Using genetic selection in the breeding of
`dairy cattle
`Julius van der Werf, University of New England, Australia and Jennie Pryce, Department
`of Economic Development, Jobs, Transport and Resources (Government of Victoria) and
`La Trobe University, Australia
`
`1
`
`Introduction
`
`2 Breeding programmes: AI, progeny testing, embryo transfer and in vitro
`fertilization
`
`3 The structure of dairy breeding programmes
`
`4 The exchange and selection of genetic material
`
`5 Genomic selection
`
`6 Multi-trait selection
`
`7 Breeding objectives
`
`8 Genomic selection for functional traits
`
`9 Conclusion
`
`10 Where to look for further information
`
`11 Acknowledgements
`
`12 References
`
`1 Introduction
`
`Genetic change in dairy populations has been dramatic over the last five decades. Over
`the long term, selective breeding had been one of the most powerful tools to change
`the constitution and productivity of the dairy herd. Two main technologies have shaped
`dairy breeding programmes during the last five decades. Artificial insemination (AI) has
`been applied since the 1950s and had an important impact on the selection intensity
`of males and the dissemination of the best genetic material across the population.
`Genomic selection was introduced in the last decade and has been a second wave
`of influential technology that has affected the rates of gain in breeding programmes.
`Herd recording of an increasing number of dairy characteristics and genetic evaluation
`methods, notably best linear unbiased prediction (BLUP) (Henderson, 1973), have
`developed alongside these breeding and selection technologies and as such have
`become very powerful instruments in the genetic selection of dairy populations. AI, and
`
`http://dx.doi.org/10.19103/AS.2016.0005.15
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`2
`
`Using genetic selection in the breeding of dairy cattle
`
`to a smaller extent the female reproductive technologies, have made dairy breeding
`a very dynamic activity where genetic material can easily be acquired or exchanged
`over long distances. This has made possible the large-scale introduction of Holstein
`Friesian genes into many dairy populations throughout the world a few decades ago,
`while there is currently still a large open world market for genetic material, which is
`supported by international genetic evaluation systems (Banos, 2010). With it, of course,
`has developed the threat that the genetic diversity of the world’s dairy population has
`been narrowing to suboptimal levels.
`The genetic change of dairy populations has largely been driven by the increased
`milk productivity per cow. The milk production per cow has more than doubled since
`the introduction of AI in the 1950s, with an annual increase of about 2%. In the United
`Kingdom, the average milk production per cow per year was 5151 kg in 1990, whereas
`it was 7899 kg in 2014 (AHDB Dairy), which is a 53% increase. In the United States, the
`production per cow per lactation increased by 72% between the years 1985 and 2015
`(USDA, Fig. 1). The genetic trend between 1990 and 2000 was 1092 kg (VanRaden, 2004),
`implying that about 70% of the phenotypic increase was due to genetic selection. The
`emphasis on productivity increases have led to associated changes in other traits, notably
`a decrease of reproductive performance. Selection indexes have been augmented since
`the 1990s to also include more objective measurements of reproductive performance,
`health traits and efficiency and longevity.
`This chapter will emphasize two main aspects of breeding programmes in dairy. First, we
`will discuss the development of structures of breeding programmes. Then we will discuss
`the main factors that drive genetic change, in particular, the dominating progeny testing
`schemes, and how these factors have changed over time with the introduction of new
`technologies. The second part of this chapter will focus on the evolution of breeding
`objectives and to what extent this has affected selection response for the various traits
`in dairy production. This section will include some principles about selection response,
`particularly how multiple traits may change depending on the breeding objectives used
`and the recorded information that is available about the various traits. This section is
`important because there is often a lack of understanding of how a change in selection
`emphasis can be quite different from a change in selection response. Many breeding
`programmes lack good phenotypic information about non-production traits and as a result
`the genetic change is still dominated by an increase in yield, in spite of an increased
`selection emphasis on other traits.
`
`12000
`11000
`10000
`9000
`8000
`7000
`6000
`5000
`1975 1985 1995 2005 2015 2025
`
`milk yield per cow
`
`(kg/lactation)
`
`Figure 1 Milk production per cow in the United States (www.ers.usda.gov/datafiles).
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`Using genetic selection in the breeding of dairy cattle
`
`3
`
`2 Breeding programmes: AI, progeny testing, embryo
`transfer and in vitro fertilization
`
`2.1 AI and progeny testing
`An important starting point in a discussion of dairy breeding programmes is the paper
`by Rendel and Robertson (1950). They discuss the potential to achieve genetic change
`in breeding programmes, and question the progress that has been made, including
`the issues of separating progress due to selection and progress due to improved
`management. This paper also introduced the concept of selection pathways, arguing that
`genetic improvement could be predicted by adding the improvements from cow and
`bull selection and by distinguishing four different pathways of selection. On the basis of
`more intense selection, the very best bulls would be mated to the very best cows and
`from the offspring the males would be chosen to sire the herds. Cows could be selected
`on the basis of their milk performance, while bulls could be selected on the basis of the
`performance of their dam or progeny if they had some. Less intense selection could be
`practised to choose the young bulls and cows that produce the female replacements for
`the herd. This will lead to the well-known Rendel and Robertson formula, which states that
`the total genetic change is the sum of the selection differentials over the various pathways,
`divided by the sum of the generation intervals in each pathway. The selection differential
`depends on selection intensity, selection accuracy and the amount of genetic variation
`that is available. So the gain per year is
`
`where sA is the additive genetic standard deviation of the trait (or aggregate of traits)
`under selection. Rendel and Robertson predicted a maximum genetic gain of 1% per
`year due to selection. A second paper from Robertson and Rendel (1950) focused on the
`progeny testing scheme. Progeny testing was not obviously a better scheme, as earlier
`pointed out by Dickerson and Hazel (1944), because relatively large coordinated breeding
`programmes were required to allow enough test matings of young bulls. With few test
`matings and larger generation intervals, progeny testing was unlikely to be competitive.
`However, Robertson and Rendel (1950) were inspired by the upcoming AI technology and
`considered larger breeding units with 2000 dairy cows, for example, from a number of
`farms.
`
`2.2 Example: rate of genetic gain in the 4-pathway structure of a
`progeny testing scheme
`This section discusses an example of a national breeding programme that can achieve a
`rate of genetic improvement of 1.25%.
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`s
`intensity*accuracy*
`A
`
`nr_of_paths
`
`=∑1
`
`rrval
`generation_inte
`
`i
`nr_of_paths
`
`∑1
`
`i=
`
`d
`G
`year
`
`=
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`4
`
`Assumptions
`
`Using genetic selection in the breeding of dairy cattle
`
` • A commercial dairy cow population of 1 million animals
` • Fifty breeding bulls needed per year
` • Five hundred young bulls tested
` • Two thousand elite cows mated, selected out of 300 000 herd-recorded cows (30% is
`assumed to be recorded and suitable as a bull dam)
` • Five sires selected for elite matings
` • Seventy per cent of female calves kept as herd replacements
`
`Selection accuracies
`
`Let 20% of the population be used for test matings, that is, 200 000 cows, giving 400 test
`matings per young bull, giving 100 daughters per tested sire completing a first lactation.
`Selection accuracy is 0.87 for males (based on 100 progeny, heritability = 0.25) and 0.50
`for females (based on one’s own performance).
`
`Generation intervals
`
`The average age of the parents when their progeny are born: cows 4.5 years, bulls 6.5
`years.
`Table 1 summarizes the key parameters needed to predict the gain per year.
`The Rendel and Robertson (1950) formula for genetic gain in a 4-pathway breeding
`structure is
`
`
`
`Table 1 Genetic contribution for each of the four selection paths in a dairy cattle breeding programme
`
`Selection path
`
`Selection
`proportion
`
`Selection
`intensity
`(i)
`
`Selection
`accuracy
`(r)
`
`Generation
`interval
`(L)
`
`Sires for Sires
`
`5/500
`
`Dams for Sires
`
`2000/300,000
`
`Sires for Dams
`
`Dams for Dams
`
`50/500
`
`70%
`
`2.65
`
`2.79
`
`1.76
`
`0.47
`
`0.87
`
`0.50
`
`0.87
`
`0.50
`
`Total
`
`6.5
`
`4.5
`
`6.5
`
`4.5
`
`22
`
`Selection
`differential
`(in sA)
`2.31
`
`1.40
`
`1.53
`
`0.24
`
`5.47
`
`% contribution
`to genetic gain
`
`45
`
`26
`
`28
`
`4
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`
`
`intensity accuracy*
`
`*
`
`s
`A
`
`nr_of_paths
`
`=∑
`
`rrval
`generation_inte
`
`1
`i
`nr_of_paths
`
`∑1
`
`i=
`
`d
`G
`year
`
`=
`
`
`
`(i
`
`.r
`
`SS SS
`
`+
`
`=
`
`i
`
`+
`
`i
`
`.r
`DS DS
`+
`L
`L
`SS
`DS
`
`.r
`
`SD SD
`+
`L
`SD
`
`+
`+
`
`i
`.r
`
`DD DD
`L
`DD
`
`) *
`
`s
`A
`
`(
`
`=
`
`
`
`
`
`
`
`) *
`
`sA ==
`
`5 47
`.
`22
`
`s
`BO
`
`=
`
`0 25.
`
`s
`
`A
`
`
`
`
`
`+
`+
`+
`
`
`.2 65 0 87 2 79 0 5 1 76 0 87 0 47 0 5* . . * . . * . . * .
`
`+
`+
`+
`
`.6 5 4 5 6 5 4 5. . .
`
`
`
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`Using genetic selection in the breeding of dairy cattle
`
`5
`
`Hence, we may expect an annual genetic improvement equal to be one-quarter of a
`genetic standard deviation (sA). This is equal to about 1.25% of the mean (given that
`sA = h*sP and assuming a heritability (h2) = 0.25 and sP (phenotypic standard deviation) is
`10% of the mean.
`The paper by Rendel and Robertson (1950) laid the basis for progeny testing schemes
`where they pointed out that the number of cows used for test matings and the number
`of young bulls to be progeny tested could be optimized. They predicted that a rate of
`genetic gain of 1.5% was possible. They also proposed to mate young bulls to 20 of
`their daughters to test the bull for carrying deleterious recessives. With the strong growth
`of AI in the dairy industry and consequently the larger breeding units, the progeny test
`programme became a dominating feature of dairy selection in the next 60 years.
`
`2.3 Breeding programmes using embryo transfer and in vitro
`fertilization
`A second revolutionary insight in dairy breeding programme design was introduced by
`the classic paper of Nicholas and Smith (1983), and this time it was motivated by the
`introduction of female reproductive technologies. On the basis of an earlier idea for beef
`cattle schemes (Land and Hill, 1975), they proposed a dairy breeding scheme that differed
`radically from the present classical progeny testing scheme. During the early 1980s, it
`became possible that females could produce a larger number of offspring via multiple
`ovulation and embryo transfer (MOET), and therefore, fewer females needed to be
`selected for breeding. It even became possible to harvest oocytes from juvenile females
`and create embryos in vitro, after which the embryos were implanted in recipient cows:
`juvenile in vitro fertilization and embryo transfer (JIVET). Initially, this was thought to be
`more suitable for beef cattle breeding where relevant traits can be measured in both sexes
`in an early stage of their life. MOET could be used for cows used in elite matings, but
`selection intensities in dairy cows were already high. For example, if 400 rather than 2000
`cows were used in the DS path, the selection intensity would be 3.0 instead of 2.79 and the
`overall gain would only be 2% higher, which seemed not enough for a costly technology.
`Nicholas and Smith (1983), however, sought to improve upon the classical progeny test
`scheme, and pointed out that MOET schemes produce full sib families and information
`from female sibs can be used to select bulls at a much younger age. As the genetic gain is
`a balance between selection accuracy and generation interval, such a scheme could give
`more gain per year, even though the accuracy of selection is much lower than in progeny
`testing schemes. In breeding schemes using JIVET, selection could be on the average
`breeding value of the parents and generation intervals could be as low as 15 months.
`Initial estimates of additional benefits were high with a predicted 50% increase in genetic
`gain per year. However, these predictions were too optimistic. First, it became clear that
`a selection strategy that relies heavily on family information was strongly affected by
`the reduction of genetic variance as a result of that selection. This is referred to as the
`Bulmer effect (Bulmer, 1971), which shows that in a typical breeding scheme the genetic
`variation would be reduced by about 25% as a result of selection. The variance among
`selected males is drastically reduced and the accuracy of selection based on information
`on paternal half siblings is easily halved when compared to the accuracy that does not take
`selection into account (Van Arendonk and Bijma, 2003). Second, as already pointed out
`by Nicholas and Smith (1983), the MOET and JIVET schemes were likely going to lead to
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`6
`
`Using genetic selection in the breeding of dairy cattle
`
`more inbreeding and additional benefits of these schemes might be lower if inbreeding
`rates had to be constrained.
`Many scientific studies appeared in the 1980s and 1990s on optimizing MOET breeding
`schemes, showing smaller gains, roughly between 10% and 25% (Nicholas, 1997).
`However, the higher inbreeding was still a problem, and when methods to constrain
`inbreeding in breeding programmes became available (Meuwissen, 1997), much of the
`predicted gains disappeared (Van Arendonk and Bijma, 2003). Nevertheless, the in vitro
`embryo production became widespread in the 1990s; in the year 2000, more than 100 000
`embryos were transferred in dairy cattle in Europe (figure cited by Van Arendonk and
`Bijma, 2003). Many breeding companies started to use MOET to more efficiently use
`elite cows to generate young bulls for progeny testing. At that time, MOET breeding
`programmes did not replace progeny testing schemes, probably due to factors such as
`the variation in embryo yield (Nicholas, 1997), cost and the love affair of the dairy industry
`with progeny tested bulls. It was also clear that the AI companies were reluctant to market
`semen from bulls that had no progeny test information. A few companies used schemes
`where initial selection was based on the performance of siblings, although the key to
`widespread use was still large progeny groups (e.g. the MOET scheme operated by Genus
`in the United Kingdom (McGuirk, 1990)).
`It could be said that these centralized nucleus programmes arrived too soon, as one
`aspect that might have made them more popular was the increased need for a more
`balanced breeding programme, where selection based on production traits was gradually
`replaced by a wider breeding objective where functional traits such as health and fertility
`became more important. Selection for functional traits requires a large amount of well-
`recorded phenotypic data. In fact, there has been limited success in many countries in
`developing breeding values, for health traits in particular, because of scarcity of data in
`progeny test herds. Often these data are only available in a subset of herds that have a
`particular interest in data recording. It was not until around 2008 that there was a renewed
`interest due to the potential created by genomic selection that can make use of these
`data-rich herds.
`
`3 The structure of dairy breeding programmes
`
`At this stage, it is useful to consider the structure of dairy breeding programmes, as
`this is affected by reproductive technologies. The structure of breeding programmes is
`usually described as a pyramid, with a breeding nucleus at the top (Fig. 2). In the nucleus,
`selection is made on the basis of investment in the measurement of phenotype, pedigree
`and now genomic testing. The genetic mean of the nucleus improves continuously due
`to this selection process. Animals born in the nucleus, but not used as parents, can be
`transferred to lower tiers to ultimately disseminate the genetic improvement to commercial
`producers. In pig and poultry breeding schemes, the breeding programmes have a distinct
`tier structure, with a closed centralized nucleus, and often one or two multiplier tiers. The
`number of animals in the nucleus is low relative to the number of commercial animals they
`ultimately affect. This is due to the higher fecundity of these species. As a result, nucleus
`breeding programmes for pigs and poultry are run by a few large breeding companies
`that dominate the world market. Furthermore, pig and poultry breeding programmes
`use several breeding lines and ultimately sell a crossbred animal. By contrast, breeding
`programmes for cattle and sheep have a much less distinct structure.
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`Using genetic selection in the breeding of dairy cattle
`
`7
`
`The dairy breeding programme is characterized by having an open dispersed nucleus,
`consisting of elite cows that are mated to elite bulls. These elite matings result in young
`bulls and ultimately tested bulls that are sold to commercial farmers. The nucleus
`dispersed as elite cows could be located at many different dairy herds, often mixed
`with commercial dairy cows. The female offspring of elite matings often are candidates
`for elite matings themselves, but the elite commercial cows are also candidates,
`hence the term ‘open’ nucleus. The bulls are usually owned by the AI companies. At
`the time when embryo technologies became commercialized, a number of breeding
`companies also started to own females, and in essence created a centralized nucleus.
`There were two main reasons for this development: the first is that the logistics of
`running an efficient MOET programme was easily implemented when the donor cows
`were physically together in one place. The other reason was that a centralized nucleus
`allowed centralized testing of females. This was an advantage over selection of the elite
`cows from the herd recording schemes, as there was a perception that the breeding
`value of these elite cows was usually overpredicted. This was evident when the parent
`average estimated breeding value (EBV) of young bulls usually dropped once a progeny
`EBV based on their progeny test became available. In addition, it was easier to use a
`centralized nucleus to record traits that were hard to measure in commercial herds, for
`example, feed efficiency.
`As with female reproductive technologies, the number of females needed in a nucleus
`is much smaller and therefore the technology had the potential to drive dairy breeding
`programmes more towards the centralized and potentially closed nucleus programmes
`such as in poultry and pigs. In spite of these arguments, establishing commercial dairy
`nucleus herds does not seem to be sustainable; some were only temporarily profitable,
`as they were funded out of additional embryo sales to commercial farmers or large
`semen exports outside a targeted commercial population. In essence, a specialized dairy
`breeding nucleus is likely to be too expensive, although we are not aware of any published
`cost–benefit studies. Important factors are likely to include the low reproductive rate of
`females and the cost of increasing this with technology. The main way of disseminating
`the genetic superiority created is, therefore, through the sale of semen. However, the
`margins in the semen market are small, whereas the price elasticity may be high. In other
`
`(cid:31)l(cid:31)(cid:31)e(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)s(cid:31)
`(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)s(cid:31)(cid:31)es(cid:31)(cid:31)(cid:31)el(cid:31)(cid:31)e(cid:31)(cid:31)ull(cid:31)(cid:31)(cid:31)(cid:31)s
`
`Nucleus
`
`(cid:31)(cid:31)(cid:31)s(cid:31)(cid:31)es
`
`(cid:31)(cid:31)(cid:31)(cid:31)e(cid:31) c(cid:31)(cid:31)l(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)uce(cid:31)s(cid:31)(cid:31)
`
`N(cid:31)(cid:31)(cid:31)(cid:31)l(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)s(cid:31)
`(cid:31)(cid:31)e(cid:31)(cid:31)(cid:31)e(cid:31)(cid:31)(cid:31)(cid:31)s(cid:31)(cid:31)es(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)(cid:31)l(cid:31)c(cid:31)(cid:31)s
`
`Figure 2 The two-tier breeding structure.
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`8
`
`Using genetic selection in the breeding of dairy cattle
`
`words, running a complete centralized dairy nucleus breeding programme may simply be
`too expensive.
`Nicholas (1997) pointed out that there could be a role for MOET nucleus herds, even
`though the more developed dairy breeding programmes seemed to hang on to their
`progeny testing paradigm. The higher reproductive rate allows managing a smaller
`size nucleus, which is easier to maintain from a commercial point of view. If there were
`many of such breeding units, he thought that could circumvent the inbreeding problem,
`and he suggested that commercial animals could be produced by crossing animals
`from different breeding schemes. Nicholas (1997) also referred to the finding by Smith
`(1988), who pointed out that such nucleus herds might be very suitable for breeding
`programmes in developing countries as it could be a way to focus on the breeding
`programme investments more efficiently because the nucleus is relatively small and
`centralized.
`Bichard (2002) presented an ‘outsider view’ on dairy breeding programme structures
`and suggested that there might be an overemphasis on the role of progeny testing in dairy
`breeding programmes. He argued that as a result of a push for very high sire proofs, it has
`become more difficult to test young bulls, as farmers are used to highly accurate figures
`for EBV. The large number of progeny that needed to be recorded per sire also made it
`harder to have accurate sire proofs for other traits than milk production as these other
`traits are usually less easy to record. Finally, he observed that progeny testing schemes
`have resulted in large data-recording operations requiring complex statistical analysis to
`provide accurate data and that much of the intellectual power devoted to dairy genetics
`has been used to further develop such genetic evaluation systems, rather than developing
`a broader view on breeding programme alternatives, for example, where more traits are
`measured on fewer animals. Milk production data are used primarily for reasons other
`than genetic improvement, for example, for making good management decisions, and
`recorded in much larger quantities than what would be needed for a genetic improvement
`programme to select superior bulls. Whereas in most countries there has always been a
`lack of good recorded data for fertility and health traits, it is possible that in the current era
`of genomic selection, there is an opportunity to redress this imbalance, with most traits of
`economic value recorded in specialized resource herds.
`Another intriguing contrast between genetic improvement programmes in dairy and
`other livestock species is the apparent lack of crossbreeding in dairy breeding schemes.
`One exception is New Zealand, where over half of all dairy replacements are crossbred. The
`reason for the popularity of crossbreeding in New Zealand is likely to be the importance
`given to reproductive performance in pasture-based systems in order to match pasture
`availability to lactation requirements. The crisis in poor reproductive performance
`in dairy cows (especially on pasture) has generated a lot of interest in crossbreeding,
`especially in the United States, Ireland and New Zealand. Crossbreds have consistently
`higher reproductive performance than their purebred counterparts and more profitable
`performance because of lower replacement rates (Buckley et al., 2014). Crossbreeding on
`dairy farms may become increasingly popular, especially in current and likely future dairy
`markets where the global price of milk is, and will be, low and farmers look at opportunities
`to reduce the cost of production. As the amount of heterosis is greatest in the first cross,
`most of the benefits of crossbreeding are realized in the first cross. However, rotational
`crosses (of two or three breeds) are also becoming popular, with evidence to suggest that
`these are more profitable than straightbred herds in New Zealand (Lopez-Villalobos et al.,
`2000).
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`Using genetic selection in the breeding of dairy cattle
`
`9
`
`4 The exchange and selection of genetic material
`
`4.1 Exchange of genetic material
`Genetic material (bull semen, embryos and even animals) have been traded internationally
`for around 60 years. Breeding companies are also multinational and their focus is to
`provide farmers worldwide with the best bulls from around the world. National genetic
`evaluations have been around for almost as long as genetic material has been exchanged
`nationally and internationally. To help breeders and farmers compare bulls from different
`countries, Interbull was established in 1983 to support international genetic evaluations
`(Banos, 2010).
`
`4.2 Selection on merit versus genetic diversity
`On account of the internationalization of the bull semen market and across-country
`genetic evaluation, it became more common that worldwide only sons of the very best
`bulls were used. Therefore, the genetic basis of dairy breeding populations became
`rapidly smaller, especially in the Holstein Friesian breed, with estimates of the effective
`population size of the Holstein breed being less than 100 (McParland et al., 2007). Another
`way of looking at the diversity problem was that more than 80% of the young bulls tested
`in 2000 were grandsons of only five influential sires. Fortunately, selection theory was
`developed in the 1990s allowing ‘optimal contribution selection’ (Wray and Goddard,
`1994; Meuwissen, 1997), where selection was aimed at not only maximizing genetic merit,
`but also allowing for a sufficient level of genetic diversity. This strategy allows controlling
`diversity by constraining the average co-ancestry of selected parents, which is a prediction
`of half the rate of inbreeding. Meuwissen showed that about 60% more genetic gain
`could be achieved at the same rate of inbreeding with optimal contribution selection,
`compared to truncation selection on merit and imposing ‘ad hoc’ selection rules to control
`inbreeding. Although the theory for optimal selection exists, selection is rarely optimized
`at the population level, as most selection decisions are made by individual breeding
`organizations that compete with each other in the dairy bull semen market. However,
`for individual breeding programmes, there is an incentive to consider diversity, as it is
`synonymous with risk.
`
`5 Genomic selection
`
`In the 1960s, it was noted that selection could possibly be based on ‘genetic markers’ in
`the form of blood groups (Neimann-Sorensen and Robertson, 1961). When the number of
`DNA markers increased rapidly due to the development of molecular biology, the prospect
`of marker-assisted selection seemed to become a realistic addition to the breeding
`programme, and early papers showed how genetic markers could be incorporated in
`genetic evaluation (Fernando and Grossman, 1989) and into breeding programmes (Lande
`and Thompson, 1990). Selection would be based on information from genetic markers
`that are in linkage disequilibrium with genomic regions with a large effect on quantitative
`trait, the so-called quantitative trait loci (QTL) (Meuwissen and Goddard, 1996). Although
`initial estimates of QTL effects seemed promising (e.g. Georges et al., 1995), it became
`
`© Burleigh Dodds Science Publishing Limited, 2016. All rights reserved.
`
`Exhibit 1016
`Select Sires, et al. v. ABS Global
`
`

`

`10
`
`Using genetic selection in the breeding of dairy cattle
`
`clear that QTL effects were often overestimated, and the number of QTL with a large effect
`that could be useful in marker-assisted selection appeared to be disappointingly low. The
`largest QTL effect found in dairy cattle is the DGAT1 mutation (Grisart et al., 2002) with a
`major effect on milk fat and other milk constituents. For example, the DGAT1 allele that
`encodes lysine at position 232 is associated more with fat, implying that selection on this
`gene can alter fat composition of cows’ milk (Schennink et al., 2007). However, DGAT1
`is usually not a direct selection target because it is more effective to select for breeding
`values for traits predicted from multiple genetic markers simultaneously, that is, selecting
`for breeding values for fat and protein yield as target traits.
`Since the turn of the century, many studies based on increasingly dense marker panels
`have revealed that most of the observed genetic variation on most quantitative traits is due
`to a large number of genes, each with a small effect. This probably explains why the impact
`of identified QTL in marker-assisted breeding programmes has been small to negligible.
`Instead, another approach to use marker information has started to revolutionize breeding
`programmes. Meuwissen et al. (2001) proposed to use all marker information across the
`whole genome in a single analysis to predict breeding values. They showed in a simulation
`study that the reliability of a breeding value could be as high as 64% when using a large
`training population and dense genetic markers. It was not until about five years later that
`the first single nucleotide polymorphism (SNP)-chip was released by Affymetrix as a tool
`to genotype individuals for about 10 000 genetic markers. Soon followed the first Illumina
`SNP-chip that contained ~50k markers, and by 2009, Illumina released the high-density
`cattle chip with ~800k markers. A growing area is the use of low-density SNP panels
`(10k), which are especially popular for screening large numbers of young bulls and dairy
`cows (Fig. 3A). It is noteworthy that genotyping of cows is becoming particularly popular
`(Fig. 3A), as farmers begin to use genomic breeding values for management decisions,
`such as which animals to select as herd replacements. Generally, the low-density genotypes
`are imputed to 50k to calculate genomic breeding values. However, customized chips that
`do not require imputation are also being used on a large scale. For example, genotyping
`is mostly complete for around a million cattle in Ireland, with completion anticipated by
`the end of this year through the use of a customized SNP panel of the 40k that is currently
`in vogue in dairy genomic selection, and this technique obviates the need for imputation
`(Berry, 2016 personal communication).
`Many dairy bulls have been genotyped, as illustrated in Fig. 3A, and their EBVs served
`as the phenotypic information that is required for genomic predictions. By 2008, it had
`become clear that the breeding value of young bulls could be predicted with a reliability
`of at least 50% for most milk production characteristics. This was a big improvement
`over the reliability that can be achieved from an EBV based on the mean of the parents
`(about 25%, or less due to the effect of selection), although it was still short of what is
`typically achieved in a first-proof of around 50 daughters (80%). But because the genomic
`prediction could be made available at an early age, and selection of bulls at an early age
`could now be based on an EBV with reasonable predic