18. Breeding objectives
ANIMAL BREEDING = WHERE TO GO ? + HOW TO GET THERE ?
Breeding objectives are all about where to go. We often formulate this as an overall objective
to, for example:
Maximise $/head - using a different economic weight for each trait of economic importance,
as in the lecture on multi traitselection.
Maximise $/hectare or Minimise Cost per unit product - these are more appropriate as they
consider changes inresource utilisation which accommodate genetic change. [Sheep the size
of elephants would probably return more $/head -but you couldn't run so many !!].
Unfortunately these approaches are much more difficult to handle.
An outline of typical breeding objectives follows:
Breeding objectives - Meat production
1. GROWTH RATE
- fast growth to larger slaughter size for sire lines.
- fast growth to small mature size for dam lines.
2. AIM AT CARCASS SIZE DEMANDED BY THE MARKET:
- recall effect of animal size on production efficiency, from lecture 1:
Body weight: Vücut ağırlığı
Accumulated Food Intake: Eklemeli Yem Alımı
Maternal overheads: Ana verimi üzerinde olan
Line of highest efficiency: En yüksek verim hattı
Large genotype:Büyük genotip
Small genotype:Küçük genotip
Birth weights:Doğum ağırlıkları
- depends on product: veal or beef
bacon or pork
home or export
3. AIM AT DESIRED FAT COVER
- Downwards everywhere: pigs cattle sheep poultry
a) market demand
b) FCE (due to high energy cost of fat)
- Major problems with buying fat animals on the hoof.
poor market signals
- Need a minimum level of fat for :-
1. stop dehydration 2. flavour
3. hold muscles together 4. resist microbial entry
- Also fat colour, especially in poultry
4. AIM AT INCREASED LEAN PER CENT
- Difficult to do directly - CAT scanner at UNE
- Sib testing in pigs
- Progeny testing in sheep (Norway)
- Generally via dressing percent and fat cover score
5. INCREASE F.C.E.
- little variation in 'true efficiency' (Lecture 1)
- recording costs high
- central performance tests in (intensive) cattle, pigs
- on-farm tests in pigs, poultry
6. OTHER TRAITS
- Weaning rate (major importance)
Many component traits - fertility
- Adaptation to production system (especially in tropics)
Disease resistance - ticks, worms, fleece rot
Suitable conformation for extensive management (ranginess etc.)
Breeding objectives - Milk production
1. MILK YIELD - increase
2. MILK COMPOSITION - Highly dependent on product type and payment scheme.
Fat percent - increase of maintain
- polyunsaturated fats??
Protein - increase
3. FCE to MILK - Not in evidence due to costs of recording food intake.
4. MEAT TRAITS - Important in countries which have little or no beef.
Norway - 100% of beef is from dairy cattle
U.K. - 54% of beef from dairy genes
Australia. - very little of beef from dairy genes
5. FERTILITY TRAITS - not as important as for meat due to ongoing harvesting.
6. DISEASE RESISTANCE - e.g. mastitis, needs good recording as in Norway.
Lymphocyte cell counts in Sweden.
7. OTHERS - longevity; milking speed; temperament; udder defects; other type traits.
Breeding objectives - Wool production
1. CLEAN FLEECE WEIGHT - highly correlated with FCE to wool
2. FIBRE DIAMETER - smaller is better, so decrease or maintain. Modern processing
techniques require many small fibres in yarns to minimise co-incidence of weak spots and
therefore breaks on automated machinery.
3. PIGMENTATION - avoid black fibres, ginger spots. Much general colour (yellowness) is
lost in processing.
4. LENGTH and STRENGTH - coming into objective marketing?
5. VARIATION IN FIBRE FINENESS - smaller is better, relates to handle and possibly
resistance to fleece rot.
6. WEANING RATE - largely for excess young stock for meat
- ongoing harvest of wool, so less important here than with meat production.
- some problems rearing twin and trips.
7. ADAPTATION TO PRODUCTION SYSTEM
- resistance to fleece rot
- resistance to dust penetration
- low water requirement
Breeding objectives - Egg production
1. EGG NUMBER: hen day % = as a percentage
2. HIGH EGG WEIGHT (but genetic correlation with egg number = -0.5)
3. AGE AT FIRST EGG - younger gives less resource input.
4. FOOD CONVERSION EFFICIENCY: Food ---> Eggs
5. EGG QUALITY
resistance to cracking
shell colour - USA white, elsewhere brown
yolk colour - more yellow
6. DISEASE RESISTANCE - (expose and select)
7. ADAPTATION TO NEW AVIARY SYSTEMS: Animal welfare legislation in some
countries is resulting in 'fly away' aviary systems being on the horizon.
Industry structures and strategies
The following are the wool, meat, beef and dairy cattle, pigs and poultry.
Wool sheep breeding industry
STRUCTURE Mostly closed nucleus. Some open nucleus -
Merinotech companies, Grass Merinos, AMS etc., and open nucleus structure within many
USE OF Q.G. Has generally been low because:
1. Visual classing - no numbers to work on
2. Little single sire mating, date of birth, dam age, birthtype, pedigree recording.
3. Little understanding by breeders, especially on use of information from relatives etc.
But some use of testing services F.D., Yield etc.
BLUP techniques used for some elite flocks which record pedigree.
Meat sheep breeding industry
STRUCTURE: 3 Breed Cross, e.g. PD x (BL x M) is typical.
USE OF Q.G. Now quite high due to LAMBPLAN.
BLUP techniques are now in use.
LAMBPLAN across flock evaluations began in 1991.
Poor market signals limit the benefits to breeders who 'Do the Right Thing'
Beef cattle breeding industry
STRUCTURE: Purebreeding - breed societies play a big role.
Crossing now very common, female replacement problems
Synthetics Brangus, Murray Grey, Santa Getrudis etc.
Increasing use of bos indicus in suitable areas.
Stud structure prevalent.
USE OF Q.G. Moderate but increasing quite rapidly: N.B.R.S. offers use of BREEDPLAN
(with BLUP EBV's), and GROUP-BREEDPLAN (across herd evaluation). Aiming mostly at
increased growth rate. Fertility traits accommodated now, and carcass traits in 1995.
Dairy cattle breeding industry
STRUCTURE : State or national, 4 pathway
USE OF Q.G. State of the art, in most cases, using BLUP techniques.
Selection of fm (bull mothers) pathway could be more efficient in most countries.
Much pressure on conformation etc. remains.
Some use of Multiple Ovulation and Embryo Transfer.
Pig breeding industry
Mostly closed nucleus due to disease problems. Exceptions due to A.I. and hysterectomy.
Commercial product usually a backcross, 3 breed cross or similar.
USE OF Q.G. Extensive:
Specialised sire and dam lines
Optimal crossing designs
Selection indices and BLUP
'PIGBLUP' is now commercially available.
Poultry breeding industry
STRUCTURE: Breeding companies: develop and keep sire and dam lines
Producers: buy in crossbred parents and produce 3- or 4-breed crosses, or similar.
USE OF Q.G. Extensive. But techniques used are often not published.
19. Molecular genetics
Quantitative genetics uses phenotypic information to help identify animals with good
genes. Moleculargenetics techniques aim to locate and exploit gene loci which have a major
effect on quantitative traits (hence QTL -Quantitative Trait Loci).
Animal breeders have had to resort to analysis of animal performance in order to make
inferences about the value of genes carried by individuals. In only a few cases have single
loci of major effect (Quantitative Trait Loci, QTL) been identified and exploited directly.
This chapter reviews emerging techniques for detecting and using QTL in domestic animal
populations. Pedigreed performance data can be analysed to detect segregation at single QTL,
using either maximum likelihood or mixed model regression techniques. However, this is
only really useful for screening populations to find promising material for further studies
exploiting genetic markers. These genetic markers, mostly microsatelites, have been used to
construct genetic maps for the main domesticated species, and these are now being used on
data from crossbred animals in trials designed to locate QTL. There is some contention that
genetic markers used in Marker Assisted Selection programs could increase rates of genetic
gain, but this seems unlikely to become important except in cases where QTL have been
detected and are closely linked to informative markers. Where QTL have been cloned and
DNA tests made commercially available, genetic gains could come to be improved
considerably. Not only can the specific biological effects of favourable QTL alleles be
exploited directly, but modifications to the breeding program design can be made to properly
exploit direct prediction of genetic merit for milk production in bulls, meat quality in live
candidates for breeding, and future performance of embryos.
Genetic markers and the linkage map
To locate genes we can use genetic linkage studies to put 'landmarks' on the chromosomes.
We use genetic markers for this purpose. Genetic markers are loci which are easily
genotyped (i.e. the alleles each individual carries can be determined easily). Genetic markers
that are linked tend to exhibit a co-inheritance of alleles.
For example, animals with the genotype on the left will tend to give
either AB gametes or ab gametes, the proportion of Ab and aB
gametes being equal to the recombination fraction. So the pattern of
inheritance of alleles from genetic marker loci can be used quite
simply to infer linkage to each other.
Genetic markers which have been used a lot in the past include blood groups and polymorphic
enzymes. We have relatively few such markers, but this has been overcome with the advent
of new types of markers. Two of these are described briefly here:
Restriction Fragment Length Polymorhisms (RFLP's).
Restriction endonuclease enzymes cut DNA wherever they find the appropriate nucleotide
sequence (eg. Eco R1 cuts atthe 'recognition sequence' GAATTC). If there is a mutation at
this sequence, no cut is made and the resulting DNAfragment is longer. Also mutation to give
a new recognition sequence gives a pair of shorter fragments.
In this diagram we can see there are different length DNA fragments from each of these two
chromosomes, according tocutting or lack of cutting at four sites. Using a battery of
restriction endonucleases, thousands of RFLP markers can begenerated.
Microsatelites are DNA regions with variable numbers of short tandem repeats in nucleotide
sequence (CA repeats in thediagram here). Each microsatelite locus is recognised or targeted
by its primer - the unique sequence adjacent to therepeating region.
Microsatelites make good genetic markers because they each have many different 'alleles' - ie.
there can bemany different lengths of the repeat region. With many alleles, most individuals
are heterozygous, givingpower to note association between marker allele and performance in
progeny inheriting a favourable linkedQTL
This diagram represents the
electrophoretic gel in which microsatelite
DNA fragments of different size
(different alleles) have been run. Smaller
fragments of DNA can migrate further.
In this example there are four alleles, and
of course each individual can carry only
two! The genotypes of the four animals
are deduced to be bc, ad, ac and bd,
The following diagram shows a linkage map of chromosome 1 of cattle, as derived by
Barendse et al, 1994, Nature Genetics, 6: 227-235. Reproduced with permission from Bill
On the left is the linkage map derived from studies on co-inheritance of alleles. Most
loci are microsatelites or RFLPs. Distances between loci are given in units of centimorgans
(100 units is equivalent to a 50% recombination fraction).
In the centre is a representation of the chromosome with its visible banding regions.
Some loci have been localised through, for example, observing fluorescing DNA from the
gene of interest 'sticking' to particular regions.
On the right is a list of loci associated with chromosome 1 through work with somatic
cell hybrid lines. For example, alleles at these loci might all be present in a cell line
containing only chromosome 1 from cattle, the rest from, say, hamsters.
Searching for QTL
With so many mapped markers it is now possible to find QTL. There are two types of 'gene
Searching for QTL already known to exist. Examples:
· The Booroola and Inverdale genes in sheep (giving high fecundity)
· The N gene in sheep (giving carpet wool in Drysdales)
· The Halothane gene in pigs (giving increased lean percent but also stress
susceptibility). This gene has been found - it is the ryanodine receptor gene.
· The double muscling gene in cattle (giving increased muscle mass). This gene
has been found - it is the myostatin gene.
Locating these known genes is easier because we can relate their presence or absence
confidently with marker information.
Searching for unknown QTL. Locating these major genes is much more difficult because we
only have animals' phenotypes to (very weakly) detect presence or absence of a QTL. The
approaches used are illustrated next.
Screening populations for evidence of segregating QTL
Virtually all major genes currently used by animal breeders were first detected by noting
strong familial trends during inspection of recorded data. Given the potentially large
confounding influence of environment, polygenes, and segregation at the major locus,
together with the large pedigreed data sets available today, such 'eyeballing' of data can be
improved upon greatly by using computers to apply statistical techniques for major gene
detection. The diagram below shows results from a screening test using Findgene software,
described later on this page.
A number of test statistics have been developed to detect major gene segregation, and these
yield a single result per population, not per animal. Hill and Knott (1990) classify and discuss
these. Le Roy and Elsen (1992) compare the performance of 22 statistics, and suggest their
robustness may be low, especially when trait distribution is skewed. Moreover, to screen
populations in order to select likely carriers, a method is needed which allows ranking of
animals on the probability of carrying one or more major genes.
Genotype probability calculations can be very computationally demanding, with up to 3n
possible combinations of genotypes among n animals for a single locus with two alleles
segregating (Elston and Stewart, 1971; Ott, 1979). This means that using exact maximum
likelihood methods to either find the most likely combination of genotypes or to calculate
genotype probabilities is only practical for small problems involving less than about 20
animals. Animal pedigrees are generally highly looped, due to the fact that sires are mated to
a large number of dams, and involve inbreeding, which makes the problem considerably more
complex. However, Janss et al. (1993) developed an iterative method, based on that of van
Arendonk et al. (1989), and this can be used to determine genotype probabilities in large
animal breeding populations spanning several generations.
For a mixed inheritance model involving both a major gene and polygenes, exact likelihood
analysis is not feasible and approximations are needed. A numerical integration technique to
account for polygenes (e.g. Hermite integration as applied by Knott et al. (1991a,b)) yields
reasonable results. Numerical integration can only be applied to groups of independent sire
families. This restricts the value of these methods, which are not able to exploit the extra
information in data covering several generations (Janss et al., 1993).
Approximate maximum likelihood methods have been used by Hoeschele (1988), Knott et al.
(1991a,b) and Hofer and Kennedy (1993) to calculate both genotype probabilities and
estimate polygenic breeding values for simple pedigree structures. However, the flexible
iterative approach to calculating genotype probabilities (van Arendonk et al., 1989; Janss et
al., 1993) can be implemented together with a mixed-model regression step to account for the
effects of polygenes under any pedigree structure (Kinghorn et al., 1993), implemented as
This is an illustration of the method used to
arrive at converged estimates of major gene
effects, b1 and b2, and calculate genotype
probabilities for individuals. Genotype
probabilities are calculated using segregation
analysis, following the method of van Arendonk
et. al. (1989), and allele frequency estimated by
appropriate averaging of these probabilities.
These probabilities are then fitted in a regression
of phenotype on genotype probabilities and
animal breeding values, using a BLUP
framework. This regression yields estimates of
breeding value and new estimates of b1 and b2.
Phenotypes (P) are corrected for estimated
breeding values () in an attempt to reduce the
influence of polygenic effects on the next
calculation of genotype probabilities. The cycle
illustrated is repeated sufficient times to give
convergence in estimates of b1 and b2.
However, to speed operation, convergence is
first achieved with animal breeding values not fitted in the regressions, then animal breeding values
are fitted until final convergence is reached.
Both the approximate maximum likelihood methods and this regression method can lead to
estimates of genotype effects and gene frequencies for the population as well as genotype
probabilities and estimated breeding values for all individuals. However, results are generally
biased to a moderate degree, except, under some circumstances, in the absence of selection.
More recently, Monte Carlo integration techniques have been applied in segregation analysis
(Guo and Thompson, 1992, Janss et al., 1993). Janss et al. found empirically unbiased
estimates of variation due to polygenes and major gene effects in simulated data which
contained multiple loops.
The methods described in this section do not make use of genetic markers, and seem unlikely
to be able to detect genes reliably with an effect of less than about half a phenotypic standard
deviation, even with favourable combinations of population size, population structure, and
polygenic variation. However, if used routinely in parallel with genetic evaluation of
pedigreed data sets, they may prove useful in identifying individuals and families that warrant
closer scrutiny via test matings and use of genetic markers.
Marker Assisted QTL detection.
Marker assisted QTL detection studies are based on the analysis of segregation from
heterozygous individuals. This figure illustrates an hypothetical set of marker and QTL
genotypes in which the sire is heterozygous at both loci, marker allele M being associated
with favourable QTL allele Q. Phenotypes are recorded on the progeny, and an analysis for
association with marker alleles is carried out.
It is important to note that the only data available are marker locus genotypes of animals, and
their phenotypes. In practice, many marker loci are used, maybe 100 or 200, to give
reasonable coverage of the genome.
Given appropriate dam genotypes, as in this figure, allele M will be associated with
favourable phenotypes in the progeny, with a strength dependent on the effect of Q and the
closeness of linkage between the loci, reflected in the recombination fraction, r. With the
design given in the figure, it is not possible to distinguish between a loosely linked (say r =
0.2) QTL of very big effect, and a tightly linked (say r = 0.02) QTL of more moderate effect -
both of these can result in the same superiority of progeny carrying the M marker allele
inherited from the sire.
Greatest power for doing this sort of QTL detection work is achieved through one or both of
1. Set up a cross between distantly related breeds or lines to maximise the probability of
heterozygosity at both QTL and marker loci in the sires used to make the cross illustrated
above. The dams used in the cross should be one of the parental purebreeds.
2. Generate very large half-sib progeny groups from the cross illustrated above, in order
to reduce the effects of residual phenotypic variance and to test for marker-major gene
associations in individual sires. This can be done where male fecundity is high - eg with
artificial insemination in the dairy industry.
In this type of work there can be considerable difficulties in making correct inference about
the significance of results. This is because of the large number of markers tested - 5 out of 100
useless markers will give positive results at the 5% significance level! This works in both
directions - there is danger of declaring presence of a QTL when in fact it does not exist, and
there is danger of missing valuable QTL because of conservative inference.
Identifying major genes via their products
In some cases, prior knowledge of the biochemical and/or physiological basis of a phenotype
has directly led to the identification of a major gene. It is now possible to rapidly clone,
sequence and characterise a gene from a primary gene product. This approach has been most
successful for qualitative or single gene traits (Shuster et al., 1992), and the expanding
number of new genes and/or proteins isolated in eukaryotes is improving the number of
candidate genes, although not necessarily the success rate. The number of potential
candidates is very large indeed for most traits - with human and mouse research providing
massive databases to work on.
A useful refinement to this approach has been the selection of candidate genes once a gene
has been mapped to a small chromosomal region - known as positional candidate genes.
The contrast between candidate gene and linked markers approaches
The candidate gene approach to finding QTL results in direct markers, which can be used for
genotype assisted selection.
The genome scan approach uses genetic markers to search for QTL. Once a QTL has been
detected in this way, closely linked markers can be used for marker assisted selection, and in
addition, positional candidate genes can be used to try and find direct markers.
Direct markers are generally much preferred to linked markers, if they are truly markers for
major gene effects. Their biggest benefit is that they can be used without trait measurement
or pedigree recording. Despite this, there is value in having such information, to monitor the
effect of the major gene in different breeds/lines and production systems, and exploit it
However, there is some potential to incorrectly identify a candidate gene as a major gene
directly affecting the trait of interest, because of linkage disequilibrium with the true causative
gene (a reasonably consistent linkage on the chromosome with that gene) in both the original
experimental population and in re-evaluation populations. This highlights the value of re-
evaluation in distinctly different stock. There is a tendency for linkage of genes that are
related in function. [In many cases this is derived from local DNA duplication along the
chromosome, followed by some divergence in function between the resulting genes]. This
means that there is more than a random chance of close linkage and thus linkage
disequilibrium. There is also the potential, albeit small, for true direct markers to be
There is the potential for direct markers to be unreliable, especially if these are not functional
markers. This was made evident in work on the myostatin gene for double muscling - if
certain single direct markers had been adopted, there could have been false negatives in
These may turn out to be small problems in practice, and they should be identifiable during
application if a reasonable amount of trait recording is maintained for monitoing purposes.
In contrast, there is considerable need to gather trait and pedigree information for use of
linked genetic markers (see example). However, trait measurement is not required for
selection between progeny of sires which have already been tested (phase 2 in Progeny testing
to exploit detected QTL). It may also be more difficult to market the concept that bull 'X' has
a 95% chance of carrying this major gene 'Y', as opposed to a virtual guarantee from a direct
marker test. Moreover, the direct marker test will usually tell the variant of major gene
inherited from each parent, which is even less reliable with use of linked markers.
However, the fact that linked markers cover a region of chromosome means that they could be
more robust in some ways. They will be more likely to properly track a major gene than a
direct marker that turns out to be only closely linked to the causative gene. They may also
give information about a number of QTL, whether or not this is known to be the case.
Moreover, the information gathered in linked marker programs can be of direct benefit in
verifying parentage, finding direct markers, and detecting other QTL affecting the measured
In conclusion, we do not need to set up a contest between these two approaches. They should
go hand-in-hand in application, driven by commercial demands, with a natural progression
from linked markers to direct markers as more information becomes available for each case.
Exploiting Quantitative Trait Loci
Where a direct marker (DNA-test) exists for a QTL, we can use Genotype Assisted Selection
Where a only linked markers exist for a QTL, we must use Marker Assisted Selection (MAS).
Genotype Assisted Selection
Directly marked genes
Some QTL can be detected directly, usually through a DNA test. One example is the
halothane gene in pigs, which is known to be the ryanodine receptor gene (Fujii et al., 1991).
If we have a genetic marker directly inside a gene, then we have great power to exploit this
gene in breeding programs, using genotype assisted selection. This is because in most cases
we can know with 100% confidence which animals have the good variants of this gene.
Estimating QTL effects
Before proceeding to application, we must first discover the effect of each variant of the gene
on all traits of commercial importance, in each production environment, and possible in
different genetic backgrounds (eg. different breeds). This should be done in an analysis that
accounts for the effects of polygenes as well as identifiable environmental effects (Kennedy et
al., 1992). This approach can also be used to test genetic markers, including bands and
haplotypes determined from gel electrophoreses, for effects on phenotype (Blattman et al.,
1993). One remaining problem, however, is to distinguish between effects of the QTL of
interest and linked loci. Bovenhuis and Weller (1992) developed a maximum likelihood
method that accounts for both direct and linked effects and applied this method to detect
associations between milk protein polymorphisms and milk production.
Modes of exploitation
Of course there is the potential problem that the desired variant of a gene is already
widespread in the breed, with little to be gained from a genetic marker program.
Alternatively, there may be no good variants within the breed, such that some form of
crossbreeding or gene transfer is required to import them.
A big advantage with these directly marked genes is that we do not need trait or pedigree
recording for application in the field. Unfortunately, this is not the case where the gene is
marked indirectly – by one or more genetic markers which are close, but not at the same
location on the chromosome as the gene itself.
Given knowledge of the effects of major genes, decisions can be made about modes of
exploitation. Genes affecting female fertility may be of little value in breeds such as Dorset
Horn sheep, whose main impact is though their use as sires in terminal crossing systems for
meat production. On the other hand, genes increasing growth rate are more usefully
introgressed into such terminal sire breeds, rather than dam breeds, as smaller dams make for
more food-efficient meat production systems.
How should major genes be increased in frequency in target breeds or lines? A classical
selection index approach can be used (Neimann-Sorensen and Robertson 1961). This aims to
balance favourable major gene and polygene effects in order to maximise overall genetic
merit in the next generation. However, selected parents with the major gene will tend to have
poorer polygenic values - and this is a compromise that can be important where objectives are
long term in numbers of generations, as in short-generation interval species such as pigs and
Reducing the selection index weighting on a major gene will give more long-term response -
the major gene will become fixed in the long term, but with less compromise in response due
to polygenic effects. This argument also holds for selection on markers closely linked to large
QTL, and an analogous effect has been shown for crossbreeding effects plus polygenic effects
Marker Assisted Selection
The rate of detection of QTL affecting commercially important traits is increasing rapidly.
We can exploit these using linked genetic markers to make inference about QTL genotypes,
and thus undertake marker assisted selection. Marker genotypes can be used to assist
selection decisions, increasing the frequency of favourable QTL, or targeting their
introgression into other lines. The value of this depends on a number of factors:
a. Where heritability is low, the value of information on individual QTL tend to be
b. Where the trait(s) of interest cannot be measured on one sex, marker information gives
a basis to rank animals of that sex.
c. If the trait is not measurable before sexual maturity, marker information can be used to
select at a juvenile stage.
d. If a trait is difficult to measure or requires sacrifice (as with many carcass traits)
marker information can be used instead.
When direct markers exist for a QTL, Genotype Assisted Selection(GAS) can be used.
Long-term response to marker assisted selection can be disappointing.
Proper evaluation methods and mating structures are required to exploit non-additive QTL
Conclusion: Marker assisted selection can improve selection response. Its value is limited for
traits that we can breed for easily by classical methods, especially in the longer term.
However, there seems great potential for MAS to generate change in traits such as pigmented
fibers, meat quality, milk quality, and disease resistance. Biological systems are complex,
such that interaction between loci should be of importance. Given this, we will have
challenging tasks in biological modeling and breeding program design to produce ideal
There are two basic scenarios for this use of marker information: use with undetected QTL
and use with detected QTL.
MAS with undetected QTL
In this case, an analysis is carried out across the population to measure the amount of
trait variation associated with each marker locus (Lande and Thompson, 1990). Those loci
accounting for sufficient variation are included in a selection index, with positive weightings
given to alleles related to better performance. In order for there to be useful relationships
across the population between markers and undetected QTL, they must be in linkage
disequilibrium, giving a preponderance of certain linkage phases. This is likely to occur in the
generations following a wide cross, but the disequilibrium will diminish with continued
breeding. Zhang and Smith (1992) used computer simulation to show an advantage averaging
about 10% when marker information was added to a normal genetic evaluation program using
best linear unbiased prediction of breeding values. However, Lande and Thompson (1990)
showed that the advantage of using markers can be increased substantially when maternal
environmental effects are large, giving increased value in contrasting family members.
MAS with detected QTL
When QTL have been detected, it is possible to choose only closely linked markers. However,
there can be a need for appropriate population structure plus statistical analysis to exploit
QTL effects with useful reliability. Unless there is considerable linkage disequilibrium, no
one marker allele is consistently associated with a favourable QTL allele, due to
Notice that in the ram's semen, marker alleled A is not always associated with the red QTL
allele, due to recombination.
The result has been to consider marker-QTL associations within half sib families. When the
sire is heterozygous at both loci, (as in the figure above, and the one for QTL detection), and
recombination fraction is sufficiently low, then progeny phenotypes will be related to marker
alleles transmitted by the sire, and these markers can thus be used to assist selection decisions.
Progeny testing to exploit detected QTL
Progeny testing phase
The simple approach to exploiting an indirectly marked gene operates first at the level of sire
families (approach 2. above). The sire and sufficient progeny (say, 50 progeny) are
genotyped for markers flanking the known QTL - this probably means an A.I. program and/or
repeated use of a sire over years. The progeny are measured for the target traits, most
probably carcass traits.
Sacrifice of these progeny is not critical as long as there are or will be a good number of
progeny left to select amongst. Sacrificed progeny do not have to be produced out of top
quality cows, as long as this does not compromise the normal progeny testing which parallels
There are two key objectives for the 'genetic marker' aspect of this progeny test. These are to
infer whether the sire is heterozygous at the QTL (two separate variants of the QTL), and to
infer the linkage relationship between genetic markers and QTL in the sire. In addition to this
is the desirability of estimating the QTL effect as expressed in the offspring. This is not
essential if the effects of the QTL are well established as being robust across different
breeding lines and environments.
The sires for this phase should ideally be of the highest possible genetic merit, and (likely to
be) widely used, to capitalise on the costs involved.
Selection between progeny phase
Once this has been done, it is a relatively simple task to predict which variant of the QTL each
progeny carries, and this is where the value of the program is reaped. This includes future
progeny as well as progeny in the progeny test, such that overall value is increased it the
original sires are widely used. Accuracy is improved by gathering genetic marker information
on dams as well, because this helps to resolve which marker variants have been inherited from
Improvements in power
The need for large half-sib families is reduced over time, as marker and trait information is
gathered on a deeper pedigree. This is because we now have methods (approaches 3 and 4) to
use information from all relatives to make inference about which marker variant is linked to
the superior gene variants in each animal.
One major problem with this 'approach 2' is that the sires used are heterozygous - carrying
only one copy of the favourable variant of the major gene. This is because we need both
haves and have nots among the progeny to be able to pick out the ones which have inherited
the good variant from the sire. If the sire had two copies of the favourable variants, all
progeny would inherit the good variant from the sire.
So we need power to detect when the sire has two copies of the favourable variant.
Approaches 3 and 4 give some power to do this. They both use any marker and trait
information that is available, on all relatives. They also manage to make value out of
breeding designs without these large half-sib families - because of use of information from all
relatives, not just half-sibs. In practice, these approaches should come into play over time, as
this information accumulates.
Extensions to these approaches will help us to infer major gene variants inherited from both
sire and dam, across a number of different major genes. This could become of extra benefit,
as it will allow us to plan matings to give progeny predicted to have the best genetic
constitution across all major genes involved. This is important wherever there are interactions
between genes - eg the best 'marbling gene' variant might depend on what 'total fat gene'
variants are carried by the animal.
Genetic evaluation at individual QTL
Large half-sib family structures are not available in many species, leading to the need for
MAS analysis methods that can operate on general pedigrees. This could be approached by
either calculating the most likely pedigree locations for recombination between QTL and
marker loci, or calculating probabilities of recombination at each location. The latter
philosophy is more evident - Fernando and Grossman (1989) included markers in a mixed
model analysis to predict breeding values. Goddard (1992) extended this to cover linked
markers and use of pairs of markers bracketing a QTL. Van Arendonk et al
. (1994) developed an approach that can be used to exploit large numbers of QTL with little
computational overhead, and suggested methods to account for ungenotyped animals in the
data set. Meuwissen and Goddard (1997) used markers to modify transmission probabilities in
segregation analysis to calculate QTL genotype probabilities
Evaluation of Genetic value or Breeding value?
Genetic value is the value of an animal's genes to itself. Breeding value is the value of an
animal's genes to its progeny. In general, breeding value has been of much more importance
to animal breeders - it reflects the merit that can be transmitted to the next generation. It is the
sum of the average effects of alleles carried by the animal, and because of the large number of
loci classically assumed, there is no power to capitalize on anything but the average effects of
these alleles, as dominance deviations in progeny cannot be predicted under normal
However, when dealing with individual QTL we have the power to set up matings designed to
exploit favourable non-additive interaction in the progeny. See Mating structures to exploit
Five approaches to QTL evaluation
Five approaches to genetic evaluation using markers can be identified:
1. Marker association with merit across
families. This relies on population-level linkage disequilibrium. Other methods are generally
better if pedigree information is available.
2. Within-family analysis, making
inference about sires’ QTL heterozygosity and marker-QTL linkage phases, in a framework
similar to one used for marker assisted QTL detection. This leads to information for selection
3. Use of markers to infer probability
of identity by descent of contributing QTL alleles, with QTL effects treated as random and no
assumption about number of alleles at each QTL. This effectively extends 2. above to use all
pedigree information and give QTL EBV’s.
4. Use of markers to modify
transmission probabilities in segregation analysis to calculate QTL genotype probabilities.
Typically two QTL alleles are involved and QTL genotype effects are treated as fixed. This is
probably preferable where few effectively distinct alleles are known to be segregating, and
where dominance and/or epistasis are important.
5. Use of genetic markers located
within target QTL. This removes the need for trait measurements and pedigree information to
evaluate animals at QTL of known effect, leading to Genotype Assisted Selection. However,
multiple alleleism means that only complete sequence markers are fully reliable, as otherwise
QTL alleles of identical marker type can have different effects.
An analysis method which targets inference making at one or more known QTL segregating
in the population has a number of ideal features, including:
·It should make appropriate use of all available information. This may involve inference
about which meioses involve crossing over between the QTL and one or more marker loci.
·It should be able to make inference about genotype of individuals at the QTL, possibly by
calculation of probabilities for all genotype states.
·It should be able to estimate QTL genotype effects and allele frequencies
·It should be computationally feasible, especially for multiple QTL and many marker loci.
An iterative sampling approach, such as Gibbs sampling, will cover the first three above, as it
samples across the range of possible values and states, including all possible patterns of
realised QTL- marker crossing over in the population. Results are generally unbiased,
although there can be problems with unusual distributions of effects or states, for example
when genotype probabilities are sampled at 0 or 1 for small groups of related animals, and the
sampling chain becomes ‘stuck’ at these values. Unfortunately, these approaches are
The next pages describe analysis methods which are faster, and relate to approach 3 and
approach 4 above.
QTL as random effects
This approach to genetic evaluation at marked QTL calculates probabilities of identity by
descent of QTL alleles between gametes. These can be used to calculate animal EBV’s at the
QTL, much as we use coefficients of relationship to estimate ‘polygenic’ breeding values.
Marker information is used to calculate probability of identity by descent for alleles in
different individuals (Goddard, 1992) or in different gametes (Fernando and Grossman, 1989;
van Arendonk et al., 1994, Wang et al., 1995). This approach is most properly dealt with at
the gametic level. The figure gives a simple illustration of how marker genotypes can help to
more accurately build a gametic relationship matrix.
Figure. Gametic relationship matrices (GRM) for a QTL, on the right, are of dimension 6 sites
x 6 sites for the simple 3-animal pedigree shown. Elements of the GRM are probability of
identity by descent of the alleles at the prevailing pair of sites. In the upper GRM, no marker
information is available, and, for example, probability of identity by descent between sites 4
and 6 is 0.5, as site 6 (maternal) could have inherited from sites 3 or 4 with equal probability.
In the lower GRM, a marker with alleles A, B and C is available, and for example, probability
of identity by descent between sites 4 and 6 is 1, for the marker locus. If the QTL is linked
with a recombination fraction of 0.1, then the probability of identity by descent between sites
4 and 6 is 0.9, for the QTL, with a 0.1 probability (in the event of recombination) for sites 3
and 6. Special attention is required where there is ambiguity of marker allele inheritance
(Wang et al., 1995).
For any one pedigree there is a realized gametic relationship matrix, and this contains only 1’s
or 0’s as elements, simply because each pair of QTL alleles are either fully identical by
descent, or not at all. This concept has been used to test different methods for building
predicted gametic relationship matrices through simulation of replicated populations and
averaging of realised gametic relationship matrices (B.E. Clarke, unpublished). Different
methods of building the GRM give slightly different results, and the differences in resulting
EBV’s are very small indeed.
One weakness of this appraoch to evaluating animal for a QTL is that no inference is made
about QTL genotype of individuals. This can be important where there are known non-
additive effects involved (as mentioned above) , as it leads to possibilities for mating
structures to exploit QTL. more effectively. However, an approach to getting the required
QTL genotype probabilities from their estimated breeding values is given on the next page.
Genotype probabilities from the GRM method
Figure. Heights of the distribution of (EBV at QTL for animal i) at the expectations
conditional on QTL genotype are proportional to genotype probabilities.
Having used the gametic relationship matrix approach to estimate breeding values at a QTL,
there is an approach, as yet untested, to calculating genotype probability is for each QTL
genotype for each individual. For each animal, the expectation of its estimated breeding value
is calculated conditional on each possible QTL genotype. These are then related to the actual
estimated breeding value () and its error distribution as the figure. The heights of this
distribution at each QTL genotype are proportional to the genotype probabilities for this
QTL allele frequencies can then be estimated by use of a simple counting procedure using
genotype probabilities, with iteration to achieve convergence. Given QTL effects and
genotype probabilities, variance due to the individual QTL can be simply calculated, such that
QTL breeding values can be estimated based on relevant priors. This helps to overcome one
weakness in the identity-by-descent method - but without further extension, the assumption of
known QTL genotype effects remains.
QTL as fixed effects
Approach 4 listed above - to find the best fitting set of QTL genotype probabilities - shows
good promise. Maximum likelihood methods have suffered bias from simplifying
assumptions made to give computational feasibility. However, regression methods are now
giving apparently unbiased results, at least for estimation of QTL effects. These methods are
based on a 2-step iterative scheme of, firstly, calculating QTL genotype probabilities using
segregation analysis, and secondly, regressing phenotypes on these probabilities (Kinghorn et
al., 1993) or carrying out regression weighted by these probabilities (Meuwissen and
Goddard, 1997). In both cases, fixed effects and polygenic breeding values are also fitted.
Marker information is accommodated by modifying transmission probabilities at the
segregation analysis step, according to prevailing marker genotypes (Meuwissen and
Mating structures to exploit QTL
Mating structures will become more important when we have access to many QTL affecting
the trait(s) of interest. This is because interaction between loci may be more important than
we can perceive now. In this case, there will be need to generate stock with specific multi-
QTL genotypes, to exploit favourable dominance and epistatic interactions. Genetic
evaluation of type 4 provides the genotype probabilities needed for this.
Of course prediction of QTL genotype of candidates is only of real value in helping to predict
genetic values of their progeny - because the object is to improve performance of descendants.
This in turn means that the evaluation system should be intimately associated with the mate
allocation process, wherever non-additive effects are to be exploited. The combination of
animal selection and mate allocation can be termed mate selection. Application of evaluation
systems to exploit individual QTL will thus frequently involve mate selection strategies in
addition to the simpler ranking processes we are used to with selection.
One extreme example of this is where we manage to use genetic markers to identify QTL and
chromosomal regions that can contribute strongly to increased expression of heterosis in
crossbred progeny. Recurrent selection of purebreds on the performance of their crossbred
progeny has not been of great practical value - however now with extra information from
genetic markers and known QTL we have some power to breed for increased heterosis in a
Response to MAS and GAS over time
Response due to genetic change in QTL is limited – stopping when ideal genotypes have been
reached. This happens more quickly where just one or two closely marked QTL are involved
for the trait of interest, and frequencies of favourable alleles are not low. Indices that aim to
maximise response in the next generation alone are suboptimal for longer-term objectives, and
long-term benefits in genetic gain are usually less than short-term benefits.
There is an argument that normal selection would increase the frequency of favourable QTL
variants anyway. This can be true - with studies showing little long-term benefit of MAS over
normal selection under certain simple conditions. The real conclusion from this work is that
MAS is most useful for traits that are difficult or costly to measure - typically carcass and
disease traits. Moreover, QTL detection work is likely to be ongoing, such that new QTL take
over the focus of MAS programs.
Most evaluations of MAS have considered short time horizons. Kashi et al. (1990)
investigated the value of MAS of young bulls prior to entering a progeny testing scheme. The
importance of marker heterozygosity in the elite ancestral sires was critical, such that diallelic
markers were found to be of little benefit. Use of single polyallelic markers gave increases of
15 to 20% in the rate of genetic gain, and this was improved by use of haplotypes of groups of
markers. Meuwissen and van Arendonk (1992) showed that when using MAS to help rank
candidates within families, most benefit accrues in short-generation length nucleus breeding
schemes which do not benefit from the high accuracies resulting from progeny testing. They
found improvements in genetic gain of up to 25% in such schemes.
Over time, with many QTL being tracked, there will be power to model the biology of what is
happening to the traits of interest, and therefore power to predict the best combination of
variants across QTL. This may not occur for some considerable time, but when it does, we
will have considerable flexibility to specify beef genotypes that will be highly focussed on
production in defined environments and defined markets.
The potential problem of epistasis
There is a the potential danger that QTL directly detected in experimental data sets will have a
smaller true effect in other herds when they come to be exploited commercially. The extent to
which this is really a problem depends on the strength of interactions between genes, and is
essentially not predictable from a narrow information base. There is a statistical tendency for
QTL to have stronger effects in the genetic backgrounds (and production/management
environments) in which they are detected. The re-evaluation of putative QTL in unrelated
herds is an important step here.
In this pedigree, all females have been genotyped for a 2-allele locus (A,a). Light coloured
lines emanate from females. Can we deduce the genotypes of the males?
The male in the second row is quite easy. Can you explain why he must
be a heterozygote?
The male in the first row is more difficult, and the answer depends on our
prior assumption about gene frequency.
The male in the third row is even more difficult, but the answer in this
case does not depend on gene frequency. Can you explain briefly why this
is the case?
We might be able to deduce some results for simple cases like this, by
relatively simple logic. But what about bigger examples in more complex
If we have 20 ungenotyped animals we have up to 3 to the power 20 (=348,6784,401)
'possible' answers for the 2-allele, 3-genotype case. This makes solution by simple searching
methods not feasible.
Most methods for doing this sort of thing make use of information from three sources:
3.Mate(s) plus progeny,
... and use that information either recursively or iteratively over a number of cycles. Care
has to be taken not to overuse information (double dipping) or mistreat loops in the pedigree
(eg. as with inbreeding).
A larger example
A data set contains 4207 pigs in a complex
pedigree structure. Money was spent
genotyping 113 of these animals at the
Ryanodine receptor locus (whose alleles affect
expression of porcine stress syndrome). Of
these, 65 were normal homozygotes, 40 were
heterozygotes and 8 carried two copies of the
After segregation analysis, an additional 1886
animals could be excluded form one genotype class and an additional 42 animals could be
genotyped, both with 100 percent confidence. At the 90% confidence level (ie. probability of
being any given genotype > 0.9) an additional 263 animals could be genotyped.
The pig pedigree is shown here. Can you deduce the genotypes of any ungenotyped animals?
Of course not! You need a computer program to do the job for you. Richard Kerr and Brian
Kinghorn have written 'Geneprob' for this. It works well and fast on large data sets, but does
not fully account for inbreeding loops. Output is genotype probabilities (probabilities of
being AA, Aa and aa, summing to 1) for each individual.
There is the option to transfer genes between animals, breeds, species or orders by active
intervention in the laboratory to transfer genes (e.g. Palmiter et al., 1982, in mice). Most
results have been of dubious commercial value for animal breeding, largely due to the random
sites of insertion, leading to unfavourable insertion mutations, lack of control over inserted
genes and the need to inbreed to achieve homozygosity.
In particular, much effort has been made in relation to hormones which affect growth and
body composition, but high and variable levels in transgenic animals has led to physiological
problems and mortality (e.g. Rexroad et al., 1991). The use of site-targeted insertion and more
stable and controllable promoters may alleviate these problems. Gama et al. (1992) have
investigated strategies for use of transgenes in animal populations.
However, edible products from transgenic livestock seem likely to meet with much market
resistance (Ewing, 1990) - a problem less likely to be encountered by genotypes derived from
normal breeding methods which only use information from the use of DNA technology.
Other more specific applications may prove fruitful both on the farm and in the marketplace:
Examples include transferring the cysteine biosynthetic pathway to sheep from bacteria in
order to improve wool growth in periods of reduced feed supply, attempts to use plant-derived
chitinase to inhibit insect pathogens in animals, and human protein production in animal milk
and blood (see Ward, 1992).
Gene transfer method - by example
This gene construct is inserted into the mouse genome. Zinc in drinking water activates the
promoter and extra Growth Hormone is produced. It is a major effort to do this (you can look
at Palmiter et al, Nature. 300: 611-615 for more detail). The following is just a brief outline,
given to illustrate the complexity of the process (there is no need to memorise this):
1. Get messenger RNA (mRNA) for both the Growth Hormone (from pituitary tumour cells)
and Metallothionein (from mouse livers).
2. mRNA => complimentary DNA
3. Insert cDNA into an E.Coli plasmid ...
Clone this and brew up a lot.
4. Bind cDNA to rabbit (G.H.) and
mouse (Met.) chromosomes:
Note that if the gene product is known
we have a way to locate genes without
the mapping procedures described
5. Get fluorescing bits out and sequence the genomic DNA. Work out amino acids from the
triplet codes and confirm this DNA codes for what is wanted (G.H. and Met. respectively).
6. Join the Met. and G.H. genomic genes, insert the construct into an E.Coli plasmid, clone
this and brew up a lot.
7. Cut gene constructs out of plasmids and microinject:
1.Variable number of gene construct insertions.
2.Random site of insertion.
3.Growth hormone expressed in 'metallothionine active' tissues - ie. the promoter works.
4.Zinc in drinking water increases size 70%
5.Transgenes are transmitted to progeny. Transgenes are defined as DNA constructs at a
given genomic site.
1. Need inbreeding to fix transgenes:
Transgene can only 'meet itself' by inbreeding.
Future: Site targeted insertion ??
There are some indications this can be done.
This would give instant homozygotes.
2. Transgene insertion breaks host DNA - so homozygotes could be lethal.
3. Need for much testing:
Is transgene product expressed, and if so, in what tissues?
Is it controllable via the promoter?
Does it do the job desired? - at a physiological level
- overall production efficiency
Are other traits affected? Especially survival and fertility.
Will industry use resulting stock?
Any consumer resistance?
How should new techniques interact with quantitative genetics?
For further discussion on industry implementation see:
Smith 1987, "On the use of transgenes in livestock improvement" Animal Breeding Abstracts