genome-wide selection program for improvement of
Post on 12-Feb-2022
1 Views
Preview:
TRANSCRIPT
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/235766602
Genome-Wide Selection Program for Improvement of Indigenous Pigs in
Tropical Developing Countries
Thesis · January 2012
DOI: 10.13140/2.1.3287.1367
CITATIONS
2READS
416
Some of the authors of this publication are also working on these related projects:
Application of next generation genomic tools in beef: Addressing the phenomic gap View project
Development and utilization of genomic tools to exploit genetic diversity and heterosis in Canadian beef cattle populations View project
All content following this page was uploaded by Everestus Akanno on 01 June 2014.
The user has requested enhancement of the downloaded file.
Genome-Wide Selection Program for Improvement of Indigenous Pigs
in Tropical Developing Countries
by
Everestus Chima Akanno
A Thesis
presented to
The University of Guelph
In partial fulfillment of requirements
for the degree of
Doctor of Philosophy
in
Animal and Poultry Science
Guelph, Ontario, Canada
© Everestus Chima Akanno, January, 2012
ABSTRACT
GENOME-WIDE SELECTION PROGRAM FOR IMPROVEMENT OF
INDIGENOUS PIGS IN TROPICAL DEVELOPING COUNTRIES
Everestus Chima Akanno Advisor
University of Guelph, 2012 Dr. J. A. B. Robinson
Genetic improvement of indigenous pig populations in tropical developing
countries can make a significant contribution to the conservation and utilization of local
genetic resources. Designing a swine breeding program requires knowledge of genetic
parameters for economically important traits. A meta-analysis of genetic parameters
determined under tropical conditions and published from 1974 to 2009 was carried out to
provide consensus estimates of genetic parameters. Given that the data recording and
analysis infrastructure for implementing the conventional best linear unbiased prediction
(BLUP) methods is generally lacking in developing countries, Genome-wide selection
(GS) provides an approach for achieving faster genetic progress without developing a
pedigree recording system. A simulation study was carried out to evaluate the option of
using available 60 K single nucleotide polymorphism marker panel. The observed levels
of linkage disequilibrium (LD) in the tropical pig populations were simulated and
utilized. Genomic predictions were from ridge regression analysis. The results showed
that expected accuracies of genomic breeding values (GBV) were in the range of 0.31 -
0.86 for the validation set. Genome-wide selection improved accuracy of GBVs over
conventional BLUP method for traits with low heritability and in young animals with no
performance data. Crossbred training populations had higher accuracy than purebred
training populations. An assessment of the opportunities for GS in tropical pig breeding
was conducted. Genome-wide selection performed better than conventional methods by
increasing genetic gain and maintaining genetic variation while lowering inbreeding
especially for traits with low heritability, by exploiting LD and the Mendelian sampling
effects. Combining GS with repeated backcrossing of crossbreds to the selected exotic
population in moderate LD promises faster improvements of the commercial population.
A two-step selection strategy that involves the use of GS to pre-select candidates that
entered the performance test station and for selecting replacement candidates in a nucleus
swine breeding program was evaluated and compared to other conventional approaches.
Genome-wide selection generated an increase of about 38% to 172% in annual returns
compared to other conventional approaches for previously selected population in
moderate LD and about 2% to 50% increases in return for unselected population in low
LD.
iv
ACKNOWLEDGEMENT
Sincere appreciation and gratitude is extended to my advisor, Dr. J.A.B.
Robinson, for his constructive criticism, insightful guidance and immense help and
support throughout the preparation and completion of this Doctoral thesis. Thanks also go
to other members of my advisory committee, Dr. F. Schenkel, Dr. R. Friendship, for their
support and helpful discussions during my programs and more importantly to Dr. M.
Sargolzaei for developing the QMSim software which accelerated my study. Special
thanks go to Dr. L. Schaeffer for mentoring me in the course of my studies. Computing
and statistical support from Bill Szkotnicki and Dr. M. Quinton are gratefully
acknowledged. I would also like to thank all the faculty, staff and graduate students of the
Centre for Genetic Improvement of Livestock, Department of Animal and Poultry
Science, for providing a pleasant and stimulating environment over these past 4+ years.
Data assistance from the Canadian Centre for Swine Improvement is acknowledged.
The full financial support provided by Canadian Commonwealth Scholarship and
Fellowship program under the Department of Foreign Affairs and International Trade
(DFAIT), Canada, Bursary awards received from the University of Guelph and partial
funding from Federal University of Technology, Owerri, Nigeria are gratefully
acknowledged.
I thank my wife, Vivian Akanno, for continued support, encouragement, patience
and housework and my son, Great Chidubem Chukwunonye for permitting me to be
away when he needed me most. I am also grateful to my father Emmanuel Akanno, my
mother Augustina Akanno and all my siblings for being a source of special inspiration
v
and strength towards the successful completion of my studies. Support from all my
friends is appreciated.
Finally, I thank the Almighty GOD for His Grace and Blessings without which
this program would have been impossible. To Him I say ―Thus Far the Lord has helped
me‖.
Many thanks to all and may GOD bless all of you.
vi
TABLE OF CONTENTS
Chapter 1 General introduction 1
1.1 Preamble 2
1.2 Study objective 3
1.3 Justification of the study 3
1.4 Outline of thesis 4
Chapter 2 Swine genetic improvement in tropical developing countries:
Problems and Prospects 6
2.0 Abstract 7
2.1 Introduction 8
2.2 Swine genetic improvement programs in the tropics 11
2.3 Advances in swine genetic improvement 15
2.4 Genome-wide selection: a new method 17
2.5 Conclusions 28
Chapter 3 Meta-analysis of genetic parameter estimates for reproduction,
growth and carcass traits of pigs in the tropics 31
3.0 Abstract 32
3.1 Introduction 33
3.2 Materials and Methods 35
3.3 Results and Discussion 42
3.4 Conclusions and Recommendations 53
3.5 Appendix 67
vii
Chapter 4 Accuracies of genomic breeding values for simulated tropical pig
populations 83
4.0 Abstract 84
4.1 Introduction 85
4.2 Materials and Methods 88
4.3 Results and Discussion 94
4.4 Conclusions and Recommendations 106
Chapter 5 Opportunities for genome-wide selection for pig breeding in
developing countries 120
5.0 Abstract 121
5.1 Introduction 122
5.2 Materials and Methods 124
5.3 Results and Discussion 131
5.4 Conclusions and Recommendations 144
Chapter 6 Relative economic returns from selection schemes for a nucleus swine
breeding program 154
6.0 Abstract 155
6.1 Introduction 156
6.2 Materials and Methods 157
6.3 Results and Discussion 164
6.4 Conclusions and Recommendations 173
viii
Chapter 7 General Discussion and Conclusion 181
7.1 Summary 182
7.2 Recommendations 185
7.3 Future directions 186
References 190
ix
LIST OF TABLES
Table 2.1 Ranges of accuracies of genomic breeding values across various
sizes of reference population and using different statistical methods
in real dataset
30
Table 3.1 List of traits studied, abbreviations used and units of measurement
55
Table 3.2 Number of literature estimates (N), weighted least square means
(Mean±S.E), standard deviation (SD) and coefficient of variation
(CV) for reproduction, growth and carcass traits of pigs in the
tropics
56
Table 3.3 Number of literature estimates (N), combined (exotic and
indigenous) weighted mean heritability (h2), between study variance
(σ2
bs), Q statistics and 95% confidence interval (CI) for economic
traits of pigs in the tropics
57
Table 3.4 Number of literature estimates (N), weighted mean heritability for
indigenous (h2
I) and exotic (h2
E) breeds of pigs, maternal heritability
(h2
M), correlation between direct and maternal genetic effects (ram)
and common litter effect (l2) for reproduction traits of pigs in the
tropics
58
Table 3.5 Number of literature estimates (N), weighted mean heritability for
indigenous (h2
I) and exotic (h2
E) breeds of pigs, maternal heritability
(h2
M), correlation between direct and maternal genetic effects (ram)
and common litter effect (l2) for growth and carcass traits of pigs in
the tropics
59
Table 3.6 Levels of significance for the fixed factors in the analysis of
heritability estimates of some traits of pigs in the tropics
60
Table 3.7 Weighted means of estimates from literature for genetic (above
diagonal) and phenotypic (below diagonal) correlations among
reproduction traits of pigs in the tropics with the number of
estimates in parenthesis ( ) and the 95% confidence interval in
brackets [ ]
62
Table 3.8 Weighted means of genetic (above diagonal) and phenotypic (below
diagonal) correlations among growth traits of pigs in the tropics
63
x
with the number of estimates in parenthesis ( ) and the 95%
confidence interval in brackets [ ]
Table 3.9 Weighted means of genetic (above diagonal) and phenotypic (below
diagonal) correlations among carcass traits of pigs in the tropics
with the number of estimates in parenthesis ( ) and the 95%
confidence interval in brackets [ ]
64
Table 3.10 Weighted means of estimates from literature for genetic correlations
among reproduction, growth and carcass traits of pigs in the tropics
with the number of estimates in parenthesis ( ) and the 95%
confidence interval in brackets [ ]
65
Table 3.11 Weighted means of estimates from literature for phenotypic
correlations among reproduction, growth and carcass traits of pigs
in the tropics with the number of estimates in parenthesis ( ) and the
95% confidence interval in brackets [ ]
66
Table 4.1 The parameters used in the simulation program
108
Table 4.2 Correlations (r) between estimated breeding values ( ) and true
breeding values ( ) in the training (TP) and validation (VP)
populations
115
Table 4.3 Average accuracy of genomic breeding values when training and
validation population differ in structure
116
Table 4.4 Average linkage disequilibrium between adjacent marker pairs in
training and validation1 populations
117
Table 5.1 The parameters used in the simulation program
146
Table 5.2 Average cumulative responses (standard deviation units) after five
and ten generations of selection using a pedigree or genome-based
methods for different traits and breeding programs
148
Table 5.3 Average cumulated inbreeding and genetic variance after five and
ten generations of selection using a pedigree or genome-based
methods for different traits and breeding programs
149
xi
Table 6.1 Traits in the breeding objective and their parameters
175
Table 6.2 Accuracy of estimated breeding values of pigs in the training and
validation sets for traits in the selection index
176
Table 6.3 Accuracy of selection indices (rHI) for two steps selection strategies
using different methods and for different populations in a nucleus
swine breeding program
177
Table 6.4 Expected genetic change per year for different selection schemes
and populations
178
Table 6.5 Differential returns from different selection methods and for a range
of genotyping cost in a nucleus breeding program
179
Table 6.6 Differential returns from selection when GS scheme is based on pre-
selection accuracy in a nucleus breeding program
180
xii
LIST OF FIGURES
Figure 2.1 Trends in pig production in developed and developing countries
29
Figure 3.1 Funnel plot of heritability estimates for (A) LSF showing no
evidence of publication bias and (B) ADG showing classic
asymmetry suggesting the presence of a publication bias
61
Figure 4.1 Schematic representation of the simulated population structure
109
Figure 4.2 Distribution of true allelic effects simulated for the three traits
studied
110
Figure 4.3 Distribution of average minor allele frequency (MAF) for the SNP
markers in the founder populations
111
Figure 4.4 Average linkage disequilibrium as measured by r2 against distance
(cM) between a pair of marker in different founder populations
112
Figure 4.5 Effect of training population size and structure on accuracy and
persistency of accuracy following ten generations of random
mating in selected exotic (top) and unselected indigenous (bottom)
populations and for NBA
113
Figure 4.6 Effect of training population size and structure on accuracy and
persistency of accuracy following ten generations of random
mating in selected exotic (top) and unselected indigenous (bottom)
populations and for BFT
114
Figure 4.7 Decay of accuracy of GBVs in different populations and traits
under random mating for ten generations. The straight line is the
accuracy level based on trait heritability
118
Figure 4.8 Influence of trait heritability on persistency of accuracy of GBVs in
different populations under random mating
119
Figure 5.1 Schematic representation of the simulated population structure
147
Figure 5.2 Cumulative response (A), inbreeding (B), LD (C) and realised
accuracy (D) for various selection methods in an unselected
indigenous population and for ADG
150
xiii
Figure 5.3 Cumulative response (A), inbreeding (B), LD (C) and realised
accuracy (D) for different breeding programs using genome-wide
selection and for ADG
151
Figure 5.4 Cumulative responses (A), inbreeding (B), LD (C) and realised
accuracy (D) for different traits in a genome-wide selection scheme
for unselected indigenous population
152
Figure 5.5 Average minor allele frequency (MAF) changes due to (A)
different selection methods in an unselected indigenous population
for ADG, (B) different breeding programs in a genome-wide
selection scheme and for ADG and (C) different traits in a genome-
wide selection scheme for unselected indigenous population
153
Figure 6.1 Decision point in a nucleus swine breeding program 174
xiv
Glossary of Terms
BLUP Best linear unbiased predictions
DNA Deoxyribonucleic acid
GBV Genomic breeding value
GS Genome-wide selection
HWE Hardy-Weinberg equilibrium
LD Linkage disequilibrium
MAS Marker assisted selection
MAI Marker assisted introgression
MDE Mutation drift equilibrium
MAF Minor allele frequency
Ne Effective population size
NBS Nucleus breeding scheme
QTL Quantitative trait loci
SNP Single nucleotide polymorphism
1
CHAPTER 1
GENERAL INTRODUCTION
E. C. Akanno
Centre for Genetic Improvement of Livestock,
Department of Animal & Poultry Science, University of Guelph, ON, Canada
2
1.1 PREAMBLE
Genome-wide selection (GS) is currently revolutionizing livestock breeding
programs around the globe (Nejati-Javaremi, et al., 1997; Meuwissen et al., 2001).
Genome-wide selection aims to estimate the effects of dense genetic markers
simultaneously, and summing up these effects across all marker loci to predict the
genomic breeding values (GBV) of an individual. Selection decisions can then be based
on the GBV. This method relies heavily on the existence of linkage disequilibrium (LD)
between markers and putative quantitative trait loci (QTL) influencing traits of economic
importance in livestock (Habier et al., 2007; Toosi et al., 2010).
Genome-wide selection will be useful in the genetic improvement of indigenous
pigs in tropical developing countries where the infrastructures for the implementation of
conventional methods are lacking. Development of pedigree recording systems may no
longer be required but this does not remove the need to collect phenotypic data because
substantial datasets with both genotypes and phenotypes are required for training the
prediction model. The reference population could be managed under low-input conditions
and estimates of marker effects derived from this population would be used to obtain
GBV for selection candidates in a nucleus breeding herd.
Alternatively, the indigenous and exotic purebreds are crossed to produce
crossbred reference population for training marker effects and prediction of GBV in the
purebreds, thereby improving commercially relevant traits. Cryptic or transgressive
alleles (Notter, 1999) in the unselected indigenous breeds can be captured easily using
GS and their frequencies increased in this population. Other relevant adaptive genes in
3
the indigenous population necessary for pork production under tropical conditions will be
identified and incorporated into commercial population.
1.2 STUDY OBJECTIVE
The overall goal of this thesis was to evaluate the use of genome-wide selection
programs for genetic improvement of production traits in breeding programs involving
indigenous and exotic pigs in tropical developing countries. There were four specific
objectives:
1) To provide consensus estimates of genetic parameters for pig production traits
determined under tropical conditions
2) To evaluate the accuracies of genomic breeding values for simulated tropical pig
population structure
3) To assess the opportunities for genome-wide selection for pig breeding in
developing countries
4) To evaluate the relative returns from selection schemes for nucleus swine
breeding program
1.3 JUSTIFICATION OF THE STUDY
Designing an experiment to evaluate the option of implementing GS in tropical
pig breeding systems will require housing and managing a large number of reference pig
populations under tropical conditions and collecting phenotypic data on traits of interest.
These animals will need to be genotyped for thousands of marker loci based on an
existing dense marker panel. Many of these marker genotypes may not be informative
4
and some will not be segregating at all in this population. Another set of animals will
need to be produced and genotyped for the same set of markers as in the reference
population. These will be used to validate the accuracy of GBV. Currently, the cost of
genotyping for 60 K single nucleotide polymorphism (SNP) markers is still within the
neighbourhood of $120 per animal at DNA Landmarks (M. McNairnay, personal
communication, 2011) hence, implementation of GS on real populations can be very
costly.
Computer simulation can be used to conduct this experiment in a relatively short
period of time with minimum cost. Its usefulness depends on the completeness of the
assumed model and the accuracy of the parameters that go into the simulation program.
Usually, the assumed model is only an approximation of the real life situation. Therefore,
a simulation approach was chosen to evaluate the option of GS for tropical pig breeding
systems.
1.4 OUTLINE OF THESIS
The current states of knowledge about the development of genetic improvement
programs for pigs in tropical developing countries are reviewed in Chapter 2. The use of
genetic markers in pig breeding and their potential in contributing to a sustainable pig
improvement program under low-input production systems are discussed. Chapter 3
presents estimates of genetic and phenotypic parameters for pig production traits
determined under tropical conditions. This is the first time a meta-analysis of genetic
parameters was conducted for pig production traits in tropical developing countries.
Chapter 4 focuses on accuracies of GBV in populations with different breeding
5
programs and selection history. Analysis was based on simulated data assuming tropical
pig population structure. In Chapter 5, the application of genome-wide selection scheme
was evaluated and compared to conventional methods. In addition, implications for
genetic response, inbreeding rate, realized accuracy, LD and QTL minor allele frequency
change are discussed. Chapter 6 attempts to evaluate the relative returns from the
implementation of GS in a nucleus swine breeding program. A summary of all findings
from these studies are presented in Chapter 7 with some recommendations for the future.
6
CHAPTER 2
SWINE GENETIC IMPROVEMENT IN TROPICAL DEVELOPING
COUNTRIES: PROBLEMS AND PROSPECTS
E. C. Akanno, F. S. Schenkel and J. A. B. Robinson
Centre for Genetic Improvement of Livestock,
Department of Animal & Poultry Science, University of Guelph, ON, Canada
7
2.0 ABSTRACT
The increase in productivity per pig to cushion the effect of animal protein
shortage and improve the livelihood of pig producers in developing countries depends
among other factors on the development of a sustainable genetic improvement program.
This paper reviewed the current challenges limiting the implementation of swine genetic
improvement programs in tropical developing countries. The application of a genome-
wide selection program (GS), as opposed to the development of infrastructure for
implementing the traditional selection scheme, is suggested. Genome-wide selection has
the potential to increase genetic gain by increasing accuracy of genomic breeding values
for selection candidates. Genome-wide selection can be adapted to existing tropical pig
breeding system. This paper calls for the evaluation of this option using computer
simulation of pig breeding scenarios in the tropics because implementation of GS on real
populations can be very costly.
Keywords: Genetic improvement, genome-wide selection, tropical pig, developing
countries
8
2.1 INTRODUCTION
A growing concern in many tropical developing countries is the low per capita
meat consumption level created by population growth, increased climate variability, and
persistent low investment in animal production research, among others (The World Bank,
2007). Owing to recent increases in household income and associated rise in standard of
living, meat consumption in developing countries has been projected to increase from
52% currently to 63% of the world total consumption by the year 2020 (Delgado, 2003).
About 44% of this increase will be derived from pork and pork products alone (FAO,
2001).
2.1.1 Pig production in the tropics
The pig is one of the earliest domesticated animals and plays an integral part in a
number of peasant economies in the tropics. In several African and Southeast Asian
countries such as Nigeria and Thailand, intensive pig production has played a significant
role in the overall economic growth and the feeding of an expanding urban population.
Furthermore, although in some circumstances pigs do compete with humans for food
resources, they are also very versatile at utilizing byproducts and wastes that are not
suitable for human consumption (Eusebio, 1984).
Currently, exotic pigs originally imported from Europe and North America are
used in conjunction with indigenous population for commercial pork production (Olivier
et al., 2002). Importations of exotic pigs into many developing countries are known to
have occurred before 1970 (Blench, 2000). The primary reason for this importation was
to improve the performance of the indigenous pigs under tropical conditions. Forsberg et
9
al. (2005) showed that pig populations in developing countries have continued to increase
at a linear rate of 10 percent per year since early 1970s while pig populations in
developed countries have remained comparatively constant (Figure 2.1). This increase
may be connected with the introduction of exotic breeds of pigs into many tropical
developing countries around this period.
In addition to pork production, pig rearing is considered as a form of financial
security and source of independent income especially for smallholder farmers (Lekule
and Kyvsgaard, 2003). Pigs are also commonly used in social-cultural functions such as
marriage and religious ceremonies. Because an increase in the number of pigs will occur,
a major effort should be made to increase the productivity per pig while lowering the cost
per unit of production. This will help to alleviate poverty and improve the livelihoods of
resource-poor farmers in the developing world (Garett, 2000).
2.1.2 Constraints to swine production under tropical conditions
The development of a swine production system requires an understanding of the
constraints on productivity and profitability. Studies have shown that inadequate
nutrition, variable management practices, poor genetic stock, heat stress and disease
challenge are among the prime factors limiting the growth of the swine industry in
developing countries (Muys and Westenbrink, 1995; Olivier et al., 2002; van der Waaij et
al., 2002; Lekule and Kyvsgaard, 2003; Beek, 2007; Mohanty and Chhabra, 2007).
Exotic pigs require sophisticated high-input feeding and management systems in order
for the pigs to fully express their superior genetic merit for faster growth rate, larger litter
size at birth and weaning, and good carcass quality. However, medium- to low-input
10
production systems tend to dominate the swine industry in most tropical locations
(Madalena et al., 2002; Olivier et al., 2002).
In these locations, ambient temperature is frequently greater than 22 , the
evaporative critical temperature for swine (Quiniou and Noblet, 1999), and in intensive
production systems, buildings are usually open or semi-open to the outdoor ambient
environment. Also, the climate is very suitable for both bacterial multiplication and
parasite survival (Kunavongkrit and Heard, 2000). Under these conditions, the exotic pig
breeds suffer from heat stress and disease challenge resulting in lower performance than
averages reported for their counterparts in temperate environments.
Another implication of this outcome is the huge investment in housing, high
quality feed and medication which tends to compromise with profitability of pork
production in the tropics. Attempts have been made to manipulate productivity of exotic
pigs through cross-breeding with the low yielding indigenous breeds (Nguyen et al.,
1996; Lakhani and Jogi, 1999; Oseni, 2005).
2.1.3 Development of indigenous pig populations
The swine production system in tropical areas often operates on a subsistence
level that is characterised by medium- to low-inputs and a stressful environment. The
local pig population has been adapted to these harsh conditions through generations of
natural selection. This has resulted in indigenous breeds and strains that survive and
produce under the prevailing environmental stresses of the tropics (King, 1987;
Kanengoni et al., 2004). Also, this population could be a potential source of the so-called
11
transgressive or cryptic alleles. These alleles are superior genes for some productive traits
supposedly ―hidden‖ in unselected breeds that are inferior for these traits (Notter, 1999).
There is evidence that indigenous breeds of pigs are less susceptible to common
diseases and parasites than the exotic pigs (Zanga et al., 2003), and have the ability to
survive long periods of feed and water deprivation (Anderson, 2003). They are less
reliant on external inputs, unlike exotic breeds and are, therefore, a potential part of a
sustainable agricultural system for the resource-poor farmers (Chiduwa et al., 2008).
They are generally hardy, can survive and reproduce on a low plane of nutrition under
extensive production systems (Holness, 1976). Despite these attributes, there is still
neglect in the utilization of these breeds for pork production leading to a decline in their
numbers, which puts them at risk of extinction (FAO, 1998; Lekule and Kyvsgaard,
2003).
2.2 SWINE GENETIC IMPROVEMENT PROGRAMS IN THE TROPICS
2.2.1 Nucleus breeding schemes
Nucleus breeding scheme (NBS) represents a breeding program in which
selection is carried out on a small number of elite breeding animals in a nucleus that
produces germ-plasm or breeding stock for use in the field. This method has been
proposed to overcome the technical constraints limiting the implementation of genetic
improvement programs for low-input production systems in developing countries (Smith,
1988). The genetic superiority generated in the nucleus is disseminated to the whole
population. The rate of genetic progress depends, among other things, on the size of the
nucleus and the effectiveness of selection within the nucleus (Weigel, 2001). Opening the
12
nucleus to replacements from the village herds might improve the genetic progress and,
more importantly, its chances of success because it encourages more farmer participation
(Bondoc and Smith, 1993).
Although there has been some success recorded for NBS, these programmes are
often faced with extensive genotype by environmental interactions and lack of acceptance
of nucleus breeding stock in the field. A genome based selection program offers
opportunities to overcome these limitations by collecting phenotypes and
deoxyribonucleic acid (DNA) from a large sample of animals in the field, without
requiring them to be pedigreed. The resulting estimates could be used to either select
within a nucleus herd or to identify superior breeding stock in the field on the basis of
their single nucleotide polymorphism (SNP) genotypes (Dekkers, 2010).
2.2.2 Cross-breeding programs
Cross-breeding is a successful management practice for increasing pork
production in swine. By exploiting heterosis, cross-breeding is known to improve
reproduction and growth traits (Adebambo, 1986; Gerasimov and PronAE, 2006;
Nwakpu and Onu, 2007), and for breed (or line) complementarities. Generally, those
traits that suffer from inbreeding depression or have low heritability are improved
through crossbreeding (Falconer and Mackay, 1996). In practice, there are many crossing
designs employed, from two-breed crosses and backcrosses, reciprocal crosses or three-
way-terminal crosses to crosses involving more than four lines (Oseni, 2005).
Results of several cross-breeding studies in the tropics involving exotic and
indigenous breeds of pigs have indicated varying levels of heterosis (Adebambo, 1985;
13
Meade Moreno, 1987; Kanengoni et al., 2004; Oke et al., 2006; Nwakpu and Onu, 2007).
In addition to the advantage of utilising heterosis, different breeds or lines may be kept to
breed for different objectives, for example specialised sire and dam lines selected for lean
growth rate and reproduction, respectively (Visscher et al., 2000). Selection can also
proceed on the crossbred population in order to conserve useful genetic material for pork
production in this environment.
Furthermore, the heterozygosity observed across loci with the attendant linkage
disequilibrium (LD) in crossbred population has led to the mapping of loci that affects
traits of economic importance in swine (Rothschild et al., 2007). Crosses between
divergent lines have become a mating design of choice in the search for quantitative trait
loci (QTL). Consequently, F2 and backcross designs are employed in experiments aimed
at identifying and mapping QTL that associates with traits of economic importance in
swine (see for example King et al., 2003; Mote and Rothschild, 2006 ).
2.2.3 Traditional genetic improvement methods
Genetic improvement of pigs in the tropics is centred on improving reproduction,
growth and carcass quality traits with little emphasis on disease resistance due to the
difficulty associated with measuring this trait (Adebambo, 1986; Rohilla et al., 2000;
Rinaldo et al., 2003; Oseni, 2005; Chimonyo and Dzama, 2007). The approach used is
the traditional quantitative genetic method based on infinitesimal model assumptions
(Fisher, 1918). Reliable estimates of genetic and phenotypic parameters for all traits of
economic importance are required as priors in order to predict responses to selection, to
choose among various breeding plans, to estimate economic returns and to predict best
14
linear unbiased predictions (BLUP) of breeding values (Henderson, 1975) for selection
candidates.
A limitation to the successful implementation of traditional BLUP method in
developing countries is the lack of extensive and reliable pedigree information to
facilitate gene tracking (Pathiraja, 1987; Kahi et al., 2005; Chimonyo and Dzama, 2007),
inaccurate performance recording and the dearth of genetic parameter estimates for
economic traits of pigs under tropical conditions (Kahi et al., 2005). Genetic parameters
are known to be population specific and are often not estimated very precisely, even in
large datasets. A mistake often made by breeders in the tropics is to use estimates of
genetic parameters determined in a temperate environment when constructing the genetic
variance-covariance matrix required in animal genetic evaluation. When this is the case,
accuracy is compromised and genetic progress is delayed (Visscher et al., 2008).
2.2.4 Genetic and phenotypic parameters
Over the last four decades, several estimates of genetic and phenotypic parameters
for pig production traits determined under tropical conditions have been published in the
literature. These estimates were derived using a variety of methods, are of variable
quality in terms of sample size used to estimate them, and have contradictory results.
Under these circumstances, swine breeders and researchers may have difficulty making
sense of the literature estimates or taking stock of the knowledge base. This can have
negative consequences on breeding decisions and can hinder genetic progress.
An effort to harmonise these estimates by pooling results from different
populations is intuitively appealing, particularly if differences between populations are
15
mainly due to sampling error (Koots, 1994). If estimates are systematically affected by
identifiable factors, pooled estimates can be adjusted by applying appropriate statistical
models. A statistical method that is currently used for synthesis of results is ―meta-
analysis‖ (Sutton et al., 2000). This method has been used to summarise genetic
parameter estimates for beef cattle (Koots et al., 1994; Giannotti et al., 2005) and sheep
(Safari et al., 2005) production traits.
2.3 ADVANCES IN SWINE GENETIC IMPROVEMENT
2.3.1 Use of genetic markers
The advent of molecular genetics technology driven by the human genome project
has led to the study of variation at the DNA level, and the ability to generate large
numbers of polymorphic genetic markers in any species of livestock. The SNP appears to
be a marker of choice because of its abundant coverage of the genome and ease of
genotyping (Visscher et al., 2000). These markers are variations in the nucleotide base
sequence at the same point in the genome among individuals or between an individual‘s
chromosome pairs. Methods have been developed for high-throughput genotyping of
thousands of SNPs per individual animal and the cost of genotyping per individual
marker genotype is currently less than one cent (Weller, 2010). More so, a high density
porcine SNP chip (~60 K) for pigs developed by Illumina Inc in collaboration with the
Swine Genome Sequencing Consortium (SGSC) already exists (Ramos et al., 2009).
The application of the new marker technology to pig genetic improvement stems
from the search for chromosomal regions in the swine genome that are linked to putative
QTLs influencing traits of economic importance (Visscher and Haley, 1995). To date,
16
about 6344 significant (P < 0.05) QTLs have been identified in the pig, influencing over
593 traits (Retrieved April 28, 2011 from pigQTLdb: www.animalgenome.org/cgi-
bin/QTLdb/SS/summary). Chromosome 1 tends to have the highest number of QTLs
identified (about 1391) and these QTLs significantly affect most of the important
economic traits including litter size, average daily gain and back-fat thickness (Malek et
al., 2001; Milan et al., 2002; Evans et al., 2003; Kim et al., 2005; Buske et al., 2006).
Other QTLs affecting economic traits in pigs are summarised by Dekkers (2004).
2.3.2 Marker-assisted selection and introgression
Research towards utilising detected QTLs in swine breeding programmes using
marker-assisted selection (MAS) or marker-assisted introgression (MAI) has been
extensive in developed countries (Lande and Thompson, 1990; Hayes and Goddard,
2003; Dekkers and Chakraborty, 2004; Frisch and Melchinger, 2005; Hospital, 2005),
but implementation has been limited and increases in genetic gain have been small
(Dekkers, 2004). The weaknesses in the methodology of MAS and MAI lies in their
requirements for 1) QTL identification followed by estimation of effects on phenotype; 2)
linkage phase between a marker and QTL which persists only within families; 3)
decisions on how many markers to be included in genetic evaluation; and 4) use of
stringent significance tests in QTL detection.
As most polygenic traits are controlled by many genes, tracking a small number
of these through a few markers will only explain a small proportion of the genetic
variance (Goddard and Hayes, 2007). Methods that use all markers across the genome
simultaneously may therefore result in higher reliabilities of predictions of total genetic
17
merit, indicating that a larger proportion of the genetic variance is explained (Meuwissen
et al., 2001; Gianola et al., 2003).
2.4 GENOME-WIDE SELECTION: A NEW METHOD
Genome-wide selection (GS) uses a ―training population‖ of individuals with
phenotypic and genotypic records to develop a prediction model that takes genotypic data
from a ―candidate population‖ of untested individuals and produces genomic breeding
values (GBV) of an individual using the estimated additive marker effects as priors
(Meuwissen et al., 2001). This means that the infinitesimal model assumption (Fisher,
1918) implied in the prediction of BLUP breeding values may no longer hold and that a
finite number of loci underlie the variation in quantitative traits (Ewing and Green, 2000).
Using GS in animal genetic improvement will yield a considerable increase in
selection response for juvenile animals that do not have phenotypic records (Meuwissen
et al., 2001; Gianola et al., 2003; Hayes et al., 2009) and potentially can reduce the cost
of breeding program (Schaeffer, 2006). Genome-wide selection is also believed to be
advantageous for sex-limited traits, traits that are difficult or expensive to measure like
disease traits and traits with a low heritability (Goddard and Hayes, 2007).
2.4.1 Methods for genomic breeding value prediction
The development of statistical methods for the prediction of GBV is an ongoing
endeavour ( Nejati-Javaremi, et al., 1997; Whittaker et al., 1999; Meuwissen et al., 2001;
Xu, 2003; Fernando et al., 2007; Gianola et al., 2006; Hayes et al., 2009; VanRaden et
al., 2009). Several methods have been proposed to deal with the problem of over-
18
parameterization and multicollinearity between the predictors when using high density
markers as fixed effects in the prediction model. These methods differ in the prior
assumptions that they make about the distribution of marker effects, which in turn reflects
the distribution of QTL effects and has been shown to affect accuracy of GBV. In the
section that follows, three of these statistical methods that are widely used in the
implementation of GS are described briefly. In these methods, a linear mixed effect
model for the analysis of genomic data with bi-allelic genotypes can be defined as:
yi = μ + ∑j zijβjδj + ei 2.1
where yi is the phenotypic observation on an individual i, μ is the overall mean of y, zij is
0, 1, or 2 depending on the SNP genotype at the marker locus j in individual i, βj is the
allele substitution effect associated with marker j, δj is a 1 or 0 indicator variable for the
inclusion or exclusion of marker j in the estimation of breeding values, and ei is a
residual.
2.4.1.1 Ridge regression
The application of ridge regression in the analysis of marker data was first
proposed by Whittaker et al. (1999), and modified by Meuwissen et al. (2001) for
analysis of markers across the genome with a slight adjustment for the variance of the
marker loci made by Fernando et al. (2007). In general, ridge regression is a variant of
ordinary multiple linear regression where the goal is to circumvent the problem of
instability arising from collinearity of the predictors and over-parameterization. The
estimates of the regression coefficients are shrunk towards zero through the use of a
regularization parameter (λ) that controls the amount of shrinkage.
19
The ridge regression method assumes that δj given in equation 2.1 is always 1 and
that βj ~ N (0, σ2
β) for all markers. The marker variance, σ2
β, is estimated from the data by
a maximum likelihood method. The solution to equation 2.1 can be obtained by solving
the mixed model equations (Henderson, 1984) given below:
2.2
The value of λ was given by Meuwissen et al. (2001) as a function of the prior variances
and the total number of marker loci (m) expressed as while Fernando et al. (2007)
modified this expression by using the mean heterozygosity of the markers across the
genome given as . This method also assumes that marker effects are distributed
with equal variance and are somewhat similar to the traditional BLUP method except that
the later equations are based on individual animals and their relationships.
Genomic evaluation using ridge regression approach is simple and fast to
implement with minimal computations compared to other methods. Habier et al. (2007)
showed that ridge regression is more effective at capturing genetic relationships among
genotyped individuals because it fits more markers into the prediction model. In addition,
there are freely available computer programs for implementing ridge regression method
with the functionality of handling large genomic dataset (Sargolzaei et al., 2009).
2.4.1.2 Bayesian methods
Meuwissen et al. (2001) proposed two variants of the Bayesian approach for
estimating marker allele effects. They called these methods BayesA and BayesB. The
Bayesian approach attempts to model the data at two levels namely; at the level of the
20
data and at the level of the variances of the marker loci. The data model is analogous to
ridge regression except that the variances of the marker loci (σ2
βj) differ at every position
and are estimated using the combined information from the prior distribution of variances
and from the data.
The prior distribution of the variances of the marker loci was the scaled inverted
chi-square distribution, χ-2
(v, S), where S is a scale parameter and v is the degrees of
freedom. Meuwissen et al. (2001) showed how these parameters can be derived. The
difference between the two Bayesian methods lies in the assumptions about the
distribution of the SNP effects and their implementation. BayesA assumes that all marker
loci affect the traits with different variances and all loci are included in the model for
estimation, i.e. δj in equation 2.1 is always equal to 1. Gibbs sampling algorithm is
convenient for estimating effects and variances for this method.
BayesB assumes that there are many loci with zero genetic variance and a few
segregating loci affecting the trait of interest. The marker effects are modelled by
excluding those loci that are not affecting the trait from the estimation model and
therefore uses a prior that has a high density, π, at σ2
βj = 0 and has an inverted chi-square
distribution for σ2
βj > 0. This means that δj is equal to zero with probability π when σ2
βj =
0 and δj is equal to one with probability 1 – π when σ2
βj > 0. The probability π of non-
segregating loci is assumed to be known a priori and a Metropolis-Hasting algorithm is
used to determine for each locus whether it has an effect on the phenotype or not. If it has
not, the effect of that locus is set to zero (Meuwissen et al., 2001).
In a related study by Kizilkaya et al. (2010), another Bayesian technique called
BayesCπ was used. As in BayesB, in BayesCπ, δj is equal to zero with probability π but π
21
is estimated assuming a uniform prior distribution between 0 and 1. In addition, BayesCπ
assumes that the prior variance for the effects of all markers for which δ j = 1 is equal.
That is, the effect βj is zero when δj = 0 or βj ~ N (0, σ2
β) when δj = 1.
The implementation of the Bayesian procedures like BayesB in genomic
evaluation have resulted in slightly higher accuracy than ridge regression (Meuwissen et
al., 2001) because BayesB is more effective at capturing LD between markers and QTL
that are sufficiently strong to allow their accuracy to persist longer than those from ridge
regression (Habier et al., 2007). In addition, the over-parameterization problem of ridge
regression is smaller with BayesB and different variances are assumed for each marker.
More so, BayesB requires more computer time than ridge regression and the assumption
of markers with zero effect is equivalent to a decrease in the genome size and
concentration of responses from a few QTLs which results in higher accuracy in this
method (Smaragdov, 2009).
The marker effects estimated by either ridge regression or the Bayesian methods
can be applied to the SNP genotypes of selection candidates to predict their GBV using
the standard model:
GBVi = 2.3
2.4.1.3 GBLUP method
An alternative for implementing genomic selection to the methods described
above is to predict breeding values using a genomic relationship matrix in place of the
pedigree derived relationship matrix (see for example Habier et al., 2007; Hayes et al.,
2009; VanRaden et al., 2009; Hayes and Goddard, 2010). This model is equivalent to
22
predicting individual SNP effects and calculating GBV as the sum of these effects,
provided the SNP effects are assumed to be normally distributed.
The model in equation 2.1 can be re-written in matrix form assuming that δj = 1 as
follow:
y = 1nμ + Zg + e 2.4
where y is a vector of phenotypes of size n, μ is the overall mean, 1n is a vector of ones,
Z is a design matrix allocating records to breeding values, g is a vector of breeding values
and e is a vector of random errors ~ N (0, σ2
e). Then g = Wu where uj is the effect of jth
SNP, and variance V (g) = WW´σ2
u. Elements of matrix W are wij for the ith animal and
jth SNP, where wij = 0 – 2pj if the animal is homozygous 11 at the jth SNP, 1 – 2pj if the
animal is heterozygous 12 or 21 at the jth SNP, and 2 – 2pj if the animal is homozygous
22 at the jth SNP. The diagonal elements of WW´ will be , where m is
the number of SNPs. If WW´ is scaled such that G = nWW´/ then V (g) = G
σ2
g. The breeding values for both phenotyped and non-phenotyped individuals can then
be predicted by solving the equations:
2.5
Using this method in animal genetic evaluation requires replacing the additive
relationship matrix with the genomic relationship matrix in the existing genetic
evaluation program (Hayes and Goddard, 2010). The method can also be useful for
countries without a good pedigree recording system as the genomic relationship matrix
will capture this relationship information among the genotyped individuals.
23
2.4.2 Comparison of methods based on accuracies of genomic predictions
The accuracies of genomic predictions in simulation studies utilizing the
statistical methods described above ranged from 0.46 – 0.88 for the ridge regression
method (Meuwissen et al., 2001; Muir, 2007; Habier et al., 2007; Fernando et al., 2007;
Saatchi et al., 2010), 0.56 – 0.98 using different versions of the Bayesian methods
(Meuwissen et al., 2001; Solberg et al., 2008; Habier et al., 2009; Ibanz-Escriche et al.,
2009; Kizilkaya et al., 2010), and 0.59 – 0.96 using GBLUP (VanRaden, 2008;
Christensen and Lund, 2010) for a range of trait heritabilities. These results suggest
similarity in the expected accuracy of GBV based on different methods of estimation.
Consideration of appropriate method for use in genetic evaluation will depend on the
persistency of these accuracies across multiple populations and generations,
computational requirements of the method and time constraint.
Table 2.1 gives a summary of ranges of GBV prediction accuracies obtained from
studies that use real dataset from dairy cattle and swine reference population across
different traits and statistical methods. The accuracies of GBV observed from these
empirical studies are moderate to high and somewhat comparable to results from
simulation studies. Methods that assume all marker effects are distributed with equal
variance like ridge regression are as accurate as methods that assume some markers do
not explain any variance (e.g BayesB). Note that assuming all marker effects are drawn
from the same distribution does not mean the effects are all equal but that they are all
equally shrunken towards zero (Jannink et al., 2010).
The GBLUP method appears to have higher and slightly narrower range of GBV
accuracies than other methods across several traits. This could justify the higher number
24
of studies that used this method in genetic evaluation. More so, the GBLUP method
allows you to use existing genetic evaluation programs by replacing the additive
relationship matrix with the genomic relationship matrix and the estimates of GBV from
this method are BLUP (Hayes and Goddard, 2010).
Beside statistical methods used in estimating marker effects, other factors known
to affect accuracies of GBV are the marker types and density (Solberg et al., 2008), the
size and number of generations in the reference population (Muir, 2007), the population
structure and history, the trait heritability and the LD between markers and QTL across
generations (Habier et al., 2007; Toosi et al., 2010).
2.4.3 LD and accuracy of GBV across populations and generations
The accuracies of GBV rely heavily on the existence of LD between QTL and
markers in the population of interest and on the strength of predicted individuals‘
relatedness to the reference population, much as it will do when using pedigree
information to perform predictions. As the population diverges, accuracy of GBV
declines due to recombination between QTL and marker across multiple populations
unless dense markers are used to maintain sufficient LD (de Roos et al., 2009).
Conceptually assuming large reference population, the more dense the markers and
greater the population-wide LD, the better the method works.
The declining effective population size (Ne < 100) and strong directional selection
that occurs in exotic pig populations have resulted in high LD between QTL and markers.
For two SNPs separated by 1 cM distance, the squared correlation of the SNP alleles (r2)
averaged over all possible intervals across the genome ranged from 0.20 – 0.35
25
(Harmegnies et al., 2006; Du et al., 2007; Amaral et al., 2008; Jafarikia et al., 2010;
Huisman et al., 2010) in the exotic population which means that approximately 2,300
evenly spaced markers are required in order to capture QTLs in a genome scan utilizing
LD assuming an estimated porcine genome size of approximately 2,300 cM (Rohrer et
al., 1996). However, Meuwissen (2009) showed by simulation that, to achieve very
accurate GBV, 10NeL markers are required, where L is the length of the genome in
Morgans. Assuming a porcine genome size of 23 Morgans and approximately Ne = 100,
then a marker density of 23,000 SNPs is required to implement GS within breed in swine.
Indigenous pig populations on the other hand are seldom researched. Effective
population size may be larger in this type of population than that observed in exotic pig
populations and LD may be very small creating a need for more markers in order to apply
GS. In a study with indigenous cattle populations in West Africa, Thévenon et al. (2007)
observed Ne of >1000 with average r2 of 0.01 for markers < 3 cM apart. They concluded
that a large number of markers will be required for a whole genome scan in this
population. Moreover, preliminary assessment of LD in indigenous Chinese breeds of
pigs showed that LD of 0.3 magnitude occurred between 0.005 – 0.05 cM distance on
average (Amaral et al., 2008). This suggests mild selection pressures in the indigenous
pig populations and confirms that more markers will be required to implement GS in
those populations.
To implement GS in any given population, it is important to assess the extent of
LD in that population. The most commonly used measure of LD for SNP markers is the
squared correlation of alleles at two locations in the genome (r2) as proposed by Hill and
Robertson (1968). This is defined as:
26
2.6
where DAB = DA1B1 = pA1B1 – pA1pB1, pA1 is the frequency of allele A1 at locus A, pB1 is the
frequency of allele B1 at the locus B and pA1B1 is the frequency of haplotype A1B1 in the
population. The breakup of LD between markers and QTL caused by recombination is
expected to lead to a reduction of accuracy of GBV across generations of selection. This
is one factor that determines the optimal frequency to re-estimate SNP effects (Calus
2010). Some studies showed that depending on the frequency of re-estimating the effects,
some genomic evaluation methods may be more suitable to obtain SNP effects whose
accuracy is more persistent across generations. For example, methods that can exploit LD
like the Bayesian approach showed better persistency of GBV accuracy across
generations (Habier et al., 2007; Zhong et al., 2009; Calus, 2010).
2.4.4 Evaluation of genome-wide selection in tropical swine breeding
Computer simulations have proved to be a powerful and effective tool for
investigating the design and efficiency of breeding programs (Hill and Caballero, 1992).
In reality, because of the cost of maintaining a breeding population for multiple
generations, the need for replicated tests and the risk of sub-optimal genetic improvement
of livestock, it is prudent to simulate any proposed breeding method or technique prior to
implementation on real populations. Consequently, several simulation studies evaluating
approaches for implementing GS in livestock populations have been reported
(Meuwissen et al., 2001; Muir, 2007; Odegard et al., 2009).
In general, these studies have evaluated GS assuming a single population that is
either derived from mutation drift equilibrium (MDE; Meuwissen et al., 2001) or under
27
Hardy-Weinberg equilibrium (HWE; Muir, 2007). Commercial swine populations
typically fall between the extreme states of MDE and HWE (Muir, 2007). Genome-wide
selection will be efficient in populations under HWE but unstable if Ne becomes small
(Smaragdov, 2009). For populations starting at MDE, Ne value is not important and
accuracy of GS will be stable for many generations (Habier et al., 2007). Other studies
have considered a multiple population context (de Roos et al., 2009; Ibánẽz-Escriche et
al., 2009; Toosi et al., 2010) but these studies did not take into account the impact of
evolutionary pressures such as selection on long term genetic response.
The application of GS for genetic improvement of pigs in tropical developing
countries has been proposed (Goddard and Hayes, 2007). Genome-wide selection will be
useful to tropical pig breeding programs in a number of ways. Development of a pedigree
recording system may no longer be required (Habier et al., 2007), but this does not
remove the need to collect phenotypic data because substantial datasets are required for
training a prediction model. The reference population could be managed under low-input
conditions and estimates of marker effects derived from this population are used to
predict GBV for selection candidates in a nucleus breeding herd (Dekkers, 2010).
Alternatively, the indigenous and exotic pigs found in the tropics can be crossed
to produce crossbred reference population used for estimating marker effects and
predicting GBV of purebreds for selection (Toosi et al., 2010), thereby improving
commercially relevant traits. Cryptic alleles in the unselected indigenous population can
be captured in a selective sweep (Dekkers, 2010) of QTLs in LD with markers using GS
and their frequencies increased in this population. Other relevant adaptive genes in the
28
tropical indigenous population will be identified by genome analysis and incorporated
into the commercial population.
As large numbers of SNPs are being discovered, cost effective methods to
genotype these SNPs have become commercially available (Ramos et al., 2009), so
genomic selection will become feasible for the average pig breeder and the genetic gain
expected will compensate for any additional cost.
2.5 CONCLUSIONS
Improving the productivity of pigs in tropical developing countries by genetically
making them more efficient for the continuously changing conditions is important.
Contrary to conventional wisdom, it may be reasonable to bypass the development of
traditional BLUP method to implement a genome-wide selection program. Published
results indicate that GS promises to increase genetic gain by increasing accuracy of GBV
for selection candidates.
Considerable challenges are expected for implementing GS in breeding programs
involving indigenous and exotic pigs under tropical conditions. These include the
applicability of available 60 K porcine SNP panel, lack of infrastructure for swine genetic
improvement, possible cross-breeding between existing populations and computational
challenges. Given the cost associated with implementing GS on real population, it is
important to evaluate this option with computer simulation of existing scenarios.
29
Figure 2.1 Trends in pig production in developed and developing countries
Source: FAO statistics, 2002 In Forsberg et al. 2005
30
Table 2.1 Ranges of accuracies of genomic breeding values across various sizes of reference population and using different statistical
methods in real dataset
Statistical methods Reference size No. of SNPs used Dairy Cattle Swine References
Ridge regression 1945 7372 0.56 to 0.71 Moser et al. (2009)
Ridge regression 1847 42576 0.15 to 0.64 Moser et al. (2010)
Ridge regression 332-637 3090-4369 0.42 to 0.73 Hayes et al. (2009)
BayesA 332-637 3090-4369 0.37 to 0.74 Hayes et al. (2009)
BayesA 1945 7372 0.56 to 0.71 Moser et al. (2009)
BayesA 2000 46741 0.50 to 0.82 Cleveland et al. (2010)
BayesA 781 39048 0.44 to 0.71 Hayes et al. (2009)
BayesA 287 39048 0.37 to 0.67 Hayes et al. (2009)
BayesB 250 18991 0.13 to 0.61 Luan et al. (2009)
BayesB 781 39048 0.44 to 0.70 Hayes et al. (2009)
BayesB 287 39048 0.37 to 0.65 Hayes et al. (2009)
BayesCπ 2000 46741 0.33 to 0.61 Cleveland et al. (2010)
Bayesian 3600 48000 0.52 to 0.82 de Ross et al. (2009)
Bayesian 1238 38055 0.55 to 0.85 Lund and Su (2009)
GBLUP 5335 38416 0.44 to 0.79 VanRaden et al. (2009)
GBLUP 781 39048 0.44 to 0.62 Hayes et al. (2009)
GBLUP 287 39048 0.41 to 0.72 Hayes et al. (2009)
GBLUP 2490 44146 0.63 to 0.82 Harris et al. (2008)
GBLUP 3966-4127 38416 0.44 to 0.79 Schenkel et al. (2009)
GBLUP 596 42598 0.56 to 0.71 Berry et al. (2009)
GBLUP 250 18991 0.13 to 0.61 Luan et al. (2009)
31
CHAPTER 3
META-ANALYSIS OF GENETIC PARAMETER ESTIMATES FOR
REPRODUCTION, GROWTH AND CARCASS TRAITS OF PIGS IN
THE TROPICS
E. C. Akanno1, F. S. Schenkel
1, V. M. Quinton
1, R. M. Friendship
2 and J. A. B.
Robinson1
1Centre for Genetic Improvement of Livestock,
Department of Animal & Poultry Science, University of Guelph, ON, Canada
2Department of Population Medicine, Ontario Veterinary College,
University of Guelph, Guelph, ON, Canada
32
3.0 ABSTRACT
The design of swine breeding programs for the tropics requires knowledge of
genetic parameters for economically important traits determined under tropical
conditions. A literature review found a total of 468 heritability (h2) estimates, 368 genetic
correlations and 254 phenotypic correlations for 35 traits across 117 peer-reviewed
articles published from 1974 to 2009 and covering tropical Africa, Southeast Asia, the
Caribbean and Latin America. A model that incorporated between and within study
variance component was used to obtain weighted means and variances for all parameter
estimates. Weighted means and standard errors of direct and maternal heritability,
common litter effects and the correlation between direct and maternal effects are
presented for various reproduction, growth and carcass traits. The weighted means and
confidence intervals for the genetic and phenotypic correlations between these traits are
also presented. Weighted least-squares analyses of the h2 estimates were performed by
fitting a number of fixed effects and covariates for each trait as appropriate for the data.
Breed, data origin, estimation method, data age and location of study were found to be
significant (P < 0.05) for at least one of the traits analysed. These results indicate the
relevance of having local, population-specific genetic parameter estimates for the tropics.
The weighted mean estimates of genetic parameters presented here are recommended for
use when reliable estimates are not available for a specific tropical pig population. This is
quite appropriate because it uses the vast resource of published genetic parameter
estimates effectively.
Keywords: Meta-analysis, heritability, genetic and phenotypic correlations, pigs, tropics
33
3.1 INTRODUCTION
Genetic parameters like heritability, genetic and phenotypic correlations and other
components of total phenotypic variance have wide application in genetic and animal
breeding research. Accurate estimation of genetic parameters in every population of
interest requires large across-generation data sets with reliable pedigree information
which are not always available. The development of animal models using REML and
Bayesian procedures (Sorensen and Gianola, 2002; Thompson, 2008) and the advances in
computing capacity have all resulted in the publication of what are expected to be better
estimates of these parameters. The advances in procedures have allowed for estimation of
maternal heritability, common litter effects and the correlation between direct and
maternal genetic effects for many traits.
Over the last four decades, several estimates of genetic and phenotypic parameters
for pig production traits determined under tropical conditions have been published. These
estimates were derived using a variety of methods, are of variable quality in terms of
sample size used to estimate them, and have contradictory results. Under these
circumstances, swine breeders and researchers may have difficulty making sense of the
literature estimates or taking stock of the knowledge base. This can have negative
consequences on breeding decisions and can hinder genetic progress.
The motivation for this study was the need for a quantitative synthesis of previous
findings in order to provide summary estimates of genetic parameters for the
development of breeding objectives for swine genetic improvement in tropical
developing countries. A statistical method known as ―meta-analysis‖ was utilized (Sutton
et al., 2000). Koots et al. (1994a) summarised estimates of genetic parameters for beef
34
production traits using a fixed effects meta-analysis method and identified significant
study characteristics that affect estimates. A random effect meta-analysis model was used
to obtain weighted estimates of genetic parameters in sheep (Safari et al., 2005) and in
beef cattle (Giannotti et al., 2005). No such review has yet been carried out for pigs
raised in tropical environments.
Using a fixed-effect method to combine estimates of genetic parameters makes
the assumption that each published estimate is estimating the same underlying true
parameter (i.e. no heterogeneity between published estimates (Barron et al., 2008) and
any difference observed is due to chance factors or sampling variation within each study.
If this assumption is invalid, as shown by Koots et al. (1994a), then a preferred
alternative method would be to fit a random effects meta-analysis model. A random-
effects model assumes that each estimate from the various studies is estimating different
underlying true parameters, and takes into account the extra variation implied in making
this assumption. This underlying true parameter is assumed to vary at random and its
effect is believed to be normally distributed. Hence, this model accounts for the between
and within-study sources of variation and has been proposed as a more conservative
method of meta-analysis that yields results that can be generalised (Berkey et al., 1995;
Sutton et al., 2000; Safari et al., 2005; Barron et al., 2008).
The objective of this study was to use a random-effect meta-analysis model to
arrive at consensus estimates of genetic and phenotypic parameters of economically
important traits of pigs determined under tropical conditions and to test for factors that
can affect these parameters in the tropics.
35
3.2 MATERIALS AND METHODS
3.2.1 Description of study scope and traits
A systematic search of the literature using electronic and non-electronic media
was conducted to identify all references reporting estimates of genetic and phenotypic
parameters for pig production traits in the tropics. A total of 117 peer-reviewed articles
with full description of methods and published between 1974 and 2009 were reviewed.
These papers reported original research that was carried out in Southeast Asia, Latin
America, The Caribbean and Tropical Africa. These regions lie in the tropics between
latitude 23.5 north and latitude 23.5 south of the equator. The climate is hot and humid
all year round with daytime temperature above 25 and relative humidity of 85%, on
average. The annual rainfall usually exceeds 2000 mm.
Traits included in the study were those considered to be economically important
for pork production in the tropics and are likely to be included in breeding objectives.
These traits were grouped into three broad categories: reproduction, growth and carcass.
A comprehensive list of the traits, units of measurement and abbreviations used are
presented in Table 3.1. Reproduction traits included litter size at farrowing (LSF),
number born alive (NBA), litter size at 21 days (LS21), litter size at weaning (LSW),
litter weight at farrowing (LWF), litter weight at 21 days (LW21), litter weight at
weaning (LWW), pre-weaning average daily gain (PDG), maternal ability (MA
= ), survival rate (SRT = ), age at first farrowing (AFF),
age at first conception (AFC), gestation length (GLT), weaning to service interval (WSI),
farrowing interval (FIT), pre-weaning mortality (MOR) and daily feed intake during
36
lactation (DFI). Weaning of piglets occurred between 5 – 8 weeks of age for all studies
with this information.
Growth traits were defined as weights at various ages: individual birth weight
(IBW), individual weight at 21 days (IW21), individual weight at weaning (IWW, 5-8
weeks of age), post-weaning individual weight (PIW, up to 6 months of age), mature
weight (MWT, > 6 months of age) and average daily gain (ADG). Other growth traits
included are days to 90 kg weight (D90), feed conversion ratio (FCR) and average daily
feed intake (ADFI).
Carcass traits included back-fat thickness on live animals using ultrasound
devices (BFU), adjusted for live weight (BFA) and on carcasses (BFT). Most fat
measurements were taken at the P2 position which is about 65 mm from the dorsal
midline along the last rib. Estimates for carcass weight (CWT), lean meat yield (LEAN),
dressing percentage (DPT), carcass length (CLT), loin eye area (LEA) and ham yield
(HYD) were also included.
3.2.2 Data recorded and analyses
The dataset constructed from the literature included direct and maternal
heritabilities, common litter effects as a proportion of the total variance, correlations
between direct and maternal genetic effects when available for some traits, as well as
genetic and phenotypic correlations among traits and published standard errors of
parameter estimates. Other information recorded were trait means and phenotypic
standard deviations, coefficients of variation reported or derived from the trait means and
37
standard deviations, number of records used, number of parents (sires and dams), author
reference and a variety of fixed factors (See model below).
In the case where estimates from the same population were reported in different
publications only the most recent was included in the analysis. All published estimates
including negative heritabilities and those published in languages other than English were
extracted and meta-analysed as a check against publication bias (Fleiss and Gross, 1991).
The number of estimates of genetic parameters published during this period in the tropics
may be greater than that available for review. Owing to the fact that most tropical
countries are still developing, this information may not be available to the public.
All parameter estimates recorded for all traits in this review were originally
analysed assuming normality. This assumption was investigated for a number of traits
with >25 estimates and their distributions were approximately normal. Potential outliers
introduced during data entry or from other sources were identified and corrected during
analysis. A meta-analysis of published results generally proceeds by weighting each
estimate of a parameter from the various studies by the amount of information used in
estimating them. Two weighting factors namely; sample size and inverse of sampling
variance were compared for a few traits using a random- effect meta-analysis model and
the results showed that both methods gave similar weighted mean heritability as observed
by Safari et al. (2005), but the standard error of estimates were smaller when the inverse
of sampling variances was used. Hedges and Olkin (1985) and Marin-Martinez and
Sanchez-Meca (2010) have shown that the optimal weight for a meta-analysis is the
inverse of the sampling variance. In addition, the inverse of sampling variance may have
38
some advantages because the statistical significance of the estimates for each trait is
determined not only by the sample size, but also by the level of variation in the trait.
Weighted mean for the trait means, standard deviations and coefficient of
variation were obtained for all traits using sample size as weights while the inverse of the
sampling variance were used as weights for obtaining weighted mean heritability,
maternal heritability and common litter effects. For some published estimates no standard
errors were reported. Approximate standard errors were derived for these estimates by
using the combined variance method (Sutton et al., 2000). First, a weighted mean
standard deviation (SDW) was obtained from the reported standard errors using Equation
3.1, where si is the standard error and ni is the number of records for the ith estimate (i =
1, 2, . . ., y). The predicted standard error for estimates with no reported standard error
was calculated by dividing SDW by the square root of the number of records in the
estimate. The number of records was available for all studies.
3.1
In order to calculate a weighted estimate of phenotypic and genetic correlations
(r) among a combination of traits and the correlation between direct and maternal genetic
effects, the published values were first transformed to an approximate normal scale by
using a Fisher‘s Z transformation (Steel and Torrie, 1960) in Equation 3.2, with the
standard error from Equation 3.3, where n is the number of records for a phenotypic
correlation and the number of sires for a genetic correlation. The weighted mean for the Z
transformed correlations was calculated across studies and back transformed using
Equation 3.4, where rw is the weighted mean correlation (phenotypic or genetic) and z is
the weighted mean for the Z transformed correlations.
39
3.2
3.3
3.4
The standard errors generated using Equation 3.3 were highly correlated with those in
studies that reported standard errors for genetic and phenotypic correlations. Similar
observations were made by other researchers (Koots et al., 1994b; Safari et al., 2005).
3.2.3 Test of heterogeneity of estimates
The first analyses performed for the published genetic parameter estimates was a
test of heterogeneity of the estimates i.e. a test of the null hypothesis that the true
parameter estimate is the same in all studies versus the
alternative (H1) that at least one of the estimates differs from the rest. Essentially, this is
testing whether it is reasonable to assume that all the studies combined are estimating a
single underlying population parameter, and whether variation in study estimates is likely
to be random. The Q test statistic devised by Cochran (1954) and described below was
used for this purpose.
3.5
where k is the number of parameter estimates ( ), and
is the weighting factor taken as the inverse of the sampling variance (s2)
(Hedges and Olkin, 1985). Q is approximately distributed as a χ2 distribution with k - 1
degrees of freedom under H0.
3.2.4 Estimation of weighted means of genetic parameters
40
The weighted means of the parameter estimates were obtained by fitting a
random-effect model for all traits studied. As mentioned earlier, this model allowed
estimates to represent random sub-populations with differing underlying true parameters
drawn from the overall population. This took into account the extra variation implied in
this assumption by including between study and within study components of variance and
used Residual Maximum Likelihood approach (ASREML program, Gilmour et al., 2009)
to determine the relative contribution of the between and within study variances to the
weights.
The random-effect model was:
θi = µ + ui + ei 3.6
where θi is the estimate of a parameter in the ith study, µ is the population mean
(weighted mean of the estimates), ui is the between study component of the deviation
from the mean and ei is the within study component due to sampling error in the ith
estimate. The components ui and ei are normally distributed with mean zero and variances
and V, respectively. For the trait means, standard deviation and coefficient of
variation, V = where ni is the number of records. For direct and maternal
heritability, and common environmental effects, V = where wi = .
For the correlations V = , where ni is the number of sires for genetic
correlation and the number of records for phenotypic correlation and correlation between
direct and maternal genetic effects. The standard errors for the weighted means and the
95% confidence intervals of the estimates were derived from V, with those for
correlations being back transformed using Equation 3.4.
41
3.2.5 Weighted least squares analyses
A weighted least squares analysis was performed for those traits with sufficient
data in order to identify important study characteristics that may affect heritability
estimates in the tropics. The fixed linear model was:
= Bi + Dj + Mk + Ll + Am + eijklmn 3.7
where is the published estimate of heritability which is assumed to be
independently and identically normally distributed (µ, V), Bi (i = 1, 2) is the breed of sire
effect grouped as exotic breeds (such as Large White, Landrace, Duroc and Hampshire
breeds) and indigenous breeds (consisting of a variety of local pigs that are indigenous to
the tropics), Dj (j = 1, 2) is the effect of data origin such as data from experimental
stations and field data, Mk (k = 1, 2) is the effect of estimation methods grouped as intra –
class correlation of half-sibs procedure or the animal model, Ll is the absolute latitude
which represents the distance of a study location from the equator and was fitted as a
covariate that accounts for variation in rainfall, humidity, environmental mycobacteria
that may produce natural immunity, temperature, and other factors that may have
biological implications on the performance of pigs under tropical conditions (Berkey et
al., 1995), Am is the age of data effect (a covariate that was obtained as number of years
of the data used in the study relative to 2009), and eijklmn is the random error also
assumed to be independently and identically normally distributed (0, V), where V =
and is . The PROC GLM procedure of the Statistical Analysis System
(SAS Institute Inc., 2009) was used to test the significance of fixed effects and covariates.
42
3.3 RESULTS AND DISCUSSION
3.3.1 Summary statistics
The weighted descriptive statistics for 35 traits considered in this review are
presented in Table 3.2. In general, litter traits such as litter size and weight at farrowing
and weaning, and some growth traits as well as back fat thickness on carcass are all
equally likely to be recorded which suggest the importance of these traits in swine
production systems in the tropics. Exotic breeds of pigs have more estimates of these
statistics than indigenous breeds. Their performance for many of these traits was two
times greater and differed significantly (P < 0.05) from that reported for indigenous
breeds. This may be connected with the general acceptability of exotic breeds of pigs for
commercial pork production in the tropics (Egbunike, 1986).
The weighted standard deviation (SD) and coefficient of variation (CV) shows
more variability in reproduction and growth traits than in carcass traits (Table 3.2).
Carcass traits are less affected by environmental variations and have high heritability.
The SD and CV for litter traits tend to increase as the animal ages but this trend is in the
opposite directions for growth traits. This may be due to competition among piglets as
lactation advances which introduces lots of variability within litter. Because the statistics
presented in Table 3.2 are subjected to wide environmental variation, they may not be
appropriate for breeding decisions unless there is stability in management and uniformity
in environmental effects across generations, populations and countries which are
43
unrealistic in the tropics. Local means and phenotypic standard deviations are therefore
recommended for use where possible.
3.3.2 Heritability and 95% confidence intervals
The number of contributing estimates, weighted mean heritability and standard
error (h2±S.E), between study variance (σ
2bs) expressed as a percentage of the total
variance, Q statistics and 95% confidence intervals (CI) for 35 traits studied are showed
in Table 3.3. Litter traits, average daily gain, and individual weight at farrowing and
weaning and back-fat thickness all have > 10 estimates of heritability contributing to the
weighted mean. Reproduction traits had low (<0.15) heritability estimates for ten traits,
moderate (>0.15 to <0.40) heritability estimates for six traits and high (>0.40) estimates
of heritability for one trait. In the growth category, one trait had a low estimate of
heritability while the rest in the group were moderately heritable. Carcass traits have high
heritability estimates except for two traits that were moderate. The standard errors of the
weighted mean heritability for many of the traits in this study were low and consistent
with the results obtained by Safari et al. (2005) in a related study involving sheep. This
means that the weighted mean heritability estimates from this study are precise.
There were significant differences (P < 0.05) among published estimates of
heritability for 21 traits with appreciable amounts of between study variations (Table
3.3). This suggests the importance of accounting for this variation in the model used to
estimate weighted means in order to obtain estimates that are at least 95% precise. The
confidence intervals for the weighted mean heritability were however, wider than when a
fixed effect model was used. This was due to the inclusion of additional sources of
44
variation in deriving these intervals. In some cases, these intervals were outside the
boundary of expected intervals of heritability estimates.
The low heritability estimates observed for most reproduction traits were
consistent with our knowledge about the inheritance of these traits. These traits are
largely influenced by uterine (Lee and Haley, 1995) and maternal environment effects
(Chimonyo et al., 2008) as well as management practices rather than an individual‘s
additive genetic value. For many of the traits in this category, the heritability estimates
presented in Table 3.3 were within the range reported for reproduction traits in a review
by Rothschild (1996). Growth traits in general were moderately heritable (0.15±0.023 –
0.38±0.039) except for heritability of individual weight at 21 days which was low
(0.08±0.029). The heritability estimates for body weight at various ages tended to
increase from birth to adult (Table 3.3) approximating to a growth curve. The increase in
heritability with age has been attributed to certain genes being turned on and off as an
animal ages (Meyer, 1998). Similar estimates of heritability for some of the growth traits
and the observed trend have been reported in a review by Clutter and Brascamp (1998)
and also in a related study in sheep (Safari et al., 2005).
The weighted mean heritability for carcass traits in this study was moderate to
highly heritable and range from 0.33±0.045 to 0.60±0.117. The estimates for these traits
were in close agreement with the values reported in a review by Sellier (1998). The high
heritability estimates of carcass traits observed in this study means that relatively small
numbers of estimates are required for precision of genetic merit and an individual‘s
phenotypic performance can be a good predictor of its genetic background. Moreover,
measurement of carcass traits often requires the slaughter of animals resulting in fewer
45
estimates of genetic parameters in literature. In recent times, the use of ultrasound
(Wilson, 1992) for predicting back-fat thickness in live animals has increased the
frequency of evaluating for this trait. The values of heritability estimates of back-fat
thickness obtained by this method (BFU, 0.33±0.045) were slightly lower than that
obtained on carcass (BFT, 0.44±0.037). Adjustment of live back-fat thickness (BFA) for
live weight considerably increased the heritability estimates on live animals to
0.46±0.050.
3.3.3 Heritability within breed, maternal heritability and common environmental
effects
The weighted mean heritability obtained within breed groups for all traits studied
are presented in Tables 3.4 and 3.5. Exotic breeds of pigs have higher heritability
estimates for litter size at various ages than indigenous breeds while the reverse is the
case for litter weight at various ages. The numbers of estimates contributing to the mean
heritability for reproduction traits in indigenous breeds are fewer. The exotic breed group
includes Large White and Landrace which are known for high reproduction performance
in the tropics (Dan and Summers, 1996). In addition, indigenous breeds have
demonstrated good mothering ability which could lead to heavier litters in later lactations
(Adebambo and Dettmers, 1982).
Heritability estimate was low for IBW (0.10±0.016) but high for PIW
(0.49±0.199), MWT (0.41±0.046) and FCR (0.42±0.020) in the indigenous breeds while
exotic breeds had moderate heritability estimates for growth traits except for IW21 which
was low (0.02±0.014). The high heritability for FCR in the indigenous breeds supports
46
the finding that they are better converters of feed to pork than exotic breeds (Mushandu et
al., 2005), although this estimate came from only two studies. Estimate of heritability for
ADG were higher in indigenous (0.37±0.073) than in exotic (0.28±0.022) breeds. The
higher value of heritability estimate for ADG observed in the indigenous breeds may be
connected to the limited selection for this trait that have occurred in this population and
suggest room for genetic improvement. Also, their inherent ability to use less feed for
pork production is an economic advantage that can be exploited.
For the various breed groups, heritability estimates of carcass traits tend to be
higher for indigenous breeds (0.36±0.089 – 0.82±0.092) than for their exotic counterparts
(0.25±0.018 – 0.49±0.082). More importantly is the higher inherent tendency for
indigenous breeds to lay down body fat (0.63±0.208) which brings to question the quality
of their carcass given the health implication of fat to humans. Genetic improvement
programs aimed at reducing back-fat thickness in this population have been strongly
advocated (Adebambo, 1986; Chimonyo and Dzama, 2007).
The weighted maternal heritability estimates for reproduction traits were low and
ranged from 0.01±0.005 in LW21 to 0.13±0.049 in PDG (Table 3.4). For live weight at
various ages, maternal heritability tended to increase with age contrary to what is
expected (Table 3.5). For single studies that included maternal genetic effect in the model
for evaluating carcass traits, the estimates were low and ranged from 0.03±0.002 in CWT
to 0.11±0.001 in CLT. The correlations between direct and maternal genetic effects in
this study were mostly negative and strong for most traits analysed (Tables 3.4 and 3.5)
except for strong and positive correlations observed for LSF (0.64±1.407), NBA
(0.73±0.837), LW21 (0.85±2.530) and PDG (0.44±1.220), respectively. However, these
47
estimates were associated with large standard errors. A similar observation was made by
Safari et al. (2005) in sheep.
Maternal genetic effects define the influence of the dam‘s genotype on the
expression of its progeny phenotypes. In an animal model context maternal estimate
depends largely on the data structure and the model used for analysis (Meyer, 1992).
Several workers have reported an inflated direct heritability for litter traits in swine when
maternal effects were included in the model (Irgang et al., 1994; Chimonyo et al., 2008) .
Also, inclusion of maternal genetic effects for the evaluation of carcass traits was studied
by Chimonyo and Dzama (2007). The increase in maternal heritability with age of
progeny as observed for growth traits in this study (Table 3.5) may be due to data
structure used in the primary studies. A negative correlation between direct and maternal
genetic effects has been shown to be a product of both actual genetic antagonism and sire
by year interaction (Konstantinov and Brien, 2003; Maniatis and Pollott, 2003) and can
be influenced by data structure (Maniatis and Pollott, 2003).
Tables 3.4 and 3.5 also showed significant common litter effects for many of the
traits analysed. Litter-bearing species usually have a large number of non-additive
relationships (Norris et al., 2007) and common litter effects should be accounted for in
the model for analysis of traits (Chimonyo et al., 2006; Chimonyo and Dzama, 2007;
Ilatsia et al., 2008).
3.3.4 Factors affecting heritability estimates
The result of the weighted least squares analysis of heritability estimates for nine
traits with sufficient data are presented in Table 3.6. Data origin, breed group, data age,
48
absolute latitude and estimation method significantly (P < 0.05) affected most of the traits
analysed. The R2 values for the model ranged from 0.25 – 0.84. Data origin affected
heritability estimates for NBA, LW21, ADG and BFT. This was expected for some of
these traits because environmental variation is better accounted for and controlled in
experimental data than in field data (Baker et al., 1984). Breed group affected LW21,
IBW and BFT. Exotic breeds of pigs such as Large White are known to produce heavier
litters at birth while indigenous breeds generally produce fatty carcasses (Adebambo,
1986). Methods for estimating heritability affected LS21, LWW and IWW. Animal
model methodology yielded better estimates than paternal half sib correlation because it
utilises full pedigree information, allows for isolation of other sources of variation and
also accounts for possible selection in the population (Thompson, 2008).
The inclusion of covariates such as data age relative to 2009 and location of a
study defined by absolute latitude in the model for analysis significantly (P < 0.05)
affected heritability estimates for most traits analysed. Usually, an estimate of heritability
is defined for a particular trait measurement, taken at a particular time or age, in a given
population. Variation in any of these measurement standards will affect heritability
estimates. Differences in estimates may also be connected with the level of technological
advancement in each location, management practices or other environmental variation.
Use of weighted mean heritability estimates for genetic evaluation will be
appropriate if factors contributing to variation in heritability estimates are essentially
random and unidentifiable. However, a random effect model as used in this study
accounts for the existence of other sources of variation besides random error. The
49
weighted mean heritability estimates in Table 3.3 are therefore, recommended for use
because they give the most weight to estimates with the highest accuracy.
3.3.5 Genetic and phenotypic correlation
The weighted mean genetic and phenotypic correlations among all traits studied
are presented in Tables 3.7 to 3.11. Of the important trait combinations, a large number
of correlations were not available in the literature. The number of correlation estimates
contributing to the weighted mean genetic and phenotypic correlation for the available
trait pairs ranged from 1 – 22. There were more estimates contributing to genetic
correlations than phenotypic correlations. The various trait categories showed varying
degrees of positive genetic correlations among themselves than between groups.
Reproduction vs. growth tended to have more positive and high genetic correlations
among pairs of traits than growth vs. carcass and reproduction vs. carcass which had
negative and low estimates, on average. The trend was similar for phenotypic correlation
estimates.
Genetic and phenotypic correlations among important trait combinations are
required for constructing complete variance – (co)variance matrices for joint genetic
evaluation of reproduction, growth and carcass traits. The numbers of estimates for
possible pairs of traits available in the literature for tropical pigs were very few. About
one-fifth of the estimates came from a single study. Correlations among reproduction
traits tend to have more estimates followed by growth traits and then carcass traits.
The weighted genetic and phenotypic correlations among reproduction traits
ranged from -0.96 to 0.87 and -0.67 to 0.98, respectively (Table 3.7). Similar results have
50
been observed for mean correlations among reproduction traits in sheep (Safari et al.,
2005) and beef cattle (Koots et al., 1994b). Important genetic correlations were those
between LSF and LS21 (0.80), LS21 and LSW (0.87), and LW21 and LWW (0.76)
which were high and positive. These estimates suggest that selecting for litter size at
farrowing will lead to higher numbers of post-natal live piglets as well as more viable
piglets at weaning. At 21 days a pig starts to develop active immunity against diseases in
its production environment (Kyriazakis and Whittemore, 2006) which has a huge bearing
on its performance at weaning and post-weaning.
Growth traits had high estimates of genetic (-1.00 – 1.00) and phenotypic (-0.97 –
0.88) correlations among themselves (Table 3.8). This was expected because many of the
growth traits were measures of growth at different ages. The weighted genetic
correlations were higher for live weights at adjacent age classes and increased with age
from birth to adult. Weighted phenotypic correlations follow a similar trend. Individual
birth weight had a high genetic correlation with weights at 21 days (0.44) and at weaning
(0.53), while association with post-weaning individual weight was moderate (0.19). An
exception to the genetic correlations among growth traits is that between ADG and FCR
(-0.78) and between ADFI and FCR (0.68). The genetic correlation was negative and
high, and positive and high, respectively, because an autocorrelation between a ratio and
its denominator is expected to yield negative values and that between a ratio and its
numerator are positive (Pearson, 1900). In addition, fast growing animals are generally
more feed efficient.
Carcass traits tend to have more negative genetic and phenotypic correlations
among member traits (Table 3.9). Most negative correlations occurred between BFT and
51
other carcass traits. The ranges for weighted mean genetic and phenotypic correlations
between BFT and other carcass traits were -0.82 to 0.31 and -0.48 to 0.03, respectively.
BFT had negative correlation with LEAN (-0.82), CWT (-0.18) and HYD (-0.11) but
positively correlated with DPT (0.31) and LEA (0.21). DPT was highly and positively
correlated to CWT (0.91), LEA (0.68) and CLT (0.56). The weighted mean phenotypic
correlations were generally similar or lower than the corresponding genetic correlations.
These findings were in agreement with comparable traits in previous studies (Koots et al.,
1994b; Sellier, 1998; Safari et al., 2005)
The weighted genetic correlations among reproduction vs. growth traits were
mostly negative for litter size and live weights at various ages but positive for litter
weight and live weights at various ages (Table 3.10). The genetic antagonism between
litter size and live weights at various ages means that selecting for increased litter size at
farrowing will lead to lower live weight at later ages. However, because litter size is
largely controlled by environment, the reduction in live weight at later ages may be
amenable to management practices. The association between ADG and reproduction
traits were weak suggesting independence of these traits. The correlations among
reproduction vs. carcass traits were few in the literature. The available genetic
correlations were those that occurred between PDG and BFT (0.05), PDG and CWT
(0.08), and PDG and CLT (0.35) (Table 3.10). The phenotypic correlations among groups
of traits are presented in Table 3.11 and were low and negligible for many of the trait
combinations. The phenotypic correlations between growth and carcass traits ranged
from -0.50 to 0.68. The genetic and phenotypic correlations observed between growth
52
and carcass traits in this study suggest that these traits should be considered
simultaneously in a selection index.
The confidence intervals for the weighted mean genetic and phenotypic
correlation estimates (Tables 3.7 to 3.11) were wide for many of the trait pairs. This
reflects the general large variation between published estimates and the relative small
datasets that resulted in large standard errors for the individual estimates.
3.3.6 Examination of publication bias in heritability estimates
Publication bias is a term used to define what happens when a sample of
published estimates of a given parameter from the literature are not true representatives
of all possible estimates from completed studies. This may have an adverse effect on the
validity of the results from a meta-analytic review (Rothstein et al., 2005). In the case of
a genetic parameter like heritability, this bias may arise from several sources among
which includes; (1) non-acceptance by reviewers and journal editors of genetic parameter
estimates from small samples; (2) non-publication by authors of estimates with negative
variance components often referred to as ‗the file drawer problem‘; 3) publication of
results in a non-conventional language (language bias); (4) non-publication of results
from PhD and MSc dissertations and results presented at conferences (Sutton et al.,
2000); and (5) non-publication of results that are against the interest of funding agencies.
In this study, effort was made to identify and include in the meta-analysis,
estimates of heritability from a range of sources that were judged to have been peer-
reviewed and including conference proceedings if appropriate. Negative heritability
estimates were also included. Usually negative variance components will arise from
53
methods that are not constrained like ML and REML procedures to yield positive values
within the parameter space. Figures 3.1 gives a visual assessment of the presence or
absence of publication bias for heritability estimates involving two traits with sufficient
data for assessment using a funnel plot (Sutton et al., 2000) . Heritability estimates were
plotted against the inverse of sampling variance to give a shape like a funnel, if there is
no publication bias (Figure 3.1A; Macaskill and Walter, 2001). Generally, estimates from
smaller samples will be more widely spread around the mean estimate because of large
random errors. A skewed or asymmetric shape of the funnel suggests a likely publication
bias (Figure 3.1B). Because the existence of a bias in this study was negligible no effort
was made to investigate them.
3.4 CONCLUSIONS AND RECOMMENDATIONS
Among the important population parameters, heritability and genetic correlations
are the most difficult to estimate due to their requirements for sophisticated computing
techniques. The problem of variations in literature estimates of these parameters for
animal genetic evaluation has been stressed. Improving the precision of parameter
estimates by using a meta-analysis is possible. In most tropical developing countries,
there is an urgent need to provide accurate estimates of genetic parameters and in
particular correlations among groups of production traits with high economic value, such
as reproduction, growth and carcass traits. These parameters are required for accurate
genetic evaluation of pigs and the development of an optimum breeding objective and
selection indices that will have outcome that can be predicted with confidence. The
54
results from this study will address some of these needs and will serve as a basis for
comparing genetic parameter estimates in the tropics in the future.
The weighted mean estimates of genetic parameters presented here are
recommended for use when reliable estimates are not available for a specific tropical pig
population. Where estimates within a population are available, these could be combined
with the weighted literature average by simply weighting each estimate by the inverse of
the sampling variance. This would yield a population specific estimate that incorporates
prior values. The result presented in this study should allow researchers in developing
countries to extract genetic parameter estimates suitable for a wide range of tropical
conditions. The data base of published parameters from this environment and the
accompanying references are available for extraction of specific estimates in portable
format from the author.
55
Table 3.1 List of traits studied, abbreviations used and units of measurement
Traits Abbreviations Units
Reproduction
Litter size at farrowing LSF piglets
Number born alive NBA piglets
Litter size at 21 days LS21 piglets
Litter size at weaning (5 – 8 weeks) LSW piglets
Litter weight at farrowing LWF kg
Litter weight at 21 days LW21 kg
Litter weaning weight (5 – 8 weeks) LWW kg
Pre-weaning daily gain PDG g/d
Pre-weaning mortality MOR piglets
Maternal ability MA ratio
Survival rate SRT %
Age at first farrowing AFF d
Age at first conception AFC d
Gestation length GLT d
Weaning to service interval WSI d
Farrowing interval FIT d
Daily feed intake during lactation DFI Kg/d
Growth
Average daily gain ADG g/d
Individual birth weight IBW kg
Individual weight at 21 days IW21 kg
Individual weaning weight (5 – 8 weeks) IWW kg
Post weaning individual weight ( 6 months) PIW kg
Mature weight (> 6 months) MWT kg
Days to 90 kg weight D90 d
Feed conversion ratio FCR ratio
Average daily feed intake ADFI kg/d
Carcass
Back fat ultrasonically measured BFU mm
Back fat adjusted for live weight BFA mm
Back fat thickness on carcass BFT mm
Percent lean meat LEAN %
Carcass weight CWT kg
Carcass length CLT cm
Dressing percentage DPT %
Loin eye area LEA cm2
Ham yield HYD kg
56
Table 3.2 Number of literature estimates (N), weighted least square means (Mean±S.E1),
standard deviation (SD) and coefficient of variation (CV) for reproduction, growth and carcass
traits of pigs in the tropics
Traits Indigenous breed Exotic breed Overall
N Mean±S.E N Mean±S.E N Mean SD CV±S.E (%)
Reproduction LSF 2 7.35±0.97
a 23 9.77±0.16b 25 9.94 2.88 28.71±1.53
NBA 1 8.17±2.24 17 9.24±0.14 18 9.23 2.78 29.64±0.82
LS21 - - 15 9.17±0.16 15 9.17 2.52 27.55±1.58
LSW 3 8.90±0.43 17 8.19±0.28 20 8.39 1.80 24.44±3.33
LWF 2 7.18±2.16a 15 14.74±0.48
b 17 14.38 4.93 32.12±1.51
LW21 1 18.66±20.03 14 51.41±1.71 15 51.17 11.48 25.74±2.96
LWW 3 39.44±18.24 11 62.01±6.25 14 59.64 16.43 29.24±3.81
PDG 2 156.53±21.09 6 199.91±6.96 8 195.64 56.31 29.82±6.26
MA 1 0.13 1 0.90 2 0.60 0.10 16.03±0.64
SRT - - 1 84.60 1 84.60 21.18 25.03
AFF - - 1 387.01 1 387.01 3.11 0.80
AFC - - 2 254.43±7.64 2 254.43 23.17 5.93±4.76
GLT - - 3 114.50±1.00 3 114.50 4.50 3.93±1.62
WSI - - 6 7.16±0.89 6 7.16 5.98 85.70±4.95
FIT 1 140.90±13.75 3 169.26±2.84 4 168.10 47.34 28.02±3.58
DFI - - 2 5.51±0.55 2 5.51 1.37 25.94±1.95
Growth
ADG 2 335.43±171.58a 16 792.51±29.06
b 18 779.77 101.79 15.48±1.97
IBW 5 0.69±0.05a 19 1.43±0.03
b 24 1.28 0.34 32.86±9.09
IW21 1 2.20±0.95a 4 5.64±0.26
b 5 5.39 1.05 20.72±3.96
IWW 5 6.26±0.71a 18 10.23±0.46
b 23 9.03 2.04 24.46±2.32
PIW 1 19.23±6.33a 5 45.99±4.09
b 6 38.10 8.67 24.26±3.61
MWT - - 4 94.91±7.70 4 94.91 10.77 17.67±3.00
D90 - - 7 156.99±8.87 7 156.99 14.00 8.74±0.61
FCR - - 4 2.86±0.04 4 2.86 0.35 11.65±1.37
ADFI - - 1 2.71 1 2.71 0.04 1.40
Carcass
BFT 2 21.47±8.09 11 18.58±1.46 13 18.68 4.57 25.62±2.59
BFA - - 8 11.63±0.85 8 11.63 2.79 23.36±3.84
BFU 1 13.15±0.55 2 12.11±0.41 3 12.47 1.90 16.02±1.42
LEAN - - 3 54.25±2.99 3 54.25 3.63 6.53±0.11
CWT 2 23.81±4.74 1 48.90±14.42 3 26.26 3.85 10.73±5.08
CLT 3 52.55±2.20a 4 96.99±2.32
b 7 73.58 3.67 5.90±1.82
DPT 3 63.74±2.54a 2 80.91±2.37
b 5 72.92 3.92 5.91±2.87
LEA 2 21.37±2.46a 4 33.71±1.24
b 6 31.22 3.85 12.28±1.92
HYD 1 15.86±18.36 3 18.56±6.22 4 18.28 1.44 10.30±4.76 ab
means within row differ significantly (P < 0.05); 1S.E = standard error
57
Table 3.3 Number of literature estimates (N), combined (exotic and indigenous)
weighted mean heritability (h2), between study variance (σ
2bs), Q statistics and 95%
confidence interval (CI) for economic traits of pigs in the tropics
Traits N h2±S.E σ
2bs
1 Q 95% CI for h
2
Reproduction
LSF 32 0.08±0.015 0.23 142.91*** 0.05 to 0.11
NBA 29 0.08±0.008 0.83 91.44*** 0.06 to 0.10
LS21 19 0.07±0.014 0.06 28.29ns
0.04 to 0.10
LSW 19 0.07±0.012 0.49 27.99ns
0.04 to 0.10
LWF 23 0.11±0.024 0.45 53.97*** 0.06 to 0.16
LW21 19 0.13±0.014 0.18 84.97*** 0.10 to 0.16
LWW 14 0.07±0.024 0.00 31.02** 0.02 to 0.12
PDG 9 0.19±0.053 1.13 155.10*** 0.06 to 0.32
MA 2 0.27±0.308 9.09 3.97* -
SRT 3 0.62±0.177 0.00 1.55ns
-
AFF 2 0.23±0.203 9.09 2.58ns
-
AFC 1 0.21±0.030 - - -
GLT 4 0.25±0.024 0.00 1.48ns
0.17 to 0.33
WSI 6 0.08±0.028 1.99 24.01*** 0.01 to 0.15
FIT 4 0.02±0.003 0.00 0.65ns
0.01 to 0.03
DFI 3 0.08±0.037 9.10 7.12* -
MOR 3 0.20±0.070 - 9.29** -
Growth
ADG 26 0.28±0.020 0.32 79.00*** 0.24 to 0.32
IBW 38 0.15±0.023 0.22 173.17*** 0.10 to 0.20
IW21 9 0.08±0.029 0.25 136.35*** 0.01 to 0.15
IWW 32 0.22±0.025 0.33 195.87*** 0.17 to 0.27
PIW 11 0.38±0.039 0.00 4.17ns
0.29 to 0.47
MWT 8 0.23±0.041 0.00 8.19ns
0.13 to 0.33
D90 9 0.28±0.035 4.70 38.48*** 0.20 to 0.36
FCR 11 0.32±0.039 - 25.12** 0.23 to 0.41
ADFI 1 0.37±0.090 - - -
Carcass
BFT 17 0.44±0.037 0.00 37.63** 0.36 to 0.52
BFU 10 0.33±0.045 1.09 73.14*** 0.23 to 0.43
BFA 8 0.46±0.050 0.03 153.77*** 0.34 to 0.58
LEAN 4 0.42±0.023 0.00 0.49ns
0.35 to 0.49
CWT 4 0.41±0.128 0.00 7.63ns
0.00 to 0.82
CLT 10 0.39±0.054 - 31.52*** 0.27 to 0.51
DPT 6 0.60±0.117 - 10.96ns
0.30 to 0.90
LEA 9 0.51±0.062 4.24 6.82ns
0.37 to 0.65
HYD 7 0.45±0.075 - 37.04*** 0.27 to 0.63 1Expressed as a percentage of the total variation;
ns non-significant (P > 0.05); * P < 0.05; ** P <
0.01;*** P < 0.001
58
Table 3.4 Number of literature estimates (N), weighted mean heritability for indigenous (h2
I) and exotic (h2
E) breeds of pigs, maternal
heritability (h2
M), correlation between direct and maternal genetic effects (ram) and common litter effect (l2) for reproduction traits of
pigs in the tropics
Traits N h2
I±S.E N h2
E±S.E N h2
M±S.E N ram±S.E N l2±S.E
Reproduction
LSF 2 0.02±0.001 30 0.09±0.017 5 0.01±0.005 5 0.64±1.407 3 0.06±0.027
NBA 3 0.07±0.020 26 0.08±0.008 8 0.03±0.004 7 0.73±0.837 3 0.01±0.012
LS21 - - 19 0.07±0.014 8 0.01±0.005 8 -0.66±0.855 8 0.04±0.014
LSW 3 0.06±0.031 16 0.07±0.012 1 0.03±0.020 - - - -
LWF 2 0.22±0.049 21 0.11±0.026 4 0.03±0.011 3 -0.98±3.074 3 0.12±0.010
LW21 1 0.18±0.119 18 0.13±0.014 3 0.01±0.003 3 0.85±2.530 3 0.30±0.244
LWW 3 0.19±0.028 11 0.07±0.026 1 0.06±0.002 - - - -
PDG 2 0.18±0.032 7 0.22±0.070 5 0.13±0.049 4 0.44±1.220 4 0.03±0.010
MA 1 0.05±0.013 1 0.71±0.330 - - - - - -
SRT - - 3 0.62±0.177 - - - - -
AFF - - 2 0.23±0.203 - - - - - -
AFC - - 1 0.21±0.030 - - - - - -
GLT - - 4 0.25±0.024 - - - - - -
WSI - - 6 0.08±0.028 - - - - - -
DFI - - 3 0.08±0.037 - - - - - -
MOR - - 3 0.20±0.070 - - - - - -
59
Table 3.5 Number of literature estimates (N), weighted mean heritability for indigenous (h2
I) and exotic (h2
E) breeds of pigs, maternal
heritability (h2
M), correlation between direct and maternal genetic effects (ram) and common litter effect (l2) for growth and carcass
traits of pigs in the tropics
Traits N h2
I±S.E N h2
E±S.E N h2
M±S.E N ram±S.E N l2±S.E
Growth
ADG 4 0.37±0.073 22 0.28±0.022 4 0.02±0.004 4 -0.35±0.110 6 0.12±0.033
IBW 8 0.10±0.016 30 0.18±0.032 4 0.03±0.017 4 -0.19±0.647 4 0.08±0.044
IW21 2 0.17±0.034 7 0.02±0.014 4 0.06±0.031 3 -0.01±2.859 4 0.06±0.031
IWW 8 0.18±0.027 24 0.24±0.036 3 0.23±0.014 2 0.09±0.678 2 0.09±0.050
PIW 3 0.49±0.199 8 0.36±0.038 1 0.32±0.001 1 -0.01±5.920 1 0.06±0.040
MWT 2 0.41±0.046 6 0.22±0.044 - - - - 2 0.20±0.050
D90 - - 9 0.28±0.035 1 0.09±0.001 - - 2 0.18±0.090
FCR 2 0.42±0.020 9 0.30±0.045 - - - - - -
ADFI - - 1 0.37±0.090 - - - - - -
Carcass
BFT 3 0.63±0.208 14 0.44±0.039 - - - - 2 0.05±0.045
BFU 3 0.52±0.001 7 0.25±0.018 1 0.04±0.001 1 -0.28±0.006 2 0.06±0.050
BFA - - 8 0.46±0.050 1 0.08±0.001 - - 1 0.08±0.065
LEAN - - 4 0.42±0.023 - - - - - -
CWT 2 0.66±0.340 2 0.34±0.070 1 0.03±0.002 1 -0.28±0.047 1 0.18±0.002
CLT 4 0.36±0.089 6 0.42±0.067 1 0.11±0.001 1 -0.48±0.009 1 0.04±0.001
DPT 3 0.82±0.092 3 0.42±0.113 - - - - - -
LEA 3 0.58±0.099 6 0.49±0.082 - - - - - -
HYD 3 0.48±0.014 4 0.42±0.136 - - - - - -
60
Table 3.6 Levels of significance for the fixed factors in the analysis of heritability estimates
of some traits of pigs in the tropics
Traits N R2 Data Origin Breed Group Method Data Age Latitude
LSF 32 0.25 *
NBA 29 0.53 ** * ***
LS21 19 0.57 * ** *
LW21 19 0.83 *** * ** **
LWW 14 0.84 ** * ***
ADG 26 0.61 * * *
IBW 38 0.28 * *
IWW 32 0.54 ** *
BFT 17 0.50 ‡ ‡ ‡
‡ P < 0.1; * P < 0.05; ** P < 0.01; *** P < 0.001
61
Table 3.7 Weighted means of estimates from literature for genetic (above diagonal) and phenotypic (below diagonal) correlations among reproduction
traits of pigs in the tropics with the number of estimates in parenthesis ( ) and the 95% confidence interval in brackets [ ]
Traits LSF LS21 LSW LWF LW21 LWW WSI MOR PDG AFF AFC GLT DFI MA SRT
LSF 0.80 (6)
[0.62, 0.90]
0.70 (11)
[0.54, 0.81]
0.55 (11)
[-0.16, 0.89]
0.55 (10)
[0.33, 0.71]
0.08 (3)
[0.03, 0.14]
0.05 (3)
[-0.04, 0.14]
0.64 (5)
[-0.23, 0.94]
0.42
(1)
0.19
(1)
-0.48
(1)
-0.17
(1)
0.19
(1)
- -
LS21 0.82 (2)
[-1.00, 1.00]
0.87 (3)
[-0.91, 1.00]
0.69 (4)
[0.45, 0.83]
0.85 (4)
[0.78, 0.90]
0.52 (1) - -0.21 (1) - - - - - - -
LSW 0.74 (7)
[0.21, 0.93]
0.98 (3)
[0.89, 1.00]
0.55 (5)
[0.45, 0.64]
0.74 (5)
[-0.30, 0.98]
0.52 (5)
[-0.58, 0.95]
0.29 (1) -0.35 (4)
[-0.87, 0.58]
- - - 0.34
(1)
0.27
(1)
0.62
(1)
0.37 (2)
[-0.46,
0.86]
LWF 0.73 (7)
[0.55, 0.86]
0.50 (2)
[0.21, 0.71]
0.68 (3)
[0.37, 0.85]
0.65 (8)
[0.56, 0.73]
0.11 (3)
[-0.24, 0.43]
0.09 (1) 0.19 (4)
[-0.36, 0.64]
- 0.14
(1)
- - - -0.10
(1)
-
LW21 0.59 (3)
[0.21, 0.82]
0.81 (3)
[0.49, 0.94]
0.76 (2)
[-1.00, 1.00]
0.62 (3)
[0.40, 0.77]
0.76 (2)
[-0.91, 1.00]
- -0.64 (4)
[-0.92, 0.07]
- - - - - - -
LWW 0.54 (5)
[0.10, 0.81]
0.81 (1) 0.86 (4)
[0.75, 0.93]
0.46 (3)
[-0.43, 0.90]
0.81 (2)
[-0.78, 1.00]
0.05 (1) -0.48 (1)
- - - - - -0.21
(1)
0.42 (2)
[0.30,
0.54]
WSI 0.03 (2)
[-0.28, 0.33]
- 0.00 (1) 0.01 (1) - 0.00 (1) - - - - - - - -
MOR 0.17 (1) -0.59 (1) -0.28 (1) -0.28 (1) -0.65 (1) -0.67 (1) - - - - - - - -
PDG - - - - - - - 0.35
(1)
- -0.96
(1)
- -
AFF -0.03 (1) - - 0.00 (1) - - - - - - - - -
AFC - -0.05 (1) - - - 0.07 (1) - - - - - - -
GLT -0.06 (1) - 0.05 (1) - - - - - - - - 0.10
(1)
- -
DFI 0.02 (1)
- 0.11 (1) - - - - - 0.33
(1)
- - 0.02
(1)
- -
MA - - -0.34 (1) - -0.19 (1) -0.26 (1) - - - - - - - -
SRT - - 0.48 (2)
[0.11, 0.73]
- - 0.48 (2)
[-0.61, 0.94]
- - - - - - - -
62
Table 3.8 Weighted means of genetic (above diagonal) and phenotypic (below diagonal) correlations among growth traits of pigs in the
tropics with the number of estimates in parenthesis ( ) and the 95% confidence interval in brackets [ ]
Traits ADG IBW IW21 IWW PIW MWT D90 FCR ADFI
ADG 0.65 (1) - 0.87 (2)
[-1.00, 1.00]
- 1.00 (1) -1.00 (1) -0.78 (10)
[-0.91,-0.54]
0.27 (1)
IBW - 0.44 (4)
[-0.01, 0.74]
0.53 (13)
[0.13, 0.78]
0.19 (8)
[-0.21, 0.54]
0.44 (7)
[0.17, 0.65]
- - -
IW21 - 0.45 (3)
[-0.24, 0.84]
0.86 (3)
[0.45, 0.97]
0.78 (2)
[-1.00, 1.00]
0.95 (1) - - -
IWW 0.11 (1) 0.43 (12)
[0.34, 0.51]
0.71 (3)
[0.19, 0.92]
0.67 (8)
[0.48, 0.81]
0.58 (7)
[0.10, 0.84]
- - -
PIW - 0.23 (8)
[-0.06, 0.47]
0.66 (2)
[-1.00, 1.00]
0.49 (8)
[0.22, 0.70]
0.86 (6)
[0.56, 0.96]
- - -
MWT 0.98 (1) 0.19 (7)
[-0.01, 0.36]
0.88 (1) 0.43 (7)
[0.13, 0.66]
0.69 (6)
[0.39, 0.86]
-1.00 (1) - -
D90 -0.96 (1 - - - - -0.97 (1) - -
FCR -0.83 (4)
[-0.97, -0.79]
- - - - - - 0.68 (1)
ADFI 0.63 (1) - - - - - - 0.34 (1)
63
Table 3.9 Weighted means of genetic (above diagonal) and phenotypic (below diagonal) correlations among carcass traits of pigs in
the tropics with the number of estimates in parenthesis ( ) and the 95% confidence interval in brackets [ ]
Traits BFT LEAN CWT CLT DPT LEA HYD
BFT -0.82 (4)
[-0.94, -0.51]
-0.18 (2)
[-0.89, 0.89]
0.03 (6)
[-0.35, 0.41]
0.31 (3)
[-0.81, 0.90]
0.21 (4)
[-0.91, 0.96]
-0.11 (3)
[-0.44, 0.25]
LEAN -0.48 (3)
[-0.74, -0.10]
- 0.31 (2)
[-1.00, 1.00]
- 0.41 (1) -0.23 (1)
CWT - - 0.50 (2)
[-0.98, 0.99]
0.91 (1) 0.70 (2)
[-1.00, 1.00]
-
CLT -0.47 (1) 0.15 (1) - 0.56 (3)
[-0.63, 0.96]
0.25 (5)
[-0.44, 0.75]
-0.03 (2)
[-0.39, 0.34]
DPT - -0.05 (1) - - 0.68 (4)
[0.19, 0.90]
-
LEA -0.33(2)
[-1.00, 1.00]
0.48 (1) - -0.30 (2)
[-1.00, 0.99]
- 0.30 (2)
[-0.49, 0.81]
HYD 0.03 (2)
[-0.22, 0.27]
0.25 (1) - -0.35 (2)
[-0.62, 0.00]
- 0.00 (2)
[-1.00, 1.00]
64
Table 3.10 Weighted means of estimates from literature for genetic correlations among reproduction, growth and carcass traits of pigs in the
tropics with the number of estimates in parenthesis ( ) and the 95% confidence interval in brackets [ ]
Traits ADG IBW IW21 IWW BFT CWT CLT LEAN DPT LEA HYD
LSF -0.04 (1) -0.66 (4)
[-0.87, -0.23]
-0.70 (1) -0.53 (4)
[-0.73, -0.25]
-0.02 (1) - - - - - -
LS21 - -0.25 (1) 0.04 (1) -0.55 (1) - - - - - - -
LSW - -0.45 (4)
[-0.67, -0.16]
0.34 (1) -0.24 (5)
[-0.43, -0.03]
- - - - - - -
LWF - 0.31 (2)
[-0.92, 0.98]
0.03 (2)
[-0.10, 0.16]
-0.36 (1) - - - - - - -
LW21 - 0.11 (2)
[-0.58, 0.71]
0.53 (2)
[-0.44, 0.93]
0.15 (1) - - - - - - -
LWW - 0.04 (1) 0.58 (1) 0.29 (3)
[-0.01, 0.54]
- - - - - - -
PDG 0.38 (2)
[-0.99, 1.00]
0.87 (1) 0.85 (1) - 0.05 (1) 0.08 (1) 0.35 (1) - - - -
SRT - - - -0.60 (2)
[-1.00, 1.00]
- - - - - - -
AFC -0.04 (1) - - -0.02 (1) - - - - - -
WSI -0.08 (1) - - -0.27 (1) - - - - - -
MOR - -0.21 (2)
[-1.00, 1.00]
-0.78 (1) -0.30 (1) - - - - - - -
ADG - - - - -0.35 (22)
[-0.47, -0.21]
0.57 (1) 0.58 (1) 0.50 (1) 0.00 (1) 0.02 (4)
[-0.60, 0.62]
0.14 (3)
[-0.27, 0.51]
MWT - - - - 0.16 (2)
[-1.00, 1.00]
- 0.43 (1)
- - - -
D90 - - - - 0.18 (7)
[0.06, 0.29]
- - - - - -
FCR - - - - 0.22 (10)
[0.10, 0.36]
- 0.06 (2)
[-0.61, 0.68]
-0.53 (1) - -0.04 (2)
[-0.50, 0.50]
-0.07 (2)
[-0.31, 0.18]
ADFI - - - - 0.49 (2)
[-0.96, 0.99]
- - - - - -
65
Table 3.11 Weighted means of estimates from literature for phenotypic correlations among reproduction, growth and carcass traits of pigs in
the tropics with the number of estimates in parenthesis ( ) and the 95% confidence interval in brackets [ ]
Traits ADG IBW IW21 IWW BFT CWT CLT LEAN DPT LEA HYD
LSF -0.04 (1) -0.35 (4)
[-0.83, 0.41]
0.27 (1) -0.54 (3)
[-0.83, -0.03]
-0.01 (1) - - - - - -
LS21 - -0.10 (1) -0.03 (1) -0.09 (1) - - - - - - -
LSW - -0.30 (4)
[-0.53, -0.03]
0.01 (1) -0.24 (5)
[-0.52, 0.07]
- - - - - - -
LWF - -0.20 (1) -0.08 (1) -0.11 (1) - - - - - - -
LW21 - 0.07 (1) 0.33 (1) -0.14 (1) - - - - - - -
LWW - 0.23 (1) - 0.22 (3)
[0.04, 0.38]
- - - - - - -
PDG -0.09 (1) - - - -0.01 (1) 0.01 (1) 0.04 (1) - - - -
SRT - - - 0.33 (2)
[0.06, 0.56]
- - - - - - -
AFC -0.04 (1) - - -0.02 (1) - - - - - -
WSI -0.04 (1) - - -0.06 (1) - - - - - -
MOR - -0.22 (1) -0.23 (1) -0.31 (1) - - - - - - -
ADG - - - - 0.09 (13)
[-0.03, 0.22]
0.68 (1) 0.07 (4)
[-0.62, 0.69]
0.09 (1) -0.08
(1)
0.00 (3)
[-0.22, 0.23]
0.23 (3)
[-0.57, 0.81]
MWT - - - - 0.00 (1) - - - - - -
D90 - - - - 0.27 (3)
[-0.35, 0.73]
- - - - - -
FCR - - - - 0.06 (5)
[-0.13, 0.25]
- 0.12 (2)
[-1.00, 1.00]
-0.09 (1) - -0.22 (2)
[-0.97, 0.93]
-0.50 (2) [-1.00, 1.00]
ADFI - - - - 0.27 (2)
[-0.80, 0.93]
- - - - - -
66
Figure 3.1 Funnel plot of heritability estimates for (A) LSF showing no evidence of
publication bias and (B) ADG showing classic asymmetry suggesting the presence of a
publication bias
A
B
67
3.5 Appendix
3.5.1 List of literature used to construct meta-analysis dataset of genetic and phenotypic
parameters for pig production traits in the tropics
1. Adebambo, O. A. and Onakade, A. D. 1983. Growth and carcass traits of
crossbred boars of Exotic and Nigerian Indigenous pigs. Journal of Animal
Production Research 3:71-87.
2. Aina, A. B. T. 1982. Litter characteristics and related parameters of the Large
White and Landrace at the University of Ibadan Teaching and Research Farm p.
99.
3. Almeira Neto, P. P. de, Oliveira, A. I. G., Almeida, A. J. L., Lima, J. A., Silva, M.
and Costa, C. N. 1993. Genetic and phenotypic parameters of carcass traits in
pigs. Revista da Sociedade Brasileira de Zootecnia 22:624-633.
4. Alves, R. G. O., Almeida e Silva, M. de, Milagres, J. C. and Costa, P. M. A. 1978.
Economic and genetic analysis of fattening performance and carcass characters of
Landrace and Duroc pigs. Revista da Sociedade Brasileira de Zootecnia 7:245-
257.
5. Alves, R. G. O., Silva, M. A., Pereira, J. A. A., Sancevero, A. B. and Barbosa, A.
S. 1987. Effect of environmental and genetic factors on pig litter size and weight
at birth and at 21 days of age. Revista da Sociedade Brasileira de Zootecnia
16:540-549.
6. Argañosa, V. G., Penalba F. F. and Alcantara P. F. 1975. Genetic and Phenotypic
parameters of traits associated with sow productivity. Philippine Agriculturist
58:99-307.
68
7. Barbosa, L., Lopes, P. S., Regazzi, A. J., Torres, R. A., Santana Junior, M. L. and
Veroneze, R. 2008a. Estimação de parâmetros genéticos em suínos usando
Amostrador de Gibbs. Revista Brasileira De Zootecnia 37:1200-1206.
8. Barbosa, L., Lopes, P. S., Regazzi, A. J., Torres, R. A., Santana Junior, M. L. and
Veroneze, R. 2008b. Estimação de parâmetros genéticos em tamanho de
leitegada de suínos utilizando análises de características múltiplas. Revista
Brasileira de Zootecnia 37:1947-1952.
9. Barbosa, A. S., Alves, R. G. O., Bergmann, J. A. G., Sancevero, A. B. and Silva
Filho, J. M. 1984. Sources of variation and co variation in litter size and weight
of Landrace, Duroc, Large White and Hampshire pigs. 1. Litter size at birth, at
21 days, at weaning and at 56 days of age. Arquivo Brasileiro de Medicina
Veterinária e Zootecnia 36:65-76.
10. Barbosa, A. S., Araújo, C. V., Aquino, L. H., Oliveira, A. I. G. and Gonçalves, T.
M. 1998. Estimates of heritability and selection index in pigs of the "Piau" breed.
Arquivo Brasileiro de Medicina Veterinária e Zootecnia 50:753-758.
11. Bardoloi, T. and Sharma, P. K. 1997. Estimates of genetic parameters of body
weights in a closed herd of Landrace pigs. Indian Veterinary Journal 74:312-315.
12. Benevides Filho, I. M., Schlindwein, A. P. and Lôbo, R. B. 1985. Genetic study
on the performance of Piau pigs from birth to weaning. Arquivo Brasileiro de
Medicina Veterinária e Zootecnia 37:497-506.
13. Bhatia, S. S. and Sharma, P. K. 1979. Genetic studies on some economic traits in
Landrace pigs. I. Individual body weight at birth and at weaning. Indian
Veterinary Journal 56:45-48.
69
14. Cavalcante Neto, A., Lui, J. F., Sarmento, J. L. R., Ribeiro, M. N., Monteiro, J.
M. C. and Tonhati, H. 2008. Environmental factors and heritability estimate for
the weaning-estrus interval in sows. Revista Brasileira de Zootecnia 37:11
15. Cavalcante Neto, A., Lui, J. F., Sarmento, J. L. R., Ribeiro, M. N., Monteiro, J.
M. C., Fonseca, C. and Tonhati, H. 2009. Estimation models of variance
components for farrowing interval in swine. Brazilian Archives of Biology and
Technology 52:69-76.
16. Chansomboon, C., Elzo, M. A., Suwanasopee, T. and Koonawootrittriron, S.
2009. Genetic and Environmental Factors Affecting Weaning-to-First Service
Interval in a Landrace-Large White Swine Population in Northern Thailand.
Kasetsart Journal of Natural Sciences 43:669-679.
17. Chansomboon, C., Elzo, M. A., Suwanasopee, T. and Koonawootrittriron, S.
2010. Estimation of Genetic Parameters and Trends for Weaning-to-first
Service Interval and Litter Traits in a Commercial Landrace-Large White Swine
Population in Northern Thailand. Asia-Australian Journal of Animal Science
23:543-555.
18. Chauhan, V. P. S., Arora, R. L., Chhabra, A. K. and Bhat, P. N. 1993. Litter traits
and live weights of Landrace pigs in India. Indian Veterinary Journal 70:29-31.
19. Chhabra, A. K., Bhatia, S. S., Sharma, N. K. and Dutta, O. P. 1990. Genetic study
of litter traits in Large White Yorkshire pigs. Indian Veterinary Journal 67:715-
718.
20. Chhabra, A. K., Gaur, G. K, Ahlawat, S. P. S. and Paul, S. 1999. Inheritance of
carcass traits in Desi pigs. Indian Veterinary Journal 76:403-407.
70
21. Chimonyo, M. and Dzama, K. 2007. Estimation of genetic parameters for growth
performance and carcass traits in Mukota pigs. Animal 1:317-323.
22. Chimonyo, M., Dzama, K. and Bhebhe, E. 2006. Genetic determination of
individual birth weight, litter weight and litter size in Mukota pigs. Livestock
Science 105:69-77.
23. Chimonyo, M., Dzama, K. and Bhebhe, E. 2008. Genetic determination of
mothering ability and piglet growth in indigenous Mukota sows of Zimbabwe.
Livestock Science 113:74-80.
24. Costa, A. R. C., Lopes, P. S., Torres, R. A., Regazzi, A. J., Silva, M. A.,
Euclydes, R. F. and Pires, A. V. 2001. Estimação de parâmetros genéticos em
características de desempenho de suínos das raças Large White, Landrace e
Duroc. Revista Brasileira De Zootecnia 30:49-55.
25. Costa, C. N., Saralegui, W. H., Fávero, J. A. and Leitão, G. R. 1986. Genetic
parameters and selection indices for pigs. Revista da Sociedade Brasileira de
Zootecnia 15:124-131.
26. Darfour-Oduro, K. A., Naazie, A., Ahunu, B. K. and Aboagye, G. S. 2009.
Genetic parameter estimates of growth traits of indigenous pigs in Northern
Ghana. Livestock Science 125:187-191.
27. Das, A. K., Gaur, G. K., Bist, G.S. 1999. A note on genetic analysis of carcass
traits in indigenous pigs. Indian Journal of Animal Research 33:139-140.
28. Das, A. K., Mishra R. R. 1993. Genetic analysis of carcass traits of indigenous
pigs. Journal of the Assam Veterinary Council 3:29-31.
71
29. Deka , D., Kalita , D., Das , D., Goswami, R. N. 2002. Estimation of genetic
parameters for body weight at different ages and response to selection in halfbred
and graded pigs. Cheiron 31 (5&6):131-133.
30. Deo, S., Bhat, P. N., Raina, B. L. 1981. Effect of Genetic and Non-Genetic
Factors on Body Weights at Different Ages in Landrace, Large White and their
Half-Breds. Indian Journal of Animal Science 51:469-478.
31. Díaz Rodríguez, F. 1984. Genetic parameters for traits included in the pig
selection index. Preliminary study. Ciencia y Técnica En La Agricultura, Ganado
Porcino 7:103.
32. Diéguez, F. J., Trujillo, G., Fredeen, H. T., Weiss, G. M. 1983. Genetic
correlations of growth traits with backfat thickness in gilts performance-tested in
the field in Cuba. Memoria, Asociacion Latinoamericana De Produccion Animal
18:157.
33. Duc, N. V. 1999. Heritabilities and genetic correlations between production traits
in Vietnamese pigs; Rising to the challenge - breeding for the 21st Century
Customer. Proceedings of the Thirteenth Conference Association for the
Advancement of Animal Breeding and Genetics, Mandurah, Western Australia,
4th-7th July 1999. pp. 157-160.
34. Duc, N. V., Graser, H. U., Kinghorn, B. P. 1998. Heritability and genetic and
phenotypic correlations for number born alive between parities in Vietnamese
Mong Cai and Large White. Proceedings of the 6th World Congress of Genetics
Applied to Livestock Production 23:543-546.
72
35. Dzama, K., Mungate, F. 1998. Some genetic parameters of commercially raised
pigs in Zimbabwe. Proceedings of the 6th
World Congress of Genetics Applied to
Livestock Production.
36. Fernandez Giné, G. A., Freitas, Rilke Tadeu Fonseca de, Oliveira, Antonio Ilson
Gomes de, Pereira, I. G., Gonçalves T. de M. 2004. Estimativa de parâmetros
genéticos para características de carcaça em um rebanho de suínos Large White.
Revista Brasileira De Zootecnia 33:337-343.
37. Gourdine, J. L., Mandonnet, N., Naves, M., Bidanel, J. P., Renaudeau, D., 2006.
Genetic parameters of rectal temperature in sows in a tropical humid climate and
its association with performance during lactation: Preliminary results. In:
Anonymous Instituto Prociência, Minas Gerais, pp. 16-09.
38. Guerra, D., Diéguez, F. J., Santana, I., Gerardo, L., Trujillo, C. 1992. Genetic and
phenotypic parameters of a pig paternal synthetic breed. Cuban Journal of
Agricultural Science 26:11-15.
39. Gupta, R. N., Dhingra, M. M., Parmar, S. N. S. 1983. A study on litter weight at
birth and weaning in Large White Yorkshire pigs. Indian Veterinary Journal
60:381-383.
40. Gupta, R. N., Parmar, S. N. S, Johar, K. S. 1982a. Note on the effect of sire, sex
and year on weaning weight of Large White Yorkshire pigs. Indian Journal of
Animal Science 52:986-987.
41. Gupta, R. N., Parmar, S. N. S, Johar, K. S. 1982b. Note on variation in early gain
in weight in Middle White Yorkshire piglets. Indian Journal of Animal Science
52:985-986.
73
42. Gupta, R. N., Parmar, S. N. S, Johar, K. S, Dhingra, M. M. 1982c. Effect of sire,
sex and year on birth weight of Large White Yorkshire pigs. Indian Veterinary
Journal 59:289-291.
43. Ilatsia, E., Githinji, M., Muasya, T., Okeno, T., Kahi, A. 2008. Genetic parameter
estimates for growth traits of Large White pigs in Kenya. South African Journal
of Animal Science pp. 38.
44. Iloeje, M. U. 1987. Estimation of additive genetic, non-additive genetic, and
permanent environmental sources of variation in preweaning traits in swine using
the expectation-maximization algorithm for restricted maximum likelihood. Beitr
Tropical Landwirtsch Veterinarmed 25:433-438.
45. Imboonta, N., Rydhmer, L., Tumwasorn, S. 2007. Genetic parameters and trends
for production and reproduction traits in Thai Landrace sows. Livestock Science
111:70-79.
46. Institute de Ciencia Animal, Cuba. 1985. Swine. pp. 64.
47. Irgang, R., Fávero, J. A., Kennedy, B. W. 1994. Genetic parameters for litter size
of different parities in Duroc, Landrace, and Large White sows. Journal of Animal
Science 72:2237.
48. Irgang, R., Robison, O. W. 1984a. Heritability estimates for ages at farrowing,
rebreeding interval and litter traits in swine. Journal of Animal Science 59:67-73.
49. Irgang, R., Robison, O. W. 1984b. Heritability estimates for ages at farrowing,
rebreeding interval and litter traits in swine. Journal of Animal Science 59:67-73.
50. Jayarajan, S., Ulaganathan, V. 1992. Non-genetic factors affecting performance of
pigs. Indian Journal of Animal Science 62:899-903.
74
51. Jogi, S., Johar, K. S. 1992. Genetic study of dressing percentage of common
Indian pigs. Indian Journal of Animal Research 26:100-102.
52. Jogi, S., Johar, K. S. 1994. Factors affecting weaning weight in desi pigs. Indian
Journal of Animal of Research 28:35-39.
53. Jogi, S., Johar, K. S., Arora, J. S. 1992. Genetic study of back fat thickness of
common Indian pigs. Pashudhan Anushandhan 2:15-20.
54. Jogi, S., Johar, K. S., Arora, J. S. 1993. Genetic study of dressing percentage,
carcass length and back fat thickness of common Indian pigs. Indian Veterinary
Journal 70:227-231.
55. Jogi, S., Johar, K. S., Vyas, R., Arora, J. S. 1991a. Genetic study on loin eye area
of common Indian pig's carcass. Indian Veterinary Medical Journal 15:198-201.
56. Jogi, S., Johar, K. S., Vyas, R., Joshi, S. K. 1991b. Study of ham, shoulder and
loin yields from carcass of common Indian pigs. Indian Journal of Animal
Research 25:92-94.
57. Johar, K. S., Saibaba, P., Gupta, R. N. 1974. Heritability and variability of
weaning weight of Middle White Yorkshire pigs. Indian Veterinary Journal
51:591-593.
58. Johar, K. S., Saibaba, P., Gupta, R. N. 1975. A study on litter size and litter
weight at birth and weaning in Middle White Yorkshire pigs. Indian Veterinary
Journal 52:707-710.
59. Lakhani, G. P. 1989. Studies on some preweaned economic traits in middle White
Yorkshire pigs. Indian Journal of Animal Research 23:16-18.
75
60. Lakhani, G. P. 1992. Heritability and variability of weaning weight of Large
White Yorkshire pigs. Indian Journal of Animal Research 26:6-8.
61. Lakhani, G. P, Bhadauria, S. S. 1988a. Heritability and variability in the birth
weight of indigenous pigs. Indian Journal of Animal Research 22:49-50.
62. Lakhani, G. P., Bhadauria, S. S. 1988b. Heritability and variability of weaning
weight of indigenous pigs. Indian Veterinary Journal 65:1134-1136.
63. Lakhani, G. P., Jogi, S. 1998. Genetic study of birth weight in desi pigs and their
Large White Yorkshire grades. Indian Journal of Animal Research 32:47-50.
64. Lakhani, G. P., Jogi, S. 1999. Studies on dressing percentage in indigenous pigs
and their Large White Yorkshire grades. Indian Journal Animal Research 33:29-
31.
65. Lakhani, G. P, Nema, R. K. 1989. Studies on birth weight and weaning weight of
Large White Yorkshire pigs. Indian Journal Animal Research 23:13-15.
66. Lal, K., Sharma, G. C., Mishra, R. R., Prasad, S. 1989. Studies on litter size at
weaning of Indian pigs. Indian Journal of Animal Production and Management
5:80-81.
67. Leigh, A. O. 1981. Factors affecting the gestation period of pigs in Nigeria.
Tropical Animal Health and Production 13:87-93.
68. Leite, D. M. G., Pereira, C. S, Oliveira, H. N. D. 1989. Estimates of genetic
parameters of litter and piglet weight in Landrace pigs. Anais do IV Congreso
Brasileiro De Veterinárias Especialistas Em Suínos, Itapema, SC, Brazil, 1989 pp.
75.
76
69. Lourenço, F. F., Dionella, N. J. L., Medeiros, G. C. R., Rosa, V. C. da. 2008a.
Genetic characterization of litter traits in purebred Large White swine, in the state
of Rio Grande do Sul, Brazil. Revista Brasileira de Zootecnia 37(11):1959-1964.
70. Lourenço F. F., Dionello N. J. L., Medeiros G. C. R., Rosa V.C. da. 2008b.
Genetic characterization of litter traits in purebred Landrace swine, in the state of
Rio Grande do Sul, Brazil. Revista Brasileira de Zootecnia 37(9):1601-1606.
71. Ludwig, A., Alves, R. G. O., Saralegui, W., Irgang, R. 1979. Estimation of
genetic and phenotypic parameters of some fattening and carcass characters in
Landrace pigs, and proposals for a selection index. Revista Da Sociedade
Brasileira De Zootecnia 8:473-487.
72. Milagres, J. C., Fedalto, L. M., Pereira, J. A. A., Costa, P. M. da. 1981a. Sources
of variation in litter size and weight at birth and at 21 days of age in Duroc,
Landrace and Large White pigs. 2. Piglet and litter weights. Revista da Sociedade
Brasileira de Zootecnia 10:672-691.
73. Milagres, J. C., Fedalto, L. M., Silva, M. da E., Euclydes, R. F. 1981b. Sources of
variation in litter size and weight at birth and 21 days of age in Duroc, Landrace
and Large White pigs. 5. Genetic parameters. Revista da Sociedade Brasileira de
Zootecnia 10:720-734.
74. Milagres, J. C., Fedalto, L. M., Silva, M. da E., Pereira, J. A. A. 1981c. Sources of
variation in litter size and weight at birth and at 21 days of age in Duroc,
Landrace and Large White pigs. 1. Litter size. Revista da Sociedade Brasileira de
Zootecnia 10:653-671.
77
75. Mishra, R. R., Lal, S. P. K. 1990. Factors affecting litter size in Large White
Yorkshire pigs. Indian Veterinary Journal 67:517-519.
76. Mishra, R. R., Lal, K., Prasad, S. 1989. Factors affecting birth weight of Landrace
piglets. Indian Journal of Animal Production and Management 5:124-125.
77. Mishra, R. R., Lal, K., Prasad, S. 1990a. Factors affecting litter weight of
Landrace pigs. Indian Journal of Animal Production and Management 6:205-206.
78. Mishra, R. R., Lal, K., Sharma, G. C., Prasad, S. 1990b. Studies on litter weight at
weaning of local pigs maintained in India. Indian Veterinary Journal 67:552-553.
79. Mishra, R. R., Prasad, S., Lal, K. 1992. Factors influencing body weight of
Landrace piglets at weaning. Indian Veterinary Journal 69:189-190.
80. Mishra, R. R., Sharma, G. C. 1990a. Studies on reproductive traits in Landrace
pigs in India. Indian Veterinary Journal 67:425-429.
81. Mishra, R. R., Sharma, G. C., Lal, K., Prasad, S. 1990b. A study on carcass traits
of Landrace pigs in humid tropical region. Indian Veterinary Journal 67:729-733.
82. Nascimento, J. D. d. 1988. Genetic and environmental factors affecting litter size
in Duroc pigs. Arquivo Brasileiro De Medicina Veterinária e Zootecnia 40:159-
160.
83. Nguyen, N. H., McPhee, C. M., Wade, C. M. 2003. Genetic parameters for
reproduction traits in sows of lines selected for growth rate on restricted feeding.
In: 50 Years of DNA, July 2003, Melbourne, Australia.
84. Nguyen, N. H., McPhee, C. P. 2005. Genetic parameters and responses of
performance and body composition traits in pigs selected for high and low growth
rate on a fixed ration over a set time. Genetics Selection Evolution 37:199-213.
78
85. Nguyen, N. H., McPhee, C. P., Wade, C. M. 2006. Genetic variation and
responses in reproductive performance of sows in lines selected for growth rate
under restricted feeding. Animal Science 82:7-12.
86. Nguyen, N. H., McPhee, C. P. 2005. Responses in production and body
composition traits in ad libitum fed pigs from lines selected for high and low
growth rate on a fixed ration. Livestock Production Science 97:101-106.
87. Norris, D., Varona, L., Visser, D., Theron, H., Voordewind, S., Nesamvuni, E.
2007. Estimation of the additive and dominance variances in South African
Landrace pigs. South African Journal Animal Science 36:261.
88. Norris, D., Varona, L., Ngambi, J. W., Visser, D. P., Mbajiorgu, C. A.,
Voordewind, S. F. 2010. Estimation of the additive and dominance variances in
SA Duroc pigs. Livestock Science 131, 144-147.
89. Nwakpu, P. E., Onu, P. N. 2007. Correlations Between Some Weaning Traits in
Four Strains of Piglet Genotypes. Journal of Animal and Veterinary Advances
6:1238-1241.
90. Okon, B., Ngere, L. O., Ibom, L. A. 2009. Estimates of Genetic Parameters of
Body Weights of Different Classes of Pigs. Journal of Agriculture, Forestry and
the Social Sciences pp.6.
91. Oliveira, Antonio Ilson Gomes de, Saenz, E. A. C., Ferraz, B. S., Fialho, E. T.
1997. Genetic parameters for swine carcass traits. 1. Estimates of heritability.
Revista Brasileira de Zootecnia 26:54-60.
79
92. Orheruata, A. N., Izerbigbe, O., Nwosu, E. 1993. Heritability and correlation
studies of body weight and litter size at birth and weaning in swine. Nigerian
Journal of Animal Science 35:118.
93. Paul, А. К., Kundu, M. G., Singh, S., Singh, P. 2009. Heritability of growth
curve parameters of pigs. Indian Journal of Animal Science 79:716-719.
94. Pena de Borsotti, N., Verde, O. 1985. Factors affecting growth traits in piglets.
Memoria Asociacion Latinoamericana de Produccion Animal 18:154.
95. Perez, T., Dieguez, F. J., Menchaca, M., Vachal, Y. J. 1981. Estimation of genetic
parameters in the on-farm performance testing of Duroc pigs. Ciencia y Tecnica
En La Agricultura, Ganado Porcino 4:23-37.
96. Pholsing, P., Koonawootrittriron, S., Suwannasopee, T., Elzo, M. A. 2008.
Genetic association between age and litters traits at first farrowing in commercial
Pietrain-Large White population raised in an open-house system in Thailand. ,
Joint ADSA-ASAS Meeting, Indianapolis, Indiana, July 7-11, 2008. Journal of
Animal Science 86 (Suppl. 2):203.
97. Pires, A. V., Lopes, P. S, Torres, R. da, Euclydes, R. F., Costa, André Ribeiro
Corrêa da. 2000. Estimação de parâmetros genéticos de características
reprodutivas em suínos. Revista Brasileira De Zootecnia 29:1698-1705.
98. Pita, Fabiano Veraldo da Costa, Albuquerque L. G. d. 2001a. Comparação de
Diferentes Modelos para Avaliação Genética de Características de Desempenho
Pós-desmama em Suínos. Revista Brasileira De Zootecnia 30:1720-1727.
80
99. Pita, Fabiano Veraldo da Costa, Albuquerque L. G. d. 2001b. Efeitos da utilização
de diferentes covariáveis na avaliação do ganho de peso médio diário em suínos.
Revista Brasileira de Zootecnia 30:736-743.
100. Rico, C. 1988. Reproductive performance of the Hampshire breed. Cuban Journal
of Agricultural Science 22:17-24.
101. Rico, C., Gomez, J. 1982. Interrelationships between reproductive performance
traits in swine. Cuban Journal of Agricultural Science 16:161-166.
102. Rico, C, Menchaca, M. 1985. Field performance test with Duroc pigs. 1.
Environmental influences and genetic parameters of the selection index traits.
Cuban Journal of Agricultural Science 19:243-248.
103. Roso, V. M., Fries, L. A., Martins, E. S. 1995. Genetic parameters affecting
growth and carcass quality in Duroc pigs. Revista Da Sociedade Brasileira de
Zootecnia 24:310-316.
104. Schlindwein, A. P., Duarte, F. A. M., Spers, A., Bezerra, L. A. F. 1982.
Correlations between preweaning traits in Duroc swine. Revista Brasileira
Genetica 5:363-371.
105. Schlindwein, A. P., Torres, J. R., Carneiro, G. G. 1979. Heritability estimates of
litter size and pig birth weight in Duroc swine. Arquivos da Escola de
Veterinaria da Universidade Federal De Minas Gerais 31:169-178.
81
106. Sharma, G. C., Mishra, R. R. 1989. Studies on reproductive traits of Large White
Yorkshire pigs. Indian Journal of Animal Production and Management 5:185-187.
107. Siewerdt, F., Cardellino, R. A. 1996. Genetic parameters of mortality to 21 days
of age in Landrace piglets. Revista Da Sociedade Brasileira de Zootecnia
25:902-909
108. Siewerdt, F., Cardellino, R. A., Rosa, V. C. d. 1995. Genetic parameters of litter
traits in three pig breeds in Southern Brazil. Revista Brasileira De Genética
18:199-205.
109. Singh, R. N., Pandey, R. S., Biswas, S. C. 1977. A study on some of the economic
characters of Yorkshire pigs under farm conditions. Indian Veterinary Journal
54:175-179.
110. Suwanasopee, T., Mabry, J. W., Koonawootrittriron, S., Sopannarath, P.,
Tumwasorn, S. 2005. Estimated Genetic Parameters of Non-Productive Sow Days
Related to Litter Size in Swine Raised in Thailand. Thai Journal of Agricultural
Science 38:87-93.
111. Tinh, N. H., Vien, N. T., Van, Giai D., Hai, L. T. Genetic parameters and trends
for reproduction and performance traits in Yorkshire, Landrace and Duroc pigs in
South Vietnam. www.iasvn.org/uploads/files/genetic_trend_0521081944.pdf
82
112. Valarezo, J. M., Quijandria, B. 1978. Genetic parameters for production
characters in Duroc piglets. Memoria Asociacion Latinoamericana De
Produccion Animal 13:152.
113. Van, V. T. K., Duc, N. V. 1999. Heritabilities, genetic and phenotypic correlation
between reproductive performance in Mong Cai and Large White breeds. , 153.
114. Vasundharadevi, M., Govindaiah, M. G., Narasimhamurthy, H. N., Krishnappa S.
B. 1996. Genetic and non-genetic variability of weaning weight in Large White
Yorkshire pigs. Journal of Veterinary and Animal Sciences 27:82-85.
115. Vecchionacce, H., Verde, O., Muller-Haye B. 1976. Genetic and environmental
effects on growth in piglets. Memoria Asociacion Latinoamericana de Produccion
Animal 11:31-32.
116. Venanzi, J. d, Verde, O. 1995. Genetic and environmental factors affecting
production traits in young breeding pigs. Archivos Latinoamericanos De
Producción Animal 3:91-98.
117. Venanzi, J. d, Verde, O. 1996. Genetic and environmental factors affecting litter
traits in 2 pig herds in Venezuela. Archivos Latinoamericanos de Producción
Animal 4:55-71.
83
CHAPTER 4
ACCURACIES OF GENOMIC BREEDING VALUES FOR
SIMULATED TROPICAL PIG POPULATIONS
E. C. Akanno1, F. S. Schenkel
1, M. Sargolzaei
13, R. M. Friendship
2 and J. A. B.
Robinson1
1Centre for Genetic Improvement of Livestock,
Department of Animal & Poultry Science, University of Guelph, Guelph, ON, Canada
2Department of Population Medicine, Ontario Veterinary College,
University of Guelph, Guelph, ON, Canada
3L'Alliance Boviteq, Saint-Hyacinthe, Quebec, Canada
84
4.0 ABSTRACT
Genetic improvement of pigs in tropical developing countries has focused on
imported exotic populations subjected to intensive breeding. A declining effective
population size (Ne) resulted in moderate linkage disequilibrium (LD) in this population.
Presently, indigenous pig population with limited selection and low LD are being
considered for improvement. Given that the infrastructure for genetic improvement using
the conventional BLUP selection methods are lacking, a genome-wide selection (GS)
program was proposed for developing countries. A simulation study was conducted to
evaluate the option of using 60 K single nucleotides polymorphism panel and observed
level of LD in the indigenous and exotic pig populations. Several scenarios were
evaluated, including different trait heritability, different size and structure of training and
validation populations and different selection methods. The training set included
previously selected exotic pigs, unselected indigenous pigs and their crossbreds. The
ridge regression method was used to train the prediction model. The results showed that
accuracies of genomic breeding values (GBVs) in the range of 0.31 - 0.86 for different
trait heritabilities are expected for validation populations with varying LD levels as long
as high density marker panels are utilized. The GS method improved accuracy of GBVs
better than pedigree-based selection for traits with low heritability and in young animal
with no performance data. Crossbred training set had higher accuracy of GBVs than
purebreds. Genome-wide selection holds promise for genetic improvement of pigs in the
tropics.
Keywords: Genome-wide selection, accuracy, tropical pigs, developing countries
85
4.1 INTRODUCTION
Genetic improvement of pigs in tropical developing countries has focused on
imported exotic populations subjected to intensive breeding. A declining effective
population size (Ne) and strong directional selection resulted in moderate linkage
disequilibrium (LD) between quantitative trait loci (QTL) and close by markers (Henson,
1992). In practice, indigenous pig populations that are bred in extensive system or under
medium- to low-input and stressful environments are also used for pork production. As
expected, Ne in the indigenous populations is larger than that observed in commercial
populations resulting in very low LD across the genome (Thévenon et al., 2007; Amaral
et al., 2008). Moreover indigenous pig populations have been mildly selected due to their
poor performance for economic traits (Shenglin, 2007).
Although genetic improvement is desired in indigenous pig populations, a
limitation to this effort in developing countries is the lack of necessary infrastructure, i.e.
lack of programs that record phenotype on pedigreed animals in an extensive breeding
system (Kahi et al., 2005). A common approach to genetic improvement of pigs is to
carry out selection in nucleus herds and to combine the indigenous and exotic populations
in crossbreeding programs in order to produce crossbred market pigs (Oseni, 2005;
Nwakpu and Onu, 2007).
Meuwissen et al. (2001) proposed a method for genetic improvement of livestock
which was called genome-wide selection (GS). Owing to recent developments in high
throughput, high density single nucleotide polymorphism (SNP) panel technology, GS
can currently be applied. Genome-wide selection uses a ―training population‖ of
individuals with phenotypic and genotypic information to develop a prediction model.
86
The prediction model then uses genotypic data from a ―candidate population‖ of untested
individuals and produces genomic breeding values (GBV) of an individual by using the
estimated additive marker effects from the training population. Selection decisions are
based on the GBV. Using GS in swine genetic improvement is expected to yield a
considerable increase in selection responses for juvenile animals that do not have
phenotypic records (Meuwissen et al., 2001; Gianola et al., 2003; Hayes et al., 2009) and
potentially can reduce the cost of breeding programs (e.g. Schaeffer, 2006). Genome-
wide selection is also believed to be advantageous for sex-limited traits, traits that are
difficult to measure like disease and longevity traits with low heritability and traits
requiring sacrifice of the animal (Goddard and Hayes, 2007).
The success of GS depends mainly on the prediction accuracy of GBVs which is
directly proportional to the response to selection (Falconer and Mackay, 1996). Several
authors have reported high accuracies of GBV in the range of 0.45 – 0.88 in theoretical
(Meuwissen et al., 2001; Muir, 2007; de Roos et al., 2009; Toosi et al., 2010) and
empirical studies (Hayes et al., 2009; VanRaden et al., 2009; Cleveland et al., 2010) in
livestock species. Accuracy of GBV relies on the existence of LD between QTL and
markers in the population of interest and on the strength of predicted individuals‘
relatedness to the reference population, much as it will do when using pedigree
information to perform predictions (Jannink et al., 2010). As the population diverges
from the training set, accuracy of GBV declines because of recombination between QTL
and markers unless densely spaced markers are used to maintain sufficient LD (de Roos
et al., 2009). Conceptually assuming a large reference population, the more dense the
marker spacing and greater the population-wide LD, the better GS works.
87
Statistical methods for training prediction models are being investigated
(Whittaker et al., 1999; Meuwissen et al., 2001; Xu, 2003; Gianola et al., 2006; Fernando
et al., 2007). Methods that assume all marker effects are distributed with equal variance,
like ridge regression (Whittaker et al., 1999), have been found to be as accurate as
methods that assume some markers do not explain any variance (e.g. BayesB; Meuwissen
et al., 2001). Assuming all marker effects are drawn from the same distribution does not
mean all the effects are equal but that they are all equally shrunken toward zero (Jannink
et al., 2010). Habier et al. (2007) also showed that ridge regression was more effective at
capturing genetic relationships among families because it fits more markers into the
prediction model.
Therefore, the objective of this study was to evaluate long term application of GS
in populations characterised by small LD, large Ne and mild selection pressure, and in
constant hybridization with populations of opposite characteristics, assuming truncated
selection pressure. This allows the assessment of the long term application of GS in
swine genetic improvement in tropical developing countries based on the availability of a
60 K SNP chip panel for this environment. Accuracies of genomic predictions were
investigated and compared for different training datasets, traits and breeding structures.
Accuracy of a genome-based model was contrasted to a pedigree-based model for
breeding value prediction and the decay of accuracy of GBV under random mating was
also evaluated. The study was based on computer simulations of relevant tropical pig
population structures.
88
4.2 MATERIALS AND METHODS
4.2.1 Simulation software
The QMSim simulation software developed by Sargolzaei and Schenkel (2009)
was used for this study. This forward-in-time simulation program was designed to
simulate large genomes and complex pedigree structures that mimic existing livestock
populations. In this program, breeding values can be estimated using different approaches
and various selection events can be imposed on the populations. Table 4.1 gives a
description of the parameters used for the simulation.
4.2.2 Population structure
All pigs found in tropical developing countries were assumed to come from a
common base of unrelated individuals which served as ancestors to the exotic and
indigenous populations. The common base population was mated based on a union of
gametes randomly sampled from the male and female gametic pools for 1,000
generations with Ne of 5,000 breeding individuals (Figure 4.1). This ancestral population
was later contracted to Ne of 1,050 individuals consisting of 300 breeding males at the
end of another 1000 generations in order to generate an initial historic LD. Effective
population size in livestock populations generally shows a declining trend due to
domestication (Henson, 1992). The ancestral population was split into 2 unequal sized
purebred subpopulations (referred to as exotic and indigenous populations) at generation
2001.
The exotic pig population was created by drawing the top ranking 50 males and
top ranking 250 females based on their phenotypes from the common base population
89
resulting in Ne of 167 and subjected to phenotypic selection for the trait under
consideration for 50 generations with an intensity of 10% for males and 50% for females.
Each female produced 4 offspring (2 males and 2 females) resulting in 1,000 breeding
animals in each generation from which selection was made. The remaining members of
the common base population consisting of 250 males and 500 females were allowed to
mate at random concurrently for 50 generations resulting in the indigenous population
with Ne of 667. The Ne assumed in this study was chosen to yield a simulated level of LD
observed in real life. The actual Ne might differ slightly.
At generation 2049 in the simulation, randomly selected sires from the exotic
population were mated to randomly selected dams from the indigenous population to
produce F1 crossbreds and their reciprocals which were also generated by selecting sires
from the indigenous population and dams from the exotic population and randomly
mating them. Therefore, at generation 2050, four different populations namely: exotic,
exotic x indigenous, indigenous x exotic and indigenous were available to conduct the
selection experiment. The impact of these assumptions on the level and pattern of LD
will be addressed in the discussion section.
The resulting populations in generation 2050 were used in the different breeding
schemes in the ten subsequent generations (T1 – T10) that were produced (Figure 4.1).
Pure-breeding was carried out for the exotic and indigenous populations separately. The
synthetic population was produced by mating F1 sires to their reciprocal crossbred dams
followed by inter se mating. For the backcross breeding program, repeated backcrossing
was conducted for generations T1 - T5, followed by inter se mating of the backcross
population from generations T6 – T10. All lines in the breeding schemes were kept at a
90
constant size of 1,000 breeding candidates and 5% of males and 50% of females were
randomly selected in each line per generation. Throughout the selection experiment, all
sires and dams were replaced each generation thus discrete generations were assumed.
The breeding structure described above allowed for the evaluation of GS in previously
selected and unselected pure lines and their crosses used for training and validation. The
goal was to mimic the selected exotic and unselected indigenous pig populations
currently used for pork production in tropical developing countries.
4.2.3 Genome structure
The simulated genome consisted of 5 chromosomes of 150 cM each with 250
multi-allelic segregating QTL and a marker density of 22 bi-allelic loci per cM evenly
spaced across the genome (Table 4.1). This was chosen to make the simulation
computationally tractable and in such a way that the marker density represents
approximately 60 K SNP chip currently available (Ramos et al. 2009). To establish
mutation drift equilibrium (MDE) in the ancestral population after 2,000 generations,
about 3 and 20 times as many markers and QTL, respectively, were simulated with equal
starting allele frequencies and a recurrent mutation rate of 2.5 x 10-3
for markers and
infinite-allele mutation rate of 2.5 x 10-5
for QTL.
The required number of QTL and markers were drawn at random from
segregating loci with a minor allele frequency (MAF) of ≥ 0.05 at generation 2000 before
creating the exotic and indigenous populations. No new mutation was simulated in the
recent populations, resulting in many of the markers and QTL not segregating in the
exotic population after 50 generations of phenotypic selection. Mutation rate per
91
generation in mammals is estimated to be about 2.5 x 10-9
(Kumar and Subramanian,
2002) and may not have resulted in a significant number of new mutants after 50
generations in real populations. The Haldane mapping function with no interference and
Mendelian inheritance of all loci was assumed. The allelic effects of the QTL including
original and new mutations were sampled from a gamma distribution with a shape
parameter of 0.4 (Hayes and Goddard, 2001) and scaled such that the sum of the QTL
variances in the last historical generation equals the input QTL variance. The QTL was
randomly distributed across the genome. Figure 4.2 showed an L-shaped gamma
distribution of the true QTL effects for the three traits studied.
The magnitude and extent of LD in the founder populations at generation 2000
and 2050 were estimated as average correlations between pairs of marker alleles (r2)
4.1
(Hill and Robertson, 1968), where DAB = DA1B1 = pA1B1 – pA1pB1, pA1 is the frequency of
allele A1 at locus A, pB1 is the frequency of allele B1 at the locus B and pA1B1 is the
frequency of haplotype A1B1 in the population. Only markers with a MAF ≥ 0.10 were
included in this analysis. The r2 value averaged across the genome was compared with
their expectation based on Sved 1971: E (r2) , where c was the
recombination rate calculated as and
Ne was as before.
4.2.4 Phenotypes
To simulate data for each animal with a record at generation 2000, a finite
additive locus model was assumed such that the true breeding value of an individual was
92
equal to the sum of the 250 QTL effects scaled on the basis of the specified heritability
(h2) and phenotypic variance (σ
2p). Given the genetic variance, environmental effects
were generated from a normal distribution with a mean of zero and a constant variance
set to give the desired h2. The phenotype was assumed to be the sum of the genotypic and
environmental effects, i.e. no pleiotropy or genotype-environment interactions were
simulated. For the traits considered, h2 and σ
2p used in the simulation were obtained from
a previous meta-analysis study carried out for the tropical environment (Chapter 3).
Three traits, namely number born alive (NBA), average daily gain (ADG) and back-fat
thickness (BFT), were simulated independently assuming a normal distribution. All
individuals within the various genetic groups in the last 10 generations were genotyped
for the available markers. In addition, for training the prediction models, individuals in
the training dataset were assumed to be recorded for all traits studied.
4.2.5 Model for comparison
The phenotypic information from the training population was used to calculate the
estimated breeding values (EBV) for the pedigree-based model according to Henderson
(1984). The model used was
y = μ1n + Za + e 4.2
where y is observed phenotype of size n, μ is the overall mean, 1n is a vector of n ones, a
is a vector of random additive genetic effects of each individual ~N (0, Aσ2
a), A is the
additive relationship matrix, Z is a design matrix for the additive genetic effect and e is a
vector of random residuals ~N (0, Iσ2
e). The variance ratio required in the mixed model
equations was (1-h2)/h
2 where the true heritability at each generation was used.
93
For the genome-based model, the marker effects were estimated using ridge
regression described in Meuwissen et al. (2001) and extended by Fernando et al. (2007).
The statistical model was
y = μ1n + Wg + e 4.3
The mixed model equations (MME) were:
4.4
where y is observed phenotype of size n, g is a vector of allele substitution effects due to
the ith genotype ~N (0, I ), W is a design matrix that has a 0, 1, and 2 for the number
of alleles of type gi present in the jth animal, p is the base allele frequency at ith locus, q
is (1-p), σ2g is the true genetic variance at each generation and m is the total number of
markers simulated. The other parameters are as describe previously.
Once estimates of marker effects were obtained from the training data set, the
genomic breeding values of kth animal (GBVk) in both the training and validation data set
and for the future selection candidates in the genomic selection scheme was computed as
GBVk = 4.5
where and are the recorded genotype and the estimated marker allele substitution
effect at the ith locus, respectively, and m is the total number of segregating markers. For
estimating GBV of an individual in all of the different populations simulated, about
14,500 segregating markers distributed across five chromosomes with MAF ≥ 0.10 and
based on the exotic population marker panel were used.
Five sets of training data from the previously selected and unselected populations
of size 275, 775, 1,000, 1,275 and 2,275, and consisting of a single and multi-
94
generational population structure made up of males and females were initially used to
train the prediction model in order to evaluate the required reference population size and
structure for implementing GS and the persistency of the accuracy of GBV. In the pure
breeding scenario, the validation population consisted of the generation following the
training population and were made up of 1,000 breeding candidates while in another
scenario both the training and validation sets were of different population structures.
4.2.6 Analysis of results
Each scenario was replicated 10 times starting each replicate from the same base
population up to generation 2050. Comparisons between methods were based on the
correlation between the estimated and true breeding values. Standard errors were
computed as the standard deviation of average accuracy across the 10 replicates, divided
by . All analyses were coded in R statistical software (Ihaka and Gentleman, 1996)
and the Statistical Analysis System (SAS Institute Inc., 2009) was used for least squares
analysis of estimates.
4.3 RESULTS AND DISCUSSION
4.3.1 Simulation assumptions
Table 4.1 shows the parameters used to simulate the datasets for this study. As
described in the methods section, these parameters were chosen to make the simulation
computationally tractable, while providing a realistic dataset for the target populations.
The assumption of 5 chromosomes rather than 18 in the case of swine are not expected to
change substantially the main conclusions of the research as long as the number of
95
markers used are dense enough to represent the 60 K SNP markers evenly distributed
across the 18 pairs of chromosome and the number of QTL affecting the trait are fairly
large. The results from this study were generally similar to the case where only one
chromosome was simulated with the same number of markers and QTL (data not shown).
In reality, the number of QTL contributing to a given trait will not be known in advance.
Also, using a dense marker map has an important effect on the accuracy of GBVs
because a higher number of markers, when equally distributed across the genome, will
increase the probability that each QTL is in high LD with at least one marker (Calus et
al., 2008; Goddard, 2009)
The distributions of the true additive QTL effects for various traits studied are
presented in Figure 4.2. In general, there were many QTLs with small effects and a few
QTL with large effects which follows the L-shaped gamma distribution that Hayes and
Goddard (2001) suggest as the true distribution of QTL effects. The implication of this
assumption on the method used for estimating marker effects based on marker-QTL
association will be addressed later. Non-additive genetic effects were not simulated in
this study, but they play a role in a litter-bearing species like swine.
The scenarios simulated in this study mimics the exotic and indigenous pig
populations that are used for pork production in tropical developing countries (Pathiraja,
1986; Pathiraja, 1987). The exotic pigs were originally imported into these regions to
upgrade the genetic merit of the indigenous population (Blench, 2000). Due to the
improved performance of exotic pig under tropical conditions they have continued to
dominate the swine industry with intensive breeding. Because the conditions in the
tropics have failed to improve, the performance of the exotic pigs is dwindling
96
necessitating the integration of the indigenous pigs in pork production (Chimonyo and
Dzama, 2007). However, as the infrastructure for genetic improvement of indigenous pig
using the traditional method is lacking in developing countries (Kahi et al., 2005), the
goal of this study was to evaluate a scenario where GS was applied and compared with
the results from a conventional approach.
4.3.2 Extent of LD in founders
The MAF distributions for the SNP markers in the founder populations are given
in Figure 4.3. This distribution of MAF showed that about 30% of all loci in the exotic
founders have become fixed for alternate SNP alleles while the majority of the loci are
still segregating in the indigenous and crossbred founders. Consequently, evaluation of
LD in the founders and genomic predictions using the SNP markers was carried out for
MAF ≥ 0.10.
In this study, divergence between the founder populations was created by drift
and differential selection pressures imposed on the founders (Figure 4.1). The extent of
LD in the founders used to produce all training populations showed that the decay of
average r2 with genomic distance differed between them (Figure 4.4). LD was
significantly higher on short distances (0 – 2 cM) for the exotic founders compared to the
indigenous, whereas the long range LD (>10 cM) was quite similar for both groups. The
LD levels used in this simulation study are consistent with observations in real swine data
after several generations of divergence, domestication and selection events (Harmegnies
et al., 2006; Du et al., 2007; Amaral et al., 2008; Jafarikia et al., 2010).
97
Note that for the exotic and indigenous founders, average LD was similar to the
expected LD based on Sved‘s (1971) formula for effective population sizes of 167 and
667, respectively. In Sved (1971), the expected level of LD assumes a closed random
mating population with constant historical effective population size, which are not the
same assumptions in the current study. Nevertheless, the estimated constants give a good
indication that the effective number of founders and, therefore, the extent and decline of
LD with distance, differed substantially between the exotic and indigenous populations.
4.3.3 Requirements for training marker effects
The implementation of GS requires a reference population (RP) with phenotypic
and genotypic data for training the prediction model. For a breeding system that uses
populations with differing selection histories, there is a need to know the appropriate
breeding structure and size of the RPs. The RP should provide estimates of marker effects
for predicting GBVs in different populations with high accuracy and with minimal
frequency of re-estimation of marker effects.
Accuracies of GBVs with alternative sizes and structure of RP and tested in
previously selected exotic and unselected indigenous populations were investigated and
the results are given in Figure 4.5. Accuracy increased as RP size increased and improved
with a multigenerational data structure. An exciting observation is that GS works well for
traits with low heritability both for a previously selected population in moderate LD and
for an unselected population in low LD. From Figure 4.5, a trait of low heritability in a
previously selected population can achieve accuracy greater than the heritability of the
trait for at least three generations with a multi-generational training population of 2,275
98
individuals. An unselected population in low LD achieves accuracy greater than the
heritability of the trait for at least five generations with a single progeny generation of
1,000 individuals.
For a trait with high heritability in both exotic and indigenous populations (Figure
4.6), regardless of the size and structure of the RP, the accuracy cannot exceed the
heritability in the generation used for training. For similar heritability, Meuwissen et al.
(2001) found an accuracy of 0.732 using the same method of estimation and RP of 2,200
phenotypes/genotypes. Accuracies of this magnitude or higher were only observed in the
unselected indigenous population in low LD. The selected exotic population due to
previous selection events have dissipated much of the genetic variability. Thus, in the
presence of moderate LD in the exotic population, accuracies of GBVs were lower than
expected. Accuracy of GBVs is not only a function of how close the markers are to the
QTL to be able to explain the variation in the trait, but also a function of how much of the
genetic variability in the trait of interest is left to be explained. In addition, moderate to
large-sized QTL effect would be in high frequency of the favourable allele or even fixed,
making more difficult the predictions.
Another striking observation from Figures 4.5 and 4.6 is that, regardless of RP
size and structure, the decay of accuracy over generations under random mating is faster
for an unselected population in low LD level than for a selected population in moderate
LD which strengthens the importance of LD in GS schemes. Clearly, accuracy of GBVs
using a multigenerational RP with size of not less than 2,000 individuals is superior to a
single generation RP. Similar conclusions have been made by other researchers
(Meuwissen et al., 2001; Muir, 2007). However, if resources are a limitation in choosing
99
appropriate RP, then a population size of not less than 1,000 individuals should be
utilized if a considerable level of accuracy is desired, but this will require re-estimation of
marker effects after 2 generations of selection.
4.3.4 Accuracy of GBVs in training and validation populations
Although, the training population size of 2,275 individuals showed better
accuracy of GBVs (Figures 4.5 and 4.6), the results of genomic evaluation presented
from this section forward were based on a training set of 1,275 individuals chosen
because of the anticipated cost of genotyping. The correlations between estimated and
true breeding values of individuals in the different training populations (TP) and in their
first generation progeny population used for validation (VP) and for different traits and
methods of EBV estimation are presented in Table 4.2. When validation of GS was in the
first generation progeny of the training individuals, accuracy of GBVs was reduced by
3% in exotic, 10% in indigenous, 10% in synthetic and 27% in backcross populations for
a trait with low heritability. For a trait with high heritability, accuracies of GBVs were
reduced in the different validation populations except for the synthetic population where
accuracy was increased by 3%, on average. Accuracy of GBVs was higher in the
unselected indigenous purebreds than in the previously selected exotic purebreds for all
traits. More so, as the proportion of the exotic alleles in the crossbred individuals
increased, accuracy of GS increased substantially. Backcross populations had higher
accuracy for GS than the synthetic and purebred lines.
A change of selection method from a pedigree-based approach to GS resulted in
an increase in accuracy of breeding values by 11% in the TP and 61% in the VP, on
100
average (Table 4.2). In traits with low heritability where the pedigree-based scheme
needed help, accuracy of GS was increased by 16% in the TP and 82% in the VP, on
average. The gain in accuracy for a trait with high heritability was 7% for TP and 48%
for VP, on average. The application of a genome-based scheme in a crossbred
populations resulted in an increase in accuracy of EBVs by 13% in the TP and 82% in the
VP while purebred population gained an increase in accuracy of EBVs by 7% in TP and
30% in VP, on average.
The high accuracies of GBVs observed when training in the unselected
indigenous population in low LD may be due to the method of ridge regression BLUP, as
used in this study, which fitted all markers simultaneously and thus captured more
genetic relationships resulting in higher accuracy. Habier et al. (2007) indicated that the
ridge regression method exploits genomic relationships among individuals and LD
between markers and QTL to improve accuracy of GBVs. In addition, the indigenous
population was not subjected to directional selection in the past so many of the markers
were still segregating and could capture more of the QTL effects. Note that the same
marker map used in the exotic population was applied in all other population including
the unselected indigenous population. Furthermore, the genetic variance for the traits in
the indigenous population was still very large to allow for higher accuracy of GBV.
The lower accuracy observed in the exotic population was because of the previous
selection that resulted in many of the markers not segregating and the loss of genetic
variation due to selection leading to a small amount of genetic variance being explained
by marker-QTL association. Luo et al. (1997) noted that the greater the LD between a
pair of loci like marker and QTL, the larger is the variance that is associated with the
101
marker. However, there is a limitation to the amount of genetic variation that can be
explained. Another reason for the low accuracy of GBV in the selected exotic population
was that the major QTL alleles which normally will be in high frequency had their effects
estimated with difficulty due to low average MAF compared to the unselected indigenous
population where they segregate with higher frequency.
The accuracy of GBVs when the training and validation population differ in
structure and composition are presented in Table 4.3. Training in the same breed as the
validation population resulted in greater accuracy for majority of the cases. Accuracies
tended to be lower if the training and validation population differ in structure, but
reductions in accuracy were not large in some cases and depended on the breed
composition of the training individuals. Training in the selected exotic population
resulted in lower accuracy when validation was in the unselected indigenous population
but accuracy increased slightly when validation was in the crossbred. A similar trend was
observed when training in the indigenous population and validating in the exotic and
crossbred populations, respectively. The reason been that different QTLs were fixed or at
low MAF in the exotic and indigenous populations such that QTL that explain more
variation in the traits are different in the two populations. In the crossbred, QTL alleles
from both populations that control variations in trait are present resulting in higher
accuracy than in the purebreds.
For all traits studied, training in the crossbred population resulted in GBV
accuracy of 0.45 – 0.97 when validation was in the crossbred and a considerable range of
accuracy (0.12 – 0.36) when validation was in the purebreds (Table 4.3). All crossbred
individuals consisted of varying proportions of the purebred alleles resulting in greater
102
accuracy of GBV when validation was in the purebred populations. By estimating allele
substitution effects based on crossbred phenotypes, the effects of purebred alleles will be
estimated against the genetic background in which they will be expressed. This might be
a potential practice of GS in the tropics in future, for example, marker allele effects can
be trained in a range of RPs made up of different breed compositions, admixtures and
nondescript breeds. The estimates of marker effects for a defined SNP panel can then be
used for genomic predictions of purebreds followed by selection within breed in the
nucleus.
The drop in accuracy when the training and validation population differed in
structure can also be attributed to the differences in the level of LD between the two
populations and differences in allele frequency. Whereas the exotic population in this
study, for example, were in moderate LD, the indigenous population had very low LD
with differing MAF of SNP markers. Thus, even though the same marker panel was
utilized for GS in both populations, accuracy was very low. The marker-QTL linkage
phase may be different in both populations. Although this phenomenon was not evaluated
in the present study, marker-QTL linkage phase will need to be consistent in both
populations for GS to work (Dekkers and Hospital, 2002; Goddard et al. 2006). As
distance in time since divergence between this two population is high (about 50
generations, Figure 4.1), there is a greater chance for recombination to break down the
LD that was present in the ancestral population and drift to create new LD within each
subpopulation which may change the marker-QTL phase (Hill and Robertson, 1968;
Goddard et al., 2006). The same explanation could apply to the case of training in
103
purebreds and validating in crossbreds and vice versa. The use of denser marker panel for
GS implementation in future will ensure consistency in marker-QTL phase.
To further explain the differences in accuracy of GBVs between the purebred and
crossbred training populations described earlier, LD in the different training populations
was examined by comparing the average r2 between adjacent marker pairs for an average
distance of 0.05 cM (Table 4.4). There were significant differences in the amount of LD
between the training and validation populations. Figure 4.4 depicts how LD decayed with
distance in the founders of the training populations and showed differences in the rate of
breakdown of LD between the populations. Whereas LD in the purebred population was a
result of historic LD in the founders and drift (Sved, 1971; Du et al., 2007; Amaral et al.,
2008), LD in the crossbreds came from two sources (Lo et al., 1993). The first part is the
average LD that existed within the parental populations, referred to as old LD, and the
second part is the LD generated in the cross as a result of differences in gene frequencies
between the parental lines, referred to as new LD because it was created by a recent event
(Toosi et al., 2010). Old LD is usually confined to shorter distances due to accumulation
of historic recombination events while new LD extends over longer intervals. The LD at
short intervals is a function of Ne in the distant past, whereas LD at longer intervals
reflects Ne in the recent past (Hayes et al., 2003).
The phenomenon described above explains the higher accuracy of GBVs
observed in the crossbred population when compared to results from the purebreds.
Similar observations were documented by Toosi et al. (2010). In addition, alleles in a
crossbred line originated from purebred parental lines. If these purebred lines are not
closely related as in this study, the SNP allele effects will depend on their line of origin.
104
Thus, a model that accounts for line of origin effects may be warranted to avoid
overestimation of allele substitution effects. Ibanez-Escriche et al. (2009) carried out a
simulation study to evaluate the effect of using a model with breed-specific SNP allele
effects versus classical genomic model with across-breed SNP genotype effects. They
concluded that in implementing GS in crossbred populations, models that fit breed-
specific SNP allele effects may not be necessary if high density markers are utilized to
trace ancestor alleles with precision ( Ibanez-Escriche et al., 2009). As a result a breed-
specific SNP model was not entertained in the present study.
In concluding this section, training the prediction model based on crossbred
performance in the field for genetic improvement of purebreds in the nucleus holds some
promise. Although the results from this study showed lower accuracy of GBVs for this
kind of scenario due to lower level of LD in one of the parental lines, if LD was greater
or if marker density can be increased, this will be a potential option in the future. Traits
with low heritability such as reproduction and disease resistance, which are a challenge to
improve with conventional methods, may show greater genetic progress using a genome-
wide selection scheme.
4.3.5 Persistency of accuracy of GBVs
The decay of accuracy of GBVs in different populations over ten generations of
random mating for all traits studied are presented in Figure 4.7. The results showed that
for the purebred populations, accuracy tended to decrease faster under random mating in
the early generations and stabilises in the later generations. For the synthetic line, there
was stability in accuracy across generations. The backcross population exhibited the same
105
pattern as in the purebreds. From Figure 4.7, a trait with low heritability in all populations
studied can achieve accuracy greater than the heritability of the trait for at least five to
seven generations without re-estimation of marker effects. Also, for a trait with moderate
to high heritability, marker effects should be re-estimated after two generations if
accuracy greater than the trait heritability is desired.
Training the genomic prediction model in generations closer to the validation
dataset will lead to more accurate GBVs compared to training in more distant
generations. With increasing generations, the relationships between the training and
validation population will become weak due to recombination events and changes in
marker-QTL phase, and a reduction in the LD level between markers and QTLs. Muir
(2007) observed that several generations following the estimation of marker effects the
accuracy of GBVs was reduced and suggested that these effects should be re-estimated
after three to four generations of selection. For crossbred population, while
recombination breaks down existing LD between markers and QTL, recent crosses create
new LD which helps to stabilise accuracy. Traits with low heritability are known to
benefit from genome-wide selection by exploiting the Mendelian sampling effects
(Goddard and Hayes, 2007; Daetwyler et al., 2007).
Figure 4.8 presents the influence of trait variance on the persistency of GBV
accuracy in different populations over ten generation of random mating. In the exotic
population with a previous history of intense directional selection that could lead to loss
of genetic variability, accuracy for a low heritability trait like number born alive (NBA)
was stable over ten generations due to moderate LD between markers and QTL and more
genetic variance explained. For other traits in this population, accuracy decreased due to
106
loss of genetic variance (Bulmer, 1976). For the unselected indigenous population,
accuracy decreased at a faster rate in each generation of random mating for all traits. This
decrease in accuracy is due to low initial LD in this population and recombination
between markers and QTL which breaks down LD. For the crossbred populations, there
was appreciable stability in accuracy over generations of random mating for all traits.
New LD and higher allele frequency created by recent crosses in the crossbred replaced
LD that was loss by recombination between markers and QTL, thus accuracy of GBV
was maintained.
4.4 CONCLUSIONS AND RECOMMENDATIONS
The accuracy of GBVs can be very high for populations in moderate to low LD
regardless of their selection history. High accuracy requires high density markers that are
well distributed across the genome to allow for marker-QTL association. For best result,
a reference population with genotypic and phenotypic data for training the prediction
model should have a multigenerational data structure with greater than 2000 individuals.
Young animals with no phenotypic records can have GBVs early in life with high
accuracy for selection decision based on their genotypic data. Crossbred training
population performed better than purebred training set when validation was in
populations with similar structure as the training set.
In situations where infrastructures for swine genetic improvement are lacking, a
genome-wide selection program can circumvent the need to develop facilities for a
conventional selection scheme and will guarantee faster genetic progress especially for
traits with low heritability. In a system where purebreds in the nucleus are crossed to
107
produce crossbred market pigs, the prediction model can be trained based on crossbred
records in the field for genetic improvement of purebreds in the nucleus.
108
Table 4.1 The parameters used in the simulation program
Parameters Values
Genome size per chromosome 150 cM
Number of chromosomes 5
Marker density per cM 22
Number of segregating QTL per chromosome 50
Mutation rate of Marker locus 2.5 x 10-3
Mutation rate of QTL locus 2.5 x 10-5
Minor allele frequency 0.05
Distribution of additive QTL effects Gamma (shape = 0.4)
Traits
Number born alive (NBA) = 0.08; = 7.73
Average daily gain (ADG) = 0.28; = 10361.20
Back fat Thickness (BFT) = 0.63; = 20.88
Number of animals with phenotypic or genotypic records, or both
Training 1275
Prediction/Validation 1000
Number of replicates per scenario 10
109
Table 4.2 Correlations1 (r) between estimated breeding values ( ) and true breeding
values ( ) in the training (TP) and validation (VP) populations
Traits Training Methods r ( , ) r ( , ) Δ2(%)TP Δ(%)VP
NBA Exotic Pedigree 0.28 0.20
Genomics 0.31 0.31 10.71 55.00
Indigenous Pedigree 0.45 0.36
Genomics 0.45 0.41 0.00 13.89
Synthetic Pedigree 0.33 0.24
Genomics 0.37 0.34 12.12 41.67
Backcross1 Pedigree 0.56 0.23
Genomics 0.72 0.59 28.57 156.52
Backcross2 Pedigree 0.56 0.21
Genomics 0.69 0.52 23.21 147.62
ADG Exotic Pedigree 0.36 0.27
Genomics 0.40 0.35 11.11 29.63
Indigenous Pedigree 0.69 0.51
Genomics 0.70 0.60 1.45 17.65
Synthetic Pedigree 0.52 0.41
Genomics 0.61 0.62 17.31 51.22
Backcross1 Pedigree 0.85 0.42
Genomics 0.91 0.78 7.06 85.71
Backcross2 Pedigree 0.84 0.42
Genomics 0.91 0.76 8.33 80.95
BFT Exotic Pedigree 0.45 0.31
Genomics 0.51 0.45 13.33 45.16
Indigenous Pedigree 0.84 0.56
Genomics 0.85 0.68 1.19 21.43
Synthetic Pedigree 0.59 0.44
Genomics 0.69 0.70 16.95 59.09
Backcross1 Pedigree 0.95 0.54
Genomics 0.96 0.86 1.05 59.26
Backcross2 Pedigree 0.95 0.54
Genomics 0.97 0.84 2.11 55.56 1Correlations are least square means of 10 replicates per scenario (Standard error ≈ 0.020)
2Percentage change in accuracy due to change in selection method
110
Table 4.3 Average accuracy1 of genomic breeding values when training and validation population differ in structure
Traits
Training
Validation population
Exotic Indigenous Synthetic Backcross1 Backcross2
NBA Exotic 0.31 (0.039) 0.07 (0.026) 0.20 (0.049) 0.22 (0.047) 0.22 (0.062)
Indigenous 0.08 (0.036) 0.45 (0.055) 0.13 (0.053) 0.13 (0.024) 0.16 (0.012)
Synthetic 0.21 (0.054) 0.12 (0.028) 0.37 (0.038) 0.45 (0.027) 0.47 (0.039)
Backcross1 0.30 (0.024) 0.13 (0.046) 0.72 (0.019) 0.72 (0.016) 0.77(0.015)
Backcross2 0.26 (0.050) 0.19 (0.034) 0.68 (0.029) 0.77 (0.025) 0.69 (0.014)
ADG Exotic 0.40 (0.062) 0.02 (0.022) 0.20 (0.057) 0.23 (0.055) 0.27 (0.056)
Indigenous 0.03 (0.062) 0.70 (0.014) 0.27 (0.038) 0.22 (0.068) 0.22 (0.061)
Synthetic 0.22 (0.058) 0.21 (0.025) 0.61 (0.015) 0.78 (0.012) 0.79 (0.019)
Backcross1 0.35 (0.024) 0.29 (0.046) 0.85 (0.007) 0.91 (0.009) 0.86 (0.012)
Backcross2 0.24 (0.049) 0.32 (0.020) 0.86 (0.012) 0.90 (0.008) 0.91 (0.009)
BFT Exotic 0.51 (0.029) 0.05 (0.019) 0.27 (0.066) 0.26 (0.046) 0.35 (0.069)
Indigenous 0.06 (0.019) 0.85 (0.014) 0.27 (0.022) 0.27 (0.043) 0.25 (0.021)
Synthetic 0.22 (0.020) 0.18 (0.031) 0.68 (0.012) 0.86 (0.006) 0.86 (0.007)
Backcross1 0.26 (0.019) 0.33 (0.028) 0.88 (0.012) 0.96 (0.002) 0.89 (0.004)
Backcross2 0.20 (0.030) 0.36 (0.014) 0.92 (0.005) 0.94 (0.002) 0.97 (0.001) 1Values are average correlations between estimated and true breeding values and their standard errors (s.e) over 10 replicates per scenario
111
Table 4.4 Average linkage disequilibrium between adjacent marker pairs in training and
validation1 populations
Populations Exotic Indigenous Synthetic Backcross1 Backcross2
Training 0.30 0.05 0.09 0.13 0.06
Validation 0.30 0.06 0.10 0.18 0.05
1Validation population are the immediate progeny of the training population
112
Figure 4.1 Schematic representation of the simulated population structure
Selection strategy began from T1 - T10 and based on Ne = 91 for all populations simulated. Inter se
mating of backcross population took place from T6 - T10
113
Figure 4.2 Distribution of true allelic effects simulated for the three traits studied
Distribution of allele effects for ADG
Allele effects
Fre
quency
0 5 10 15
0100
200
300
400
500
Distribution of allele effects for NBA
Allele effects
Fre
quency
0.00 0.05 0.10 0.15 0.20
0100
200
300
400
Distribution of allele effects for BFT
Allele effects
Fre
quency
0.0 0.2 0.4 0.6 0.8 1.0 1.2
0100
200
300
400
114
Figure 4.3 Distribution of average minor allele frequency (MAF) for the SNP markers in
the founder populations
115
Figure 4.4 Average linkage disequilibrium (LD) as measured by r2 against distance
(cM) between a pair of marker in different founder populations
Ne _167 and Ne _667 are LD expectations based on Sved (1971): E (r2) ≈ 1/ (1 + 4Nec),
where Ne = 167 or 667 are the effective population size for the exotic and indigenous
population, respectively, and c is the recombination rate calculated as 0.5(1 – exp(-2*map
distance)). The graph was based on average r2 over 10 replicates plotted against distance
116
Figure 4.5 Effect of training population size and structure on accuracy and persistency
of accuracy following ten generations of random mating in selected exotic (top) and
unselected indigenous (bottom) populations and for NBA
Population structure includes founders (F_275), founders and progeny (F_P_775 and F_P_1275),
progeny (P_1000) or multigenerational dataset (M_G_2275). The straight line is the accuracy
level based on trait heritability
117
Figure 4.6 Effect of training population size and structure on accuracy and persistency
of accuracy following ten generations of random mating in selected exotic (top) and
unselected indigenous (bottom) populations and for BFT
Population structure includes founders (F_275), founders and progeny (F_P_775 and F_P_1275),
progeny (P_1000) or multigenerational dataset (M_G_2275). The straight line is the accuracy
level based on trait heritability
118
Figure 4.7 Decay of accuracy of GBVs in different populations and traits under random
mating for ten generations. The straight line is the accuracy level based on trait
heritability
119
Figure 4.8 Influence of trait heritability on persistency of accuracy of GBVs in different
populations under random mating
120
CHAPTER 5
OPPORTUNITIES FOR GENOME-WIDE SELECTION FOR PIG
BREEDING IN DEVELOPING COUNTRIES
E. C. Akanno1, F. S. Schenkel
1, M. Sargolzaei
13, R. M. Friendship
2 and J. A. B.
Robinson1
1Centre for Genetic Improvement of Livestock,
Department of Animal & Poultry Science, University of Guelph, Guelph, ON, Canada
2Department of Population Medicine, Ontario Veterinary College,
University of Guelph, Guelph, ON, Canada
3L'Alliance Boviteq, Saint-Hyacinthe, Quebec, Canada
121
5.0 ABSTRACT
Genetic improvement of exotic and indigenous pigs in tropical developing
countries is desired. Implementations of traditional selection methods on tropical pig
populations are limited by lack of necessary infrastructure. Genome-wide selection (GS)
provides an approach for achieving faster genetic progress without developing a pedigree
recording system for this environment. The implications of GS on long term gain and
inbreeding should be studied prior to actual implementation especially where low linkage
disequilibrium (LD) is anticipated in the target population. A simulation case-study of
this option was carried out based on the available 60 K single nucleotide polymorphism
panel for porcine genome. Computer simulation was used to explore the effects of
various selection methods, trait heritability and different breeding programs when
applying GS. Genomic predictions were based on ridge regression method. Genome-wide
selection performed better than BLUP and phenotypic selection methods by increasing
genetic gain and maintaining genetic variation while lowering inbreeding especially for
traits with low heritability. Indigenous pig populations in low LD can be improved by
using GS if high density marker panel are available for use. Combining GS with repeated
backcrossing of a crossbred population to previously selected exotic pig population in
moderate LD promises to rapidly improve the genetic merit of the commercial
population. Application of this novel method on a real population will need to be carried
out to validate these results.
Keywords: Genome-wide selection, indigenous population, swine, developing countries
122
5.1 INTRODUCTION
Genetic selection is the process that determines which individuals become
parents, how many offspring they may produce and how long they remain in the breeding
population (Bourdon, 1997). This method has been successful in making long-term
genetic change in animals and is widely accepted as the primary cause of adaptive
evolution within natural populations. For swine populations in many tropical developing
countries, the approach has been to improve by basic phenotypic selection (Pathiraja,
1986; Kahi et al., 2005). In phenotypic selection, a potential candidate is selected based
on their own performance, disregarding pedigree or the performance of sibs or progeny.
Theory shows that estimated breeding value (EBV) for an individual as a selection
criterion is equal to = h2 ( - ), where and are the individual‘s EBV and
performance record and is the population average for the trait. A problem with EBV
derived in this way is that the estimates are confounded by other extraneous factors that
need to be separated. Unless trait h2
is high, accuracy is generally low, leading to slower
genetic progress. Moreover, the reduction of genetic variance is minimal and rate of
inbreeding (ΔFG) is low for this method, because co-selection of relatives is unlikely
except for the case of high h2 (Daetwyler et al., 2007).
Henderson (1984) developed a pedigree-based method that uses information on
relatives to provide EBVs that are adjudged to be best linear unbiased predictions
(BLUP). When BLUP EBVs were used in ranking candidates and truncating the
distribution to choose those with the highest values, genetic change (ΔG) was increased,
but so was ΔFG (Belonsky and Kennedy, 1988) with a reduction in genetic variance
(Bulmer, 1976). BLUP leads to co-selection of sibs and higher rates of inbreeding
123
especially when h2 is low
(Sonesson et al., 2005). Although, the BLUP method has been
successful for swine genetic improvement in developed countries (Belonsky and
Kennedy, 1988; Kennedy et al., 1996; Gibson et al., 2001), implementation in developing
countries is limited to small nucleus herds with pedigree recording (Smith, 1988; Kahi et
al., 2005). The breeding value of an individual ( ) estimated using BLUP consists of
three components as given by the equation = (0.5) + 0.5 ( ) + , where and
are the sire and dam EBVs, respectively, and is the Mendelian sampling term.
The accuracy of the BLUP EBV for an individual depends mainly on the accuracy of its
parent EBVs, number of own phenotypic observations and collateral relatives including
progeny (Daetwyler et al., 2007). The more information on the parents and relatives that
is included in the genetic model, the higher the accuracy of EBV for the individual,
especially for lowly heritable traits.
Today, swine breeders in developing countries are interested in selection methods
that are straight forward to implement at reasonable cost. The focus is on the
identification of quantitative trait loci (QTL) associated with variation in the traits of
interest. The infinitesimal model assumed by phenotypic and BLUP selection is expected
to give way to genomic models that will enable the dissection of phenotypic variation and
unravel the interplay between genes and environment (Visscher et al., 2008). Methods
that guarantee a sustainable increase in ΔG while maintaining genetic variation and
keeping the level of ΔFG low are preferred long-term by swine breeding industry. In
schemes that allow a young animal to be selected before phenotypic performance can be
measured, phenotypic selection will be difficult to implement and BLUP selection will
have low accuracy for an individual‘s EBV based only on parent average.
124
One possible solution for selection in tropical pig production is genome-wide
selection (GS) based on genomic breeding values (GBV) predicted from dense single
nucleotide polymorphism (SNP) marker genotyping (Meuwissen et al., 2001; Muir,
2007). The ability of GS to use genetic markers to exploit Mendelian sampling effect is
promising (Daetwyler et al., 2007; Nielsen et al., 2010). Exploitation of Mendelian
sampling variation is the major source of increased accuracy which directly increases ΔG
and reduces the level of ΔFG for GS over conventional approaches.
Therefore, the objective of this study was to quantify how much gain is to be
expected in production traits using GS compared to conventional methods of selection for
pig breeding programs in developing countries. The impacts of various selection methods
on inbreeding, genetic variance, realized accuracy, linkage disequilibrium and QTL allele
frequency change were also investigated.
5.2 MATERIALS AND METHODS
5.2.1 Simulation software
A modified version of the QMSim simulation software developed by Sargolzaei
and Schenkel (2009) was used for this study. This version of the software allows for the
marker effects to be estimated externally using any preferred method and GBVs of
selection candidates are calculated and provided to the program for selection purpose. In
addition, other methods of breeding value estimation such as BLUP and phenotypic
selection can be applied and various selection events can be imposed on the populations.
Table 5.1 gives a description of the parameters used for the simulation.
125
5.2.2 Population structure
All pigs found in tropical developing countries were assumed to come from a
common base of unrelated individuals which served as ancestors to the exotic and
indigenous populations. The common base population was mated based on the union of
gametes randomly sampled from the male and female gametic pools for 1,000
generations with effective population size (Ne) of 5,000 breeding individuals (Figure 5.1).
This ancestral population was later contracted to Ne of 1,050 individuals consisting of
300 breeding males at the end of another 1,000 generations in order to generate initial
historic linkage disequilibrium (LD). This is typical of Ne in livestock populations which
have been declining due to domestication (Henson, 1992). The ancestral population was
split into two unequal sized purebred subpopulations (referred to as exotic and indigenous
populations) at generation 2001.
The exotic population was created by drawing the top ranking 50 males and top
ranking 250 females based on their phenotypes from the common base population
resulting in Ne of 167 and subjected to phenotypic selection for the trait under
consideration for 50 generations with an intensity of 10% for males and 50% for females.
Each female produced 4 offspring (2 males and 2 females) resulting in 1,000 breeding
animals in each generation from which selection was made. The remaining members of
the common base population consisting of 250 males and 500 females were allowed to
mate at random concurrently for 50 generations resulting in the indigenous population
with Ne of 667. The Ne assumed in this study was chosen to yield a simulated level of LD
observed in real life. The actual Ne might differ slightly.
126
At generation 2049 in the simulation, randomly selected sires from the exotic
population were mated to randomly selected dams from the indigenous population to
produce F1 crossbreds and their reciprocals which were also generated by selecting sires
from the indigenous population and dams from the exotic population and randomly
mating them. Therefore, at generation 2050, four different populations namely: exotic,
exotic x indigenous, indigenous x exotic and indigenous were available to conduct the
selection experiment.
The resulting populations in generation 2050 were used in the different breeding
schemes in the ten subsequent generations (T1 – T10) that were produced (Figure 5.1).
Pure-breeding was carried out for the exotic and indigenous populations separately. The
synthetic population was produced by mating F1 sires to their reciprocal crossbred dams
followed by inter se mating. For the backcross breeding program, repeated backcrossing
was conducted for generations T1 - T5, followed by inter se mating of the backcross
population from generations T6 – T10. All lines in the breeding schemes were kept at a
constant size of 1,000 breeding candidates and 5% of males and 50% of females were
randomly selected in each line per generation. Throughout the selection experiment, all
sires and dams were replaced each generation, thus discrete generations were assumed.
The goal of the breeding structure described above was to mimic the selected exotic and
unselected indigenous pig populations and their crosses currently used for pork
production in tropical developing countries.
127
5.2.3 Genome structure
The simulated genome consisted of 5 chromosomes of 150 cM each with 250
multi-allelic segregating QTL and a marker density of 22 bi-allelic loci per cM evenly
spaced across the genome (Table 5.1). This was chosen to make the simulation
computationally tractable and in such a way that the marker density represents
approximately 60 K SNP marker map currently available (Ramos et al., 2009). To
establish mutation drift equilibrium in the ancestral population after 2000 generations,
about 3 and 20 times as many markers and QTL, respectively, were simulated with equal
starting allele frequencies and a recurrent mutation rate of 2.5 x 10-3
for markers and
infinite-allele mutation rate of 2.5 x 10-5
for QTL.
The required number of QTL and markers were drawn at random from
segregating loci with a minor allele frequency (MAF) of ≥ 0.05 at generation 2000 before
creating the two recent purebred populations. No new mutation was simulated in the
recent populations, resulting in many of the markers and QTL not segregating in the
exotic population after 50 generations of phenotypic selection. Mutation rate per
generation in mammals is estimated to be about 2.5 x 10-9
(Kumar and Subramanian,
2002) and may not have resulted in a significant number of new mutants after 50
generations in real populations. The Haldane mapping function with no interference and a
Mendelian inheritance of all loci was assumed. The allelic effects of the QTL including
original and new mutations were sampled from a gamma distribution with a shape
parameter of 0.4 (Hayes and Goddard, 2001) and scaled such that the sum of the QTL
variances in the last historic generation equals the input QTL variance. The QTL was
randomly distributed across the genome.
128
The magnitude of LD in the exotic and indigenous founders at generation 2050
were estimated as average correlations between adjacent pairs of markers (r2)
5.1
(Hill and Robertson, 1968), where DAB = DA1B1 = pA1B1 – pA1pB1, pA1 is the frequency of
allele A1 at locus A, pB1 is the frequency of allele B1 at the locus B and pA1B1 is the
frequency of haplotype A1B1 in the population. Only markers with a MAF ≥ 0.10 were
included in this analysis.
The level of genetic diversity present in the simulated populations was
investigated using Wright‘s fixation indices such as FIS which defines the inbreeding
coefficient of individuals relative to the subpopulations and FST which is a relative
measure of where the subpopulations are in allelic fixation (Wright, 1965). The Fstat
program developed by Goudet (2001) was used to calculate these statistics. Genotypes at
200 loci from a random sample of 300 individuals from each of the two simulated
purebred founder populations from generation 2050 were used to estimate fixation
indices.
5.2.4 Phenotype
To simulate data for each animal with a record at generation 2000, a finite
additive locus model was assumed such that the true breeding value of an individual was
equal to the sum of the 250 QTL effects scaled on the basis of the specified heritability
(h2) and phenotypic variance (σ
2p). Given the genetic variance, environmental effects
were generated from a normal distribution with a mean of zero and a variance set to give
the desired h2. The phenotype was assumed to be the sum of the genotypic and
129
environmental effects, i.e. no pleiotropy or genotype-environment interactions were
simulated. For the traits considered, h2 and σ
2p used in the simulation were obtained from
a previous meta-analysis study carried out for the tropical environment.
Three traits namely: number born alive (NBA), average daily gain (ADG) and
back fat thickness (BFT) were simulated independently assuming a normal distribution.
All individuals within the various genetic groups in the last 10 generations were
genotyped for the available markers. In addition, for training the prediction models,
individuals in the training dataset were assumed to be recorded for all traits considered.
5.2.5 Model for genetic evaluation
The phenotypic information from the different breeding populations in each
generation of selection was used to calculate the EBV from the traditional BLUP model
according to Henderson (1984). The model was
y = μ1n + Za + e 5.2
where y is data vector of size n, μ is the overall mean, 1n is a vector of n ones, a is a
vector of random additive genetic effects of each individual ~N (0, Aσ2
a), A is the
additive relationship matrix, Z is a design matrix for the additive genetic effect and e is a
vector of random residuals ~N (0, Iσ2
e). The variance ratio required in the mixed model
equations was (1-h2)/h
2 where the true heritability at each generation was used.
In the GS scheme, marker effects were estimated using ridge regression model
described in Meuwissen et al. (2001) and extended by Fernando et al. (2007). The
statistical model was
y = μ1n + Wg + e 5.3
130
The mixed model equation (MME) was:
5.4
where y is data vector, g is a vector of allele substitution effects due to the ith genotype
~N (0, I ), W is a design matrix that has a 0, 1 and 2 for the number of alleles of type
gi present in the jth animal, p is the base allele frequency at ith locus, q is (1-p), σ2
g is the
true genetic variance at each generation and m is the total number of markers. The other
parameters are as described previously.
For training the marker effects in all of the different breeding programs simulated,
only the segregating markers with MAF ≥ 0.10 and based on the exotic population
marker panel were used. The training set consisted of 1,000 breeding individuals and
marker effects were re-estimated in each of the last ten generations. Once estimates of
marker effects were obtained, the genomic breeding values of kth selection candidates
(GBVk) in each generation was computed as
GBVk = 5.5
where and are the recorded genotype and the estimated marker allele substitution
effect at the ith locus, respectively, and m is the total number of markers.
5.2.6 Selection strategies
For the various breeding programs investigated in this study, individuals were
selected for breeding on the basis of their phenotype, BLUP EBV, GBV and at random
for each trait. In each generation, 5% of males and 50% of females for the different
breeding programs were selected resulting in Ne of 91.
131
5.2.7 Analysis of results
Each combination of breeding scenario and selection method was replicated 50
times, each replicate beginning with the same base population up to generation 2050. The
average level of true breeding values for the traits and average level of inbreeding
calculated based on pedigree were produced for each replicate and generation. Genetic
gains for the traits were calculated as differences in true genetic level from generation T1
to T10 and standardized by dividing by the true genetic standard deviation in each
generation. Cumulative response per generation was obtained by summing up the annual
gains in each generation.
Comparisons between methods were based on cumulative response, inbreeding
level, adjacent marker LD, realised accuracies and QTL allele frequency change in each
generation of selection. Realised accuracy was calculated as
5.6
where G (t) is the mean genotypic value in generation t, σA (t) is the additive genetic
standard deviation in generation t and 1.40 is the average selection intensity for the
proportions of males and females selected in each generation (Falconer and Mackay,
1996). All analyses were coded in R statistical software (Ihaka and Gentleman, 1996).
5.3 RESULTS AND DISCUSSION
5.3.1 LD and population differentiation
To show the differences between the exotic and indigenous populations which
served as founders for the various breeding programs in this study, some genetic
132
properties were examined at generation 2050. The mean heterozygosity in selected exotic
founders was 0.326 while that of the unselected indigenous founders was 0.468. The
average LD based on adjacent marker pairs (r2) was 0.286 in the exotic and 0.048 in the
indigenous for markers separated by 0.05 cM, on average. Wright‘s fixation indices were
used to quantify the amount of divergence between the simulated populations in
generation 2050. The estimated FST and FIS were 0.169 (SE = 0.014) and 0.004 (SE =
0.020), respectively, when populations were separated in generation 2000.
The mean heterozygosity in the exotic and indigenous populations implies that the
amount of genetic variation present in both populations differed significantly. Whereas
the exotic population has dissipated much of the genetic variability due to previous
selection, the unselected indigenous population still has their genetic variability intact. In
addition, due to differences in Ne for the exotic (167) and indigenous (667) populations,
the level of LD in the two populations differed markedly. Markers in LD with putative
QTL are valuable for the implementation of GS. Linkage disequilibrium levels of 0.2 and
above are considered appropriate for using genomic information in breeding programs
(Meuwissen et al., 2001; Du et al., 2007).
The estimate of FIS of 0.004 implies only a minor deficit of heterozygosity within
populations. This value corroborates with the estimate of average inbreeding level in the
indigenous population (0.030) based on pedigree information at generation 2050, but not
to the same extent as estimate in the exotic population (0.147) with a previous selection
history. The FST estimate of 0.169 shows that about 17% of the total genetic variability in
the whole population can be attributed to the differences among populations (e.g., Canon
et al. 2001), or that about 17% of shared allelic diversity was lost within each population
133
since they were separated (Toosi et al., 2010). Recently, Sollero et al. (2008) published
global FST values for indigenous and commercial breeds of pigs in Brazil based on 28
microsatellite loci. They found that 14% of the total genetic variation observed was due
to differences between populations. Therefore, the simulated founder populations in this
study had enough divergence to represent current exotic and indigenous pig populations.
5.3.2 Selection response
Average cumulative response in standard deviation units at generations 5 and 10
for different traits, breeding strategies and methods of selection are given in Table 5.2. As
expected, all breeding programs in which directional selection was practiced showed
response for all traits studied. In the pure breeding scenario, genetic progress was greater
with GS than with BLUP and differences were higher in generation 5 than in generation
10 especially for the exotic population in moderate LD. For crossbreeding program, GS
also performed better than BLUP except for backcrossing of crossbreds to the indigenous
population where BLUP showed higher response than GS.
At generation 5, response to GS averaged across population was greater than
BLUP by 41%, 31% and 21% for number born alive (NBA), average daily gain (ADG)
and back-fat thickness (BFT), respectively. The result at generation 10 shows that
response to GS was greater than BLUP by 49%, 33% and 23% for NBA, ADG and BFT,
respectively, on average. Among the purebreds, response to GS averaged across traits
was 120% and 64% higher than BLUP at generations 5 and 10, respectively, in the
previously selected exotic population while the unselected indigenous population was
134
only 15% and 19% higher in selection response to GS than BLUP at generations 5 and
10, respectively.
One of the objectives of this study was to compare response to selection from
conventional BLUP and GS methods in breeding programs involving selected exotic and
unselected indigenous pig populations. Formulas usually used to predict selection
response such as ΔG = iρσA (Falconer and Mackay, 1996); where i is selection intensity,
ρ is accuracy of prediction method and σA is the additive genetic standard deviation, over-
estimate response because it does not account for reduction of genetic variability from
directional selection and inbreeding. Consequently, increases in ΔG anticipated from GS
methods due to higher accuracies of GBV that were measured in randomly mated
populations may not reflect expectations with selection because selection changes allele
frequencies and reduces genetic variance (Bulmer, 1976).
At all levels of trait heritability and in the pure breeding scenario, GS was clearly
superior to BLUP both in the short and long term. GS achieves this increase in selection
response over BLUP for two reasons: 1) GS selects directly on QTL controlling the traits
through its LD with molecular markers, 2) GS exploits Mendelian sampling effects to
improve accuracy of selection (Daetwyler et al., 2007). The BLUP method on the other
hand can increase accuracy of EBV for a selection candidate by capturing additional
information on ancestors and collateral relatives, however, when truncated selection is
practiced this leads to an increase in ΔFG. In a crossbreeding program, GS can improve
selection response through the creation of new LD in the crossbred population in addition
to the LD in the parental breeds.
135
For traits with low heritability where BLUP methods need help, GS can improve
genetic gain by exploiting genetic variation within families through the exploitation of
Mendelian sampling effects (Daetwyler et al., 2007). This reduces the chances of
selecting sibs as well as inbreeding. Another important observation is the reduction in
selection response at generation 10 compared to generation 5 when using GS. This
reduction is a result of reduction in accuracy due to recombination breaking the
association between markers and QTL. Using denser marker panel for implementing GS
will ensure that marker-QTL association is sustained in later generation thus improving
accuracy of selection.
5.3.3 Inbreeding and genetic variance
Table 5.3 showed the average cumulated inbreeding and genetic variance at
generations 5 and 10 for different traits, breeding strategies and selection methods.
Inbreeding was higher with BLUP selection than GS at generations 5 and 10. The
inbreeding level at generation 10 across traits and populations was twice the value at
generation 5 for all selection methods. At generation 10, average inbreeding coefficients
for BLUP ranged from 0.04 to 0.27 across traits and populations while the range was 0.03
to 0.12 in the GS scheme. Inbreeding decreased with increasing trait heritability for both
selection methods. As expected, inbreeding was zero at generation 5 for most of the
crossbred populations due to crossbreeding which creates more heterozygosity in the
population. Genetic variances were considerably more reduced with BLUP selection
across traits and populations than with GS. The rate of reduction was slower for traits
with low heritability than for moderate to highly heritable traits as expected. Using GS in
136
the pure breeding scenario reduced genetic variances faster in the unselected indigenous
population than in the selected exotic population.
In this study, inbreeding was calculated based on pedigree information and this is
an expectation assuming neutral loci, therefore, two alleles at the same neutral locus on
two homologous chromosomes have an equal chance of being selected. This ignores the
existence of linkage between marker allele and a QTL affecting a trait under selection.
With the availability of genomic information, inbreeding can be calculated such that the
expectation is adjusted with identity-by-state probabilities at the marker loci to yield
actual inbreeding at specific locations across the genome (Pong-Wong et al., 2001; Liu et
al., 2002). This suggestion could make a lot of difference in the inbreeding level reported
here. In addition, if GS can result in shorter generation interval in swine breeding then the
rate of inbreeding per unit of time might differ from what is expected.
The increase in inbreeding rate associated with BLUP selection is particularly
problematic for traits with low heritability. Under BLUP selection at low heritability
more weight is placed on relatives of the selection candidate and less weight is placed on
the individual‘s own record. This causes related candidates to have similar EBV and it is
more likely that relatives will be selected. Genome-wide selection can offset this
tendency through exploiting Mendelian sampling effects (Daetwyler et al., 2007). High
levels of inbreeding can result in inbreeding depression if non-additive genetic effects are
important. This study was simulated under a finite additive locus model and inbreeding
depression was not considered, however, it could be a real concern for some traits like
NBA in the nucleus population. Commercial pig production is based primarily on
crossbred sows and market pigs, and inbreeding depression in the parental lines, if it
137
occurs, should be recovered through heterosis in the crosses (Visscher and Haley, 1995).
Increased inbreeding contributed to reduced genetic variation. Reductions in
genetic variances are expected with directional selection methods. This reduction is due
to gametic phase disequilibrium and inbreeding and both as a result of Ne and selection.
One advantage of using GS is the exploitation of new genetic variance through the
estimation of Mendelian sampling effects (Daetwyler et al., 2007). New variation is
expected to balance the loss of genetic variability due to gametic phase disequilibrium;
however, because the Mendelian sampling variance itself is also reduced by the loss of
alleles due to inbreeding, this equilibrium can never be reached in a dynamic finite
population.
5.3.4 Comparison of different selection methods
Figure 5.2 compared the various selection methods based on response, inbreeding,
LD and realized accuracy by generation in an unselected indigenous population for ADG.
Patterns of change were similar for the selected exotic populations. Genetic response
increased in each generation when directional selection was implemented on the
population, and was greater for GS, followed by BLUP and phenotypic selection (Figure
5.2A). As expected, there was no change in the randomly selected population.
No method of selection is free of inbreeding because inbreeding is a function of
Ne and can increase with selection (Figure 5.2B). In generations 1 and 2, the inbreeding
level was the same for all methods of selection, but increased much more quickly for
BLUP followed by GS. The rate of inbreeding per generation for phenotypic selection
was slightly higher than that of random selection. At the same level of inbreeding at
138
generation 2, GS had slightly higher selection response than BLUP and phenotypic
selection (Figure 5.2A). The high inbreeding that occurs with BLUP selection is because
this method achieves high accuracy of EBV estimation through the use of additional
information on ancestors and collateral relatives resulting in sibs having similar EBV and
are therefore likely to be selected together leading to an increase in ΔFG. Thus, GS
provides a method for achieving both short-term goal of increased and sustained ΔG and
long-term needs for maintaining genetic variation while keeping inbreeding low.
LD is caused by gametic phase disequilibrium due to genetic drift and selection.
The results of average LD for the various selection methods showed that GS and BLUP
can move a population in approximate linkage equilibrium to a state of LD within a very
short time (Figure 5.2C). Similarly, phenotypic and random selection can create LD in
the population. Genome-wide selection can exploit the high LD between QTL and
molecular markers to increase genetic gain, but this is not true for BLUP and phenotypic
selection. Selection based on BLUP relies on accuracy of EBV which depends on trait
heritability and the amount of information available for EBV estimation, while
phenotypic selection depends solely on trait heritability in order to make genetic
progress. The presence of gametic phase disequilibrium created by BLUP can actually
slow down genetic progress (Bulmer, 1976), as evident in Figure 5.2A.
Accuracy of selection increased for all methods of selection in the first 2
generations but decreased thereafter (Figure 5.2D). The rate of reduction in accuracy after
generation 2 was faster for BLUP, but slower for GS and phenotypic selection. The slight
reduction in accuracy overtime for GS was attributed to the exploitation of Mendelian
sampling effects (Daetwyler et al., 2007) and LD. The reduction in accuracy for BLUP
139
and phenotypic selection methods was due to a reduction in variances (Bulmer, 1976).
Accuracy was zero or near zero for random selection as expected.
5.3.5 Comparison of different breeding programs in a GS scheme
The implications of using GS in different breeding programs for improving ADG
are presented in Figure 5.3. Patterns of change are similar but differ in magnitude for
other traits. Genome-wide selection resulted in an increase in genetic response in pure
breeding and crossbreeding systems and in the short and long terms (Figure 5.3A).
Indigenous purebreds responded to GS better than the exotic purebreds and surpassed all
other populations at generation 10. The synthetic line had higher response to selection
than the purebreds in the short term. Backcrossing crossbreds to a previously selected
exotic population leads to higher genetic response while repeated backcrossing to the
inferior unselected indigenous population results in a negative response during the
backcrossing phase but a positive response during inter se mating with gains that are
comparable to the exotic purebreds obtained at generation 10.
Using GS in breeding programs aims to exploit markers in LD with QTL
affecting traits of interest. In an introgression scheme, favourable QTL in the donor line
are potentially selected and introgressed into a recipient population. After three
generations of repeated backcrossing, only about 6.25% (0.54) of the neutral alleles from
the donor line will be left in the backcross population. However, by using GS, any
favourable QTL alleles from the donor line that are rare or nonexistent in the recipient
line may be preserved at a frequency much higher than the expected value (Odegard et
al., 2009). Recently, Oseni (2005) compared different breeding schemes involving
140
Nigerian indigenous pigs and exotic Large White pigs for litter size. The result showed
that backcrossing crossbreds to the exotic Large White breed gave higher performance
for litter size than purebreds.
The relative advantage of using crossbred lines depends on how far the parental
lines have diverged. With a longer time since divergence of lines, the larger the
advantage of crossbreeding in GS scheme (Toosi et al., 2010). Combining GS with
backcrossing of crossbreds to an unselected indigenous population is not expected to
yield appreciable response especially when the indigenous population is inferior for the
trait under selection and had very low LD for the markers. The higher genetic progress
observed in unselected indigenous population in this study was due to high
heterozygosity, genetic variance and trait heritability leading to increased accuracy of
selection and faster genetic gain than the selected exotic population (Figure 5.3A). The
use of dense markers as in this study will ensure that markers are in close association
with QTL. This finding suggests that as long as a high density marker panel is utilized
and the reference population size is large, faster genetic progress is anticipated for
indigenous pigs found in developing countries when implementing a genome-wide
selection scheme.
The rate of inbreeding by generation when GS was applied in different breeding
programs is presented in Figure 5.3B. Inbreeding rate was similar in the purebreds in the
first three generations, but increased faster in the indigenous purebreds at later
generation. The increased inbreeding in the indigenous population was associated with
the faster genetic progress achieved in this population by using GS. Genome-wide
selection can lead to fixation of favourable QTL and homozygosity. Inbreeding was zero
141
in the crossbred lines undergoing repeated backcrossing and increased slightly in a linear
manner in synthetic lines.
Figure 5.3C presents the average LD between adjacent markers associated with
the different breeding programs when applying GS. Starting from a relatively low level of
LD at the initial crossbred generation, average LD for the backcross1 line was slightly
similar to LD for the exotic line from generations 3 upwards. The synthetic line also
showed increased LD, but did not exceed that of the exotic line. The purebred indigenous
and backcross2 line both had similar patterns of LD increases. The high LD level
observed in the backross1 line in the GS scheme explains the higher response to selection
observed in this population (Figure 5.3A). Odegard et al. (2009) studied the combination
of an introgression scheme with genomic selection via simulation. They concluded that a
backcross lines with similar or better performance than a purebred line can be obtained
within three to five generations of repeated backcrossing. The general increase in LD
level in the crossbreds arises from old LD in the parental lines and new LD created due to
crossing.
Realised accuracy over generations in a genome-wide selection program was
slightly stable in the purebreds but decreased rapidly in the synthetic and backcross
populations (Figure 5.3D). Genetic variation was slightly reduced in the GS scheme for
breeding programs simulated (data not shown). The slight reduction in genetic variance
was attributed to exploitation of Mendelian sampling effects which help to minimize
inbreeding and genetic variance (Daetwyler et al., 2007).
142
5.3.6 Impact of trait heritability in a GS scheme
The influence of different trait heritability on cumulative response, inbreeding,
LD and realised accuracy when GS was applied in an indigenous population is given in
Figure 5.4. The traits studied included number born alive (NBA), average daily gain
(ADG) and back-fat thickness (BFT). These traits were chosen because they represent a
range of heritability estimates and also affect profitability of pig production in developing
countries. The genetic parameters used to simulate data for each trait are weighted
estimates based on a previous meta-analysis of genetic parameters for pig production
traits in tropical developing countries (Chapter 3). Selecting for low BFT based on GBVs
leads to higher genetic response and is followed by ADG and NBA (Figure 5.4A). The
higher response obtained for BFT is due to higher accuracy of phenotype used for
training the marker effects and higher heritability of the trait leading to more reliable
GBVs. In contrast, traits like NBA have faster rate of change when using GS due to
exploitation of Mendelian sampling variation at low heritability.
Inbreeding decreased slightly with increased trait heritability under a GS scheme
(Figure 5.4B). As in traditional BLUP selection where high emphasis is placed on sib
information for low heritability traits leading to co-selection of sibs and higher inbreeding
rate, traits with low heritability have higher inbreeding rate than moderate to highly
heritable traits in a GS scheme. Linkage disequilibrium in an indigenous population
starting at very low level increases slowly in a GS scheme and the trend was
comparatively similar for all traits (Figure 5.4C). Whereas recombination breaks down
LD, selection and genetic drift creates new gametic phase disequilibrium among marker
and QTL alleles. Realised accuracy was stable across generations and the trend was
143
similar for all traits (Figure 5.4D). This suggests that genetic variation is maintained in a
GS scheme.
5.3.7 QTL minor allele frequency changes due to different factors
Figure 5.5A shows the effect of various selection methods on average QTL minor
allele frequency (MAF). A slight change in MAF when random selection was invoked is
due to genetic drift. For BLUP and phenotypic selection, small changes in QTL MAF
were observed. The BLUP and phenotypic selection methods are based on an
infinitesimal model which assumes that quantitative traits are controlled by large number
of genes each with a small effect on the trait (Fisher, 1918); consequently, changes in
allele frequency due to selection are expected to be small. A remarkable change in QTL
MAF was observed for GS. This is because GS acts directly on the QTL alleles in LD
with marker alleles that are utilised for selection and can cause more inbreeding at QTL
location compared to what can be captured by pedigree inbreeding analysis. When MAF
was similar for all directional selection methods at generations 1 and 2, the cumulative
response, inbreeding and LD levels in these generations were also similar (Figure 5.2).
This is because all of the population-wide variables studied depend on allele frequency
which can be changed significantly by any method that has a direct impact on QTL MAF.
The distribution of average MAF of QTLs for the different breeding programs
studied over 10 generations of genome-wide selection are presented in Figure 5.5B. The
change in MAF was relatively faster in the unselected indigenous purebreds than in the
previously selected exotic purebreds because at allele frequency close to 0.5, rate of
change is expected to be higher. Synthetic line showed a downward trend in MAF until
144
generation 5 when MAF stabilised. This pattern of MAF distribution in synthetic lines
was due to genetic drift in this population. Average MAF was similar at generations 4
and 5 in the indigenous and backcross2 lines. This agrees with earlier findings that
generation 4 is an appropriate cut-off for a repeated backcrossing program (Odegard et
al., 2009), because QTL allele frequency is expected to be similar in the recipient and
backcross line at this time.
Changes in average MAF of QTLs are expected for various traits undergoing a
genome-wide selection program (Figure 5.5C). Selection on any trait can drive
favourable QTL alleles to fixation while reducing the frequency of the unfavourable
alleles.
5.4 CONCLUSIONS AND RECOMMENDATIONS
This study has evaluated the application of GS in pig breeding programs
involving imported exotic and existing indigenous pig populations typical of tropical
developing countries. Given that the data recording and analysis infrastructure for
implementing the traditional BLUP selection method is generally lacking in developing
countries, GS provides an approach for achieving faster genetic progress without
developing a pedigree recording system. Genome-wide selection did better than
traditional BLUP and phenotypic selection methods by increasing genetic gain and
maintaining lower levels of inbreeding especially for traits with low heritability.
Genome-wide selection achieved this by exploiting LD and the Mendelian sampling
effects which provides opportunity for higher genetic progress while controlling rates of
inbreeding per generation.
145
Indigenous pig populations in the tropics that have a low level of population-wide
LD can be improved genetically by using the available high density 60 K SNP panel
developed for the exotic populations to implement GS. This will ensure that marker-QTL
associations capture most of the genetic variability within a trait. Combining GS with
repeated backcrossing of crossbreds to the selected exotic population in moderate LD
promises to improve the genetic merit of the commercial population more rapidly than
the other methods considered in this study. This possibility will lead to the introgression
of favourable QTLs required for tropical pork production from the indigenous population
into the commercial line.
146
Table 5.1 The parameters used in the simulation program
Parameters Values
Genome size per chromosome 150 cM
Number of chromosomes 5
Marker density per cM 22
Number of segregating QTL per chromosome 50
Mutation rate of Marker locus 2.5 x 10-3
Mutation rate of QTL locus 2.5 x 10-5
Minor allele frequency 0.05
Distribution of additive QTL effects Gamma (shape = 0.4)
Traits
Number born alive (NBA) = 0.08; = 7.73
Average daily gain (ADG) = 0.28; = 10361.20
Back fat Thickness (BFT) = 0.63; = 20.88
Number of animals with phenotypic or genotypic records, or both
Training 1275
Prediction/Validation 1000
Number of replicates per scenario 50
147
Table 5.2 Average cumulative responses (standard deviation units) after five and ten
generations of selection using a pedigree or genome-based methods for different traits
and breeding programs
Traits Populations Methods G5 G10 Δ1(%)G5 Δ(%)G10
NBA Exotic Pedigree 0.89 2.49
Genomics 2.02 4.20 126.97 68.67
Indigenous Pedigree 2.26 4.86
Genomics 2.77 6.19 22.57 27.37
Synthetic Pedigree 2.51 4.98
Genomics 3.64 6.50 45.02 30.52
Backcross1 Pedigree 4.28 5.36
Genomics 4.86 7.22 13.55 34.70
Backcross2 Pedigree -1.25 1.38
Genomics -1.19 2.54 -4.80 84.06
ADG Exotic Pedigree 1.05 3.02
Genomics 2.29 4.91 118.10 62.58
Indigenous Pedigree 3.41 7.21
Genomics 3.89 8.48 14.08 17.61
Synthetic Pedigree 3.80 6.72
Genomics 4.55 7.70 16.48 14.58
Backcross1 Pedigree 4.83 6.00
Genomics 5.38 8.03 11.39 33.83
Backcross2 Pedigree -1.10 3.21
Genomics -1.14 4.37 -3.64 36.14
BFT Exotic Pedigree -1.19 -3.39
Genomics -2.57 -5.44 115.97 60.47
Indigenous Pedigree -4.43 -9.39
Genomics -4.77 -10.39 7.68 10.65
Synthetic Pedigree -4.66 -7.89
Genomics -4.97 -8.47 6.65 7.35
Backcross1 Pedigree -5.11 -6.50
Genomics -5.53 -8.46 8.22 30.15
Backcross2 Pedigree 0.86 -4.97
Genomics 1.13 -5.38 -31.40 8.25 1Percentage change in genetic gain due to change in selection method
148
Table 5.3 Average cumulated inbreeding and genetic variance after five and ten
generations of selection using a pedigree or genome-based methods for different traits
and breeding programs
Traits
Populations
Methods
Inbreeding Genetic variance
G5 G10 G5 G10
NBA Exotic Pedigree 0.14 0.27 0.37 0.30
Genomics 0.05 0.11 0.35 0.29
Indigenous Pedigree 0.08 0.19 0.64 0.55
Genomics 0.06 0.12 0.64 0.50
Synthetic Pedigree 0.07 0.19 0.73 0.43
Genomics 0.04 0.10 0.60 0.37
Backcross1 Pedigree 0.00 0.15 0.40 0.31
Genomics 0.00 0.04 0.40 0.31
Backcross2 Pedigree 0.00 0.10 0.67 0.60
Genomics 0.00 0.05 0.71 0.57
ADG Exotic Pedigree 0.14 0.24 15.64 12.82
Genomics 0.04 0.09 15.29 12.27
Indigenous Pedigree 0.06 0.14 44.75 35.89
Genomics 0.05 0.10 41.63 31.52
Synthetic Pedigree 0.04 0.15 42.21 19.14
Genomics 0.03 0.08 32.46 15.99
Backcross1 Pedigree 0.00 0.14 17.80 13.81
Genomics 0.00 0.04 17.75 12.73
Backcross2 Pedigree 0.00 0.06 50.02 49.63
Genomics 0.00 0.04 52.24 43.63
BFT Exotic Pedigree 0.12 0.22 0.65 0.50
Genomics 0.04 0.08 0.59 0.47
Indigenous Pedigree 0.05 0.11 2.71 2.13
Genomics 0.04 0.09 2.67 1.96
Synthetic Pedigree 0.03 0.13 2.08 0.70
Genomics 0.03 0.07 1.79 0.66
Backcross1 Pedigree 0.00 0.13 0.72 0.56
Genomics 0.00 0.03 0.73 0.51
Backcross2 Pedigree 0.00 0.04 3.29 3.27
Genomics 0.00 0.04 3.67 3.28
149
Figure 5.1 Schematic representation of the simulated population structure
Selection strategy began from T1 - T10 and based on Ne = 91 for all populations simulated. Inter se
mating of backcross population took place from T6 - T10
150
Figure 5.2 Cumulative response (A), inbreeding (B), LD (C) and realised accuracy (D) for various selection methods in an unselected
indigenous population and for ADG
151
Figure 5.3 Cumulative response (A), inbreeding (B), LD (C) and realised accuracy (D) for different breeding programs using
genome-wide selection and for ADG
152
Figure 5.4 Cumulative responses (A), inbreeding (B), LD (C) and realised accuracy (D) for different traits in a genome-wide selection
scheme for unselected indigenous population
153
Figure 5.5 Average minor allele frequency (MAF) changes due to (A) different selection methods in an unselected indigenous
population for ADG, (B) different breeding programs in a genome-wide selection scheme and for ADG and (C) different traits in a
genome-wide selection scheme for unselected indigenous population
154
CHAPTER 6
RELATIVE ECONOMIC RETURNS FROM SELECTION SCHEMES
FOR A NUCLEUS SWINE BREEDING PROGRAM
E. C. Akanno1, F. S. Schenkel
1, M. Sargolzaei
1, R. M. Friendship
2 and J. A. B.
Robinson1
1Centre for Genetic Improvement of Livestock,
Department of Animal & Poultry Science, University of Guelph, Guelph, ON, Canada
2Department of Population Medicine, Ontario Veterinary College,
University of Guelph, Guelph, ON, Canada
3L'Alliance Boviteq, Saint-Hyacinthe, Quebec, Canada
155
6.0 ABSTRACT
The objective of this study was: 1) to evaluate a two-step selection strategy
involving genome-wide selection (GS) versus phenotypic and pedigree-based selection
methods, and 2) to compare the differential returns from selection schemes in a nucleus
swine breeding program. Annual genetic gain and return per dollar invested were used as
evaluation criteria. Data for the study was generated by simulation of relevant target
populations typical of swine population structure in developing countries and assuming
that genetic improvement was concentrated on purebreds in a nucleus breeding program.
Selection of replacement candidates were based on selection indices constructed for all
traits in the breeding objectives for pig production in developing countries. Traits in the
selection index included number born alive, average daily gain, feed conversion ratio and
loin eye area. The result shows that accuracy of genomic indices for the pre-selection and
selection strategy ranged from 0.38 to 0.66 and were higher than the accuracy of
conventional selection indices (0.10 – 0.64). Genome-wide selection generated an
increase of about 38% to 172% in annual returns compared to other conventional
approaches for a population in moderate linkage disequilibrium. The expected return per
dollar invested in a GS scheme was comparable to conventional selection methods when
the cost of genotyping for 60 K single nucleotide polymorphism panel was $30 per pig.
This genotyping cost is realisable using genotype imputation techniques with lower
density, lower cost genotyping panels.
Keywords: Genome-wide selection, nucleus breeding program, differential returns,
simulation, developing countries
156
6.1 INTRODUCTION
Nucleus breeding schemes (NBS) describe a breeding program in which selection
is carried out on a small number of elite breeding animals in a nucleus that produces
germ-plasm or breeding stock for use in a multiplier or commercial population. This
program has been proposed to overcome the technical constraints limiting the
implementation of genetic improvement of pigs for low-input production systems in
developing countries (Smith, 1988). The genetic superiority generated in the nucleus is
disseminated to the whole population. Genetic evaluation to select replacement stock is
implemented after a performance testing program has been completed for all possible
candidates (de Roo, 1987).
Performance testing is utilized to prevent biases in estimated breeding value
(EBV) caused by preferential treatment (Kuhn et al., 1994) and to avoid the impact of
heterogeneous variances in different environments (Garrick and van Vleck, 1987). The
performance testing of young boars and gilts is associated with a substantial cost, usually
an average of $55 per tested pig based on Canadian prices (B. Sullivan, personal
communication, 2011). This price is based on costs for animal registration, performance
testing, cost for marketing the non-selected intact males, costs for raising purebred
females and cost for genetic evaluation (B. Sullivan, personal communication, 2011).
Owing to the cost of performance testing, swine breeders often pre-select the animals to
be tested. In developing countries, pre-selection of candidates for performance testing is
based on knowledge of the phenotypic performance of their parents and when they have
attained certain age or body weight (Dzama, 2002). In addition, selection of replacement
157
stock from tested candidates is strictly on the pigs own performance excluding records of
its relatives (Aker and Kennedy, 1988: Dzama, 2002).
With the advent of genome-wide selection (GS) based on genomic breeding
values (GBV) predicted from dense single nucleotide polymorphism (SNP) marker
genotyping (Meuwissen et al., 2001), a genomic index (Dekkers 2007) of traits in the
breeding objective can be constructed for all possible candidates and used for pre-
selection of individuals entering the test station and for selection of replacement
candidates. Although, a variety of factors influence the outcome of a swine nucleus
breeding scheme (see for e.g. Figure 1 in de Roo (1987)), the objective of this simulation
study was not to consider all of these factors, but; 1) to evaluate a two-step selection
strategy involving genome-wide selection versus phenotypic and pedigree-based
selection methods, and 2) to compare the differential returns from selection schemes in a
nucleus swine breeding program. Two scenarios involving previously selected
populations in moderate linkage disequilibrium (LD) and unselected populations in low
LD were studied.
6.2 MATERIALS AND METHODS
6.2.1 Simulation of populations
Data for this study were simulated following the approach described in section 5.2
of Chapter 5. The simulated genome consisted of 5 chromosomes of 150 cM each with
250 multi-allelic segregating quantitative trait loci (QTL) and a marker density of 22 bi-
allelic loci per cM evenly spaced across the genome to represent approximately 60 K
SNP marker panel that is currently available (Ramos et al., 2009). Two purebred lines
158
consisting of a previously selected population in moderate LD between adjacent markers
(r2 = 0.29) and unselected population in low LD (r
2 = 0.05) served as the source of
foundation stock for the nucleus and for the two scenarios investigated. A 250 sow
operation that is capable of producing on average two litters per sow per year was
assumed. A total of 250 sows and 25 boars were randomly chosen from each of the
purebred lines to establish the foundation stock for the nucleus. These were randomly
mated to produce 20 piglets per sow per year resulting in a total of 5,000 piglets annually
assuming litter size and litter rate are already corrected for fertility and piglet survival.
Besides the nucleus population, another population was simulated for each trait
starting from the same base population as in the nucleus. This population was used as a
training set to estimate the marker effects to be used as priors for the application of GS in
the nucleus breeding scheme. This design eliminates the effect of close relationship
between the training set and the prediction set in the nucleus.
6.2.2 Traits in the model
The traits included in the breeding objective were number born alive (NBA),
average daily gain (ADG), feed conversion ratio (FCR) and loin eye area (LEA). The
genetic parameters used for simulating the data are presented in Table 6.1. Traits were
simulated independently assuming a finite additive locus model. No pleiotropy or
genotype-environment interaction was simulated. The genetic parameter estimates used
for simulation were from a meta-analysis study carried out for pig production traits in
tropical developing countries (Chapter 3). However, the economic values for each trait
159
were based on Canadian prices obtained from Canadian Centre for Swine Improvement
(CCSI) website. All cost figures are presented in Canadian dollars.
6.2.3 Model for genetic evaluation
In this era of genomics, it is reasonable to assume the availability of a repository
of marker effect estimates defined for specific locations on a given SNP panel and for a
given population. Using the same SNP panel, individuals can be genotyped at any time
and their GBVs obtained for selection decision based on prior estimates of marker effects
from a similar training population. The assumption is that marker density is high enough
to capture useful LD between markers and QTL leading to high accuracy of selection.
This is akin to the conventional method where genetic parameter estimates from other
sources are used as priors in genetic evaluation of a novel similar population in order to
obtain estimated breeding values (EBVs) for selection purposes.
The separate population used as training set to estimate the marker effects was
also evaluated using a pedigree-based best linear unbiased prediction (BLUP) model for
comparison. Comparisons were based on accuracy between true and estimated breeding
values for the training and validation populations. The training set consisted of 2,275
individuals and 15,100 segregating markers with minor allele frequency (MAF) ≥ 0.10,
while the validation set consisted of 5,000 breeding candidates produced annually in the
nucleus.
The pedigree-based BLUP model assumed was according to Henderson (1984).
The model used was
160
y = μ1n + Za + e 6.1
where y is observed phenotype of size n, μ is the overall mean, 1n is a vector of n ones, a
is a vector of random additive genetic effects of each individual ~N (0, Aσ2
a), A is the
additive relationship matrix, Z is a design matrix for the additive genetic effects and e is a
vector of random residuals ~N (0, Iσ2
e). The variance ratio required in the mixed model
equations was (1-h2)/h
2. The ASREML program (Gilmour et al., 2009) was applied for
estimation of genetic effects and variances.
For the genome-based model, the marker effects were estimated using ridge
regression described in Meuwissen et al. (2001) and extended by Fernando et al. (2007).
The statistical model was
y = μ1n + Wg + e 6.2
The mixed model equations (MME) were:
6.3
where y is observed phenotype of size n, g is a vector of SNP allele substitution effects
~N (0, I ), W is a design matrix that contains 0, 1 or 2 for the number of one of the
two alleles in the SNP genotypes, is the average base SNP heterozygosity, σ2
g is the
genetic variance estimated using ASREML program and m is the total number of
markers. The other parameters are as describe previously.
Once estimates of marker effects were obtained from the training data set, the
genomic breeding values of kth animal (GBVk) in the training and for the future
candidates in the nucleus (prediction set) was computed as
161
GBVk = 6.4
where and are the recorded genotype (0, 1 or 2) and the estimated marker allele
substitution effect at the jth locus, respectively, and m is the total number of markers. For
predicting GBV of an individual in the two population scenarios, only about 15,100
segregating markers with MAF ≥ 0.10 in the selected population marker panel were used.
6.2.4 Pre-selection strategy
All individuals in the rearing piggery were pre-selected to enter into the test
station for performance testing (Figure 6.1). Three pre-selection strategies were applied
namely: 1) phenotypic selection which uses a phenotypic index (PI) constructed from
parent average phenotypic performance of all individual‘s in the rearing piggery as
follows:
PI = 6.5
where vl is the economic weight of trait l, Pl is an individual‘s parent average of
phenotypic performance for trait l, is the mean parent average of phenotypic
performance for trait l, is the trait heritability and t is the number of traits in the index,
2) pedigree selection based on selection index (SI) calculated for each individual as
follows:
SI = 6.6
where vl is the economic weight of trait l, PAl is the predicted parent average breeding
values of an individual for trait l obtained by using the BLUP model in equation 6.1 and t
is the number of traits in the index, and 3) genome-wide selection based on a genomic
index (GI) constructed for all individuals in the rearing piggery as follows:
162
GI = 6.7
where vl is the economic weight of trait l, GBVl is the genomic breeding values of an
individual for trait l predicted using equation 6.4 assuming that estimates of marker
effects were available from a similar training population and t is the number of traits in
the index. All piglets in the rearing piggery were assumed to be genotyped for the same
SNPs as in the training population. A total of 1,000 top ranking individuals made up of
male/female candidates were pre-selected for performance testing based on the selection
indices described previously.
6.2.5 Selection criteria
In order to predict genetic gain per year for the whole operation to compare
between selection schemes implemented, records were assumed to be available for all
performance tested candidates at the end of the test and for all traits studied. Three
selection criteria were applied namely; 1) phenotypic index based on an individual‘s own
phenotype, 2) selection index based on BLUP evaluation of animals with observation and
3) genomic index based on marker effects re-estimated for all traits using phenotype from
performance testing. The accuracy of selection was determined as correlation between the
index and the aggregate genotype (H: the product of the economic value and true
breeding value for each trait summed across all traits in the breeding objective) for all
tested candidates.
In another scenario for the genome-wide selection scheme, accuracy of GI for
pre-selection strategy was applied to predict genetic gain per year. This allowed for the
163
evaluation of a strategy that completely eliminates performance testing program
assuming that marker effects will continue to be available from other sources.
6.2.5 Analysis of results
Assuming that 5% of the young boars and 50% of gilts were selected annually as
replacement stock from all tested candidates, expected genetic gain per pig per year was
predicted as follow:
6.9
where i is the selection intensity derived based on the proportion of each sex selected, rHI
is the correlation between the aggregate genotype and the selection index in males and
females, L is the generation interval which was assumed as 1.8 years based on Canadian
average and σH is the standard deviation of the aggregate genotype which was derived by
solving equation 6.9 using parameters from Table 6.1 and assuming that covariance
between traits is zero.
6.9
where v is the economic weight and is the genetic variance of traits in the aggregate
genotype. The expected standard deviation of the aggregate genotype (σH) worked out to
be $178.95. Return per year was calculated by multiplying by 5,000 assuming
that the nucleus produces 5,000 improved breeding stocks annually. Depending on the
scenario evaluated, total cost involved in the operation included cost of genotyping 5,000
breeding candidates for pre-selection and the cost of performance testing 1,000 pre-
selected candidates. The return on investment was obtained as total returns minus total
164
cost while return per dollar invested was obtained by dividing return on investment by
total cost. All analyses were coded in R statistical software (Ihaka and Gentleman, 1996)
and each scenario was replicated ten times.
6.3 RESULTS AND DISCUSSION
6.3.1 Accuracy of selection
The accuracy of EBVs in the training and validation sets when using the two
different selection methods (pedigree and genomics) and for traits in the breeding
objective is given in Table 6.2. Accuracy tended to increase with increasing heritability
of the traits across methods and populations. Genome-wide selection had higher accuracy
than pedigree-based method in the training and validation sets and for all traits in the
breeding objective. The validation result for the pedigree method was based on parent
average of individuals in the nucleus. Previously selected population in moderate LD had
lower accuracy than unselected population in low LD for different traits and selection
methods due to differences in selection pressure which reduced the genetic variation in
the selected population (Bulmer, 1976) and also due to difficulty in the estimation of
marker allele substitution effects at low MAF in the selected population. In addition, GS
method had higher accuracy than the pedigree method because GS acts directly on the
QTL in LD with markers to improve accuracy (Habier et al., 2007). The accuracies of
genetic evaluation presented in this study for different trait heritabilities were in
agreement with results from previous study (Chapter 4).
Table 6.3 presents the correlation (rHI) between aggregate genotype and various
selection indices applied for pre-selection and selection strategies. Accuracy of genomic
165
index (GI) for pre-selection strategy was similar to the average accuracy based on a
single trait (Table 6.2). For a GS scheme, the accuracy of the GI is expected to be slightly
lower than the accuracy based on a single trait (Sitzenstock et al., 2010) because of the
weighting effect of the economic values on the trait breeding values. Similar trend was
observed when comparing accuracies of selection index to the average accuracy for a
single trait in the pedigree-based BLUP method. The accuracy of GI in the selected
population was slightly lower than the values observed for unselected population due to
differences in selection pressure. The accuracy of GI for the unselected population was
similar to the values reported by König and Swalve (2009) and Sitzenstock et al. (2010)
in simulation studies.
For pre-selecting candidates entering the test station, accuracy of GI used was
higher than the accuracy of conventional methods in the selected and unselected
populations (Table 6.3). The pre-selection of candidates using the conventional approach
had lower accuracy because they were constructed based on parent average information.
In real life, candidates for pre-selection have no records of their own to allow for
construction of selection or phenotypic indices. Swine breeders in developing countries
usually pre-select candidate for performance testing based on knowledge of their parents
performance (Dzama, 2002). For example, a breeder will pre-select a candidate for
performance test if it comes from a sow with greater than 10 piglets in a litter or gilt with
greater than 7 piglets in a litter. The availability of genomic information early in life of
selection candidates offers the opportunity to make selection decision that is accurately
informed on the basis of an individual‘s genetic merit.
166
Furthermore, accuracy of genomic index for pre-selection strategy was greater
than the accuracy of phenotypic and BLUP selection indices when records became
available in the nucleus for genetic evaluation (Table 6.3). This is because GS uses dense
marker panel to exploit LD between marker and QTLs affecting traits in the index
leading to improved accuracy of selection. When marker allele substitution effects were
re-estimated for all traits using the record of performance of 1,000 breeding candidates
from the test station, accuracy of GI was higher than the accuracy of conventional
selection indices in the selected population but slightly lower than accuracy of BLUP
selection index in unselected population. This decrease in rHI value in the unselected
population when marker effects were re-estimated was due to the sample size used in the
re-estimation (1,000 vs. 2,275 individuals used previously for training) such that QTL
allele effects were likely estimated with lower accuracy due to lower average MAF.
Accuracy of GS is expected to increase when re-estimating marker allele effects from a
large reference dataset of greater than 2,000 individuals (Chapter 4).
The expected accuracy of phenotypic index currently employed in nucleus test
stations in developing countries was lower than the accuracy for conventional BLUP and
genome-wide selection methods and was greatly reduced in the selected population than
in the unselected population. For all methods of selection, the accuracy of selection
indices was similar in males and females because neither the young boars nor the gilts
have progeny with records. When parent records were included in the evaluation, males
had higher accuracies than females due to more information on the sire pathway (result
not shown). Under a GS scheme, genetic evaluation can proceed on selection candidates
based on their own genotype and phenotype information without requiring knowledge of
167
their parent records (Meuwissen et al., 2001) which is not the case for a pedigree-based
method.
As in developed countries, non successful attempts have been made over the past
two decades to institute a pedigree recording system that will allow traditional selection
strategies for swine improvement in developing countries (Kahi et al., 2005), NBS was
proposed as an alternative by selecting within a nucleus using conventional approach
(Smith, 1988). Performance testing of individuals in the nucleus was an additional
strategy to compare the genetic merit of potential candidates for selection of replacement
stock. Under a performance testing program, the use of an individual's own performance
as a measure of its genetic merit is reliable for traits of high heritability like loin eye area
and average daily gain. An individual‘s estimated genetic merit can further be improved
by including the records of its relatives in the genetic evaluation program. However, it
becomes problematic when dissecting an animal genetic merit for a trait with low
heritability like number born alive. More records would need to be collected in order to
accurately estimate breeding value for such a trait. In addition, developing a recording
system for implementing a conventional BLUP method can be very costly and may be
impracticable for developing countries.
As shown in Tables 6.2 and 6.3, the availability of genetic markers will help to
improve the accuracy of selection for traits with low heritability and will provide GBV
early in life of selection candidates which offers an opportunity to reduce the cost of
performance testing. In real life more records will need to be collected from performance
tested candidates and added to the records from training set for re-estimation of marker
effects in order to improve accuracy of selection. As an alternative, GS can be
168
implemented based on accuracy of pre-selection strategy assuming that estimates of
marker effects will continue to be available from other sources which eliminates
completely the practice of performance testing by keeping only individuals with GI that
are above the threshold for selection. The selected candidates are then raised to
reproductive maturity and can be mated early in life, thus reducing both generation
interval and the cost of housing and performance testing. In addition, by preselecting
individuals that enter the test station, potential candidates can further be evaluated for
secondary traits such as conformation and health. However, will farmers accept genomic
indices as a selection criterion without having a prior knowledge of the performance of
selection candidates in a test station? This will be the crucial point for practical
implementation of GS in a nucleus swine breeding program in future.
6.3.2 Expected genetic change per pig per year
The expected genetic gain per pig per year in the nucleus is instrumental in
showing the differences between the various selection methods employed (Table 6.4).
For the selected population in moderate LD, using GS based on re-estimated marker
effects to select replacement stock will achieve about 132% and 17% greater returns per
year compared to current phenotypic selection method and a possible pedigree-based
approach, respectively. The expected increase in returns of 17% was less than the 37%
increase in returns reported by Simianer (2009) using a deterministic approach to
evaluate GS. When GS was used to replace phenotypic selection in an unselected
population in low LD, annual returns was increased by 43% but was reduced by 3% when
substituting for a pedigree BLUP method. This outcome supports the fact that having
169
sufficient level of LD in the population of interest is central to the functionality of GS.
For population with low LD, a higher marker density than 60 K SNP panel current
employed in this study may lead to more accurate GBV.
Under a nucleus breeding program, it is possible to select replacement stock
directly from the rearing piggery based on accuracy of GS used in the pre-selection
strategy. The expected change in genetic gain from such a GS approach will be 172% and
38% greater returns compared to phenotypic and pedigree BLUP method, respectively, in
selected population in moderate LD (Table 6.4). For unselected population in low LD, a
greater return of 50% and 2% compared to phenotypic and pedigree BLUP methods,
respectively, can be expected when GS is applied based on pre-selection accuracy. In
addition, the possibility of re-estimating marker allele substitution effects based only on
small data set from the test station will lead to an expected reduction in annual returns of
15% and 5% for selected and unselected populations, respectively. Because
recombination events due to exchange of genetic material during meiosis can change
marker-QTL phase configuration leading to a reduction in accuracy of GS and
consequently, a reduction in the expected annual returns, marker effects should be re-
estimated with large datasets (>2000 animals, Chapter 4) whenever possible and after two
generations of selection.
In the study by Schaeffer (2006), annual genetic gain was doubled because of a
substantial reduction of generation intervals in three of the four pathways of selection in
dairy cattle. Reduction of generation interval might be unfeasible with swine but by
increasing the accuracy of genomic indices through the use of denser marker panel,
annual genetic gain can be increased and this will offset the cost of genotyping animals.
170
6.3.3 Differential returns from selection schemes in a nucleus breeding program
By taking into account the cost of genotyping service and performance testing, the
differential returns from various selection methods in a nucleus swine breeding program
were compared (Tables 6.5 and 6.6). Depending on the selection method used, the pre-
selection of 1,000 pigs that entered the test station out of 5,000 pigs produced annually in
the nucleus offered an opportunity to reduce the cost of performance testing. The
remaining 4,000 pigs were expected to be sold to multipliers as additional returns.
However, since accuracy of pre-selection differed with the method applied, it is possible
to choose different breeding candidates to be performance tested by these methods.
Consequently, good breeding candidates may be sent to the market rather than been fully
performance tested. This is why it is important to establish an accurate method for pre-
selection. Genome-wide selection offers an opportunity to have accurate GBV early in
life of selection candidates and this will have an impact on expected rate of annual
returns.
The total cost involved in the GS scheme when marker effects were re-estimated
consisted of the cost of genotyping 5,000 pigs in order to obtain GBV for pre-selection
and the cost of performance testing 1,000 pre-selected candidates. Starting at the current
cost of genotyping for 60 K SNP panel which is around $120 per pig at DNA Landmarks
(M. McNairnay, personal communication, 2011), several genotyping cost were evaluated
by dropping the price by $30 up to a minimum price of $15 per pig. This provides an
opportunity to find the optimum genotyping cost that will guarantee faster return per
dollar invested when using GS at the current marker density of 60 K SNP chip. Cost of
genotyping for thousands of marker is known to have reduced by a rate of 33% over a
171
three year period (Simianer, 2009). In this study prior estimates of marker effects were
assumed available at no extra cost but in reality there is a cost. In the scenario where GS
was applied based on accuracy of pre-selection strategy, cost of performance testing was
ignored.
The total cost involved in the phenotypic and pedigree-based methods was only
the cost of performance testing valued at about $55 per pig for 1,000 tested candidates
based on prices from a Canadian test station (B. Sullivan, personal communication,
2011). For the pedigree-based method, it was assumed that the infrastructure for
implementing this method already existed. In tropical developing countries, where the
infrastructure for swine genetic improvement using the conventional BLUP method are
lacking, instituting a pedigree recording system for this environment will be a source of
additional cost which can be very substantial.
The results presented in Table 6.5 suggests that if genotyping cost for 60 K SNP
chip currently at $120 per pig can be reduced to $15 per pig, GS scheme will generate
similar returns per dollar invested compared to phenotypic selection in a nucleus breeding
program that uses a previously selected population in moderate LD. For the unselected
population in low LD, GS method achieved a lower return than phenotypic selection at
the same genotyping cost of $15 per pig. This suggests the importance of having LD in
the population in order to apply GS. Also for all range of genotyping costs per pig, the
GS scheme had lower returns per dollar invested compared to the pedigree BLUP method
for the selected and unselected populations, respectively.
In the scenario where GS was applied based on accuracy of pre-selection strategy
(Table 6.6), reduction of genotyping cost to $15 per pig will lead to an increase of $4.32
172
and 11 cents per dollar invested compared to phenotypic and pedigree BLUP methods,
respectively, in previously selected population. At a genotyping cost of $30 per pig, GS
can generate similar expected returns per dollar invested compared to currently practiced
phenotypic selection in previously selected population. For unselected population,
genotyping cost will need to be reduced to $15 per pig for the expected returns from GS
to be higher than phenotypic selection by $1.14. Presently, the cost of genotyping can be
reduced by genotype imputation (Howie et al., 2009; Sargolzaei et al., 2011). In a recent
study in dairy cattle, Sargolzaei et al. (2011) imputed 50 K genotypes from 6 K SNP
panel with accuracy of about 99%. The cost of genotyping an animal for the 6 K SNP
panel is around $40 at DNA Landmarks (M. Sargolzaei, personal communication, 2011).
This technology can be applied to swine breeding, thus reducing the genotyping cost.
A major concern while genome-wide selection is conjectured to be disadvantaged
compared to conventional approach is the cost of genotyping animals for millions of
marker loci. As genotyping cost decreases and marker density that is greater than 60 K
SNP panel assumed in the current study become available in future, QTL allele effects
are expected to be estimated with more precision due to higher LD and leading to higher
accuracy of selection and greater returns per dollar invested. The overall results from this
study indicates that, although the returns per dollar invested are greater for conventional
BLUP selection method ignoring cost of instituting a pedigree recording system for
developing countries, the marginal rates of returns to selection using GS are expected to
be greater as cost of genotyping for 60 K SNP panel become reduced to $30 per pig or as
next generation sequencing technology become feasible for swine breeding in future.
173
6.4 CONCLUSIONS AND RECOMMENDATIONS
Relative economic returns from the implementation of genome-wide selection in a
nucleus swine breeding program are possible. Benefits arise from an increase in accuracy
of genomic indices and the opportunity to eliminate performance testing costs.
Genotyping costs are critical to the return on investment in a GS scheme. Through the use
of lower density, lower cost of genotyping panels and imputation of genotypes equivalent
to a high density panel, genotyping costs can be controlled with no appreciable loss in
genetic response provided the population is in reasonable LD. For unselected populations
in low LD, denser marker panels will need to be used in order to increase accuracy of
selection. In addition, individuals entering a performance test station in a nucleus
breeding program can be pre-selected with higher accuracy by using a genomic index
compared to selection indices that are based on parent average information.
In a nucleus breeding scheme and in countries where a recording system for
implementing a conventional BLUP selection scheme are lacking, genome-wide selection
can supplant the development of this infrastructure. Increasing the accuracy of genomic
indices in the future through the use of denser marker panels and large training
population sizes will have a greater impact on the economic efficiency of GS and this
will offset the cost of genotyping animals. Selection of breeding candidates from the
rearing piggery based on genomic indices in a pre-selection strategy can possibly be
implemented. However, will farmers accept genomic indices as a selection criterion
without having a prior knowledge of the performance of selection candidates in a test
station? This will be the crucial point for practical implementation of GS in a nucleus
swine breeding program in future.
175
Table 6.1 Traits in the breeding objective and their parameters
Variables NBA ADG FCR LEA
Heritability 0.08 0.28 0.32 0.51
Standard deviation (SD) 2.78 101.79 0.35 3.85
Economic value, $ 18.07 3.02 -5.17 1.13
NBA = number born alive; ADG = average daily gain; FCR feed conversion ratio;
LEA = loin eye area
176
Table 6.2 Accuracy1 of estimated breeding values of pigs in the training and validation
sets for traits in the selection index
Populations Traits Methods Training set Validation set2
Selected NBA Pedigree 0.36±0.011 0.10±0.024
Genomics 0.49±0.021 0.43±0.029
ADG Pedigree 0.41±0.023 0.15±0.018
Genomics 0.54±0.016 0.47±0.016
FCR Pedigree 0.40±0.019 0.13±0.019
Genomics 0.51±0.019 0.46±0.021
LEA Pedigree 0.45±0.022 0.19±0.023
Genomics 0.58±0.020 0.52±0.027
Unselected NBA Pedigree 0.47±0.018 0.19±0.018
Genomics 0.55±0.016 0.44±0.014
ADG Pedigree 0.69±0.012 0.40±0.024
Genomics 0.77±0.011 0.67±0.018
FCR Pedigree 0.70±0.007 0.42±0.024
Genomics 0.78±0.006 0.68±0.016
LEA Pedigree 0.80±0.005 0.50±0.021
Genomics 0.85±0.003 0.74±0.008 1Correlation between true and estimated breeding values averaged over 10 replicates
2Validation result for pedigree method is based on parent average of individuals in the nucleus
177
Table 6.3 Accuracy of selection indices (rHI) for two steps selection strategies using
different methods and for different populations in a nucleus swine breeding program
Populations Methods Group Pre-selection Selection
Selected Phenotypic Males 0.10±0.006 0.17±0.014
Females 0.11±0.009 0.16±0.012
Pedigree Males 0.12±0.018 0.33±0.046
Females 0.13±0.020 0.33±0.042
Genomics Males 0.45±0.019 0.39±0.026
Females 0.47±0.019 0.38±0.028
Unselected Phenotypic Males 0.33±0.019 0.44±0.015
Females 0.32±0.013 0.42±0.034
Pedigree Males 0.36±0.024 0.64±0.020
Females 0.36±0.024 0.64±0.021
Genomics Males 0.65±0.019 0.62±0.016
Females 0.66±0.018 0.62±0.019
178
Table 6.4 Expected genetic change per year for different selection schemes and
populations
Populations Methods ΔG/pig/year ($) 1 Return ($) Δ (%)
2 Δ (%)
3
Selected Phenotypic 47.54 237,705.25 131.58 172.48
Pedigree 93.83 469,147.25 17.33 38.06
Genomics 110.09 550,470.08 - -
PreGBV 129.54 647,699.58 - -
Unselected Phenotypic 123.52 617,576.33 42.72 50.27
Pedigree 181.97 909,861.33 -3.12 2.00
Genomics 176.29 881,428.17 - -
PreGBV 185.61 928,054.58 - - 15% of the males and 50% of the females were assumed to be selected annually
2Percent change in annual returns due to the use of GS based on marker effects re-estimation
(Genomics) 3Percent change in annual returns due to the use of GS based on pre-selection accuracy (PreGBV)
179
Table 6.5 Differential returns from different selection methods and for a range of genotyping cost in a nucleus breeding program
Populations
Methods
Genotyping
cost/pig ($)
Total returns ($)
Total costs ($)1
Return on
investment ($)
Return per dollar
invested ($)
Selected Phenotypic 0 237,705.25 55,000.00 182,705.25 3.32
Pedigree 0 469,147.25 55,000.00 414,147.25 7.53
Genomics 15 550,470.08 130,000.00 420,470.08 3.23
Genomics 30 550,470.08 205,000.00 345,470.08 1.69
Genomics 60 550,470.08 355,000.00 195,470.08 0.55
Genomics 90 550,470.08 505,000.00 45,470.08 0.09
Genomics 120 550,470.08 655,000.00 -104,529.92 -0.16
Unselected Phenotypic 0 617,576.33 55,000.00 562,576.33 10.23
Pedigree 0 909,861.33 55,000.00 854,861.33 15.54
Genomics 15 881,428.17 130,000.00 751,428.17 5.78
Genomics 30 881,428.17 205,000.00 676,428.17 3.30
Genomics 60 881,428.17 355,000.00 526,428.17 1.48
Genomics 90 881,428.17 505,000.00 376,428.17 0.75
Genomics 120 881,428.17 655,000.00 226,428.17 0.35 1Total cost for GS scheme includes cost of performance testing 1000 candidates
180
Table 6.6 Differential returns from selection when GS scheme is based on pre-selection accuracy in a nucleus breeding program
Populations
Methods
Genotyping
cost/pig ($)
Total returns ($)
Total costs ($)1
Return on
investment ($)
Return per dollar
invested ($)
Selected Phenotypic 0 237,705.25 55,000.00 182,705.25 3.32
Pedigree 0 469,147.25 55,000.00 414,147.25 7.53
PreGBV 15 647,699.58 75,000.00 572,699.58 7.64
PreGBV 30 647,699.58 150,000.00 497,699.58 3.32
PreGBV 60 647,699.58 300,000.00 347,699.58 1.16
PreGBV 90 647,699.58 450,000.00 197,699.58 0.49
PreGBV 120 647,699.58 600,000.00 47,699.58 0.08
Unselected Phenotypic 0 617,576.33 55,000.00 562,576.33 10.23
Pedigree 0 909,861.33 55,000.00 854,861.33 15.54
PreGBV 15 928,054.58 75,000.00 853,054.58 11.37
PreGBV 30 928,054.58 150,000.00 778,054.58 5.19
PreGBV 60 928,054.58 300,000.00 628,054.58 2.09
PreGBV 90 928,054.58 450,000.00 478,054.58 1.06
PreGBV 120 928,054.58 600,000.00 328,054.58 0.55 1Total cost for GS scheme excludes cost of performance testing
181
CHAPTER 7
GENERAL DISCUSSION AND CONCLUSION
E. C. Akanno
Centre for Genetic Improvement of Livestock,
Department of Animal & Poultry Science, University of Guelph, ON, Canada
182
7.1 SUMMARY
Genetically improving the productivity of indigenous swine populations in the
tropics is important. This improvement can ultimately contribute to poverty alleviation
and food security for the ever increasing human population in developing countries. A
lack of knowledge of genetic improvement strategies has remained a major limitation to
genetically improving indigenous swine populations by limiting the decision-making
capability of both national and regional improvement agencies. This thesis has focused
on the evaluation of genome-wide selection as a strategy to overcome such constraints
that limit the genetic improvement of indigenous pig populations in tropical developing
countries.
The development of swine genetic improvement programs for tropical developing
countries requires an understanding of the constraints on productivity and profitability.
Chapter 2 reviews the current challenges limiting the implementation of genetic
improvement programs for tropical swine populations. Given that the infrastructure for
implementing traditional BLUP selection methods is lacking in these regions, the
application of a genome-wide selection program (GS) based on the available 60 K single
nucleotide polymorphism (SNP) marker panel was proposed. Genome-wide selection can
be adapted to existing tropical pig breeding systems and has the potential to increase
genetic gain by increasing the accuracy of genomic breeding values (GBV) of potential
selection candidates.
In Chapter 3, genetic parameters for economically important traits of indigenous
and exotic pigs determined under tropical conditions were meta-analyzed. A random
effect model that incorporated ―between‖ and ―within‖ study variance components was
183
used to obtain weighted means and variances for all parameter estimates. The weighted
mean estimates of genetic parameters presented in this study are recommended for use
when reliable estimates are not available for a specific indigenous swine population.
Where estimates within a population are available, these could be combined with the
weighted literature average by simply weighting each estimate by the inverse of the
sampling variance. This would yield a population-specific estimate that incorporates prior
values based on much more data. Also, these results should allow researchers in
developing countries to extract genetic parameter estimates suitable for a wide range of
tropical conditions. The data base of published parameters from the tropics and the
accompanying references are available for extraction of specific estimates in portable
format from the author.
The application of a genome-wide selection program as opposed to the
development of infrastructure for implementing a traditional pedigree and performance-
based selection method for pig genetic improvement in developing countries was
evaluated in Chapters 4 and 5 using simulation.
The results of genomic predictions presented in Chapter 4 showed that
accuracies of GBVs in the range of 0.31 - 0.86 are expected for validation populations
with low linkage disequilibrium (LD) levels and a previous selection history as long as a
high density panel of markers was utilized. High accuracy in the indigenous population
will require a multi-generational reference population with a size of not less than 2,000
individuals and marker density of greater than 60 K SNP panel. However, if resources are
a limitation in choosing an appropriate reference population, then a population size of at
least 1,000 individuals is recommended, but this will require re-estimation of marker
184
effects after two generations of selection. Genome-wide selection improved accuracy of
breeding value prediction over pedigree-based methods for traits with low heritability and
for young animals with no performance data. Crossbred training populations had higher
accuracy than purebred training populations. One way to implement genome-wide
selection in the tropics is to train the prediction model based on crossbred performance in
the field for genetic improvement of purebreds in the nucleus.
Chapter 5 presents an assessment of the opportunities for genome-wide selection
scheme in tropical pig breeding programs. Genome-wide selection performed better than
existing conventional methods by increasing genetic gain and maintaining genetic
variation while lowering inbreeding especially for traits with low heritability. Genome-
wide selection achieves this by exploiting LD and the Mendelian sampling effects. An
indigenous pig population with very low LD level can be improved by using GS if a high
density marker panel is available for use. Combining GS with repeated backcrossing of
crossbreds to the selected exotic population in moderate LD promises faster genetic
improvement of the commercial population.
A two-step selection strategy that involves the use of GS to pre-select candidates
that entered the performance test station and for selecting replacement candidates in a
nucleus swine breeding program was evaluated and compared to other conventional
approaches in Chapter 6. Benefits from a GS scheme arise from an increase in accuracy
of genomic indices. However, a dramatic reduction in the cost of genotyping for the 60 K
SNP panel and potential elimination of performance testing cost for a population in
moderate LD is required for a cost effective implementation of a GS scheme. Individuals
entering a performance test station in a nucleus breeding program can be pre-selected
185
with higher accuracy by using a genomic index compared to selection indices based on
parent average information. Genome-wide selection generated an increase of 38% to
172% in annual returns compared to other conventional approaches for previously
selected populations in moderate LD and about 2% to 50% increases in return for
unselected population in low LD. Selection of breeding candidates directly from the
rearing piggery based on genomic indices for a pre-selection strategy can be
implemented. However, will farmers accept genomic indices as a selection criterion
without having a prior knowledge of the performance of selection candidates in a test
station? This will be the crucial point for practical implementation of GS in a nucleus
swine breeding program in future.
7.2 RECOMMENDATIONS
From the foregoing, this thesis makes the following recommendations:
1. Despite the large number of genetic parameter estimates reported in the literature
for pig production traits in the tropics, some economically important traits like
carcass and meat quality traits have fewer parameters reported, and there is a need
for a complete set of genetic parameters, especially for the indigenous pig
populations.
2. There are good prospects for genome-wide selection in pig genetic improvement
in developing countries as breeding values of selection candidates can be
estimated with high accuracy without using pedigree information.
3. Genome-wide selection schemes can supplant the development of conventional
BLUP selection methods for genetic improvement of unselected indigenous pig
186
populations, but this will require a denser than 60 K SNP marker panel and a
multi-generational training set of at least 1,000 individuals.
4. Combining genome-wide selection with repeated backcrossing of the crossbred
population to the selected exotic pig promises rapid genetic improvement of the
commercial population.
5. In a nucleus breeding program, individuals entering a performance test station can
be pre-selected with higher accuracy using a genomic index compared to a
conventional selection index that is based on parent average information.
6. Potential breeding candidates can be selected directly from the rearing piggery
without undergoing performance testing but this assumes that there would be a
suitable training set for providing the estimates of marker effects for a given SNP
panel.
7.3 FUTURE DIRECTIONS
There are a number of challenges to the widespread implementation of GS in
developing countries. The feasibility of GS for genetic improvement of indigenous and
exotic pig populations found in tropical environments needs to be evaluated by
considering both the increase in genetic gain that is achievable with GS over
conventional selection methods and the extra cost involved in genotyping individuals for
a high density marker panel. The economic returns associated with genetic progress must
outweigh the cost of genotyping and phenotyping for GS to be profitable.
In a swine population, assuming selection intensity is constant, increases in
annual genetic gain from GS are expected to come from increases in accuracy of GBVs
187
used for selecting parents of future generation. The potential of GS to appreciably
decrease generation interval in swine is limited because generation interval is already
short. Therefore, in the application of GS on real populations, methods that will guarantee
higher accuracy of GBVs should be explored.
Predictive formulae have shown that a large multigenerational dataset of greater
than 2,000 individuals is required as a training set to get high GS accuracy in the
prediction population (Chapter 4 this thesis; Hayes et al., 2009). Usually only a limited
number of phenotypes and genotypes are available, particularly for traits of low
heritability like reproduction and health traits. However, it is exactly these traits for
which GS could create the greatest benefit. Increasing the sample size of the training set
requires consideration of the cost of both genotyping and phenotyping. In developing
countries where a coordinated system for data collection is lacking, it may be difficult to
collect a large number of accurate phenotypic records to achieve high GS accuracy. An
increased effort to obtain large numbers of accurate phenotypes in order to reach high GS
accuracy is desirable.
The accuracy of GS can be increased by using high density markers that are well
distributed across the genome to allow for marker-QTL association. While the cost of
genotyping for a high density marker panel is expected to decrease in future, it still may
be too costly to justify widespread genotyping in swine populations where annual genetic
gain is dependent on accuracy of GS alone. Lower cost solutions that will enable swine
breeders in developing countries to benefit from the expected extra gains from GS
method should be developed. One such solution is to use a low cost, low density panel
and then impute missing genotypes to recreate a higher density panel. Recent studies
188
showed that a high density panel can be imputed from a low density panel with 99%
accuracy (Howie et al., 2009; Sargolzaei et al., 2011). Genotyping cost would be reduced
because selection candidates would only need to be genotyped with a low cost, low
density panel. Therefore, genotype imputation could provide a lower cost solution to
increase the number of genotypes. However, successful imputation of genotypes requires
knowledge of the pedigree of genotyped individuals. In developing countries, absence of
coordinated pedigree recording system may limit this effort. In swine nucleus breeding,
the typical use of a relatively small number of founders would likely create a population
structure suitable for imputation (Daetwyler, 2009).
The model used in the simulation of traits for genomic evaluation assumes finite
additive locus effects. No dominance or epistatic effect at each locus or pair of loci was
considered. As is often the case, this simplified model was necessary to make the
simulation computationally tractable. In litter bearing species like swine, and with
crossbreeding programs, dominance effects are known to exist for traits with low and
moderate heritability (Oseni, 2005; Ilatsia et al., 2008). The inclusion of dominance
effects in models for the prediction of GBVs could increase the accuracy of genomic
predictions if the data structure allows it (Wellmann and Bennewitz, 2010). Moreover,
the predicted dominance or epistatic effects could be used to choose mating pairs to
maximise heterozygosity thereby recovering inbreeding depression and utilizing possible
overdominance (Wellmann and Bennewitz, 2010).
Based on results from the current study (Chapters 4 and 5), higher accuracy of
genomic predictions were associated with higher genetic variation in the training
populations. For indigenous pig populations in tropical developing countries that have
189
experienced mild selection pressure in the past, sufficient genetic variation exists to drive
selection (Chapter 3). Consequently, using the indigenous pig population as a reference
population will increase accuracy of GS if implemented in a population with similar
structure. In addition, crossbreeding of indigenous and exotic pig populations will lead to
an increase in heterozygosity in the crossbreds with an attendant increase in genetic
variation. Using such a crossbred population as a training set promises to increase
expected accuracy of GS.
In conclusion, genome-wide selection can increase genetic gain in swine
populations and there is great potential to reorganize the current swine breeding program
in tropical developing countries in order to exploit GS benefits. While the best
implementation strategy on a real swine population is only suggested by this work,
significant potential exists for GS to eliminate the need for a performance testing
program and to adapt itself to tropical pig breeding systems in developing countries.
190
REFERENCES
Adebambo, O. A. 1986. Selective breeding of the Nigerian indigenous pigs for the rural
producer. 3rd World Congress on Genetics Applied to Livestock Production,
Lincoln, Nebraska, USA July 16 - 22, 1986.
Adebambo, O. A. 1985. Crossbreeding the Nigerian indigenous pigs: Effect of crossing
and early weaning on sow performance. Turrialba 35(3): 221-227.
Adebambo, A. O. and Dettmers, A. E. 1982. Efficiency of feed and milk utilization by
litters of indigenous and exotic pure and crossbred pigs. Tropical Animal
Production 7(3): 204-208.
Aker, C. and Kennedy, B. W. 1988. Estimated Breeding Value: A tool for genetic
improvement of swine. In: FACTSHEET. Ontario ministry of agriculture, food
and rural affairs, order number 88 – 014.
Amaral, A. J., Hendrik-Jan, M., Crooijmans, R. P. M. A., Heuven, H. C. M. and Groenen,
M. A. M. 2008. Linkage disequilibrium decay and haplotype block structure in
pig. Genetics 179: 569-579.
Anderson, S. 2003. Animal genetic resources and sustainable livelihoods. Ecological
Economics 45: 331-339.
Baker, R. L., Wickham, B. W. and Morris, C. A. 1984. Relationship between central
performance test for growth in Hereford bulls and growth and carcass traits of
progeny. Animal Production 39:371-382.
191
Barron, U. G., Bergin, D. and Butler, F. 2008. A meta-analysis study of the effect of
chilling on prevalence of salmonella on pig carcasses. Journal of Food Protection
71(7): 1330-1337.
Beek, V. T. 2007. Should there be fear of tropical pig diseases? Pig Progress 23(5): 16-
17.
Belonsky, G. and Kennedy, B. 1988. Selection on individual phenotype and best linear
unbiased predictor of breeding value in a closed swine herd. Journal of Animal
Science 66(5): 1124.
Berkey, C. S., Hoaglin, D. C., Mosteller, F. and Colditz, G. A. 1995. A random-effects
regression model for meta-analysis. Statistics in Medicine 14(4): 395-411.
Berry, D. P., Kearney, F. and Harris, B. L. 2009. Genomic selection in Ireland.
Proceedings of the Interbull International Workshop - Genomic Information in
Genetic evaluation, Uppsala, Sweden, Bulletin no. 39.
Blench, R. M. 2000. A history of pigs in Africa. In: The origins and development of
Africans livestock; archeology, genetics, linguistics, and ethnography. Routledge
Oxon, UK. 546 pages
Bondoc, O. L. and Smith, C. 1993. Deterministic genetic analysis of open nucleus
breeding schemes for dairy cattle in developing countries. Journal of Animal
Breeding and Genetics 110:194-208.
Bourdon, R. M. 1997. Understanding animal breeding. 1st ed. United State of America,
Prentice-Hall, Inc.
192
Bulmer, M. 1976. The effect of selection on genetic variability: A simulation study.
Genetics Research 28(02): 101-117.
Buske, B., Sternstein, I. and Brockmann, G. 2006. QTL and candidate genes for
fecundity in sows. Animal Reproduction Science 95(3-4): 167-183.
Calus, M. P. L. 2010. Genomic breeding value prediction: methods and procedures.
Animal 4 (2): 157-164.
Calus, M. P. L., Meuwissen, T. H. E., de Ross, A. P. W. and Veerkamp, R. F. 2008.
Accuracy of genomic selection using different methods to define haplotypes.
Genetics 178:553-561.
Canon, J., Alexandrino, P., Bessa, I., Carleos, C., Carretero, Y., Dunner, S., Ferran, N.,
Garcia, D., Jordana, J., Laloe, D., Pereira, A., Sanchez, A. and Moazami-
Goudarzi, K. 2001. Genetic diversity measures of local European beef Cattle
breeds for conservation purposes. Genetics Selection Evolution 33:311-332
Chiduwa, G., Chimonyo, M., Halimani, T. E., Chisambara, S. R. and Dzama, K. 2008.
Herd dynamics and contribution of indigenous pigs to the livelihoods of rural
farmers in a semi-arid area of Zimbabwe. Tropical Animal Health and Production.
40 (2): 125-136.
Chimonyo, M., Dzama, K. and Bhebhe, E. 2006. Genetic determination of individual
birth weight, litter weight and litter size in Mukota pigs. Livestock Science 105(1-
3): 69-77.
193
Chimonyo, M. and Dzama, K. 2007. Estimation of genetic parameters for growth
performance and carcass traits in Mukota pigs. Animal 1(3): 317-323.
Chimonyo, M., Dzama, K. and Bhebhe, E. 2008. Genetic determination of mothering
ability and piglet growth in indigenous Mukota sows of Zimbabwe. Livestock
Science 113(1): 74-80.
Christensen, O. F. and Lund, M. S. 2010. Genomic prediction when some animals are not
genotyped. Genetics Selection Evolution 42:2
Cleveland, M. A., Forni, S., Garrick, D. J. and Deeb, N. 2010. Prediction of genomic
breeding values in a commercial pig population. 9th World Congress on Genetics
Applied to Livestock Production, Leipzig, Germany.
Clutter, A. C. and Brascamp, E. W. 1998. Genetics of Performance Traits. In The
Genetics of the Pig. Rothschild, M. F. and Ruvinsky, A. eds. CAB International,
New York, USA p 427.
Cochran, W. G. 1954. The combination of estimates from different experiments.
Biometrics 10: 101-129.
Daetwyler, H. D. 2009. Genome-wide evaluation of populations. Doctoral dissertation,
Wageningen University.
Daetwyler, H. D., Villanueva, B., Bijma, P. and Woolliams, J. A. 2007. Inbreeding in
genome-wide selection. Journal of Animal Breeding and Genetics 124(6): 369-
376.
194
Dan, T. T. and Summers, P. M. 1996. Reproductive performance of sows in the tropics.
Tropical Animal Health Production 28(3): 247-256.
de Roos, A. P. W., Hayes, B. J. and Goddard, M. E. 2009. Reliability of genomic
predictions across multiple populations. Genetics. 183(4): 1545.
de Roo, G. 1987. A stochastic model to study breeding schemes in a small pig
population. Agricultural Systems 25(1):1-25.
Dekkers, J. C. 2004. Commercial application of marker - and gene - assisted selection in
livestock: Strategies and lessons. Journal of Animal Science 82 E - Supplement:
E313- 328.
Dekkers, J. C. M. 2007. Prediction of response to marker-assisted and genomic
selection using selection index theory. Journal of Animal Breeding and
Genetics 124:331-341.
Dekkers, J. C. M. 2010. Animal genomics and genomic selection. In: Adapting animal
production to changes for a growing human population. Proceedings of
International Conference, Lleida, Spain, 19-21 May 2010, p. 61 – 71.
Dekkers, J. C. and Hospital, F. 2002. The use of molecular genetics in the improvement
of agricultural populations. Nature Review Genetics 3:22-32
Dekkers, J. C. and Chakraborty, R. 2004. Optimizing purebred selection for crossbred
performance using QTL with different degrees of dominance. Genetics Selection
Evolution 36(3): 297-324.
195
Delgado, C. L. 2003. Rising consumption of meat and milk in developing countries has
created a new food revolution. Journal of Nutrition 133(11): 3907-3910.
Du, F. X., Clutter, A. C. and Lohuis, M. M. 2007. Characterizing linkage disequilibrium
in pig populations. International Journal of Biological Sciences 3(3): 166.
Dzama, K. 2002. The pig performance testing scheme in Zimbabwe. AGTR Case study,
Nairobi, Kenya, ILRI.
Eusebio, J. A. 1984. Pig production in the tropics. Longman, Essex, UK.
Egbunike, G. N. 1986. Response of European pigs to improved management under the
humid tropical climatic conditions in Ibadan, a review. Pig News and Information.
p. 315-324.
Evans, G. J., Giuffra, E., Sanchez, A., Kerje, S., Davalos, G., Vidal, O., Illan, J. L.,
Noguera, J. L. Varona, L., Velander, I., Southwood, O. I. de Koning, D. J., Haley,
C. S., Plastow, G. S. and Andersson, L. 2003. Identification of quantitative trait
loci for production traits in commercial pig populations. Genetics 164(2): 621-
627.
Ewing, B. and Green, P. 2000. Analysis of expressed sequence tags indicates 35,000
human genes. Nature Genetics 25(2): 232-234.
Falconer, D. S. and Mackay, T. F. C. 1996. Introduction to Quantitative Genetics. 4th ed.
England, Longman Group limited.
196
FAO 2001. Bulletin of statistics. 2(1)
FAO 1998. Local animal breeds still threatened by extinction p.4
Fernando, R. L, Habier, D., Stricker, C., Dekkers, J. C. M. and Totir, L. R. 2007.
Genomic selection. Acta Agriculturae Scandinavica, Section A - Animal
Sciences 57(4): 192-195.
Fisher, R. A. 1918. The correlation between relatives on the supposition of Mendelian
inheritance. Transactions of Royal Society of Edinburgh 52: 399-433.
Fleiss, J. L. and Gross, A. J. 1991. Meta-analysis in epidemiology, with special reference
to studies of the association between exposure to environmental tobacco smoke
and lung cancer: A critique. Journal of Clinical Epidemiology 44(2): 127-139.
Forsberg, C. W., Golovan, S. P., Ajakaiye, A., Phillips, J. P., Meidinger, R. G., Fan, M.
Z., Kelly, J. M. and Hacker, R. R. 2005. Genetic opportunities to enhance
sustainability of pork production in developing countries: A model for food
animals. Applications of Gene-based technologies for improving animal
production and health in developing countries, p. 429-446.
Frisch, M. and Melchinger, A. E. 2005. Selection theory for marker-assisted
backcrossing. Genetics. 170(2): 909-917.
Garett, J. L. 2000. Overview. In: Achieving urban food and nutrition security in the
developing world, 2020 Focus 3.
197
Garrick, D. J. and van Vleck, L. D. 1987. Aspects of selection for performance in several
environments with heterogeneous variances. Journal of Animal Science 65:409-
421.
Gerasimov, V. and PronAE, E. 2006. Industrial crossing of pigs - a main method of
production of commercial pigmeat. Svinovodstvo (Moskva) p. 5-7.
Giannotti, J. D. G., Packer, I. U. and Mercadante, M. E. Z. 2005. Meta-analysis for
heritability estimates of growth traits in beef cattle. Revista Brasileira De
Zootecnia 34: 1173-1180.
Gianola, D., Fernando, R. L. and Stella, A. 2006. Genomic-assisted prediction of genetic
value with semi parametric procedures. Genetics 173(3): 1761- 1776.
Gianola, D., Perez-Enciso, M. and Toro, M. A. 2003. On marker-assisted prediction of
genetic value: Beyond the ridge. Genetics 163(1): 347.
Gibson, J.P., Quinton, V.M. and Simedrea, P. 2001. Responses to selection for growth
and back fat in closed nucleus herds of Hampshire and Duroc pigs. Canadian
Journal of Animal Science 81(1): 17 – 23.
Gilmour, A. R., Gogel, B. J., Cullis, B. R. and Thompson, R. 2009. ASReml user guide
release 3.0.3
Goddard, M. E., Hayes, B. J. McPartlan, H. and Chamberlain, A. J. 2006. Can the same
genetic markers be used in multiple breeds? Proceedings of the 8th
World
Congress on Genetics Applied to Livestock Production, Belo Horizonte, Brazil
198
Goddard, M. E. and Hayes, B. J. 2007. Genomic selection. Journal of Animal Breeding
and Genetics 124(6): 323-330.
Goddard, M. E. 2009. Genomic selection: prediction of accuracy and maximisation of
long term response. Genetica 136:245-257.
Goffinet, B. and Gerber, S. 2000. Quantitative trait loci: A meta-analysis. Genetics
155(1): 463-473.
Goudet, J. 2001. FSTAT, a program to estimate and test gene diversities and fixation
indices (Version 2.9.3).
Habier, D., Fernando, R. L. and Dekkers, J. C. M. 2007. The impact of genetic
relationship information on genome-assisted breeding values. Genetics 177(4):
2389.
Habier, D., Fernando, R. L. and Dekkers, J. C. M. 2009. Genomic selection using low-
density marker panels. Genetics 182:343-353.
Harmegnies, N., Farnir, F., Davin, F., Buys, N., Georges, M. and Coppieters, W. 2006.
Measuring the extent of linkage disequilibrium in commercial pig populations.
Animal Genetics 37:225-231.
Harris, B. L., Johnson, D. L. and Spelman, R. J. 2008. Genomic selection in New Zealand
and the implications for National genetic evaluation. In Proceedings of the 36th
ICAR Biennial Session held in Niagara Falls, U.S.A. p. 325 -330.
199
Hayes, B. and Goddard, M. E. 2001. The distribution of the effects of genes affecting
quantitative traits in livestock. Genetics Selection Evolution 33(3): 209-229.
Hayes, B. J. and Goddard, M. E. 2003. Evaluation of marker assisted selection in pig
enterprises. Livestock Production Science 81(2-3): 197-211.
Hayes, B. J., Visscher, P. M., McPartlan, H.C. and Goddard, M. E. 2003. Novel multi-
locus measure of linkage disequilibrium to estimate past effective population size.
Genome Research 13:635-643.
Hayes, B. J., Bowman, P. J., Chamberlain, A. J. and Goddard, M. E. 2009. Invited
review: Genomic selection in dairy cattle: Progress and challenges. Journal of
Dairy Science 92(2): 433-443.
Hayes, B. J., Bowman, P. J., Chamberlain, A. C., Verbyla, K. and Goddard, M. E. 2009.
Accuracy of genomic breeding values in multi-breed dairy cattle populations.
Genetics Selection Evolution 41:51
Hayes, B. J. and Goddard, M. E. 2010. Genome-wide association and genomic selection
in animal breeding. Genome 53:876-883.
Hedges, L. V. and Olkin, V. 1985. Statistical methods for meta-analysis. London,
Academic Press.
Henderson, C. R. 1975. Best linear unbiased estimation and prediction under a selection
model. Biometrics 31(2): 423-447.
200
Henderson, C. R. 1984. Applications of linear models in animal breeding. University of
Guelph Guelph, Canada.
Henson, E. L. 1992. In situ Conservation of Livestock and Poultry. In: (2nd edn ed.),
FAO Animal Production and Health Paper 99, Food and Agricultural
Organization of the United Nations.
Hill, W. G. and Robertson, A. 1968. Linkage disequilibrium in finite population.
Theoretical and Applied Genetics 38:226-231
Hill, W. G. and Robertson, A. 1968. The effects of inbreeding at loci with heterozygote
advantage. Genetics 60(3): 615.
Hill, W. G., and Caballero, A. 1992. Artificial selection experiments. Annual Review of
Ecological System 23: 287-310.
Holness, D. H. 1976. The role of the indigenous pig as a potential source of protein in
Africa - a review. Rhodesia Agricultural Journal 73(3): 59-62.
Howie, B. N., Donnelly, P. and Marchini, J. 2009. A flexible and accurate genotype
imputation method for the next generation of genome-wide association studies.
PLoS Genetics 5(6): e1000529. doi:10.1371/journal.pgen.1000529.
Hospital, F. 2005. Selection in backcross programmes. Philosophical Transactions of
Royal Society of London B Biological Sciences 360(1459): 1503-1511.
201
Huisman, A. E., Cherel, P. and van Schriek, M. 2010. Linkage disequilibrium and
signatures of selection on chromosome 1 in commercial sire and dam line. 9th
World Congress on Genetics Applied to Livestock Production, Leipzig, Germany.
Ibánẽz-Escriche, N., Fernando, R. L., Toosi, A. and Dekkers, J. C. M. 2009. Genomic
selection of purebreds for crossbred performance. Genetics Selection Evolution
41(1): 12.
Ihaka, R. and Gentleman, R. 1996. R: A language for data analysis and graphics. Journal
of Computational and Graphical Statistics 5(3): 299-314.
Ilatsia, E., Githinji, M., Muasya, T., Okeno, T. and Kahi, A. 2008. Genetic parameter
estimates for growth traits of Large White pigs in Kenya. South African Journal
of Animal Science 38(3)
Irgang, R., Fávero, J. A. and Kennedy, B. W. 1994. Genetic parameters for litter size of
different parities in Duroc, Landrace, and Large White sows. Journal of Animal
Science, 72(9): 2237.
Jafarikia, M., Maignel, L., Wyss, S. and Sullivan, B. 2010. Linkage disequilibrium in
Canadian swine breeds. 9th
World Congress on Genetics Applied to Livestock
Production, Leipzig, Germany.
Jannink, J. L. 2010. Dynamics of long-term genomic selection. Genetics Selection
Evolution. 42(1): 35.
202
Jannink, J. L., Lorenz, A. J. and Iwata, H. 2010. Genomic selection in plant breeding:
From theory to practice. Briefings in Functional Genomics. 9(2): 166.
Kahi, A. K., Rewe, T. O. and Kosgey, I. S. 2005. Sustainable community-based
organizations for the genetic improvement of livestock in developing countries.
Outlook Agriculture 34(4): 261-270.
Kanengoni, A. T., Dzama, K., Chimonyo, M., Kusina, J. and Maswaure, S. M. 2004.
Growth performance and carcass traits of Large White, Mukota and Large White
x Mukota F1 cross given graded levels of maize cob meal. Animal Science: An
International Journal of Fundamental and Applied Research p.61-66.
Kennedy, B.W., Quinton, V.M. and Smith, C. 1996. Genetic changes in performance
tested pigs for fat depth and growth rate. Canadian Journal of Animal Science,
76 (2): 41
Kim, J. J., Zhao, H. H., Thomsen, H., Rothschild, M. F. and Dekkers, J. C. M. 2005.
Combined line-cross and half-sib QTL analysis of crosses between out bred lines.
Genetic Research 85(3): 235-248.
King, A. H., Jiang, Z., Gibson, J. P., Haley, C. S. and Archibald, A. L. 2003. Mapping
quantitative trait loci affecting female reproductive traits on porcine chromosome
8. Biology of Reproduction 68(6): 2172-2179.
King, J. W. B. 1987. Improvement of pigs in the tropics: General principles: Animal
genetic resources, strategies for improved use and conservation. Proceedings of
the 2nd meeting of the FAO/UNEP expert panel, Warsaw, Poland, June 1986 with
203
proceedings of the EAAP/PSAS Symposium on Small Animal Populations of
Domestic Animals.
Kizilkaya, K., Fernando, R. L. and Garrick, D. J. 2010. Genomic prediction of simulated
multibreed and purebred performance using observed fifty thousand single
nucleotide polymorphism genotypes. Journal of Animal Science, 88: 544-551.
Konstantinov, K. V. and Brien, F. D. 2003. Influence of sire by year interactions on the
direct-maternal genetic correlation for weaning weight of Western Australian
Merino sheep. Australian Journal of Agricultural Research 54(7): 723-729.
König, S. and Swalve, H. H. 2009. Application of selection index calculations to
determine selection strategy in genomic breeding programs. Journal of Dairy
Science 92:5292-5303.
Koots, K. R. 1994. Studies on the genetic and economic parameters required for beef
cattle improvement. Doctoral dissertation, University of Guelph.
Koots, K. R., Gibson, J. P., Smith, C. and Wilton, J. W. 1994a. Analyses of published
genetic parameter estimates for beef production traits. 1. Heritability. Animal
Breeding Abstracts. 62(5): 309-338.
Koots, K. R., Gibson, J. P. and Wilton, J. W. 1994b. Analyses of published genetic
parameter estimates for beef production traits. 2. Phenotypic and genetic
correlations. Animal Breeding Abstracts 62(11): 825-853.
204
Kuhn, M. T., Boettcher, P. J. and Freeman, A. E. 1994. Potential bias in transmitting
abilities of females from preferential treatment. Journal of Dairy Science 77:2428-
2437.
Kumar, S. and Subramanian, S. 2002. Mutation rates in mammalian genomes.
Proceedings of the National Academy of Sciences (USA) 99:803-808
Kunavongkrit, A. and Heard, T. W. 2000. Pig reproduction in Southeast Asia; Animal
reproduction Research and practice II. Proceedings of the 14th International
Congress on Animal Reproduction, Stockholm, Sweden, 2-6 July, 2000. Animal
Reproduction Science 60/61: 527-533.
Kyriazakis, I. and Whittemore, C. T. 2006. Whittemore's science and practice of pig
production. Blackwell Pub.
Lakhani, G. P. and Jogi, S. 1999. Studies on dressing percentage in indigenous pigs and
their Large White Yorkshire grades. Indian Journal Animal Research 33(1): 29-
31.
Lande, R. and Thompson, R. 1990. Efficiency of marker-assisted selection in the
improvement of quantitative traits. Genetics 124(3): 743-756.
Lee, G. J. and Haley, C. S. 1995. Comparative farrowing to weaning performance in
Meishan and Large White pigs and their crosses. Animal Science 60(2): 269-280.
205
Lekule, F. P. and Kyvsgaard, N. C. 2003. Improving pig husbandry in tropical resource-
poor communities and its potential to reduce risk of porcine cysticercosis. Acta
Tropical 87(1): 111-117.
Liu Y., Jansen G. B., Lin, C. Y. 2002. The covariance between relatives conditional on
genetic markers. Genetics Selection Evolution 34:657-678
Lo, L. L., Fernando, R. L. and Grossman, M. 1993. Covariance between relatives in
multi-breed populations: Additive model. Theoretical Applied Genetics 87:423-
430
Luan, T., Woolliams, J. A., Lien, S., Kent, M., Svendsen, M. and Meuwissen. T. H. E.
2009. The accuracy of genomic selection in Norwegian Red Cattle assessed by
cross-validation. Genetics 183:1119-1126.
Lund, M and Su, G. 2009. Genomic selection in the Nordic countries. Proceedings of the
Interbull International Workshop - Genomic Information in Genetic evaluation,
Uppsala, Sweden, Bulletin no. 39.
Luo, Z. W., Thompson, R. and Woolliams, J. A. 1997. A population genetics model of
marker-assisted selection. Genetics 146:1173-1183.
Macaskill, P. and Walter, S. D. 2001. A comparison of methods to detect publication bias
in meta-analysis. Statistics of Medicine 20(4): 641-654.
Madalena, F., Agyegman, K., Cardellino, R. and Jain, G. 2002. Genetic improvement in
medium – to low – input system of animal production; Experiences to date. 7th
World Congress on Genetic Applied to Livestock Production.
206
Malek, M., Dekkers, J. C. M. Lee, H. K., Baas, T. J., Prusa, K. Huff-Lonergan, E. 2001.
A molecular genome scan analysis to identify chromosomal regions influencing
economic traits in the pig. II. meat and muscle composition. Mammalian Genome.
12(8): 637-645.
Maniatis, N. and Pollott, G. E. 2003. The impact of data structure on genetic (co)
variance components of early growth in sheep, estimated using an animal model
with maternal effects. Journal of Animal Science 81(1): 101.
Marin-Martinez, F. and Sanchez-Meca, J. 2010. Weighting by inverse variance or by
sample size in random-effects meta-analysis. Educational and Psychological
Measurement 70(1): 56-73.
Meade Moreno, G. 1987. Evaluation of productivity of crossbred pigs from 4 breeds
using two systems of crossbreeding. Veterinaria México. 18(3): 280.
Meuwissen, T. H. E. 2009. Accuracy of breeding values of 'unrelated' individuals
predicted by dense SNP genotyping. Genetics Selection Evolution 41:35
Meuwissen, T. H. E., Hayes, B. J. and Goddard, M. E. 2001. Prediction of total genetic
value using genome-wide dense marker maps. Genetics 157(4): 1819-1829.
Meyer, K. 1992. Variance components due to direct and maternal effects for growth traits
of Australian beef cattle. Livestock Production Science 31(3-4): 179-204.
Meyer, K. 1998. Estimating covariance functions for longitudinal data using a random
regression model. Genetics Selection Evolution 30(3): 221-240.
207
Milan, D., Bidanel, J. P., Iannuccelli, N., Riquet, J., Amigues, Y., Gruand J. Le Roy, P.,
Renard, C. and Chevalet, C. 2002. Detection of quantitative trait loci for carcass
composition traits in pigs. Genetics Selection Evolution 34(6): 705-728.
Mohanty, G. P. and Chhabra, A. K. 2007. Heat stress and ameliorative measures in
crossbred pigs. Indian Veterinary Journal 84(4): 430-431.
Moser, G., Tier, B., Crump, R. E., Khatkar, M. S. and Raadsma, H. W. 2009. A
comparison of five methods to predict genomic breeding values of dairy bulls
from genome-wide SNP markers. Genetics Selection Evolution 41:56.
Moser, G., Tier, B., Khatkar, M. S., Hayes, B. J. and Raadsma, H. W. 2010. Accuracy of
direct genomic values in Holstein bulls and cows using subset of SNP markers.
Genetics Selection Evolution 42:37.
Mote, B. E. and Rothschild, M. F. 2006. SNP detection and linkage mapping for pig
genes involved in growth. Animal Genetics 37(3): 295-296.
Muir, W. M. 2007. Comparison of genomic and traditional BLUP - estimated breeding
value accuracy and selection response under alternative trait and genomic
parameters. Journal of Animal Breeding and Genetics 124(6): 342-355.
Mushandu, J., Chimonyo, M., Dzama, K., Makuza, S. M. and Mhlanga, F. N. 2005.
Influence of sorghum inclusion level on performance of growing local Mukota,
Large White and their F-1 crossbred pigs in Zimbabwe. Animal Feed Science
Technolology, 122(3-4): 321-329.
208
Muys, D. and Westenbrink, G. 1995. Pig Husbandry in the Tropics. Meinderts J. ed.
Wageningen, Netherlands, Agromisa.
Nejati-Javaremi, A., Smith, C. and Gibson, J. P. 1997. Effect of total allelic relationship
on accuracy of evaluation and response to selection. Journal of Animal Science
75:1738-1745.
Nielsen, H. M., Sonesson, A. K. and Meuwissen, T. H. E. 2011. Optimum contribution
selection using traditional BLUP and genomic breeding values in aquaculture
breeding schemes. Journal of Animal Science 89:630-638
Nguyen, T., Phung T. V., Pham, H. D., Nguyen, N., Nguyen K. Q. and Vo, T. H. 1996.
Improvement of productivity and meat quality of pigs in the Red River Delta
region by crossbreeding. In: Exploring Approaches to Research in the Animal
Sciences in Vietnam: A workshop held in the city of Hue, Vietnam, 31 July-3
August, 1995.
Norris, D., Varona, L., Visser, D., Theron, H., Voordewind, S. and Nesamvuni, E. 2007.
Estimation of the additive and dominance variances in South African Landrace
pigs. South African Journal Animal Science 36(4): 261.
Notter, D. 1999. The importance of genetic diversity in livestock populations of the
future. Journal Animal Science 77(1): 61.
Nwakpu, P. E., and Onu, P. N. 2007. Utilization of Nigerian native pig in breeding 1. A
genetic assessment of crossbred heterosis in growth and litter size. Research
Journal of Animal Sciences. 1(4): 139-147.
209
Odegard, J., Yazdi, H., Sonesson, A. and Meuwissen, T. H. E. 2009. Incorporating
desirable genetic characteristics from an inferior into a superior population using
genomic selection. Genetics 181:737-745.
Oke, U. K., Ibe, S. N., Ologbose, F. I. and Amaefula, K. U. 2006. Effect of breed of sire
on growth performance of exotic crossbred pigs in a humid tropical environment.
Journal of Animal and Veterinary Advances 5(9): 744-748.
Olivier, J. J., Moyo, S., Montaldo, H. H., Thorpe, W., Zárate, A. V. and Trivedi, K. R.
2002. Integrating genetic improvement into livestock development in medium-to
low-input production systems. 7th World Congress on Genetics Applied to
Livestock Production
Oseni, S. 2005. Evaluation of the F1 and backcrosses of Nigerian local pigs and the Large
White for litter characteristics in Southwest Nigeria. Livestock Research for Rural
Development 17(4): article 44.
Pathiraja, N. 1986. Improvement of pig-meat production in developing countries. 1.
Exploitation of hybrid vigour (heterosis). World Animal Review, : 18-25.
Pathiraja, N. 1987. Improvement of pig-meat production in developing countries. 2.
Selection schemes. World Animal Review p. 2-10.
Pearson, K. 1900. Mathematical contributions to the theory of evolution. VII. on the
correlation of characters not quantitatively measurable. Philosophical
210
Transactions of the Royal Society of London. Series A, Containing Papers of a
Mathematical or Physical Character, p. 1-405.
Pong-Wong, R., George, A. W., Woolliams, J. A., Haley, C. S. 2001. A simple and rapid
method for calculating identity-by-descent matrices using multiple markers.
Genetic Selection Evolution 33:453-471.
Quiniou, N. and Noblet J. 1999. Influence of high ambient temperature on performance
of multiparous lactating sows. Journal of Animal Science 77: 2124-2134.
Ramos, A.M., Crooijimans, R. P., Affara, N. A., Amaral, A. J., Archibald, A. L.,
Beever, J. E., Bendixen, C., Churcher, C., Clark, R., Dehais, P., Hansen, M. S.,
Hedegaard, J., Hu, Z. L., Kerstens, H. H., Law, A. S., Megens, H. J., Milan, D.,
Nonneman, D.J., Rohrer,G.A., Rothschild, M. F., Smith, T. P., Schnabel, R.
D., Van Tassell, C. P., Taylor, J. F., Wiedmann, R. T., Shook L. B. and
Groenen,M.A. 2009. Design of high density SNP genotyping assay in the using
of SNPs identified and characterized by nextgeneration sequencing technology.
PLoS One 4: e6524.
Rinaldo, D., Canope, I., Christon, R., Rico, C. Ly, J. and Dieguez, F. 2003. Creole pigs in
Guadeloupe and Cuba: A comparison of reproduction, growth performance and
meat quality in relation to dietary and environmental conditions. Pig News
Information 24(1): 17N-26N.
211
Rohilla, P. P., Bujarbaruah, K. M., Kumar, M. and Singh, G. 2000. Carcass traits of
Large White Yorkshire, Hampshire and Naga local pigs. Indian Journal of Animal
Science 70(3): 307-308.
Rohrer, G. A., Alexander, L. J., Hu, Z. Smith, T. P. Keele, J. W. and Beattie, C. W.
1996. A comprehensive map of the porcine genome. Genome Research 6(5): 371.
Rothschild, M. F. 1996. Genetics and reproduction in the pig. Animal Reproduction
Science 42(1 - 4): 143-151.
Rothschild, M. F., Hu, Z. L. and Jiang, Z. H. 2007. Advances in QTL mapping in pigs.
International Journal of Biological Sciences. 3(3): 192-197.
Rothstein, H. R., Sutton, A. J. and Borenstein, M. 2005. Publication bias in meta-
analysis. Publication Bias in Meta-Analysis, Prevention, Assessment and
Adjustments. Chichester: Wiley, p. 1-7.
Saatchi, M., Miraei-Ashtiani, S. R., Nejati Javaremi, A., Moradi-Shahrebabak, M. and
Mehrabani-Yeghaneh, H. 2009. The impact of information quantity and strength
of relationship between training set and validation set on accuracy of genomic
estimated breeding values. African Journal of Biotechnology 9 (4): 438-442.
Safari, E., Fogarty, N. and Gilmour, A. 2005. A review of genetic parameter estimates for
wool, growth, meat and reproduction traits in sheep. Livestock Production
Science 92(3): 271-289.
212
Sargolzaei, M. and Schenkel, F. S. 2009. QMSim: A large-scale genome simulator for
livestock. Bioinformatics 25(5): 680.
Sargolzaei, M., F. S. Schenkel and P. M. VanRaden. 2009. gebv: Genomic breeding
value estimator for livestock. Technical report to the Dairy Cattle Breeding and
Genetics Committee. October 07, 2009.
Sargolzaei, M., Schenkel, F. and Chesnais, J. 2011. Accuracy of imputed 50k genotypes
from 3k and 6k chips using FImpute version 2. Technical report to the Dairy
Cattle Breeding and Genetics Committee Meeting, September 13, 2011.
SAS Institute Inc. 2009. SAS/STAT ® 9.2 User‘s guide, second edition.
Schaeffer, L. R. 2006. Strategy for applying genome-wide selection in dairy cattle.
Journal of Animal Breeding and Genetics 123(4): 218-223.
Schenkel, F. S., Sargolzaei, M., Kistemaker, G., Jansen, G. B., Sullivan, P., Van
Doornmaal, B. J., VanRanden, P. M. and Wiggans, G. R. 2009. Reliability of
genomic evaluation of Holstein cattle in Canada. Proceedings of the
Interbull International Workshop - Genomic Information in Genetic evaluation,
Uppsala, Sweden, Bulletin no. 39.
Sellier, P. 1998. Genetics of Meat and Carcass Traits. In: The Genetics of the Pig.
Rothschild, M. F., and Ruvinsky, A. eds. CAB International, New York, USA p.
143.
213
Shenglin, Y. 2007. Genetic diversity of Thai Indigenous pig, wild boar and Chinese
Qianbei black pig based on microsatellite DNA and sequence polymorphism of
mitochondria DNA cytochrome b gene. Doctoral dissertation, Suranaree
University of Technology.
Simianer, H. 2009. The potential of genomic selection to improve litter size in pig
breeding programs. Proceedings of the 60th
Annual Meeting of European
Association of Animal Production, Barcelona, Spain.
Sitzenstock, F., König, S., Ytournel, F. and Simianer, H. 2010. Evaluation of genomic
selection for functional traits in horse breeding programs. Proceedings of the 9th
World Congress on Genetics Applied to Livestock Production, Leipzig,
Germany
Smaragdov, M. G. 2009. Genomic selection as a possible accelerator of traditional
selection. Russian Journal of Genetics 45(6): 633-636.
Smith, C. 1988. Genetic improvement of livestock in developing countries using nucleus
breeding units. World Animal Review. 65: 2–10.
Solberg, T. R., Sonesson, A. K., Woolliams, J. A. and Meuwissen, T. H. E. 2008.
Genomic selection using different marker types and densities. Journal of Animal
Science 86(10): 2447-2454.
Sollero, B. P., Paiva, S. R., Faria, D. A., Guimaraes, S. E. F., Castro, S. T. R., Egito, A.
A., Albuquerque. M. S. M., Piovezan, U., Bertani, G. R. and Mariante, A. da S.
214
2009. Genetic diversity of Brazilian pig breeds evidenced by microsatellite
markers. Livestock Science 123:8-15.
Sonesson, A. K., Gjerde, B. and Meuwissen, T. H. E. 2005. Truncation selection for
BLUP-EBV and phenotypic values in fish breeding schemes. Aquaculture 243:
61–68.
Sorensen, D. and Gianola, D. 2002. Likelihood, bayesian and MCMC methods in
quantitative genetics. New York, Springer-Verlag.
Sutton, A. J., Abrams, K. R., Jones, D. R., Sheldon, T. A. and Song, F. 2000. Methods for
Meta-Analysis in Medical Research. J. Wiley.
Steel, R. G. D. and Torrie, J. H. 1960. Principles and procedure of statistics. New York,
McGraw-Hill.
Sved, J. A. 1971. Linkage disequilibrium and homozygosity of chromosome segments in
finite populations. Theoretical Population Biology 2(2): 125-141.
The World Bank. 2007. World development report 2008: Agriculture for development p.
369.
Thévenon, S., Dayo, G. K., Sylla, S., Sidibe, I., Berthier, D. Legros H., Boichard, D.,
Eggen, A. and Gautier M. 2007. The extent of linkage disequilibrium in a large
cattle population of western Africa and its consequences for association studies.
Animal Genetics 38(3): 277-286.
215
Thompson, R. 2008. Estimation of quantitative genetic parameters. Proceedings of the
Royal Society B 275(1635): 679.
Toosi, A., Fernando, R. L. and Dekkers, J. C. M. 2010. Genomic selection in admixed
and crossbred populations. Journal of Animal Science 88(1): 32.
van der Waaij, E. H., Bijma, P. Bishop, S. C. and van Arendonk, J. A. 2002. Using
genetic markers for disease resistance to improve production under constant
infection pressure. Journal of Animal Science 80(2): 322-329.
VanRaden, P. M. 2008. Efficient methods to compute genomic predictions. Journal of
Dairy Sciences. 91:4414-4423.
VanRaden, P. M., Van Tassell, C. P., Wiggans, G. R., Sonstegard, T. S., Schnabel, R. D.,
Taylor, J. F. and Schenkel, F. S. 2009. Invited review: Reliability of genomic
predictions for North American Holstein bulls. Journal of Dairy Science 92(1):
16.
Visscher, P. M., Hill, W. G. and Wray, N. R. 2008. Heritability in the genomics era —
concepts and misconceptions. Nature Reviews Genetics 9(4): 255-266.
Visscher, P. M., Pong-Wong, R., Whittemore, C. and Haley, C. S. 2000. Impact of
biotechnology on (cross) breeding programmes in pigs. Livestock Production
Science 65(1): 57-70.
Visscher, P. M. and Haley, C. S. 1995. Utilizing genetic markers in pig breeding
programmes. Animal Breeding Abstracts. 63(1): 1-8.
216
Weigel, K. A. 2001. Controlling inbreeding in modern breeding programs. Journal of
Dairy Science. 84(1): E177-E184.
Weller, J. I. 2010. Whole genome marker-assisted selection. CAB Reviews: Perspectives
in Agriculture, Veterinary Science, Nutrition and Natural Resources. 5(023): 1-
10.
Wellmann, R. and Bennewitz, J. 2010. Considering dominance in genomic selection. 9th
World Congress on Genetics Applied to Livestock Production, Leipzig,
Germany.
Whittaker, J. C., Thompson, R. and Denham, M. C. 1999. Marker-assisted selection using
ridge regression. Annals of Human Genetics 63(04): 366-366.
Wilson, D. E. 1992. Application of ultrasound for genetic improvement. Journal of
Animal Science 70(3): 973-983.
Wright, S. 1965. The interpretation of population structure by F-statistics with special
regard to systems of mating. Evolution 19(3): 395-420.
Xu, S. 2003. Estimating polygenic effects using markers of the entire genome. Genetics
163(2): 789.
Zanga, J., Chimonyo, M., Kanengoni, A., Dzama, K. and Mukaratirwa. S. 2003. A
comparison of the susceptibility of growing Mukota and Large White pig
genotypes to infection with ascaris suum. Veterinary Research Communication
27: 653-660.
217
Zhong, S., Dekkers, J. C. M., Fernando, R. L. and Jannink, J. L. 2009. Factors affecting
accuracy from genomic selection in populations derived from multiple inbred
lines: a Barley case study. Genetics 182:355-364.
View publication statsView publication stats
top related