.9
3
214
WOG
2.4
2.00
Pat-e
23.8
8.2
3
217
BWT
2.48
2.00
Pat-e
36.3
12.3
3
225
HW
2.33
4.81
HS
3
226
FIL%
2.66
2.89
LC
3
231
WOG
2.42
4.93
HS
3
234
FNTH
2.97**
5.65
HS
3
234
WOG%
2.67
3.40
Partial
3
274
FAT%
2.07
4.45
HS
4
160
FAT
2.31
2.51
LC
2.2
4
160
FAT%
2.56
2.78
LC
5
32
FIL
2.74*
5.34
HS
5
32
FNTH
2.38
4.87
HS
5
32
HW
1.87
4.16
HS
5
32
TWM
2.41
4.91
HS
5
35
WOG
2.28
4.73
HS
5
45
TEN
2.09
1.70
Pat-e
5
48
CS
2.59
2.81
LC
5
48
FNTH
2.73*
2.30
Pat-e
5
48
FNTH%
3.90**
4.21
5
48
WM%
3.09*
5
48
WOG
5
48
5
50
0.4
0.1
-0.4
0.2
16.3
4.5
-2.8
7.6
0.4
0.1
-0.4
0.3
-1.0
0.2
0.7
-1.4
1.2
0.1
0.0
0.0
0.1
-0.2
0.1
0.0
0.1
LC
-0.4
0.1
0.0
0.2
3.35
LC
-0.4
0.1
0.1
0.2
2.10
1.70
Pat-e
WOG%
2.32
2.52
LC
-0.5
0.1
-0.1
0.2
FIL%
2.84*
3.08
LC
-0.3
0.1
0.1
0.2
29.7
0.0
-1.4
-13.9
-19.9
10.1
0.7
0.2
0.2
0.5
4.4
7.4
oo
oo
TABLE 3 Continued
5
51
TWM
2.93*
2.50
Pat-e
-9.7
3.0
5
53
FIL
2.92*
2.50
Pat-e
-8.6
2.6
5
64
FAT%
4.51***
4.84
LC
0.2
0.0
0.0
0.1
5
67
FAT
4.19**
4.51
LC
3.5
0.8
-0.4
1.4
6
48
WM%
1.76
4.00
HS
-
-
-
-
6
59
FAT
2.28
4.74
HS
-
-
-
-
6
65
FNTH
3.24*
5.99
HS
-
-
-
-
6
88
BWT
2.58
2.80
LC
47.3
15.3
-41.6
25.3
6
88
FIL
2.58
2.80
LC
10.5
3.5
-10.7
5.8
6
88
HW
2.94*
3.19
LC
45.9
13.6
-36:5
22.6
6
88
TEN
2.61*
2.84
LC
2.1
0.7
-2.4
1.2
6
88
TWM
2.80*
3.03
LC
12.7
4.1
-13.1
6.8
6
88
WOG
2.17
2.36
LC
29.9
10.9
-30.2
18.1
7
72
WOG%
1.91
1.50
Mat-e
7
82
FAT%
2.52
4.09
HS
-
-
-
-
7
85
FAT
1.93
4.26
HS
-
-
-
-
8
25
FAT
1.49
3.62
HS
-
-
-
-
8
10
27
FAT%
1.92
4.24
HS
-
-
-
-
0.3
0.1
_
4.69
TPL
2.24
HS
89
1 Pos = estimate of position relative to the consensus map
2 HW = live weight (after transport to processing facility), g; BWT = body weight (prior to transport to processing facility), g;
FAT = weight of abdominal fat pad, g; WOG = weight of carcass without giblets, g; FNTH = weight of the front half of the
carcass, g; FIL = weight of the fillet, g; TEN = weight of the tender, g; TWM = weight of the white meat, g; WM% = TWM
percent of HW; CS = conformation score; FAT% = FAT percent of HW; WOG% = WOG percent of HW; FNTH% = FNTH
percent of HW; FIL% = FIL percent of HW; TEN% = TEN percent of HW; and TPL = transport loss, g.
3 -Log P = the negative Log base ten of the comparison-wise P-value. * = 1% chromosome-wise significance, ** = 5%
experiment-wise significance, and *** =1% experiment-wise significance.
4 Var(%) = the % of phenotypic variance explained by including the QTL in the model.
5 Denotes the inheritance pattern of the QTL: Mat-e = maternal expression; Pat-e = paternal expression; Partial =
partially imprinted QTL (Full model); LC = line-cross QTL; HS = half-sib QTL; CB = combined QTL.
6 Additive and dominance estimates for HS QTL differed by sire and are not reported here.
Figure 1. QTL detected on Gga 3 with line-cross, half-sib, or
combined models.
O TEN: CB model
5% Experiment-wise threshold2
X FNTH: HS model
$ WOG and HW: HS model;
5%.Chromosome-wise.threshold
• FAT%: HS model
FNTH%: LC model1
• TWM, WM%, FIL%, FIL:
LC model
100
200
Consensus Position (cM)
1
Two of the plotted lines are representative of multiple traits. HS = half-sib model; CB =
combined model; HW = live weight (after transport to processing facility), g; WOG =
weight of carcass without giblets, g; FNTH = weight of the front half of the carcass, g;
FIL = weight of the fillet, g; TEN = weight of the tender, g; TWM = weight of the white
meat, g; WM% = TWM percent of HW; FAT% = FAT percent of HW; FNTH% = FNTH
percent of HW; FIL% = FIL percent of HW.
2 Significance thresholds averaged across models and traits.
31 = marker position
300
91
Figure 2. QTL detected on Gga 5 with line-cross, half-sib, or
combined models.
5% Experiment-wise
threshold2
O FNTH%: LC model
x FAT%, FAT: LC model
• FIL%, WOG%, CS, WM%:
LC model1
5% Chromosom
ise threshold
$ TWM, FIL, FNTH, WOG:
HS model
HW: HS model
40
50
60
70
80
90
100
110 120
Consensus Position (cM)
1
Two of the plotted lines are representative of multiple traits. HS = half-sib model; HW =
live weight, g; FAT = weight of abdominal fat pad, g; WOG = weight of carcass without
giblets, g; FNTH = weight of the front half of the carcass, g; FIL = weight of the fillet, g;
TWM = weight of the white meat, g; WM% = TWM percent of HW; CS = conformation
score; FAT% = FAT percent of HW; WOG% = WOG percent of HW; FNTH% = FNTH
percent of HW; FIL% = FIL percent of HW.
2 Significance thresholds averaged across models and traits.
31 = marker position
Figure 3. QTL detected on Gga 3 with parent of origin models.
5% Experiment-wise threshold
O HW, BWT : Pat-e model1
• FNTH: Pat-e model
5% Chromosome-wise threshold
• WM%: Pat-e model
x TEN: Pat-e model
$ TWM: Pat-e model;
WOG%: Part-e model
100
1 Pat-e
Consensus Position (cM) 200
300
= paternal expression; Part-e = partial imprinting; FNTH = weight of the front half of the carcass, g; TEN = weight of the tender, g;
HW = harvest weight, g; TWM = total white meat weight, g; WOG% = weight of the carcass without giblets percent of HW; WM% = TWM/HW;
2 Significance thresholds averaged across models and traits.
31 = marker position
93
Figure 4. QTL detected on Gga 5 with parent of origin models.
1% Chromosomewise threshold
O TWM: Pat-e model1
5% Chromosome
wise threshold
-LOG(P)
x FNTH: Pat-e model
• FIL: Pat-e model
$ TEN: Pat-e model
• WOG: Pat-e model
40
50
60
70
80
90
100
110 120
Consensus Position (cM)
1
Pat-e = paternal expression; FIL = fillet weight, g; FNTH = weight of the front half of
the carcass, g; TEN = weight of the tender, g; TWM = total white meat weight, g;
WOG = weight of the carcass without giblets, g.
2 Significance thresholds averaged across models and traits.
J | = marker position
94
CHAPTER 3
MICROSATELLITE MARKERS ASSOCIATED WITH RESISTANCE TO
MAREK'S DISEASE IN COMMERCIAL LAYER CHICKENS
A manuscript submitted to Poultry Science.
J. P. McElroy*, J. C. M. Dekkers*, J. E. Fultonî, N. P. O'Sullivanî, M. Seller#, E. Lipkin#,
W. Zhang*, K. J. Koehler*, S. J. Lament*, and H. H. Chengt
Iowa State University, Ames, Iowa 50011; +USDA, ARS, Avian Disease and Oncology
Î
Laboratory, East Lansing, Michigan 48823; Hy-Line International, P.O. Box 310, Dallas
Center, IA, 50063; ^Department of Genetics, Alexander Silberman Life Science Institute,
Hebrew University of Jerusalem, 91904 Jerusalem, Israel
ABSTRACT The objective of the current study was to identify QTL conferring resistance
to Marek's disease (MD) in commercial layer chickens. To generate the resource population,
two partially inbred lines that differed in MD-caused mortality were intermated to produce
five backcross families. Vaccinated chicks were challenged with very virulent plus MD virus
strain 648A at 6 d and monitored for MD symptoms. A recent field isolate of the MD virus
was used because the lines were resistant to commonly used, older laboratory strains.
Selective genotyping was employed using 81 microsatellites selected based on prior results
with selective DNA pooling. Linear regression and Cox proportional hazard models were
used to detect associations between marker genotypes and survival. Significance thresholds
were validated by simulation. Seven and six markers were significant based on proportion of
false positive and false discovery rate thresholds less than 0.2. Seventeen markers were
associated with MD survival considering a comparison-wise error rate of 0.10, which is
about twice the number expected by chance, indicating that at least some of the associations
represent true effects.
Thus, the present study shows that loci affecting MD resistance can
be mapped in commercial layer lines. More comprehensive studies are under way to confirm
and extend these results.
(Key words: chicken, Marek's disease, quantitative trait loci, survival, genetic resistance)
95
INTRODUCTION
Marek's disease (MD), a lymphoma caused by an avian herpesvirus, is a major
disease affecting the poultry industry.
It has been roughly estimated that, worldwide,
Marek's disease costs the poultry industry one to two billion dollars a year (Morrow and
Fehler, 2004).
The economic damage of MD is probably even greater because
immunosuppression induced by the MD virus reduces resistance to other pathogens, which
can lead to symptoms in young, market-weight broilers (Biggs et al., 1968; Abbassi et al.,
1999), and lowers feed efficiency and other production traits (Groves, 1995; Islam et al.,
2002). Vaccines have been produced that initially were effective in reducing MD incidence
(Witter, 1985), but MD virus strains have evolved to the point that commercial vaccines are
no longer fully protective.
An alternative method to reduce the incidence of MD is to genetically improve the
chicken's innate resistance to this disease.
Resistance or susceptibility to MD is a
quantitative trait, being affected by multiple genes and the environment.
Genetic
improvement of quantitative traits can be achieved by selection of individuals with favorable
phenotypic characteristics, by marker assisted selection on genomic regions that harbor genes
that confer the favorable phenotype, or both (Lande and Thompson, 1990). Marker-assisted
selection is particularly useful for lowly heritability traits and for traits that are difficult to
measure. The MD resistance falls into both these categories.
Use of marker-assisted selection requires knowledge of genes affecting a trait or of
markers tightly linked to those genes (Dekkers and Hospital, 2002). A genome scan can be
used to identify regions of the genome that harbor genes affecting a quantitative trait of
interest, so-called QTL (Soller and Beckmann, 1983; Beckmann and Soller, 1983). Once a
QTL region is identified, it can be more intensely studied to find the causative gene or a
closely linked marker for use in selection programs.
Genomic regions associated with resistance to MD have been identified in several
studies of noncommercial poultry populations. Vallejo et al. (1998) and Yonash et al. (1999)
identified QTL on chromosomes 1,2,4, 7, and 8 affecting MD resistance by using the same
¥2 cross between two White Leghorn lines (Avian Disease and Oncology Laboratory lines 6
96
and 7) that differed in MD resistance. Bumstead (1998) used a backcross of the same ADOL
lines to map a QTL for MD resistance on chromosome 1. Many studies have shown that the
MHC complex (B blood group) on chromosome 16 affects resistance to MD (Hanson et al.,
1967; Bacon et al., 1981; Bacon, 1987; Schierman and Collins, 1987; Lamont, 1989; Bacon
and Witter, 1994; S chat et al., 1994). The growth hormone gene (GH1) on chromosome 1
also has an allelic association with MD resistance (Kuhnlein et al., 1997; Liu et al., 2001b).
Microarray analysis has shown that GH1 expression is associated with differences in MD
resistance (Liu et al., 2001a) and the GH1 protein has been shown to interact with the SORF2
protein, a protein only found in virulent MD virus strains (Liu et al., 2001b).
Stem
lymphocyte antigen 6 complex locus E (LY6E) on chromosome 2 has also been identified as
an MD resistance gene through genetic, RNA, and protein analysis (Liu et al., 2003). Since
all these studies used experimental populations, it is important to confirm the association of
these QTL regions with MD in commercial populations, which will enable selection upon the
QTL in those populations. The objective of the current study, therefore, was to identify QTL
associated with MD resistance (defined as survival time following challenge) in a cross
between lines of commercial layer chickens.
MATERIALS AND METHODS
Experimental Population and Phenotyping
The population was a backcross between two partially inbred lines (as determined by
foundation from narrow genetic bases) of commercial layer chickens. In a prior screening of
the parental lines for 102 microsatellite markers, 60% were fixed in line 1 and 80% in line 2.
The lines were also fixed for different serologically typed B blood group alleles: B2 in line 1
and B15 in line 2. Prior studies also showed the parental lines to differ in susceptibility to
experimental challenge with a very virulent MD virus: percentage MD mortality was 41.4
and 21.0 percentage points higher in Line 1 than Line 2 (data not shown), defining line 2 as
the more resistant of these two lines.
97
To produce the resource population (see Figure 1), five line 1 males were pair mated
to individual line 2 females to generate five full-sib F1 families. Seven males from each Fi
family were each mated to 15 line 1 females to generate five grandsire backcross groups,
each consisting of seven half-sib sire-families.
A total of 656 backcross female chicks (85 to 160 per backcross group) were
vaccinated with 500 plaque forming units of bivalent HVT/SB-1 vaccine2 at 1 d of age and
subcutaneously inoculated with 500 plaque forming units of the vv+ (very virulent plus)
648A MD virus strain (Witter, 1997) at 6 d of age. Age at death and presence or absence of
tumors by visual examination was recorded from 30 to 140 d of age. Initiation of records at
30 d of age excludes typical early chick and brooding mortality. Animals surviving to the
end of the study (140 d) were euthanized by CO2 inhalation. Treatment of the animals met or
exceeded accepted guidelines (presented in Guidelines for the Care and Use of Agricultural
Animals in Agricultural Research and Teaching, 1988). Birds were housed and phenotypic
data were collected at Hy-Line International, Dallas Center, IA. Survival time, quantified
based on number of days of survival post experimental challenge with a virulent MD virus,
was the phenotype used for QTL mapping.
Markers and Genotyping
For DNA isolation, blood was collected from the jugular vein at 3 wk of age in
syringes containing EDTA with 22 gauge needles. The Qiagen QIAamp DNA Blood Mini
kit3 was used for DNA isolation according to the manufacturer's instructions, except that 25
uL of whole blood plus 175 uL PBS were used for the spin protocol, and the samples were
incubated at 70C.
The backcross progeny were selectively genotyped (Lander and Botstein, 1989;
Darvasi and Soller, 1992) by genotyping the 20% (n = 133) of chicks with shortest survival
times past 30 d and that had tumors and the 20% longest survivors (n = 134) for 81
microsatellite markers. Individuals in the extremes of the phenotypic distribution contain the
majority of the information needed to identify markers linked to that trait (Lander and
2
Merial Select, Gainesville, GA 30503.
98
Botstein, 1989) and maximizes power with limited genotyping (Lebowitz et al., 1987).
Presence of macroscopically visible tumors was used as a defining trait to place shortsurviving birds into the category of MD-susceptible short survivors, thus minimizing the
placement into this category of birds that died from non-MD-related causes. All paternal
grandparents and F, sires were also genotyped but genotypes were not available for dams of
the backcross chicks.
Markers used in the current study were chosen based on their associations with MD
resistance in two preliminary selective DNA pooling analyses (unpublished), following
methods described in Darvasi and Soller (1994) and Lipkin et al. (1998). Fifty-six of the 81
genotyped markers were chosen based on a selective DNA pooling analysis of 117 markers
in this population (data not shown), and the 25 additional markers were chosen from 120
tested markers based on a selective DNA pooling analysis of the reciprocal backcross
population (data not shown). Markers that were included in the initial pooling analyses were
chosen by position to get maximal genomic coverage. The 81 markers were distributed
among 17 chromosomes: 16 on chromosome 1, 14 on 2, 10 on 3, 9 on Z, 8 on 5, 7 on 4, 4 on
15, 2 each on 6, 8, and 18, and 1 each on 7, 9, 13, 17, 23, 27, and E22. The average marker
interval for chromosomes with multiple markers was 37 cM.
Statistical Analyses
Line Origin Probabilities. The objective of the statistical analyses was to identify
associations of marker alleles with survival in the backcross offspring, based on line origin of
the marker allele that was inherited from the Fi sire. Line origin could be determined for 38
markers that were fixed for alternative alleles in the parental lines and for 23 markers for
which distinct alleles were segregating in the parental lines. For 20 of the 81 markers,
however, parental lines segregated at least one common allele.
For such markers, an
offspring was not fully informative when identically heterozygous to the F% sire. On average
across these markers, 55% of the offspring were not fully informative. Although genotypes
of the backcross dams were unknown, allelic frequencies in the line were known from
previous genotyping of line 1 individuals.
3 Cat.
# 51106; Qiagen, Valencia, CA.
These frequencies were used to infer the
99
probability that a dam transmitted a given allele to the non-informative offspring and,
equivalently, the probability that the sire transmitted the alternate allele. The probability that
a backcross offspring with marker genotype
AiA2
inherited a line 1 allele from its Ft sire
(p(Li)) was then computed as:
p(L,) = p(A, = LI| A] = F l ) p (Ai = F 1 ) + p (A = L l | A = F l ) p (A = F1),
2
2
2
where p(Aj = LI ] A, = Fl) is the probability that allele Aj originated from line 1 given that it
came from the F, sire (= 1, 0.5, or 0 following Mendelian inheritance), and can be computed
based on allele frequencies among dams, f(Aj), as:
p(Aj =F1) = 1 - f(Aj) / [f(A,) + f(A2)].
Note that, since p(L,) = 1 - p(L2), all information on line origin is captured by p(Li).
Statistical Models.
Only individuals that were genotyped were included in the
analyses. The phenotypic data were right-skewed, censored (some individuals survived to the
end of the study and therefore did not have a date of death), and only phenotypic extremes
were genotyped (selective genotyping). For this reason the Cox Proportional Hazards (CPH)
model (Cox, 1972) was used for analysis, in addition to regular least squares regression
(Legendre, 1805). Aside from the proportional hazards assumption, estimation for the CPH
model is rank-based (distribution-free) and accommodates survival data with censoring
(Smith, 2002); therefore, it may be more appropriate than least squares regression for
analysis of these data. Effect estimates from the CPH model can be interpreted as natural
logarithms of ratios of hazards. The least squares regression model was also considered
because it is computationally easier to employ than the CPH model and the effect estimates
from linear regression have a more convenient interpretation with respect to mean survival
time.
In preliminary analyses under the CPH model, effects of grandsire, blood group
genotype, their interaction, and the interactions of these effects with p(L%) did not occur more
100
often than expected by chance, considering a comparison-wise error rate (CWER) of 0.05
(fewer than 5% of the tests had a CWER > 0.05 for each of the effects). Therefore, these
effects were not included in the final models for analysis.
The final model used for linear regression analysis was:
S| =y5p(L,)i +Sj,
where Sj is the survival time of animal i, in days; /? is the increase in the mean survival time
associated with inheriting the line 1 versus the line 2 marker allele; p(L])j is as defined
previously for animal i; and s, is the residual for animal i.
Following Smith (2002) the model used for CPH analysis was:
S(t;Tii) = [So(t)]^i\
where S(t;r|j ) is the probability that animal i survived at least until time t and
So (t)
is the
baseline survivor function:
so(t) = e ^ ^
where ho(t) is the baseline hazard function and
T|i =/?p(Ll)iThe corresponding hazard function is:
h(t) = exp(ri i )h 0 (t) = exp((3p(L 1 ) i )h 0 (t),
where [J is the allelic effect on the natural log of the ratio of hazards for inheriting the line
one versus line 2 marker, and p(L,)j is as defined previously. This formulation of the model
allows the use of standard CPH statistical software for estimation of /?, and the baseline
hazard is not needed to estimate /? (Smith, 2002). This approach provides an approximation
to a partial likelihood estimator for /?, as discussed in the Appendix. Simulation of data under
the null hypothesis of no QTL effect was used to ensure that the standard f-values that were
obtained from each model were appropriate. To account for multiple testing, the False
Discovery Rate (FDR) (Benjamini and Hochberg, 1995; Weller et al., 1998) and the
Probability of False Positives (PFP) (Fernando et al., 2004; Heifetz, 2004) statistics were
used as an indication of the strength of associations of markers with survival.
101
The linear regression and CPH models were compared based on their ability to
identify markers associated with MD survival and for their agreement in comparison-wise Pvalues and estimates of marker effects. The (3 coefficient in the linear regression model is
interpreted in units of days, whereas P in the CPH model is interpreted in terms of the
conditional odds of dying during a small time period after any particular time point given that
an animal survives up to that time point. Because of their different scales, correlation was
used to compare the effect estimates and P-values of the two models.
Simulation Analysis. To determine whether standard P-values were appropriate for
the analyses that were conducted, survival data with properties similar to the observed data
were simulated with a backcross model under the null hypothesis of no QTL, following
procedures described by Vincent Ducrocq (Station de Genetique Quantitative et Appliquée,
Institut National de la Recherche Agronomique, France; personal communication).
To
simulate new samples of survival times that reflect the features of the observed data, a
survival function (S(t)) was estimated from the observed data using the non-parametric
Kaplan-Meier estimator (Kaplan and Meier, 1958) of a survivor function:
M
nj
for t(k) < t < t(k+i), where j is the rank of a particular day (t) among all chronologically ordered
days in which death occurred, k is the rank of the day at which the survival function is being
evaluated, nj is the number of animals at risk on day j, and dj is the number of animals dying
on day j. Note that nj excludes any animals that either died or were censored before day j,
but includes any animal censored at day j. Estimated S(t) = 1 for 0 < t < t(i). In this
application animals were only censored at 140 d. A simulated sample of death times was
obtained by drawing a sample from a uniform [0,1] distribution and inverting the KaplanMeier estimate of the survivor function to obtain death times. Animals with death times
exceeding 140 d were censored at 140 d. Each animal was also independently assigned one
of two marker alleles based on a random draw from a binomial distribution with a 50%
102
chance of inheriting either allele, simulating a marker linked to a QTL with no effect. To
simulate selective genotyping data, the simulated animals were ranked based on survival time
and the 20% shortest and longest survivors were analyzed using the linear regression and
CPH models. Animals surviving to 140 d were considered censored for the CPH model and
as dying at 140 d for the linear regression model. This process was repeated for 1,000
replicates of 700 animals and the proportions of replicates with a f-value less than CWER
levels of 0.1, 0.05, 0.01, and 0.001 were compared to the expected false positive rate for
those levels.
A two-tailed binomial test (Miller and Miller, 1999) was used to identify
deviations from expectations.
RESULTS
Distribution of Survival Times
Survival times in the backcross population over the recording time (30 to 140 d)
ranged from 33 to 140 d (Figure 2). Survival times followed a right-skewed distribution with
a mean of 65.5 d, a median of 59.0 d, and a standard deviation of 23.9 d. Twenty-eight
individuals (4.3%) survived to the end of the study and were considered censored in the
analyses.
False Positive Rates
The percentage (of 1000 simulated replicates) of tests that had a comparison-wise Pvalue less than 0.1, 0.05, 0.01, or 0.001 from analyzing the selective genotyping survival data
simulated under the null hypothesis (no QTL effect) with the linear regression and CPH
models is shown in Table 1. None of the false positive rates were significantly different (P >
0.05) from the expected rates (the four CWER thresholds) based on a two-tailed binomial
test.
This indicates that the comparison-wise f-values obtained from the actual data
correspond to tests with valid type I error levels for both methods of analysis.
103
Marker Analysis
As mentioned above, in preliminary analyses, proportions of significant effects of
grandsire and blood group did not deviate from the expected by chance using a CWER of
0.05 under the CPH model, and therefore they were not included in the final models for
analysis. Results, summarized in Table 2 and Figure 3, showed seven and six markers that
exceeded a 0.2 threshold using PFP and FDR, respectively. Results for an additional ten
markers, that were significant relative to a CWER of 0.10, but did not reach a PFP of FDR
threshold of 0.2, are also summarized in Table 2 and Figure 3. Although the evidence of an
association of these markers with survival is not as strong, these results are included here so
that they can be compared to results from other experiments that are or will be reported in the
literature, and so that trends can be seen in identified genomic regions reported in the present
study. The corresponding locations from the consensus genetic linkage map of the chicken
are also indicated in Table 2 and Figure 3.
Correlations of Effect Estimates and P-values
The correlation between effect estimates from the linear regression and CPH analyses
of all markers was -0.96, indicating that CPH estimates can be accurately predicted from
linear regression estimates.
The negative relationship arises from the difference in the
interpretation of the parameters in the two models; a smaller expected survival time from the
regression model corresponds to a larger hazard ratio in the CPH model. The correlation
between the P-values from the two models was 0.83, suggesting good correspondence in the
degree of significance.
DISCUSSION
Comparison of Analyses
Analysis of the simulated selective genotyping survival data showed that standard
determinations of CWER resulted in valid false positive rates and, therefore, in valid P-
104
values for the CPH and linear regression models. This result was expected for the CPH
model because, assuming the proportional hazards assumption was met, the data did not
violate assumptions of the model, but not necessarily expected for linear regression because
the assumption of normally distributed data was violated (Larsen and Marx, 1990). The
CWER for the linear regression analysis appeared robust to violation of this assumption,
likely because the numbers of individuals were large enough for the central limit theorem to
become a factor (Miller and Miller, 1999), and therefore the estimated coefficients had a
large sample normal distribution, so the significance tests also had an approximate normal
distribution. This does not, however, mean that the effect estimates obtained from either of
these models are valid. It is well known that linear regression overestimates QTL effects
when selective genotyping is employed, even if phenotype is normally distributed in the
complete data set (Lander and Botstein, 1989). Darvasi and Soller (1992) and Ronin et al.
(1998) proposed methods to correct this bias for normally distributed traits, but these
methods are not appropriate for survival data because of skewness and censoring.
From analysis of the actual data, linear regression appears to be as, or more, powerful
than the CPH model for analyzing the selective genotyping survival data. There was a strong
linear relationship between estimates from the two models. An advantage of the linear
regression model over the CPH model is that estimated coefficients are much easier to
interpret. Estimates from linear regression are in days of survival, whereas CPH estimates
are in terms of an exponential function of the odds that animals die in a small time period
following some time point given that the animals survive up to that time point. Estimation of
coefficients in the CPH model is a based on the ranks of the death and censoring times,
which results in the model ignoring information in the spacing between death times. The
regression approach uses information in the spacing of death times, but it will be affected by
the handling of censoring times. Using the censoring time as a death time in the regression
analysis was not a major issue in the current study because censoring only occurred at 140 d,
and only 4.3% of the animals survived beyond 140 d.
105
Markers Associated with Marek's Disease Survival
Several markers were associated with length of survival in this post-vaccination MD
challenge using an FDR or PFP threshold of 0.2. One of these markers corresponds to a QTL
identified on chromosome 2 near the region identified for MD susceptibility by Yonash et al.
(1999) and Vallejo et al. (1998) (around 90 cM on the consensus map). Confirmation of
QTL in multiple populations is important for eliminating false positives and demonstrating
segregation of the QTL in multiple populations. In the current study, no QTL were found in
regions on chromosomes 1, 4, 7, 8, 12, and 17, which were identified as possibly harboring
QTL by Vallejo et al. (1998), Yonash et al. (1999), and
Bumstead (1998).
These
discrepancies could be due to lack of segregation of these same QTL between the lines used
in the present study, insufficient power, or false positives in the other three studies.
Discrepancies between the studies may also have arisen because a recent vv+ field isolate of
the MD virus was used in the current study, whereas the other studies used less virulent
strains. The lines of birds used here are resistant to the commonly used older laboratory
strains of the MD virus. Previous studies also did not identify the strong QTL identified in
the present study on chromosome Z. This may be due to the fact that in the previous studies,
this chromosome was only surveyed with a single marker. In an Fa cross of highly inbred
chicken lines, however, Zhou et al. (2003) identified a QTL for antibody response kinetics on
chromosome Z, near the location of the QTL for MD resistance identified in the present
study. The growth hormone receptor gene is also located on chromosome Z near the same
position. The growth hormone-1 gene (on chromosome 1) has been associated in previous
studies with MD resistance (Kuhnlein et al., 1997; Liu et al., 2001 a,b). Interactions between
markers near the growth hormone-1 and the growth hormone receptor genes affecting MD
resistance were not significant (data not shown).
For the 17 markers identified in the present study to have an association with MD
resistance with a CWER <0.10, 12 showed allele effects in the expected direction, with the
favorable allele originating from the more resistant Line 2. These QTL, therefore, explain
part of the difference in MD resistance between the two lines. However, favorable QTL
alleles were also identified as originating from the less resistant line (i.e., cryptic alleles).
106
Major Histocompatability Complex
The MHC has been shown to be associated with resistance to many diseases in
poultry, including MD (Bacon, 1987; Bacon and Witter, 1994; Bacon et al., 1981;
Lakshmanan et al., 1997; Lamont, 1989; Lamont, 1998; Steadham et al., 1987). The current
study did not, however, find an association between the MHC and MD survival. In the
current study, the MHC associated blood group alleles present were B2 and B15, and the
blood group genotypes in the backcross offspring were B2/B2 or B2/B15.
If, for the
experimental population in the current study, the B2 allele was completely dominant over the
B15 allele, a difference in survival between the two MHC genotypes would not be expected.
Epistatic interactions between the MHC and background genes that were not linked to a
marker used in this study, which could mask an MHC effect, may also explain the lack of
association between blood group alleles and MD survival. The strain of the MD virus used
in the current study (vv+ 648A) may have also impacted the role of the MHC for affecting
survival.
Significance Tests
To account for multiple testing, FDR and PFP thresholds were used. Seven and six
markers exceeded FDR and PFP threshold of 0.2 using linear regression, and five and five
using the CPH model. For PFP, a threshold of 0.2 results in the expectation that 80% of the
tests exceeding this threshold are true positives (Fernando et al., 2004). Interpretation of a
0.2 threshold for FDR is more difficult to define but is similar, though somewhat more
conservative, depending on the number of true effects present in the dataset.
Another issue to consider regarding experiment-wise thresholds is that 56 of the
markers used in the current study were selected from previous analyses of DNA pools for
117 markers on the same experimental population. Reported results assumed that the 56
markers represented a random set of markers, regardless of the early pools results; that is, on
the assumption that the pooling analyses were not at all indicative of the results of the current
analyses. If the pooling analyses were highly predictive of results of the current study, all
117 markers used in the pooling analyses would need to be considered when determining
experiment-wise significance levels using FDR and PFP, resulting in only three markers
107
exceeding the 0.2 threshold (the same three markers would exceed the FDR and PFP
threshold of 0.2 for both the CPH and linear models). This, along with the results reported,
represents the two extremes, i.e., the upper and lower bounds of the experiment-wise
thresholds to be used in the current study. A comparison of CWER P-values from individual
genotyping for the 56 markers that were selected based on the pools, however, showed rather
low correlations, ranging from 0.11 to 0.21, with P-values from the pooling analyses, in
which the pools were created within B blood group genotype. This indicates that statistical
tests based on the pooling analyses were not very predictive of the results of statistical tests
based on individual genotyping, and therefore our assumption was correct and the
experiment-wise thresholds need not be corrected for pre-selection of markers from the
pooling analysis. Note that the problem of pre-selection does not arise for the 25 markers
selected based on pool analyses of the reciprocal backcross because these were based on
different individuals and data.
The large discrepancy between results of the statistical tests from the pooling analysis
and the selective genotyping analysis are likely due to several reasons, including accuracy of
the pools and differences in traits considered and in methods for statistical analyses.
Although the correlation between frequencies of alleles estimated from pools and the actual
frequency of alleles in the individuals that contributed to the pools was high (approximately
0.90), it was not 1.00, and parental lines were not fixed for alternate alleles for 44 of the 117
markers, which could lead to errors from the pooling analyses. The trait analyzed in pooling
analyses was also different from the trait analyzed in selective genotyping analyses; for the
pooling analyses, pools were formed within blood group genotype and the number of tumors
was considered as an additional variable when selecting individuals, whereas only length of
survival was considered as the phenotype in the current analyses. In addition, individuals
surviving to the end of the study which did not have tumors were not included in the pooling
analysis, but were included in the selective genotyping analysis. Since these individuals are
the most extreme, they were likely the most informative in the selective genotyping study
and therefore contributed to the discrepancy between the pooling and individual genotyping
analyses. Finally, the statistical models used for pool analysis and those used for analysis of
the individual genotyping results were also quite different.
108
Implications
Identification of and subsequent selection upon QTL affecting MD resistance will be
useful to the poultry industry to reduce losses caused by MD virus infection (Vallejo et al.,
1998). Improving genetic resistance to Marek's disease can also improve vaccine efficacy
(Lamont et al., 2002) and possibly increase the length of time that vaccines are useful before
the virus mutates to become resistant. The QTL identified in the current experiment are
starting points for more intensive studies to precisely locate the QTL positions for utilization
in markers-assisted selection or for identification of the genes responsible for phenotypic
variation. Phenotypic selection for MD requires exposure and costly challenge of relatives of
selection candidates with the pathogenic agent to obtain phenotypic data on resistance or
susceptibility (Arthur and Albers, 2003). Direct selection on genes that affect resistance to
MD, or on linked markers, does not require a direct disease challenge of immediate relatives,
although challenge studies are needed to identify initial associations. Direct selection uses
genetic information on selection candidates rather than their relatives and can be
implemented almost immediately upon hatching, thereby potentially shortening the
generation interval. Therefore, identifying genetic regions affecting MD resistance is of
great value to the poultry industry. The current study is the first reported QTL scan for MD
resistance in commercial layers. However, the identification of the QTL on chromosome 2 in
both the current study and in studies using experimental lines also shows the usefulness of
experimental populations in identifying QTL that may also be segregating in commercial
populations. Current availability of the draft of the complete chicken genome sequence and a
2.8 million SNP map will facilitate future QTL and causative gene identification (Hillier et
al., 2004; Wong et al., 2004).
ACKNOWLEDGMENTS
Research was partly supported by a USDA National Needs Fellowship in Animal
Biotechnology (JM), by Hatch and State of Iowa funds (SJL, JCMD) Project 6680, Project
109
6674, and USDA NRICGP award #99-03307 (HHC and MS). The authors acknowledge
Laurie Molitor for her technical assistance.
APPENDIX
The hazard function h(t) for a line is proportional to the conditional probability that an
animal dies shortly after time t given that it survives to time t. The CPH model assumes that at
any time t, the hazard resulting from inheriting the line 1 allele
h(t) = ep h0(t)
is proportional to a baseline hazard h0(t) for inheriting the line 2 allele. At any time point t,
the regression coefficient (5 is the natural logarithm of the relative risk that a death occurs
shortly after time t for animals inheriting the line 1 allele versus those inheriting the line 2
allele, provided that the animals have survived through time t. When there is uncertainty about
the allele that was inherited, the hazard is an average of the hazards for the two possible alleles
h(t) = p(Li)eP h0(t) + [1- p(Li)]ho(t)
where p(L]) is the probability of inheriting the line 1 allele.
Then, a maximum partial
likelihood estimator for (3 is obtained by maximizing
n
i=l
p(Li)jexp (p) + [1 - p(L 1 ) i ]
(Al)
I (p(Li)j expCP)+[1 - p(L] ) j] )
UGR(ti)
where r is the number of observed deaths and R(tj) denotes the set of animals still alive at
the time of the i-th death. This is not a standard form of the Cox partial likelihood and it
cannot be maximized with standard statistical software. To use standard statistical software
for the Cox model, an alternative estimator is obtained by using p(Lj)j as an explanatory
variable in a standard Cox model. This yields
h(ti) = exp(Pp(L1)i)h0(ti)
110
as an approximation for
h(ti ) = [p(M )j exp((3) + l-p(L1)i]h0(t)
resulting in a partial likelihood
\
/
A
i=i
exp(Pp(LJi)
(A2)
Z (exP(p(Li)j))
/
that can be maximized with standard statistical software for the Cox model.
Using
simulation, it can be shown that maximizing (A2) provides an estimator of p that tends to be
biased a little farther away from zero than the estimator obtained by maximizing (Al). The
standard error for this estimator, however, is also slightly larger, resulting in a test of
significance that maintains nearly the correct type I error level. Note that maximizing either
(Al) or (A2) results in an increasingly more biased estimate of p as fewer animals with
intermediate failure times are genotyped. When the null hypothesis is true (|3=0), however,
standard errors of the estimators increase in a proportional manner so tests retain desired type
I error levels.
REFERENCES
Abbassi, H., F. Coudert, Y. Cherel, G. Dambrine, J. Brugere-Picoux, and M. Maciri. 1999.
Renal cryptosporidiosis (Cryptosporidium baileyi) in specific-pathogen-free chickens
experimentally coinfected with Marek's disease virus. Avian Dis. 43:738-744.
Arthur, J. A., and G. A. A. Albers. 2003. Industrial perspective on problems and issues
associated with poultry breeding. Pages 1-12 in Poultry Genetics, Breeding and
Biotechnology. W. M. Muir and S. E. Aggrey, ed. CABI Publishing, Cambridge, MA.
Bacon, L. D. 1987. Influence of the major histocompatibility complex on disease resistance
and productivity. Poult. Sci. 66:802-811.
Bacon, L. D., and R. L. Witter. 1994. B haplotype influence on the relative efficacy of
Marek's disease vaccines in commercial chickens. Poult. Sci. 73:481-487.
Ill
Bacon, L. D., R. L. Witter, L. B. Crittenden, A. Fadly, and J. Motta. 1981. B-haplotype
influence on Marek's disease, Rous sarcoma, and lymphoid leukosis virus-induced
tumors in chickens. Poult. Sci. 60:1132-1139.
Beckmann, J. S., and M. Soller. 1983. Restriction fragment length polymorphisms in genetic
improvement: methodologies, mapping and costs. Theor. Appl. Genet. 67: 35-43.
Benjamini, Y., and Y. Hochberg. 1995. Controlling the false discovery rate: A practical and
powerful approach to multiple testing. J. R. Stat. Soc. B 57:289-300.
Biggs, P.M., P.L. Long, S.G. Kenzy, and D.G. Rootes. 1968. Relationship between Marek's
disease and coccidiosis. II. The effect of Marek's disease on the susceptibility of
chickens to coccidial infection. Vet. Rec. 83:284-289.
Bumstead, N. 1998. Genomic mapping of resistance to Marek's disease. Avian Pathol.
27:S78-S81.
Cox, D. R. 1972. Regression models in life tables (with discussion). J.R. Stat. Soc. B 34:187220.
Darvasi, A., and M. Soller. 1992. Selective genotyping for determination of linkage between
a marker locus and a quantitative trait locus. Theor. Appl. Genet. 85:353-359.
Darvasi, A., and M. Soller. 1994. Selective DNA pooling for determination of linkage
between a molecular marker and a quantitative trait locus. Genetics 138:1365-1373.
Dekkers, J. C., and F. Hospital. 2002. The use of molecular genetics in the improvement of
agricultural populations. Nat. Rev. Genet. 3:22-32.
Fernando, R. L., D. Nettleton, B. R. Southey, J. C. Dekkers, M. F. Rothschild, and M. Soller.
2004. Controlling the proportion of false positives in multiple dependent tests.
Genetics 166:611-619.
Groenen, M. A., H. H. Cheng, N. Bumstead, B. F. Benkel, W. E. Briles, T. Burke, D. W.
Burt, L. B. Crittenden, J. Dodgson, J. Hillel, S. Lamont, A. Ponce de Leon, M. Soller,
H. Takahashi, and A. Vignal. 2000. A consensus linkage map of the chicken genome.
Genome Res. 10:137-147.
Groves, P. 1995. Some epidemiological aspects of Marek's disease in the Australian poultry
industry. Epidemiology Chapter Proceedings. Pages 27-35. Sydney: Australian
College of Veterinary Scientists.
112
Guide for the Care and Use of Agricultural Animals in Agricultural Research and Teaching.
1988. Consortium, Association Headquarters, 1111 N. Dunlap Avenue, Savoy, IL
61874.
Hanson, M. P., G. R. J. Law, and J. N. Van Zandt. 1967. Differences in susceptibility to
Marek's disease in chickens carrying two different B locus blood group alleles. Poult.
Sci. 46:1268 (Abstr.).
Heifetz, E. M. 2004. Thesis: Fine mapping of QTL, methods and results. The Hebrew
University of Jerusalem. Jerusalem, Israel.
Hillier, L. W., W. Miller, E. Bimey, W. Warren, R. C. Hardison, C. P. Ponting, P. Bork, D.
W. Burt, M. A. Groenen, M. E. Delany, J. B. Dodgson, A. T. Chinwalla, P. F. Cliften,
S. W. Clifton, K. D. Delehaunty, C. Fronick, R. S. Fulton, T. A. Graves, C.
Kremitzki, D. Layman, V. Magrini, J. D. McPherson, T. L. Miner, P. Minx, W. E.
Nash, M. N. Nhan, J. O. Nelson, L. G. Oddy, C. S. Pohl, J. Randall-Maher, S. M.
Smith, J. W. Wallis, S. P. Yang, M. N. Romanov, C. M. Rondelli, B. Paton, J. Smith,
D. Morrice, L. Daniels, H. G. Tempest, L. Robertson, J. S. Masabanda, D. K. Griffin,
A. Vignal, V. Fillon, L. Jacobbson, S. Keije, L. Andersson, R. P. Crooijmans, J.
Aerts, J. J. van der Poel, H. Ellegren, R. B. Caldwell, S. J. Hubbard, D. V. Grafham,
A. M. Kierzek, S. R. McLaren, I. M. Overton, H. Arakawa, K. J. Beattie, Y.
Bezzubov, P. E. Boardman, J. K. Bonfield, M. D. Craning, R. M. Davies, M. D.
Francis, S. J. Humphray, C. E. Scott, R. G. Taylor, C. Tickle, W. R. Brown, J.
Rogers, J. M. Buerstedde, S. A. Wilson, L. Stubbs, I. Ovcharenko, L. Gordon, S.
Lucas, M. M. Miller, H. Inoko, T. Shiina, J. Kaufman, J. Salomonsen, K. Skjoedt, G.
K. Wong, J. Wang, B. Liu, J. Yu, H. Yang, M. Nefedov, M. Koriabine, P. J. Dejong,
L. Goodstadt, C. Webber, N. J. Dickens, I. Letunic, M. Suyama, D. Torrents, C. von
Mering, E. M. Zdobnov, K. Makova, A. Nekrutenko, L. Elnitski, P. Eswara, D. C.
King, S. Yang, S. Tyekucheva, A. Radakrishnan, R. S. Harris, F. Chiaromonte, J.
Taylor, J. He, M. Rijnkels, S. Griffiths-Jones, A. Ureta-Vidal, M. M. Hoffman, J.
Severin, S. M. Searle, A. S. Law, D. Speed, D. Waddington, Z. Cheng, E. Tuzun, E.
Eichler, Z. Bao, P. Flicek, D. D. Shteynberg, M. R. Brent, J. M. Bye, E. J. Huckle, S.
Chatterji, C. Dewey, L. Pachter, A. Kouranov, Z. Mourelatos, A. G. Hatzigeorgiou,
113
A. H. Paterson, R. Ivarie, M. Brandstrom, E. Axelsson, N. Backstrom, S. Berlin, M.
T. Webster, O. Pourquie, A. Reymond, C. Ucla, S. E. Antonarakis, M. Long, J. J.
Emerson, E. Betran, I. Dupanloup, H. Kaessmann, A. S. Hinrichs, G. Bejerano, T. S.
Furey, R. A. Harte, B. Raney, A. Siepel, W. J. Kent, D. Haussier, E. Eyras, R.
Castelo, J. F. Abril, S. Castellano, F. Camara, G. Parra, R. Guigo, G. Bourque, G.
Tesler, P. A. Pevzner, A. Smit, L. A. Fulton, E. R. Mardis, and R. K. Wilson. 2004.
Sequence and comparative analysis of the chicken genome provide unique
perspectives on vertebrate evolution. Nature 432: 695-716.
Islam, A.F.M.F., C.W. Wong, S.W. Walkden-Brown, I.G. Colditz, K.E. Arzey, and P.J.
Groves. 2002. Immunosuppressive effects of Marek's disease virus (MDV) and
herpesvirus of turkeys (HVT) in broiler chickens and the protective effect of HVT
vaccination against MDV challenge. Avian Pathol. 31:449-461.
Kaplan, E. L., and P. Meier. 1958. Nonparametric estimation from incomplete observations.
J. Am. Stat. Assoc. 53:457-481.
Kuhnlein, U., L. Ni, S. Weigend, J. S. Gavora, W. Fairfull, and D. Zadworny. 1997. DNA
polymorphisms in the chicken growth hormone gene: Response to selection for
disease resistance and association with egg production. Anim. Genet. 28:116-123.
Lakshmanan, N., J. S. Gavora, and S. J. Lamont. 1997. Major histocompatibility complex
class II DNA polymorphisms in chicken strains selected for Marek's disease
resistance and egg production or for egg production alone. Poult. Sci. 76: 1517-1523.
Lamont, S. J. 1989. The chicken major histocompatibility complex in disease resistance and
poultry breeding. J. Dairy Sci. 72:1328-1333.
Lamont, S. J. 1998. Impact of genetics on disease resistance. Poult. Sci. 77: 1111-1118.
Lamont, S. J., M. G. Kaiser, and W. Liu. 2002. Candidate genes for resistance to Salmonella
enteritidis colonization in chickens as detected in a novel genetic cross. Vet.
Immunol. Immunopathol. 87:423-428.
Lande, R., and R. Thompson. 1990. Efficiency of marker-assisted selection in the
improvement of quantitative traits. Genetics 124:743-756.
Lander, E. S., and D. Botstein. 1989. Mapping mendelian factors underlying quantitative
traits using RFLP linkage maps. Genetics 121:185-199.
114
Larsen, R. J., and M. L. Marx. 1990. Statistics. Prentice Hall, Englewood Cliffs, NJ.
Lebowitz, R. J., M. Soller, and J. S. Beckmann. 1987. Trait-based analyses for the detection
of linkage between marker loci and quantitative trait loci in crosses between inbred
lines. Theor. Appl. Genet. 73:556-562.
Legendre, A. M. 1805. Appendix to Nouvelles Méthodes pour la détermination des orbites
des comètes: Sur la Méthode des moindres quarrés. Paris, France.
Lipkin, E., M. O. Mosig, A. Darvasi, E. Ezra, A. Shalom, A. Friedmann, and M. Soller.
1998. Quantitative trait locus mapping in dairy cattle by means of selective milk
DNA pooling using dinucleotide microsatellite markers: Analysis of milk protein
percentage. Genetics 149:1557-1567.
Liu, H. C., H. H. Cheng, V. Tirunagaru, L. Sofer, and J. Bumside. 2001a. A strategy to
identify positional candidate genes conferring Marek's disease resistance by
integrating DNA microarrays and genetic mapping. Anim. Genet. 32:351-359.
Liu, H. C., H. J. Kung, J. E. Fulton, R. W. Morgan, and H. H. Cheng. 2001b. Growth
hormone interacts with the Marek's disease virus SORF2 protein and is associated
with disease resistance in chicken. Proc. Natl. Acad. Sci. USA 98:9203-9208.
Liu, H.C., M. Niikura, J. Fulton, and H.H. Cheng. 2003. Identification of chicken stem
lymphocyte antigen 6 complex, locus E (LY6E, alias SCAT) as a putative Marek's
disease resistance gene via a virus-host protein interaction screen. Cytogen. Genome
Res. 102:304-308.
Miller, I., and M. Miller. 1999. John E. Freund's Mathematical Statistics. Prentice Hall,
Upper Saddle River, NJ.
Morrow, C., and F. Fehler. 2004. Marek's disease: a worldwide problem. Pages 49-61 in
Marek's Disease, An Evolving Problem. F. Davison and V. Nair, ed. Elsevier Ltd.,
Oxford, U.K.
Ronin, Y .1., A. B. Korol, and J. I. Weller. 1998. Selective genotyping to detect quantitative
trait loci affecting multiple traits: interval mapping analysis. Theor. Appl. Genet.
97:1169-1178
115
Schat, K. A., R. L. Taylor, Jr., and W. E. Briles. 1994. Resistance to Marek's disease in
chickens with recombinant haplotypes to the major histocompatibility (B) complex.
Poult. Sci. 73:502-508.
Schierman, L. W., and W. M. Collins. 1987. Influence of the major histocompatibility
complex on tumor regression and immunity in chickens. Poult. Sci. 66:812-818.
Schmid, M., I. Nanda, M. Guttenbach, C. Steinlein, M. Hoehn, M. Schartl, T. Haaf, S.
Weigend, R. Fries, J. M. Buerstedde, K. Wimmers, D. W. Burt, J. Smith, S. A'hara,
A. Law, D. K. Griffin, N. Bumstead, J. Kaufman, P. A. Thomson, T. Burke, M. A.
Groenen, R. P. Crooijmans, A. Vignal, V. Fillon, M. Morisson, F. Pitel, M. TixierBoichard, K. Ladjali-Mohammedi, J. Hillel, A. Maki-Tanila, H. H. Cheng, M. E.
Delany, J. Bumside, and S. Mizuno. 2000. First report on chicken genes and
chromosomes 2000. Cytogenet. Cell Genet. 90:169-218.
Smith, P. J. 2002. Analysis of Failure and Survival Data. Chapman & Hall/CRC, Boca
Raton, FL.
Soller, M., and J. S. Beckmann. 1983. Genetic polymorphisms in varietal identification and
genetic improvement. Theor. Appl. Genet. 67: 25-33.
Steadham, E. M., S. J. Lamont, I. Kujdych, and A. W. Nordskog. 1987. Association of
Marek's disease with Ea-B and immune response genes in subline and F2 populations
of the Iowa State SI Leghorn line. Poult. Sci. 66: 571-575.
Vallejo, R. L., L. D. Bacon, H. C. Liu, R. L. Witter, M. A. Groenen, J. Hillel, and H. H.
Cheng. 1998. Genetic mapping of quantitative trait loci affecting susceptibility to
Marek's disease virus induced tumors in F2 intercross chickens. Genetics 148:349360.
Wang, J. 2003. Ph.D. Dissertation: Interval mapping of QTL with selective DNA pooling
data. Iowa State University. Ames, IA.
Weller, J. I., J. Z. Song, D. W. Heyen, H. A. Lewin, and M. Ron. 1998. A new approach to
the problem of multiple comparisons in the genetic dissection of complex traits.
Genetics 150:1699-1706.
Witter, R. L. 1985. Principles of vaccination. Pages 203-250 in Marek's Disease. L. N.
Payne, ed. Martinus Nijhoff Publishing, Boston, MA.
116
Witter, R. L. 1997. Increased virulence of Marek's disease virus field isolates. Avian Dis.
41:149-163.
Wong, G. K., B. Liu, J. Wang, Y. Zhang, X. Yang, Z. Zhang, Q. Meng, J. Zhou, D. Li, J.
Zhang, P. Ni, S. Li, L. Ran, H. Li, R. Li, H. Zheng, W. Lin, G. Li, X. Wang, W.
Zhao, J. Li, C. Ye, M. Dai, J. Ruan, Y. Zhou, Y. Li, X. He, X. Huang, W. Tong, J.
Chen, J. Ye, C. Chen, N. Wei, L. Dong, F. Lan, Y. Sun, Z. Yang, Y. Yu, Y. Huang,
D. He, Y. Xi, D. Wei, Q. Qi, W. Li, J. Shi, M. Wang, F. Xie, X. Zhang, P. Wang, Y.
Zhao, N. Li, N. Yang, W. Dong, S. Hu, C. Zeng, W. Zheng, B. Hao, L. W. Hillier, S.
P. Yang, W. C. Warren, R. K. Wilson, M. Brandstrom, H. Ellegren, R. P. Crooijmans,
J. J. van der Poel, H. Bovenhuis, M. A. Groenen, I. Ovcharenko, L. Gordon, L.
Stubbs, S. Lucas, T. Glavina, A. Aerts, P. Kaiser, L. Rothwell, J. R. Young, S.
Rogers, B. A. Walker, A. van Hateren, J. Kaufman, N. Bumstead, S. J. Lamont, H.
Zhou, P. M. Hocking, D. Morrice, D. J. de Koning, A. Law, N. Bartley, D. W. Burt,
H. Hunt, H. H. Cheng, U. Gunnarsson, P. Wahlberg, L. Andersson, E. Kindlund, M.
T. Tammi, B. Andersson, C. Webber, C. P. Ponting, I. M. Overton, P. E. Boardman,
H. Tang, S. J. Hubbard, S. A. Wilson, J. Yu, and H. Yang. 2004. A genetic variation
map for chicken with 2.8 million single-nucleotide polymorphisms. Nature 432: 717722.
Yonash, N., L. D. Bacon, R. L. Witter, and H. H. Cheng. 1999. High resolution mapping and
identification of new quantitative trait loci (QTL) affecting susceptibility to Marek's
disease. Anim. Genet. 30:126-135.
Zhou, H., H. Li, and S. J. Lamont. 2003. Genetic markers associated with antibody response
kinetics in adult chickens. Poult. Sci. 82:699-708.
117
TABLE 1. False positive rates for the linear regression and Cox's proportional
hazards models from simulation under the null hypothesis of no QTL effect, for
different comparison-wise significance levels. Results are based on 1,000 replicates.
=
" = = = = C o m p a r i s o n - w i s e significance leveF
Model
0.1
0.05
0.01
0.001
Linear regression
Cox proportional
hazard
0.111
0.053
0.01
0
j0g
0 066
0 007
„
0
'Using Fisher's exact test, none of the values were different (comparisonwise P <0.05) from the expected values based on significance level.
118
TABLE 2. Markers associated (comparison-wise P < 0.10) with Marek's disease
survival
ADL0176
LEI0116
ADL0253
HUJ0005
/LC&A278
MCW0055
LEI0121
Chromo­ Position1
some
2
2
2
2
45
5
5
5
6
8
15
Z
Z
z
z
z
z
90
115
225
230
220
5
45
100
40
95
30
0
20
30
35
75
130
Linear regression
P-value
0.0143'4
0.0113'4
0.074
0.086
0.031
0.01g3
0.0073'4
0.081
0.094
0.161
0.029
0.0013'4
0.0003'4
0.0013'4
0.092
0.076
0.103
Cox's proportional hazards
Effect (d)2
-10.71
-12.01
-7.65
-7.79
9.28
10.32
20.01
7.53
7.24
-6.00
-16.13
-14.63
-15.27
-14.54
-7.22
-7.61
-7.00
P-value
0.051
0.0063'4
0.191
0.232
0.124
0.067
0.008^
0.135
0.161
0.083
0.057
oo
o o
o o
4^
Marker
0.0013'4
0.066
0.049
0.077
'Locations from the consensus genetic linkage map of the chicken (Groenen et al., 2000;
Schmid et al., 2000)
"Negative effects from the linear regression model (in days) and positive effects from the
Cox proportional hazards (in terms of the effect on the natural log of the hazard ratio)
model indicate that the favorable allele was derived from the more resistant line.
Significant based on the Proportion of False Positive rate < 0.2 [equivalent comparisonwise significance levels: 0.0183 (linear regression) and 0.008 (Cox proportional hazards)].
^Significant based on False Discovery Rate <0.2 [equivalent comparison-wise significance
levels: 0.0143 (linear regression) and 0.008 (Cox proportional hazards)].
^ There is evidence that MAXL may not be on chromosome 4 (Wang, 2003).
Effect2
0.26
0.40
0.17
0.17
-0.20
-0.25
-0.64
-0.20
-0.19
0.23
0.42
0.46
0.45
0.44
0.24
0.26
0.23
119
FIGURE 1. Population design used to generate each of the five grand-parental backcross
families. Numbers of individuals are indicated by the numbers in the circles. Five line 1
males were pair mated to five line 2 females to generate five full-sib F, families. Seven sires
from each of the five Fi families were each mated to approximately 15 line 1 females to
generate five groups (by grandsire) of backcross individuals. Length of survival was
recorded for all backcross individuals. Genotypes for 81 microsatellite markers were known
for all grandparents, F, sires, and selected backcross individuals. Genotypes were not
observed for the dams of the backcross individuals. MD = Marek's disease.
Parental Lines
Line 1
5 Males
Genotypes
Line 2
5 Females
Genotypes
Dams
Pedigree and
Genotypes Unknown
F, Sines
Genotypes
Backcross Offspring
Challenged
with MO Virus
® ® <3>
0)0)
Phenotypes and
Selected Genotypes
120
FIGURE 2. Distribution of length of survival in the backcross population. Dark
shading indicates the selectively genotyped individuals.
a
«
"i
125-
10075-
i
5025-
30
40
50
60
80
90 100 110 120 130 140
Survival (d)
121
FIGURE 3. Markers associated with Marek's disease resistance (comparison-wise P < 0.1)
on chromosomes 2, 5, and Z. Markers are indicated on the right side of each chromosome,
and their positions (cM) on the left. The stringency of the threshold is indicated by the size
of the circle. Un-filled circles indicate cryptic alleles. The model [linear regression (LR) or
Cox's proportional hazards (CPH)] in which a marker was found to be associated with
Marek's disease resistance is indicated by the location of the circle on the left or right side of
the chromosome, respectively. Asterisks (*) indicate markers exceeding probability of false
positives or false discovery rate threshold of 0.2.
Chr.2
LB
o
Chr.5
CPH
LR
5
90—*. *— fl£UL0309 *
45 -
A ADLQ176 *
115—*,
Chr.Z
CPH
o
LE/0116 *
"
ADL0253 *
"U u
83
165
AM.0300
MCW0034
225230-
100.
UL
0
20
30
35
CPH
-AOL0022
-MCW0331
-MCW0055 *
~w—MCW0259
AOL 0023
7595
274
130-
150
—ADL0273
I—
— LEI0121
144
320
198
378
165
414-
201-
Comparison-wise
•
p < = 0.1
#
p < = 0.05
#
p < = 0.01
V_y Unfilled circles indicate
cryptic alleles, i.e.
susceptible alleles from
the resistant line.
^
Arrows indicate approximate
positions of markers on
chromosomes 2, 5, and Z with
comparison-wise P ^ 0.1 for
both analyses (not all position
numbers shown)
122
CHAPTER 4
COMPARISON OF METHODS FOR ANALYSIS OF SELECTIVE
GENOTYPING SURVIVAL DATA
A manuscript in preparation for submission to Genetics, Selection, and Evolution
J. P. McElroy3, W. Zhangb, K. J. Koehlerb, S. J. Lamonf, and J. C. M. Dekkers3
^Department of Animal Science, Iowa State University, Ames, Iowa 50011; ^Department of
Statistics, Iowa State University, Ames, Iowa 50011
123
ABSTRACT - Identification of genomic regions harboring genes (QTL) affecting
economic traits using genetic markers is a primary step for improvement of agricultural
species through marker-assisted selection.
Selective genotyping of only animals with
extreme phenotypes can reduce the costs associated with QTL mapping studies. Although
the most common QTL mapping methods assume normality of phenotypes, many traits are
non-normally distributed, such as survival traits. Therefore, the objective of the present
study was to compare models for identification of QTL associated with survival traits, and
with selectively genotyped survival traits.
Data were simulated to model the survival
distribution of a population of chickens challenged with Marek's disease virus.
Cox
proportional hazards (CPH), linear regression (LR), and Weibull models were compared for
their appropriateness to analyze the data, ability to identify associations of marker alleles
with survival, and estimation of effects when all individuals were genotyped (full
genotyping) and when selective genotyping was used. Little difference in power was found
between the CPH and the LR model for low censoring cases for both full and selective
genotyping. With high censoring and full genotyping, LR was less powerful than the other
two models, but had power similar to the other two models with selective genotyping. The
simulated data did not follow a Weibull distribution and, as a result, the Weibull model
generally resulted in less power than the other two models and overestimated effects. Effect
estimates from LR and CPH were unbiased when all individuals were genotyped, but
overestimated when selective genotyping was used. Thus, LR is preferred for analyzing
survival data when the amount of censoring is low because of ease of implementation and
interpretation.
Including phenotypic data of non-genotyped individuals in selective
genotyping analysis increased power, but resulted in LR having an inflated false positive
rate, and therefore the CPH model is preferred for this scenario. Results from the research
presented herein are directly applicable to interval mapping analyses.
Keywords: survival/Cox proportional hazards/Weibull/quantitative trait loci
124
1. INTRODUCTION
Genetic association analyses are becoming a common approach in animal breeding to
identify genes or genomic regions that affect quantitative traits (e.g., Malek et al., 2001; van
Kaam et al., 1999b; Zhou et al., 2003). Most analyses utilize statistical models that assume
normality of phenotypes (Haley and Knott, 1992; Haley et al., 1994; Jansen and Stam, 1994;
Lander and Botstein, 1989; Zeng, 1993). Many phenotypic traits of interest in agriculture,
however, do not follow a normal distribution (e.g. ordinal traits such as conformation scores,
binary traits such calving/not calving, or time to success/failure traits such as survival time).
Many agriculturally important traits follow a survival distribution (e.g. survival after
infection, length of productive life, and days open). To analyze such traits, the Weibull and
Cox proportional hazards models are commonly employed (Beaudeau et al., 1995; Benard et
al., 1999; Ducrocq et al., 2000; Grohn et al., 1997; Grohn et al., 1998; Hirst et al., 2002;
Maizon et al., 2004; Rajala-Schultz et al., 2001; Roxstrom et al., 2003; Southey et al., 2001).
The Weibull model, a generalization of the exponential model, is parametric and is therefore
appropriate for only specific distributions; the Cox model is a rank-based semi-parametric
method and, therefore, should be appropriate for all distributions as long as the hazards
between groups are proportional (Smith, 2002).
Both models can appropriately handle
datasets that include data from individuals without a recorded time of death, i.e., censored
data. Weigend et al. (2001) was the only example the authors located that used survival
models to identify quantitative trait loci (QTL) in livestock.
Another distributional difficulty that is often encountered in marker/phenotype
association analyses is created by the use of selective genotyping. Selective genotyping is a
method to reduce the costs of an experiment by genotyping only individuals from the
extremes of the phenotypic distribution (Lander and Botstein, 1989; Lebowitz et al., 1987).
Individuals from the phenotypic extremes provide most of the power for a marker-trait
analysis, so power can be maximized for a limited number of genotypes by selective
genotyping. This, however, also causes the distribution of phenotypes to be non-normal.
When the phenotype of all individuals in the experiment is normally distributed, using linear
regression to analyze selective genotyping data results in effects being overestimated (Lander
and Botstein, 1989). Darvasi and Soller (1992) and Ronin et al. (1998) derived an
125
approximation to correct for this bias when the complete phenotypic data follows a normal
distribution.
Maximum likelihood can also be used to appropriately analyze such data
(Lander and Botstein, 1989; Muranty and Goffmet, 1997), if the proper distribution can be
identified. Henshall and Goddard (1999) showed that logistic regression of genotype onto
phenotype also resulted in unbiased estimates of effects; however, the original phenotype
must be normally distributed for this method as well.
McElroy et al. (submitted) used the Cox proportional hazards model and linear
regression to identify markers associated with survival in selectively genotyped layer
chickens infected with Marek's disease. Using simulation under the null hypothesis of no
QTL, they found that both linear regression and the Cox proportional hazards model resulted
in valid false positive rates for tests of association. They did not, however, compare the
power of these two models, nor did they evaluate the use of a Weibull model. Therefore, the
objective of this study was to compare the validity of false positive rates and the power of
employing the Weibull model, the Cox proportional hazards model, and the linear regression
model to analyze marker or QTL associations under full and selective genotyping of survival
data.
2. MATERIALS AND METHODS
2.1. Data Simulation
Survival data were simulated using the methods of Vincent Ducrocq (personal
communication) for a backcross under the null hypothesis of no QTL, or the alternative
hypothesis that a QTL affecting survival resides at the marker under analysis. To mimic real
data, the simulated data were generated to represent actual data from the backcross
population of Marek's disease-challenged layer chickens described by McElroy et al.
(submitted). Survival times in the real experimental population over the recording time
ranged from 33 to 140 d (Figure 1). Survival times followed a right-skewed distribution with
a mean of 65.5 d, a median of 59.0 d, and a standard deviation of 23.9 d. Twenty-eight
individuals (4.3%) survived to the end of the study and were considered censored in the
analyses. First, a survival function (S(t)xM) was estimated from the real data using the nonparametric Kaplan-Meier estimator (Kaplan and Meier, 1958):
126
S(?)km ~ ]
for ?(k) ^ t < ;(k+i), where j is the rank of a particular day (t) among all chronologically ordered
days in which death occurred, k is the rank of the day in which the survival function is being
evaluated, n, is the number of animals still alive on day j, and dj is the number of animals
dying on day j. Censored individuals, i.e. individuals surviving to day 140 (n=28 or 4.3%),
were considered to have died at day 140 for this estimation. This resulted in an estimate of
S(î)km for
each time of death (t).
Survival data for 700 backcross animals were simulated by generating a survival
probability, Soft),, for each individual i based on a proportional hazards model:
exp(/Zsxi)
S(t )i = S 0 (t)\
where S(t)j is a draw from a random uniform [0,1], /?s is the simulated QTL allele substitution
effect (Falconer and Mackay, 1996), and Xj indicates the QTL allele received from the F1
parent, which was either zero or one as drawn from a random binomial distribution with a
50% chance of getting either allele. The value of So(t)\ obtained for an animal was then
cross-referenced with the estimated Kaplan-Meier function obtained from the real data to get
the corresponding time of death. To simulate selective genotyping, only the 20% short (140
animals) and 20% long (140 animals) surviving individuals were considered for analysis and
their genotype for the marker (QTL) was assumed known.
The simulation was performed for each of four allele substitution effect levels of the
QTL: Q = 0, 0.1, 0.2, or 0.3. The largest effect was chosen to be 0.3 because all analyses had
high power for this effect. Two additional censoring scenarios (0 and 20%) were also
considered, in addition to the 4.3% censoring that was present in the real data. For the no
censoring scenario, the Kaplan-Meier function from the real data set was extended past day
140 by assuming the same risk of dying for each day past 140 as between day 101 and 139.
For 20% censoring, day 77 was considered as the last day of the study.
2.2. Models of Analysis
Simulated data were analyzed for associations of marker and phenotype using the
Weibull model, the Cox proportional hazards model, and the linear regression model.
127
The linear regression model was:
Si =^lrP(Q)> + Sj,
where S\ is the survival time of animal i, in days; /?lr is the increase in the mean survival time
associated with inheriting allele 1 versus allele 0; p(Q), is the probability of inheriting one
versus the other QTL allele for animal i; and Ej is the residual for animal i.
The Cox model used, following Smith (2002), was:
where S(t;n)\ is the probability that animal i survived at least until time t, So(t) is the baseline
survival, S0(t) = e"' h0(tJd\ where h0(t) is the baseline hazard, r]j = /?pH p(Q),, where /SPH is the
allele substitution effect of the QTL on the natural log of the hazard ratio for the Cox model,
and p(Q)i is as defined previously.
The Weibull model (Weibull, 1951) used was:
where S(t)l is the probability that animal i survived at least until time t, 0 is the shape
parameter of the Weibull distribution, a is the scale parameter of the Weibull distribution, Pw
is the allele substitution effect of the QTL on the natural log of the hazard ratio for the
Weibull model, and p(Q)i is as defined previously. For all models, p(Q)j = 0 or 1 for
genotyped individuals, depending on which QTL allele they received.
Selective genotyping data were analyzed in two ways: i) including only genotypic and
phenotypic data from genotyped individuals (SG) in the analysis, and ii) also including
phenotypic data from the ungenotyped individuals (SGI) in the analysis.
For the SGI
scenario, p(Q), = 0.5 for ungenotyped individuals, because with unknown genotypes, these
individuals were equally likely to have received either allele. These data were then analyzed
with the three models of interest. Animals dying on day 140 were considered censored for
the Weibull and Cox models, and to have died on day 140 for the linear regression model.
False positive rates (assuming no QTL effect) and power of the analyses were
calculated based on 1000 replicates for each allelic effect level for each of two significance
levels (p < 0.05 and 0.01) for the Full (all individuals genotyped), SG, and SGI genotyping
scenarios. A two-tailed binomial test was used to determine if false positive rates differed
128
significantly from expected (i.e. the a level) under the null hypothesis and a two-tailed
Fisher's exact test was used to determine if the power of the alternate models differed
significantly.
Empirical thresholds were derived and used to compute power for the Weibull model
for all genotyping scenarios and for the linear regression model for the SGI scenario for each
level of censoring because they resulted in inflated false positive rates. Empirical thresholds
were obtained from 10,000 replicates of data simulated with no QTL effect.
Average allele substitution effect estimates and their variances across each set of
1000 replicates were also calculated. Estimated effects from linear regression describe the
average difference of survival in days between the two QTL genotypes, and estimates from
the Cox and Weibull models describe the allele substitution effect on the natural log of the
hazard ratio. To make the effect estimates from all three models directly comparable, the
mean difference of survival in days between groups inheriting one versus the other QTL
allele was estimated for the Cox and Weibull models for each replicate. For the Weibull
model, estimates in terms of the hazard ratio can be directly converted into estimates in days
by:
E(T) =r(l/a + 1) (0 ep(Q)iAv)-i/a
where E(T) is the expected mean days of survival, F is a gamma function, and p(Q)i, a, pw,
and 0 are as described previously. The mean difference in days was calculated by finding the
difference between E(T) for p(Q), = 0 and 1. To obtain estimates in days from the Cox
proportional hazards model, the difference in the means of the survival functions for p(Q), =
0 and 1 was calculated across all times using the Cox proportional hazards estimates for /?phEstimates from the three models were compared based on their correlations, magnitudes, and
standard errors. The magnitudes of effect estimates from the three models were compared by
using a t-test assuming unequal variances.
129
3. RESULTS
3.1. False Positive Rates
In Table I is shown the false positive rates for the Weibull, Cox, and linear regression
models for a = 0.05. False positive rates for the Cox model were not significantly (p < 0.05)
different from expected for any genotyping scenario.
False positive rates for linear
regression were not significantly (p < 0.05) different from expected in the full and SG cases,
but were significantly (p < 0.05) different for every censoring level when SGI was used. The
Weibull model had a significantly (p < 0.05) higher number of false positives than expected
for every censoring and genotyping scenario. Differences from expected for the three models
with a = 0.01 were similar in significance to those when a - 0.05.
3.2. Power Among Genotyping Scenarios
Although not always significant, definite trends in differences of power were apparent
when comparing genotyping scenarios (Table II). For all three models, across QTL effects
and censoring, the Full genotyping scenario had more power than SG and SGI, and SGI had
more power than SG. The results for a = 0.05 is shown in Table II, but trends were similar
for a = 0.01.
3.3. Power Between Models
Although not always significant, for the Full case (a = 0.05), the Cox model was
more powerful than both linear regression and the Weibull model, and the linear regression
model was more powerful than the Weibull model, except with 20% censoring (Table III).
The linear regression model had lower power when censoring was high for the Full scenario,
but performed similarly to the Cox and Weibull models for the SG and SGI scenarios with
20% censoring (Table III).
For the SG low-censoring (0 and 0.04%) scenarios, linear
regression tended to have more power than the Cox and Weibull models (Table III). Power
was similar between the Weibull and Cox models for the SG scenario (Table III). For the
SGI scenario, the Cox model tended to have higher power than the Weibull model across
different censoring levels, and slightly lower power than the linear regression model with low
censoring (Table III).
Linear regression tended to have higher power than the Weibull
130
model with SGI scenario and 0 and 0.04% censoring (Table III). Results for the power
between models presented in Table III are for a = 0.05, but trends were similar for a = 0.01.
3.4. Effect Estimates
For the Full genotyping scenario, estimates of allele substitution effects on the
ln(hazard ratio) scale from the CPH model were never significantly (p > 0.2) different from
the simulated values (0.1, 0.2, or 0.3) (data not shown). For the Cox model, the correlation
of estimates on the ln(hazard ratio) scale with converted estimates in days (within censoring
level) was less than -0.99 for all scenarios, so the Cox estimates in days were adequate to
compare estimates in days from all models with the expected estimates that were in terms of
the ln(hazard ratio). Effect estimates in days from the Cox model for the SG and SGI
scenarios with non-zero effects simulated were significantly (p < 0.05) larger than expected
(i.e. 0.1, 0.2, and 0.3), but effects were less overestimated for the SGI scenario.
Within a censoring and genotyping scenario, mean estimates from all three models
were highly correlated (greater than 0.99). For a given model, correlations of mean effect
estimates (in days) from different genotyping scenarios were also greater than 0.99. The rest
of the results will only consider simulated effects > 0. For the Full genotyping scenario,
mean estimates from the Cox and linear regression models did not differ significantly (p >
0.05), except for the 0.2 censoring, effect = 0.3, case (Table IV). For the Full scenario, mean
estimates from the Weibull model were significantly (p < 0.05) greater than mean estimates
from the Cox model.
For the SG and SGI scenarios, mean estimates from the linear
regression model were significantly (p < 0.05) larger than mean estimates from the Cox
model for all cases.
For the Full and SG scenarios, mean Weibull estimates were
significantly (p < 0.05) larger than mean estimates from linear regression, with the exception
of the SG, no-censoring scenario. For the SGI scenario with low censoring (0 and 0.04),
mean estimates from linear regression were significantly (p < 0.05) larger than Weibull
estimates. However, for the SGI high-censoring (0.2) scenario, mean estimates from the
Weibull model were significantly (p < 0.05) larger than mean estimates from linear
regression.
131
For all three models, standard errors of the means of the effect estimates were less for
the Full scenario than for SG and SGI (Table IV). For the Cox and Weibull models, standard
errors were greater for the SG scenario than for the SGI scenario but they did not differ for
the linear regression model.
4. DISCUSSION
4.1. False Positive Rates
Significance tests under the Cox model were robust to the distribution of phenotypes
that were included in the analysis. The Weibull model had a significantly (P < 0.05) inflated
false positive rates for all scenarios, indicating that the data did not fit a Weibull distribution.
This was confirmed by significance (P < 0.01) of a lack of fit test of the survival distribution
from the real population to the Weibull distribution using the Cramer-von Mises W test
(Phillips, 1972). Tests based on linear regression were robust to the lack of normality in the
Full and SG scenarios, but had an inflated false positive rate for the SGI scenario. Further
research is needed to understand why the addition of the phenotypes of the ungenotyped
individuals causes linear regression to have an inflated false positive rate.
4.2. Power Differences Between Models
Differences in power between models varied between genotyping scenarios and
censoring levels. In the Full scenario, the Cox model had higher power than the linear
regression and Weibull models. This is likely because the Cox model is more appropriate for
this distribution. The linear regression model had less power than the Cox and Weibull
models when censoring was 20%. This was expected because the censored individuals were
considered as dying on the last day of the study in the linear regression model, and therefore
groups the extreme long-surviving individuals (which are more likely to have the favorable
genotype) with less extreme long-surviving individuals that would be expected to have a
lower frequency of the favorable genotype. This situation reduces the mean phenotypic
difference between individuals with the favorable allele and individuals with the unfavorable
allele, and thereby reduces power.
132
4.3. Power Differences Between Genotyping Scenarios
For all models and scenarios, power was somewhat (usually less than 10%) lower for
the SG and SGI scenarios than for the Full scenario. This is expected because for a normally
distributed phenotype, it has been shown that, for a fixed number of genotyped individuals,
SG is more powerful than Full (Lebowitz et. al, 1987; Lander and Botstein, 1989; Darvasi
and Soller, 1992), but when genotyping is not constrained, Full is more powerful than SG.
This is because the SG scenario excludes some of the information that is used in the Full
scenario, as does the SGI scenario. For the simulated survival data, SGI had slightly (usually
less than 5%) higher power than SG for all models and scenarios. To determine which
individuals are the most extreme for selective genotyping, phenotypes from the entire
population must be obtained, so phenotypes of ungenotyped individuals should be available
for inclusion in a selective genotyping analysis to increase power.
4.4. Effect Estimation
For the Full scenario, effects estimated with the Cox model were not significantly
different from the simulated effects.
This is likely because the survival data were
appropriately distributed for the Cox model. The linear regression model also estimated
effects accurately in the Full scenario, indicating robustness of linear regression to deviations
from normality.
The Weibull model overestimated effects in the Full scenario by
approximately 50%, likely due to lack of fit of the data to a Weibull distribution.
For the SG scenario, it is well known that linear regression overestimates effects
when the phenotypes are normally distributed, and methods have been devised to correct for
these biases (Darvasi and Soller, 1992; Ronin et al., 1998). The Cox and Weibull models
also overestimated effects in the SG scenario by approximately 100%, although the Cox
model did so significantly (p < 0.05) less than the Weibull and linear regression models. For
the SGI scenario, linear regression overestimated the true effects slightly more than for the
SG scenario, and the Cox and Weibull models overestimated the true effects slightly less.
Skewness of the distribution may cause estimated effects from linear regression to increase
when including the ungenotyped individuals with a p(Q)i = 0.5, since the mean phenotype of
133
these individuals will not fall on the regression line between the mean phenotypes of the two
extremes. With a normally distributed phenotype, the mean of the ungenotyped individuals,
when given the mean genotypic probability (0.5), is expected to fall on the regression line
between the means of the two phenotypically extreme groups, and therefore should not
change the slope of the regression line. Further investigation into the proper statistical
models for including non-genotyped individuals in the analysis of non-normal selectively
genotyped data is needed.
4.5. Conclusions
Generally, the Cox model performed better than the linear regression and Weibull
models across genotyping scenarios and censoring levels, without the need to derive
empirical significance thresholds. This was likely because the distribution of the data met
the assumptions of the Cox model better than that of the other two models. However,
survival models are much more computationally demanding than linear regression and the
effect estimates from survival models are much more difficult to interpret than the effect
estimates from linear regression. This, combined with the fact that the powers of linear
regression and the Cox models were similar, suggests that with little censoring and Full or
SG genotypic scenarios, linear regression can be the model of choice, since false positive
rates for linear regression were valid for these two scenarios. The Weibull model had an
inflated false positive rate for all scenarios and linear regression was had an inflated false
positive rate for the SGI scenario, so significance thresholds had to be determined
empirically, which is also computationally demanding.
In cases where selective genotyping is employed, more power can be obtained by
including phenotypes from non-genotyped individuals in the analysis. These phenotypes
should typically be available since they need to be recorded to determine which individuals
are extreme. The Cox model was the only model that did not require the derivation of
empirical thresholds for SGI, and, therefore, may be the best model to use in this situation.
Further research needs to be done to determine if the Cox model would also be appropriate
for normally distributed traits with the SGI scenario.
134
The research presented herein can be extrapolated to interval mapping.
The
genotypic data were entered into the models as probabilities of marker genotypes. The
genotypic data from an interval mapping analysis are also used to compute probabilities of
genotypes at a specific locus, and therefore the results from the current study are directly
applicable to interval mapping analyses.
5. ACKNOWLEDGEMENTS
Research was partly supported by a USDA National Needs Fellowship in Animal
Biotechnology and by Hatch and State of Iowa funds.
6. BIBLIOGRAPHY
Beaudeau, F., V. Ducrocq, C. Fourichon, and H. Seegers. 1995. Effect of disease on length of
productive life of French Holstein dairy cows assessed by survival analysis. J. Dairy
Sci. 78: 103-117.
Benard, H. J., K. D. Stark, R. S. Morris, D. U. Pfeiffer, and H. Moser. 1999. The 1997-1998
classical swine fever epidemic in the Netherlands—a survival analysis. Prev. Vet.
Med. 42: 235-248.
Darvasi, A., and M. Soller. 1992. Selective genotyping for determination of linkage between
a marker locus and a quantitative trait locus. Theor. Appl. Genet. 85: 353-359.
Ducrocq, V., B. Besbes, and M. Protais. 2000. Genetic improvement of laying hens viability
using survival analysis. Genet. Sel. Evol. 32: 23-40.
Falconer, D. S., and T. F. C. Mackay. 1996. Introduction to quantitative genetics. Fourth
edition. Longman Group Ltd., Essex, England.
Grohn, Y. T., V. Ducrocq, and J. A. Hertl. 1997. Modeling the effect of a disease on culling:
An illustration of the use of time-dependent covariates for survival analysis. J. Dairy
Sci. 80: 1755-1766.
Grohn, Y. T., S. W. Eicker, V. Ducrocq, and J. A. Hertl. 1998. Effect of diseases on the
culling of Holstein dairy cows in New York state. J. Dairy Sci. 81: 966-978.
135
Haley, C. S., and S. A. Knott. 1992. A simple regression method for maping quantitative trait
loci in line crosses using flanking markers. Heredity 69: 315-324.
Haley, C. S., S. A. Knott, and J. M. Elsen. 1994. Mapping quantitative trait loci in crosses
between outbred lines using least squares. Genetics 136: 1195-1207.
Henshall, J. M., and M. E. Goddard. 1999. Multiple-trait mapping of quantitative trait loci
after selective genotyping using logistic regression. Genetics 151: 885-894.
Hirst, W. M., R. D. Murray, W. R. Ward, and N. P. French. 2002. A mixed-effects time-toevent analysis of the relationship between first-lactation lameness and subsequent
lameness in dairy cows in the UK. Prev. Vet. Med. 54: 191-201.
Jansen, R. C., and P. Stam. 1994. High resolution of quantitative traits into multiple loci via
interval mapping. Genetics 136: 1447-1455.
Kaplan, E. L., and P. Meier. 1958. Nonparametric estimation from incomplete observations.
J. Amer. Stat. Assoc. 53: 457-481.
Lander, E. S., and D. Botstein. 1989. Mapping mendelian factors underlying quantitative
traits using RFLP linkage maps. Genetics 121: 185-199.
Lebowitz, R. J., M. Soller, and J. S. Beckmann. 1987. Trait-based analyses for the detection
of linkage between marker loci and quantitative trait loci in crosses between inbred
lines. Theor. Appl. Genet. 73: 556-562.
Maizon, D. O., P. A. Oltenacu, Y. T. Grohn, R. L. Strawderman, and U. Emanuelson. 2004.
Effects of diseases on reproductive performance in Swedish Red and White dairy
cattle. Prev. Vet. Med. 66: 113-126.
Malek, M., J. C. Dekkers, H. K. Lee, T. J. Baas, and M. F. Rothschild. 2001. A molecular
genome scan analysis to identify chromosomal regions influencing economic traits in
the pig. I. Growth and body composition. Mamm. Genome 12: 630-636.
Muranty, H., and B. Goffmet. 1997. Selective genotyping for location and estimation of the
effect of a quantitative trait locus. Biometrics 53: 629-643.
Phillips, D. T. 1972. Applied Goodness of Fit Testing. American Institute of Industrial
Engineers, Inc. Atlanta, GA.
136
Rajala-Schultz, P. J., W. J. A. Saville, G. S. Frazer, and T. E. Wittum. 2001. Association
between milk urea nitrogen and fertility in Ohio dairy cows. J. Dairy Sci. 84: 482489.
Ronin, Y. I., A. B. Korol, and J. I. Weller. 1998. Selective genotyping to detect quantitative
trait loci affecting multiple traits: Interval mapping analysis. Theor. Appl. Genet. 97:
1169-1178.
Roxstrom, A., V. Ducrocq, and E. Strandberg. 2003. Survival analysis of longevity in dairy
cattle on a lactation basis. Genet. Sel. Evol. 35: 305-318.
Smith, P. J. 2002. Analysis of failure and survival data. Chapman & Hall/CRC, Boca Raton,
FL.
Southey, B. R., S. L. Rodriguez-Zas, and K. A. Leymaster. 2001. Survival analysis of lamb
mortality in a terminal sire composite population. J. Anim. Sci. 79: 2298-2306.
Steel, R. G. D., and J. H. Torrie. 1980. Principles and procedures of statistics: A biometrical
approach. Second Edition ed. McGraw-Hill Publishing Company, New York.
van Kaam, J. B., M. A. Groenen, H. Bovenhuis, A. Veenendaal, A. L. Vereijken, and J. A.
Van Arendonk. 1999b. Whole genome scan in chickens for quantitative trait loci
affecting carcass traits. Poult. Sci. 78: 1091-1099.
Weibull, W. 1951. A Statistical Distribution Function of Wide Applicability," J. Appl.
Mech. 18: 293-297.
Weigend, S., S. Matthes, J. Solkner, and S. J. Lamont. 2001. Resistance to Marek's disease
virus in White Leghorn chickens: Effects of avian leukosis virus infection genotype,
reciprocal mating, and major histocompatibility complex. Poult. Sci. 80:1064-1072.
Zeng, Z. B. 1993. Theoretical basis for separation of multiple linked gene effects in mapping
quantitative trait loci. Proc. Natl. Acad. Sci. USA 90: 10972-10976.
Zhou, H., H. Li, and S. J. Lamont. 2003. Genetic markers associated with antibody response
kinetics in adult chickens. Poult. Sci. 82: 699-708.
137
Table I. False positive rates for Cox proportional hazards, linear
regression, and Weibull models for different censoring levels and
genotypic scenarios at a = 0.05 based on 1000 replicates.
False Positive Rate2
Censoring
Genotyping1 CPH3
(%)
Full
0.047
0
Full
0.046
4
Full
0.054
20
SG
0
0.048
0.048
SG
4
SG
0.045
20
SGI
0.045
0
0.047
4
SGI
0.050
20
SGI
LR3
WEI3
0.043
0.049
0.058
0.049
0.050
0.050
0.207*
0.184*
0.157*
0.169*
0.139*
0.079*
0.068*
0.081*
0.068*
0.263*
0.234*
0.163*
'Full = all individuals genotyped and used in analysis; SG = selective genotyping with ungenotyped
individuals excluded from the analysis; SGI = selective genotyping with ungenotyped individuals
included in the analysis.
2 * indicates significant difference (p < 0.05) between observed and expected false positive rates (0.05)
based on a two-tailed binomial test.
3 CPH = Cox proportional hazards; LR = linear regression; WEI = Weibull.
138
Table II. Power of different genotyping scenarios for different censoring levels
and simulated effects using Cox, linear regression, and Weibull models at a =
0.05 based on 1000 replicates.
Difference in Power2
Model1 Censoring
(%)
0
CPH
0
CPH
0
CPH
CPH
4
CPH
4
CPH
4
CPH
20
CPH
20
20
CPH
0
LR
0
LR
0
LR
4
LR
4
LR
4
LR
20
LR
20
LR
20
LR
WEI
0
WEI
0
0
WEI
WEI
4
WEI
4
WEI
4
20
WEI
20
WEI
WEI
20
1 CPH
QTL
Effect
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
0.1
0.2
0.3
M3_SG Full_SGI
0.047* 0.033
0.105* 0.056*
0.052* 0.024*
0.034
0.026
0.122* 0.061*
0.055* 0.012
0.043* 0.035
0.129* 0.077*
0.065* 0.033*
0.039* 0.017
0.080* 0.040
0.035* 0.015
0.006
0.028
0.081* 0.034
0.031* 0.024*
0.033 -0.002
0.066* 0.024
0.048* 0.016
0.029
0.020
0.026
0.022
0.022* 0.019
0.022
0.025
0.069* 0.050*
0.036* 0.022*
0.050* 0.036*
0.128* 0.080*
0.069* 0.045*
SG-SGI
-0.014
-0.049*
-0.028*
-0.008
-0.061*
-0.043*
-0.008
-0.052*
-0.032*
-0.022
-0.040
-0.020
-0.022
-0.047*
-0.007
-0.035*
-0.042
-0.032*
-0.009
-0.004
-0.003
-0.003
-0.019
-0.014
-0.014
-0.048*
-0.024
= Cox proportional hazards; LR = linear regression; WEI = Weibull.
* indicates significant difference (p < 0.05) between genotyping scenarios from a twotailed Fisher's exact test.
3 Full = all individuals genotyped and used in analysis; SG = selective genotyping with
ungenotyped individuals excluded from the analysis; SGI = selective genotyping with
ungenotyped individuals included in the analysis.
2
139
Table III. Power of Cox proportional hazards, linear regression, and
Weibull models for different censoring levels and genotypic scenarios
at a = 0.05 based on 1000 replicates.
Censoring
0
1
Effect
0.1
2
CPHi Fuli
LR Full
0.262
0.253
0.245
0.639
WEI Full CPH SG LR SG WEI SG CPH SGI LR SGI WEI SGI
0.216
0.229
0.236
0.225
0.654
0.645
0.688ab
0.694"
0.649b
0.927
0.936
0.931
0.955"
0.956"
0.934b
0.229
0.216
0.211
0.204
0.224
0.233
0.207
0.724"
0.679"
0.612
0.643
0.610
0.673"
0.690"
0.629"
0.976"
0.976"
0.955b
0.921" 0.945b
0.919"
0.964"
0.952ab
0.933b
0.1
0.211
0.200
0.215
0.168
0.167
0.165
0.176
0.202
0.179
20
0.2
0.667"
0.591b
0.663"
0.538
0.525
0.535
0.590
0.567
0.583
20
0.3
0.951"
0.923*"
0.955"
0.886
0.875
0.886
0.918
0.907
0.910
0
0.2
0.744"
0.734"
0.671b
0
0.3
0.979"
0.971"
0.953b
4
0.1
0.250
0.239
4
0.2
0.734"
4
0.3
20
0.215
0.214
CPH = Cox proportional hazards; LR = linear regression; WEI = Weibull.
Full = all individuals genotyped and used in analysis; SG = selective genotyping with ungenotyped
individuals excluded from the analysis; SGI - selective genotyping with ungenotyped individuals included
in the analysis.
3 Numbers with different letter subscripts within a row and genotyping scenario (Full, SG, SGI) are
significantly different at P < 0.05. No letter indicates no significant differences (P < 0.05) within a row and
genotyping scenario.
2
Table IV. Average estimates of effects (standard errors) in days from 1000 replicates for Cox proportional
hazards, linear regression, and Weibull models for different censoring levels and different simulated QTL effects.
Censoring
Simulated
QTL Effect
CPH1 Full2
CPH SG
CPH SGI
LR Full
LR SG
LR SGI
WEI Full
WEI SG
WEI SGI
0.03(0.06)a3
0.05(0.14/
0.05(0.11)8
0.05(0.06)'
0.03(0.15/
0.08(0.15)*
-0.02(0.09)'
0.02(0.15)"
-0.04(0.13)*
0
0.1
-2.424a
-4.78d
-3.92*
-2.46"
-5.19e
-5.46"
-3.21"
-5.03d e
-4.65'
0
0.2
-4.64a
-9.24'
-7.53»
-4.60'
-9.89'
-10.38"
-6.23"
-9.77e
-9.08'
-6.66"
-13.34d
-6.66'
-14.23e
-14.93"
-9.08"
-14.25e
-13.35'
0.07(0.06)a
0.16(0.13)"
0.16(0.13)*
-5.08"
0.04
0.04
0.04
0.04
0.3
0
0.1
nr
0
©
0
00
0
0.01(0.06)'
-0.02(0.13/
-2.20'
-4.30d
0.01(0.10):
-3.65*
-2.27"
-4.80e
-4.23a
-9.02e
-4.57'
-9.48"
-5.98"
-10.69f
-8.83'
-13.95"
-8.67"
-15.23f
-12.88'
-4.23'
-8.32""
0.3
-6.1la
-12.08d
-10.09s
-6.22'
-13.27e
0.02(0.03)'
0.02(0.05)d
0.03(0.06)'
0.02(0.03)'
0.05(0.07)"
0.05(0.07)*
0.02(0.04)a
0.13(0.20)"
0.04*
-1.06'
-2.17e
-2.17"
-1.46"
-5.83f
-2.48'
-2.06"
-4.34e
-4.35"
-2.89"
-10.46f
-4.83'
-3.15"
-6.59e
-6.63"
-4.29e
-14.14f
-7.09'
0.20
0.1
-0.99'
-1.79d
-1.90s
0.2
-2.01a
-3.88'
-3.79s
-3.03"
-5.86d
1 CPH
-5.67f
0.2
0
0.20
-3.10"
-6.99B
0.20
0.20
-0.03(0.08)' -0.04(0.17)" -0.06(0.13)*
0.3
-5.66*
= Cox proportional hazards; LR = linear regression; WEI = Weibull
= all individuals genotyped and used in analysis; SG = selective genotyping with ungenotyped individuals excluded from the analysis; SGI =
selective genotyping with ungenotyped individuals included in the analysis.
3 Different letters indicate significant differences (p < 0.05) in average effect estimates between models within censoring and effect: a, b, and c for
the Full scenario; c, d, and e for the SG scenario; and g, h, and i for the SGI scenario.
4 Standard errors were equal across effects within model and genotyping scenario.
2 Full
Figure 1. Distribution of survival times from the real population of chickens
(Data reported in McElroy et al. (submitted)).
125
JS
a
s
100
1
•a
5
£
75
1
I
50
25-L
T~
~r~
40
50
i
60
70
r~
80
90
100
Survival Time (days)
T
I
110
120
130
140
142
CHAPTER 5
GENERAL CONCLUSIONS AND DISCUSSION
5.1 CONCLUSIONS
Detected QTL
As presented in this dissertation, crosses of commercial chicken lines were successfully
utilized to identify QTL affecting economically valuable traits.
Chapter 2 detailed
identification of QTL affecting white meat percent and other growth and composition traits
in meat-type chickens (broilers). With multiple statistical models, 68 QTL associated with
various growth, carcass, and percentage traits, including white meat percent, were identified.
The QTL were segregating both between and within lines. Identification of within-line QTL
illustrates that, despite many years of selection for growth traits, genetic variation still exists
within the commercial lines. Therefore, continued improvement of growth traits is possible
through within line selection of chickens. The chromosomes with the most QTL were Gga 3
and Gga 5. White meat percent QTL were identified on Gga 2, 3, 5, and 6. The largest of
these QTL were on Gga 2 and 6, and explained 6.8% and 4% of the phenotypic variance,
respectively. This study also identified QTL with parent of origin effects, providing possible
evidence for gametic imprinting in chickens. Use of mendelian and imprinted QTL in a
selection scheme will be discussed below.
Chapter 3 described identification of QTL associated with survival to Marek's disease in a
backcross of commercial layer lines. The identified QTL were on Gga 2, 4, 5, 6, 8, 15, and
Z. The estimated effects in days of some of these QTL were quite large (20 days longer
survival for one QTL). However, the research in Chapter 4 showed that the linear regression
model overestimates effect size in selectively genotyped survival data, thus estimates from
linear regression presented in Chapter 3 were likely overestimated.
Models for QTL Detection
Several statistical models were used for the successful identification of QTL in Chapters 2
and 3. Half-sib, line-cross, combined, and parent-of-origin linear regression models were
143
utilized in Chapter 2; and linear regression and Cox proportional hazards models were used
in Chapter 3. Chapter 4 described the comparison, through simulation, of linear regression,
Cox proportional hazards, and Weibull models to detect QTL associations with a nonnormally distributed, selectively genotyped phenotype. Results from Chapter 2 showed that
including multiple QTL segregation and expression models allowed identification of QTL
that would have been missed had only a single model been used. The research in Chapter 3
showed that, for survival data, the linear regression and Cox proportional hazards models had
nearly equivalent power to identify QTL, and both models had valid false positive rates.
Estimates from linear regression are, however, much easier to interpret, and therefore, linear
regression was the preferred model for analysis of the survival data in Chapter 3.
The research presented in Chapter 4 shows that linear regression and Cox proportional
hazard models had valid false positive rates with both full and selective genotyping (when
non-genotyped individuals were excluded), indicating the robustness of linear regression to
deviations from normality.
The research in Chapter 4 also showed that with full and
selective genotyping (when non-genotyped individuals were excluded), the Cox proportional
hazards and linear regression models had similar power to detect QTL, with the exception of
the full genotyping scenario with high censoring. In this Chapter it was also shown that the
Weibull model was inappropriate for the particular phenotypic distribution that was
simulated, indicating that the distribution of data must be carefully considered when using a
Weibull model. The linear regression and Cox models accurately estimated QTL effects with
full genotyping, but overestimated effects with selective genotyping.
Full genotyping
yielded more power than selective genotyping, which was as expected since full genotyping
uses more information than selective genotyping. Given these results, the linear regression
model was preferred for analyzing full genotyping and selective genotyping (when nongenotyped individuals were excluded) datasets when censoring was low because linear
regression was more powerful than the Weibull model, had similar power to the Cox model,
was easier to implement, was less computationally demanding than the survival models, and
resulted in effect estimates that were easier to interpret than those from the survival models.
With full genotyping, censoring had a greater negative effect on power of the linear
144
regression model than on power of the survival models, and therefore the Cox proportional
hazards model was preferred when the amount of censoring is high.
All models used in the research in Chapter 4 overestimated effects with selective genotyping.
Darvasi and Soller (1992) and Muranty and Goffinet (1997) suggested methods to correct for
these biases with normally distributed selective genotyping data using ANOVA and
maximum likelihood. It would be possible to extend the maximum likelihood method to
non-normally distributed traits if the distribution of the phenotype can be specified. The
research in Chapter 4 showed that power could be gained by including ungenotyped
individuals in the selective genotyping analysis with a genotypic probability of 0.5, with the
caveat that including this caused linear regression to have an inflated false positive rate.
Because linear regression was inappropriate for analyzing selective genotyping data while
including phenotypes from the non-genotyped individuals, the Cox proportional hazards
model was preferred.
5.2 DISCUSSION
In summary, this dissertation describes the successful mapping of QTL for economic traits in
commercial chickens. Also, the research described in this thesis illustrates the need to
consider several aspects when designing and analysing an experiment. The researcher must
consider the adequacy of the structure and size of the population for providing valid answers
to the questions in which the researcher is interested. Determining the appropriate statistical
methods for analysis of the population is also not trivial. Different statistical models are
needed based on the phenotypic distribution (normal vs. non-normal) of the data, fixation of
QTL in the parental lines (F2 vs. half-sib models), population structure (F2 vs. backcross),
and mode of expression of the QTL (mendelian vs. imprinting models).
Identified QTL
For traits that are hard to measure, expensive to measure, lowly heritable, or for which an
individual needs to be killed to measure, marker assisted selection is useful (Dekkers and
Hospital, 2002). To perform marker assisted selection, genes or regions of the genome
145
harboring these genes must be identified. Therefore, mapping of QTL affecting economic
traits in commercial chickens was the focus of this thesis. All four papers presented herein
describe QTL mapping methods, and three of the papers include applications of these
methods to map QTL in commercial chickens.
Identification of QTL in commercial lines is important because QTL identified in
experimental crosses may not always be directly applicable to commercial lines (de Koning
et al., 2004). However, many QTL identified, even in commercial lines, will be specific only
to the line or population that was used to map the QTL. This is because in other lines, the
QTL may be fixed for a particular allele, background genes may epistatically alter the effect
of the QTL, or environmental differences may alter the effect of the QTL. Therefore, unless
a QTL is identified using the population in which the selection on that QTL will be
implemented, different lines will still need to be screened to determine if the QTL is
segregating within them before using the QTL for selection.
Seventeen putative QTL affecting Marek's disease were identified in the study reported in
Chapter 3. The largest of these QTL had an effect of 20 days survival time. These estimates
are likely biased upward because of the use of selective genotyping or because the effects of
significant QTL are reported (Beavis, 1998). In Chapter 4, simulated QTL having an effect
of 6 days of survival was estimated to have an effect of 13 days when selective genotyping
was used. The simulated phenotypic data in Chapter 4 was modeled after the phenotypic
data in Chapter 3. Therefore, ignoring the so called Beavis effect, it is likely that the largest
QTL identified in Chapter 3 had an effect larger than 6 days of survival, and, following this
logic, the approximate range of QTL effects identified in Chapter 3 would be from 2.5 to > 6
days of survival.
These QTL regions can be useful in selection programs designed to reduce losses from
Marek's disease (see below for methods of utilizing QTL in selection programs), although
how they should be used is debatable. Birds in the study with long survival after exposure to
Marek's disease may have survived because they were resistant to the disease, and therefore
146
overcame infection and the associated symptoms. Alternatively, birds in the study with long
survival may have survived because they were tolerant of the disease and continued showing
symptoms of disease throughout their lifetimes. The former situation would allow birds to
return to normal levels of production after clearing the disease, whereas the latter situation
would result in chronically sick birds that would have low production and could infect other
birds throughout their lifetimes. It would be beneficial to select birds that are resistant to
Marek's disease because mortality would be decreased with little effect on production.
However, it would also be beneficial to select birds that are less tolerant to Marek's disease
to remove birds from the population early that would otherwise have lowered production and
increased risk of dying from Marek's disease during the production period.
The QTL
identified in the research presented in Chapter 3 may affect survival for either or both of
these reasons. To determine whether these QTL affect Marek's disease survival because of
resistance or tolerance, other traits would need to be examined in the birds, such as tumor
presence and production. None of the markers associated with Marek's disease survival
reported in Chapter 3 are significantly (p < 0.05) associated with presence or absence of
tumors. Therefore, the QTL identified affecting for Marek's disease survival reported in
Chapter 3 likely affected tolerance instead of resistance.
The analyses described in Chapters 3 and 4 were single marker analyses, meaning that
multiple markers were not used simultaneously to separate QTL effect from QTL position
(effect and position are confounded in single marker analyses). Haley and Knott (1992)
described interval mapping (which uses multiple markers at the same time to estimate both
QTL position and effect) using linear regression, and this method was used in the research
reported in Chapter 2. The Cox proportional hazards model can also be easily extended to
interval mapping by replacing the single marker line origin probabilities that were used in
Chapter 3 by line origin probabilities for putative QTL positions along a chromosome, which
can be derived from flanking marker information. This model can then be fit for all positions
across a chromosome to determine the most likely location of a QTL.
147
The populations used to identify QTL in Chapters 2 (F2) and 3 (backcross) were crosses
designed to generate wide linkage disequilibrium within the QTL mapping generation.
Because the linkage disequilibrium in these populations extended across wide regions of the
genome, only single marker analysis was used in Chapter 3, and confidence intervals for
QTL position from QTL mapping experiments are typically quite large (> 40 cM: eg. Kim et
al., 2002; Bennejwitz et al., 2002), the exact locations of the QTL identified in Chapters 2
and 3 are unknown and many of the QTL identified may not have been very close to a
marker. As will be discussed below, markers tightly linked to the QTL may be needed to
utilize the QTL for selection.
The lines crossed to generate the mapping populations were chosen because they differed in
white meat percent (Chapter 2) or Marek's disease resistance (Chapter 3). Therefore, QTL
affecting white meat percent are expected to have different alleles or different frequencies of
alleles between the two lines, which was shown by the identification of line cross QTL
affecting white meat percent. Large effect QTL were also segregating within lines. The
white meat percent QTL with the largest estimated effect (explaining half of the phenotypic
standard deviation) showed both within and between line segregation, and the QTL with the
second largest effect was only identified as segregating within the lines. The apparent
segregation of QTL both within and between lines can be a result of the same two QTL
alleles being present in both parental lines but at different frequencies between the parental
lines, the presence of more than two QTL alleles across the parental lines, or multiple tightly
linked QTL that segregate differentially between the parental lines (e.g., one QTL fixed for
alternate alleles between the parental lines and the other QTL segregating within the lines).
In the study reported in Chapter 2, an F2 cross was used to map QTL, whereas a backcross
was reported in Chapter 3. The F2 experimental design is more powerful than the backcross
design, and additive and dominance QTL effects can be estimated from an F2 design,
whereas only the additive minus the dominance QTL effect can be estimated in a backcross
design (Soller et al. 1976). Therefore, the F2 design is preferred over the backcross design
for the anonymous marker approach of detecting QTL.
148
Parent of Origin QTL
Several QTL displaying parent of origin effects were identified in Chapter 2.
Further
analyses, or verification in other populations, is needed to determine if these QTL are truly
imprinted QTL. These parent of origin QTL may be a result of imprinted genes, or they may
be a result of alternate within-line QTL alleles being inherited differentially between the F1
males and females by chance (Thomsen et al., 2004). Small numbers of F1 parents of one or
both sexes (such as the 7 F1 sires used in this study) facilitate this chance differential
inheritance.
In Chapter 2, QTL were identified at 220 cM on Gga 3 showing paternal
expression and affecting several growth and carcass traits. Tuiskula-Haavisto et al. (2004)
also identified a QTL in this region showing paternal expression affecting egg weight, which
is additional evidence that this is truly an imprinted QTL.
Some authors propose that the mechanism for imprinting evolved for the purpose of gene
silencing due to differing optimal strategies of each parent (mother and father) for resource
allocation of the mother's resources to the offspring of polyandrous viviparous animals, i.e.
the conflict hypothesis (Moore and Haig, 1991).
This mechanism is proposed to have
evolved after the evolutionary split between birds and mammals (Yokomine et al., 2005).
Even though female birds expend some resources for their offspring (eg., body heat,
protection, feeding), the amount of resources expended by avian mothers for postzygotic
offspring that is affected by genes expressed in the offspring are likely small compared to
that of mammals. Also, if the mechanism for imprinting in mammals occurred after the
evolutionary split between birds and mammals, birds and mammals would not share a
homologous imprinting mechanism. However, if imprinting is important for producing a
larger number of viable offspring, then birds may still have developed a mechanism for
imprinting analogous to that of mammals.
Pardo-Manuel de Villena et al. (2000)
hypothesized that imprinting may have evolved because of the need to distinguish between
homologous chromosomes (one from each parent) and sister chromosomes (exact copies)
during meiosis and mitosis, and that gene silencing is just a side effect of this mechanism.
Following this hypothesis, all sexually reproducing animals would need the imprinting
149
mechanism, or some equivalent mechanism, to distinguish between homologous and sister
chromosomes.
Statistical Models
A common theme throughout this thesis was the use of multiple statistical models for QTL
detection. Chapter 2 described the identification of QTL for traits that are approximately
normally distributed, Chapter 3 described QTL detection with non-normally distributed
phenotypes, and Chapter 4 described the comparison of statistical models, through
simulation, to detect QTL with non-normally distributed phenotypes. Using multiple models
that assume different segregation and expression patterns allowed detection of QTL in
Chapter 2 that would otherwise have been missed. The research presented in Chapter 4, in
particular, demonstrated the importance of using proper statistical models to analyze
phenotypic data with a particular distribution (survival distribution), as well as a subset of the
overall distribution (selective genotyping) for QTL detection. This work showed that use of
the Weibull model, which is common model for survival data, for the data of Chapter 3,
would have resulted in a greater than expected percentage of identified QTL that were not
true QTL (false positives).
The research in Chapter 4 also shows that including the
phenotypes of the non-genotyped individuals in the analyses increased power but also
increased the false positive rate for linear regression. Further investigation or utilization of
these extra spurious QTL would result in wasted costs and resources.
The statistical methods used in the research presented herein fall into two main categories:
linear regression and survival models.
However, several additional methods have been
proposed and utilized for QTL mapping.
Maximum likelihood was the first
method
described for QTL interval mapping (Lander and Botstein, 1989). In most instances, linear
regression and maximum likelihood yield similar results (Haley and Knott, 1992), but
maximum likelihood is much more computationally demanding than linear regression. The
maximum likelihood method can handle missing data, such as in the case of the selective
genotyping data in Chapter 3. However, the distribution of the phenotypes would still need
to be specified, and a function describing the specific distribution of the phenotypes in
150
Chapter 3 is unknown. Uimari et al. (1996) presented a Bayesian method for mapping QTL
using multiple markers. Rather than determining the probability of the null hypothesis being
false, as in frequentist approaches, Bayesian methods result in a distribution describing the
probability that a parameter is equal to a specific value.
Bayesian analysis has several
desirable properties, such as taking into account the uncertainties associated with marker
genotypes and with the number of QTL on a chromosome, but is very computationally
demanding and difficult to implement (Hoeschele et al., 1997).
De Koning et al. (2003) compared a variance component method with a half-sib linear
regression analysis, which followed transmission of QTL alleles from the sire, to map QTL
using commercial pig lines with a general population structure. The linear regression method
detected more QTL than the variance component method but the latter did detect some QTL
that were not detected by linear regression. To explain the differences between the two
analyses, a subset of QTL were analyzed with half-sib linear regressing following the
transmission of alleles from the dams. The discrepancies were due to allelic states of the
QTL in the parents. For example, when the linear regression model detected QTL that were
not detected by the variance component method, only one or two sires used in the analysis
were heterozygous for the QTL, or the QTL effects were not observed in the dams. Because
the variance component method jointly utilized information from all sires and dams, QTL
effects such as these were "diluted" in the variance component analysis, and, therefore, were
not detected. In the cases where the variance component analysis identified QTL that were
not identified with linear regression, most of the QTL were detected in the half-sib analysis
of the dams. De Koning et al. (2002) found good agreement between linear regression and
variance component methods when full-sib family sizes were small. Because the dams of the
backcross individuals were unknown in the research presented in Chapter 3, the full-sib
family structure in the backcross was unknown. Therefore, it is likely that little would be
gained by analyzing the data with variance component methods. The research presented in
Chapter 2 utilized both half-sib and line cross models. Based on the work of de Koning et al.
(2003), the half-sib model would have detected QTL that would have been missed by the
variance component method and the line cross model would have identified QTL that the
151
half-sib model would have missed. Therefore, for the population used in Chapter 2, there is
likely little to be gained by using the variance component method for detecting QTL.
The minimum distance method (Wolfowitz, 1957), in which estimates of parameters are
chosen to minimize the distance between the observed data and the fitted model, for mapping
QTL was described by Perez-Enciso and Toro (1999) and compared to maximum likelihood.
They found that when the phenotypic distribution was the one assumed for the maximum
likelihood analysis, maximum likelihood performed better than the minimum distance
method with both full and selective genotyping. However, when outliers were simulated in
the data, the minimum distance method performed better than maximum likelihood with full
and selective genotyping. A published comparison between the minimum distance method
and other methods, such as linear regression and survival analysis, could not be identified,
and therefore a recommendation regarding the use of the minimum distance method with the
data from Chapters 2 and 3 cannot be made.
Meuwissen and Goddard (2000) proposed a linkage disequilibrium method to fine map QTL
using historical recombinants in 5 - 20 cM chromosomal regions. For this method, several
markers are genotyped in regions that have previously been identified from QTL studies such
as those presented in Chapters 2 and 3 of this thesis. Identity by descent (IBD) of these
regions is determined by the extent of marker allele sharing between individuals. Estimates
for the haplotype effect variance and residual variance are then obtained for each segment,
defined by the pair of markers flanking the region, by maximum likelihood, and the segment
with the highest likelihood is the most likely location of the QTL. With the distances
between the markers used in Chapters 2 and 3, this method would not be recommended.
However, in further research to fine map QTL identified in Chapter 2, this method could be
useful, with the addition of more markers.
The extent of the between-line linkage
disequilibrium would, however, hinder fine mapping of between line QTL in this population.
However, if the population was extended by outcrossing F2 individuals or by creating
advanced intercross lines, thereby decreasing the extent of linkage disequilibrium, the
Meuwissen and Goddard (2002) method could be used to fine map the QTL. This method
152
would be useful to map QTL segregating within line by examining only the haplotypes
originating from one or the other line. Since the lines used in Chapter 3 were partially inbred
and many of markers were fixed within line, this method would likely not be useful for fine
mapping in that population.
Population structure is also an important theme throughout the four papers.
Chapter 2
describes utilization of an F2 population structure, whereas Chapters 3 and 4 used a
backcross population structure. The importance of the population structure and the statistical
models used to analyze the population are illustrated in Chapter 2. Although the data were
the same for all models used in this section, statistical models assuming different population
structures (half-sib vs. line cross models) were able to identify different QTL.
The analyses described in Chapters 3 and 4 were single marker analyses, meaning that
multiple markers were not used simultaneously to separate QTL effect from QTL position
(effect and position are confounded in single marker analyses). Haley and Knott (1992)
described interval mapping (which uses multiple markers at the same time to estimate both
QTL position and effect) using linear regression, and this method was used in the research
reported in Chapter 2. The Cox proportional hazards model can also be easily extended to
interval mapping by replacing the single marker line origin probabilities that were used in
Chapter 3 by line origin probabilities for putative QTL positions along a chromosome, which
can be derived from flanking marker information. This model can then be fit for all positions
across a chromosome to determine the most likely location of a QTL.
5.3 RECOMMENDATIONS
QTL analyses are typically performed in livestock for two purposes: identification of genes
or genomic regions for use in marker assisted selection (MAS) programs for genetic
improvement of livestock and/or to identify genes involved in a biological process to
elucidate the pathways of that process. The QTL regions identified in the present study can
be utilized for both purposes.
153
Marker Assisted Selection
Successful utilization of the QTL identified in Chapters 2 and 3 for MAS programs requires
further investigation into these QTL. Markers associated with QTL can be used for selection
in other commercial populations, but the association and the linkage phase of the markers
with the QTL need to be identified in the population to be selected (Dekkers and Hospital,
2002).
Identifying marker associations with a QTL will only be possible if variation of
marker and QTL alleles exists in the population of interest.
Also, the QTL that were
identified in Chapter 2 were classified by their segregation patterns: between line, within
line, or both. The segregation patterns of the QTL identified in Chapter 2 give no indication
of the segregation patterns in other populations. If there is within line variation of a QTL in
the line to be selected, markers linked to that QTL can be included in the selection criteria to
increase the frequency of the favorable QTL allele. However, if no within line variation
exists, marker assisted introgression (Dekkers and Hospital, 2002) can be employed to move
the favorable QTL allele into the line of interest. Alternatively, marker assisted selection can
be utilized in crossbred animals if the favorable QTL allele is segregating in one of the
parental lines (Piyasatian et al., 2005). The QTL regions identified can also be saturated with
more markers for identification of markers that are very tightly linked to the causative
mutation. Linkage disequilibrium mapping (Meuwissen and Goddard, 2000) could then be
used to fine map the causative mutation, which then can be selected upon using the flanking
marker genotypes. Also, once the genomic region containing the causative mutation is fine
mapped, this region can be sequenced and all polymorphic sites can be tested across multiple
populations to determine which is most likely the causative mutation, which then can be
tested and selected upon in the population of interest. The final method of using the QTL
regions in MAS is to identify candidate genes in these regions, and test for the association of
the candidate genes with the trait in the population to be selected, or across populations to
identify very tight linkage (Rothschild and Soller, 1997). The QTL regions identified in
Chapter 2 and 3 can be utilized in all three of these methods.
The QTL identified affecting white meat percent (Chapter 2) and Marek's disease resistance
(Chapter 3) are the most interesting to include in a marker assisted selection program because
154
of the importance of these traits to the poultry industry and the difficulty in measuring these
traits. Molecular genetic information can be included in a selection program by including the
information in a selection index (Lande and Thompson, 1990; Dekkers and Arendonk, 1998),
or using the information in marker assisted BLUP (Totir et al., 2004). However, before using
these QTL in a selection program, their effects on other traits of interest must be evaluated.
It is possible that the QTL may have deleterious effects on other important traits that would
negate the benefits of selecting upon them to improve white meat percent or Marek's disease
resistance.
Imprinted QTL can be useful in MAS programs involving crosses between different parental
lines because the QTL will only be expressed when inherited from parents of one sex, and,
therefore, selection can be performed on only the lines used to create one of the parents. For
instance, if a QTL is only expressed when paternally inherited, selection on the favorable
allele only needs to be performed in the lines used to generate the male parents to get the full
QTL effect in the commercial offspring.
Elucidating Biological Pathways
With the molecular analysis tools now available, several methods are available to identify a
QTL's role in a biological pathway. One method is to test interactions of the QTL with other
genes or QTL regions (Moore and Williams, 2005). Associations between QTL and gene
expression data can also be used to determine the QTL's involvement in a particular pathway
(Jansen and Nap, 2001). Probably the most informative, but difficult, method of determining
pathways in which the QTL is involved is to perform extensive protein and RNA analyses
(Glazier et al., 2002). To perform these analyses, candidate genes for the QTL must be
identified. Then, proteins and RNA from the genes are tested for interactions with other
proteins, RNA, or DNA by binding assays, such as filter binding, gel mobility shift assays, or
coimmunoprecipitation (Weaver, 1999; Sambrook and Russell, 2001). The ideal ending to
this research would be to identify the entire biological pathway, from gene to phenotype.
The QTL identified in Chapters 2 and 3 can be used as starting points for any of these
analyses to elucidate biological pathways.
155
Chicken Genomic Sequence
The complete sequence of the chicken genome is now available (Hillier et al., 2004), and 2.8
million SNP s have been identified (Wong et al., 2004). The genomic sequence, along with
the SNP map to identify genomic variation, will facilitate the physical mapping of markers,
the identification of candidate genes corresponding to QTL locations, and the identification
of sequence variation within those candidate genes. The research presented herein utilized
microsatellite markers to trace inheritance of specific chromosomal regions, and to associate
sequence variation in these regions with phenotypes. However, with the complete chicken
genomic sequence and a large number of SNPs now identified in the chicken, as well as the
development of high throughput methods to genotype SNPs [eg. SnaPshot from Applied
Biosystems, Foster City, CA; and SNPchip from SNPchip, LLC, www.snpchip.com (2005)],
anonymous marker QTL mapping studies would benefit from using SNP markers instead of
microsatellite markers.
Use of SNP markers permits denser genomic coverage than
microsatellites, and the use of high throughput methods for SNP markers increases the rate at
which the study can progress. Although recombination rates between most SNPs have not
been determined through direct observation in chicken populations, they can be
approximated by comparing their physical locations in the genome with the physical
locations of markers which have been linkage mapped.
5.4 BIBLIOGRAPHY
Beavis, W. D. 1998. QTL analysis: power, precision, and accuracy. In: Paterson AH (ed)
Molecular dissection of complex traits. CRC Press, Boca Raton, FL, pp 145-173.
Bennewitz, J., N. Reinsch, and E. Kalm.
2002.
Improved confidence intervals in
quantitative trait loci mapping by permutation bootstrapping. Genetics 160:16731686.
Darvasi, A. and M. Soller. 1992. Selective genotyping for determination of linkage between a
marker locus and a quantitative trait locus. Theor. Appl. Genet. 85:353-359.
de Koning, D. J., Y. Nagamine, G. Evans, and C. S. Haley. 2002. QTL detection in outbred
populations; comparison of variance component and simple methods. Proceedings of
156
7th World Congress of Genetics Applied to Livestock Production. Montpellier,
France, CD-ROM communication no. 21-16.
de Koning D. J., R. Pong-Wong, L. Varona, G. J. Evans, E. Giuffra, A. Sanchez, G. Plastow,
J. L. Noguera, L. Andersson, and C. S. Haley. 2003. Full pedigree quantitative trait
locus analysis in commercial pigs using variance components. J. Anim. Sci. 81:
2155-2163.
de Koning, D. J., C. S. Haley, D. Windsor, P. M. Hocking, H. Griffin, A. Morris, J. Vincent,
and D. W. Burt. 2004. Segregation of QTL for production traits in commercial meattype chickens. Genet. Res. 83:211-220.
Dekkers, J. C. M., and J. A. M. Van Arendonk. 1998. Optimizing selection for quantitative
traits with information on an identified locus in outbred populations. Genet. Res.,
Camb. 71:257-275.
Dekkers, J. C., and F. Hospital. 2002. The use of molecular genetics in the improvement of
agricultural populations. Nat. Rev. Genet. 3:22-32.
Glazier, A. M., J. H. Nadeau, and T. J. Aitman. 2002. Finding genes that underlie complex
traits. Science 298:2345-2349.
Haley, C. S. and S. A. Knott. 1992. A simple regression method for mapping quantitative
trait loci in line crosses using flanking markers. Heredity 69:315-324.
Hillier, L. W., W. Miller, E. Bimey, W. Warren, R. C. Hardison, C. P. Ponting, P. Bork, D.
W. Burt, M. A. Groenen, M. E. Delany, J. B. Dodgson, A. T. Chinwalla, P. F. Cliften,
S. W. Clifton, K. D. Delehaunty, C. Fronick, R. S. Fulton, T. A. Graves, C.
Kremitzki, D. Layman, V. Magrini, J. D. Mcpherson, T. L. Miner, P. Minx, W. E.
Nash, M. N. Nhan, J. O. Nelson, L. G. Oddy, C. S. Pohl, J. Randall-Maher, S. M.
Smith, J. W. Wallis, S. P. Yang, M. N. Romanov, C. M. Rondelli, B. Paton, J. Smith,
D. Morrice, L. Daniels, H. G. Tempest, L. Robertson, J. S. Masabanda, D. K. Griffin,
A. Vignal, V. Fillon, L. Jacobbson, S. Kerje, L. Andersson, R. P. Crooijmans, J.
Aerts, J. J. Van Der Poel, H. Ellegren, R. B. Caldwell, S. J. Hubbard, D. V. Grafham,
A. M. Kierzek, S. R. Mclaren, I. M. Overton, H. Arakawa, K. J. Beattie, Y.
Bezzubov, P. E. Boardman, J. K. Bonfield, M. D. Craning, R. M. Davies, M. D.
Francis, S. J. Humphray, C. E. Scott, R. G. Taylor, C. Tickle, W. R. Brown, J.
157
Rogers, J. M. Buerstedde, S. A. Wilson, L. Stubbs, I. Ovcharenko, L. Gordon, S.
Lucas, M. M. Miller, H. Inoko, T. Shiina, J. Kaufman, J. Salomonsen, K. Skjoedt, G.
K. Wong, J. Wang, B. Liu, J. Yu, H. Yang, M. Nefedov, M. Koriabine, P. J. Dejong,
L. Goodstadt, C. Webber, N. J. Dickens, I. Letunic, M. Suyama, D. Torrents, C. Von
Mering, E. M. Zdobnov, K. Makova, A. Nekrutenko, L. Elnitski, P. Eswara, D. C.
King, S. Yang, S. Tyekucheva, A. Radakrishnan, R. S. Harris, F. Chiaromonte, J.
Taylor, J. He, M. Rijnkels, S. Griffiths-Jones, A. Ureta-Vidal, M. M. Hoffman, J.
Severin, S. M. Searle, A. S. Law, D. Speed, D. Waddington, Z. Cheng, E. Tuzun, E.
Eichler, Z. Bao, P. Flicek, D. D. Shteynberg, M. R. Brent, J. M. Bye, E. J. Huckle, S.
Chattel]i, C. Dewey, L. Pachter, A. Kouranov, Z. Mourelatos, A. G. Hatzigeorgiou,
A. H. Paterson, R. Ivarie, M. Brandstrom, E. Axelsson, N. Backstrom, S. Berlin, M.
T. Webster, O. Pourquie, A. Reymond, C. Ucla, S. E. Antonarakis, M. Long, J. J.
Emerson, E. Betran, I. Dupanloup, H. Kaessmann, A. S. Hinrichs, G. Bejerano, T. S.
Furey, R. A. Harte, B. Raney, A. Siepel, W. J. Kent, D. Haussier, E. Eyras, R.
Castelo, J. F. Abril, S. Castellano, F. Camara, G. Parra, R. Guigo, G. Bourque, G.
Tesler, P. A. Pevzner, A. Smit, L. A. Fulton, E. R. Mardis, and R. K. Wilson. 2004.
Sequence and comparative analysis of the chicken genome provide unique
perspectives on vertebrate evolution. Nature 432:695-716.
Hoeschele, I., P. Uimari, F. E. Grignola, Q. Zhang, and K. M. Gage. 1997. Advances in
statistical methods to map quantitative trait loci in outbred populations. Genetics 147:
1445-1457.
Jansen, R. C., and J. P. Nap. 2001. Genetical genomics: The added value from segregation.
Trends Genet. 17:388-391.
Kim, J. J., S. K. Davis, and J. F. Taylor. 2002. Application of non-parametric bootstrap
methods to estimate confidence intervals for QTL location in a beef cattle QTL
experimental population. Genet. Res., Camb. 79, 259-263.
Lande, R. and R. Thompson.
1990.
Efficiency of marker-assisted selection in the
improvement of quantitative traits. Genetics 124: 743-756.
Lander, E. S., and D. Botstein. 1989. Mapping mendelian factors underlying quantitative
traits using RFLP linkage maps. Genetics 121:185-199.
158
Moore, T., and D. Haig. 1991. Genomic imprinting in mammalian development: A parental
tug-of-war. Trends Genet. 7: 45-49.
Moore, J. H., and S. M. Williams. 2005. Traversing the conceptual divide between biological
and statistical epistasis: Systems biology and a more modern synthesis. Bioessays
27:637-646.
Muranty, H., and B. Goffmet. 1997. Selective genotyping for location and estimation of the
effect of a quantitative trait locus. Biometrics 53:629-643.
Pardo-Manuel de Villena, F., E. de la Casa-Esperon, and C. Sapienza. 2000. Natural
selection and the function of genome imprinting: Beyond the silenced minority.
Trends Genet. 16: 573-579.
Perez-Enciso, M., and M. A. Toro. 1999. Robust QTL effect estimation using the minimum
distance method. Heredity 83: 347-353.
Piyasatian, N., L. R. Totir, R.L. Fernando, and J.C.M. Dekkers. 2005. Marker-assisted
selection on multiple QTL in a crossbred population. Abstract. 2005 ASAS ADSA
Midwest Meeting, Des Moines, Iowa.
Rothschild, M. F., and M. Soller. 1997. Candidate gene analysis to detect genes controlling
traits of economic importance in domestic livestock. Probe 8: 13-20.
Sambrook, J. and Russell, D. W. (2001). Molecular Cloning: A Laboratory Manual. Cold
Spring Harbor Laboratory Press. Cold Spring Harbor, New York.
Soller, M., T. Brody and A. Genizi. 1976. On the power of experimental designs for the
detection of linkage between marker loci and quantitative loci in crosses between
inbred lines. Theor. Appl. Genet. 47:35-39.
Thomsen, H., H. K. Lee, M. F. Rothschild, M. Malek, and J. C. Dekkers. 2004.
Characterization of quantitative trait loci for growth and meat quality in a cross
between commercial breeds of swine. J. Anim. Sci. 82:2213-2228.
Totir, L. R., R. L. Fernando, J. C.M. Dekkers, S. A. Fernandez, and B.Guldbrandtsen. 2004.
The effect of using approximate gametic variance covariance matrices on marker
assisted selection by BLUP. Genet. Sel. Evol. 36:29-48.
159
Tuiskula-Haavisto, M., D. J. De Koning, M. Honkatukia, N. F. Schulman, A. Maki-Tanila,
and J. Vilkki. 2004. Quantitative trait loci with parent-of-origin effects in chicken.
Genet. Res. 84:57-66.
Uimari, P., G. Thaller, and I. Hoelchele. 1996a. The use of multiple markers in a Bayesian
method for mapping quantitative trait loci. Genetics 143: 1831-1842.
Weaver, R. F. 1999. Molecular Biology. The McGraw-Hill Companies, Inc. Boston, MA.
Wolfowitz, J. 1957. The minimum distance method. Ann. Math. Stat. 28,75-84.
Wong, G. K., B. Liu, J. Wang, Y. Zhang, X. Yang, Z. Zhang, Q. Meng, J. Zhou, D. Li, J.
Zhang, P. Ni, S. Li, L. Ran, H. Li, R. Li, H. Zheng, W. Lin, G. Li, X. Wang, W.
Zhao, J. Li, C. Ye, M. Dai, J. Ruan, Y. Zhou, Y. Li, X. He, X. Huang, W. Tong, J.
Chen, J. Ye, C. Chen, N. Wei, L. Dong, F. Lan, Y. Sun, Z. Yang, Y. Yu, Y. Huang,
D. He, Y. Xi, D. Wei, Q. Qi, W. Li, J. Shi, M. Wang, F. Xie, X. Zhang, P. Wang, Y.
Zhao, N. Li, N. Yang, W. Dong, S. Hu, C. Zeng, W. Zheng, B. Hao, L. W. Hillier, S.
P. Yang, W. C. Warren, R. K. Wilson, M. Brandstrom, H. Ellegren, R. P. Crooijmans,
J. J. Van Der Poel, H. Bovenhuis, M. A. Groenen, I. Ovcharenko, L. Gordon, L.
Stubbs, S. Lucas, T. Glavina, A. Aerts, P. Kaiser, L. Rothwell, J. R. Young, S.
Rogers, B. A. Walker, A. Van Hateren, J. Kaufman, N. Bumstead, S. J. Lamont, H.
Zhou, P. M. Hocking, D. Morrice, D. J. De Koning, A. Law, N. Bartley, D. W. Burt,
H. Hunt, H. H. Cheng, U. Gunnarsson, P. Wahlberg, L. Andersson, E. Kindlund, M.
T. Tammi, B. Andersson, C. Webber, C. P. Ponting, I. M. Overton, P. E. Boardman,
H. Tang, S. J. Hubbard, S. A. Wilson, J. Yu, and H. Yang. 2004. A genetic variation
map for chicken with 2.8 million single-nucleotide polymorphisms. Nature 432:717722.
Yokomine, T., H. Shirohzu, W. Purbowasito, A. Toyoda, H. Iwama, K. Ikeo, T. Hori, S.
Mizuno, M. Tsudzuki, Y. Matsuda, M. Hattori, Y. Sakaki, and and H. Sasaki. 2005.
Structural and functional analysis of a 0.5-mb chicken region orthologous to the
imprinted mammalian Ascl2!Mash2-Igf2-h 19 region. Genome Res. 15: 154-165.
160
CHAPTER 6
ACKNOWLEDGEMENTS
I first thank my major professors, Jack Dekkers and Susan Lamont, for their guidance,
teaching, and patience. Sue and Jack, individually, are the best possible mentors, and having
both of them as co-major professors greatly enhanced my education and development as I
ventured through my Ph.D. training. I am very grateful to them for making my Ph.D.
experience optimally educational and enjoyable. Secondly, I thank my other committee
members, Rohan Fernando, Max Rothschild, and Michael Lee, for their guidance. I thank
Matt and DeDe Abbott for their friendship, conversation, and for helping me to stay sane
throughout graduate-school (or joining me when I wasn't feeling quite sane).
I thank
Michael Kaiser and David Harry for their friendship and all of the advice that they have
given to me throughout my dissertation research. I am grateful to my office-mates, labmates, and other graduate students, both past and current, in the Animal Breeding department
for their friendship and help, and for joining me in my boat. I would specially like to thank
my past and current office-mates for putting up with my music. I thank my friends and
boxing coaches, Terry and Marge Dowd, for trying to keep me in line and in shape. And
finally, but most importantly, I thank my family. I am very thankful for my mother and
father, Kay and Paul McElroy, Jr., for bringing me into the world, caring for and raising me
well, teaching me to be curious and to question all things, and their loving support. I thank
my brother and sister, Jeff and Jessi, for their support and for challenging me while growing
up. I thank my fiancée, So Hyun Lee, for being so considerate and understanding when I was
under stress for various graduate-school related reasons. And, I thank you for taking the time
to read my acknowledgements.