J. Dairy Sci. 101:3686–3701
`https://doi.org/10.3168/jds.2017-13335
`© American Dairy Science Association®, 2018.
`Symposium review: Possibilities in an age of genomics:
`The future of selection indices1
`J. B. Cole2 and P. M. VanRaden
`Animal Genomics and Improvement Laboratory, Agricultural Research Service, USDA, Beltsville, MD 20705-2350
`
`ABSTRACT
`
`Selective breeding has been practiced since domes-
`tication, but early breeders commonly selected on
`appearance (e.g., coat color) rather than performance
`traits (e.g., milk yield). A breeding index converts
`information about several traits into a single number
`used for selection and to predict an animal’s own per-
`formance. Calculation of selection indices is straight-
`forward when phenotype and pedigree data are avail-
`able. Prediction of economic values 3 to 10 yr in the
`future, when the offspring of matings planned using
`the index will be lactating, is more challenging. The
`first USDA selection index included only milk and fat
`yield, whereas the latest version of the lifetime net
`merit index includes 13 traits and composites (weighted
`averages of other additional traits). Selection indices
`are revised to reflect improved knowledge of biology,
`new sources of data, and changing economic conditions.
`Single-trait selection often suffers from antagonistic
`correlations with traits not in the selection objective.
`Multiple-trait selection avoids those problems at the
`cost of less-than-maximal progress for individual traits.
`How many and which traits to include is not simple to
`determine because traits are not independent. Many
`countries use indices that reflect the needs of differ-
`ent producers in different environments. Although the
`emphasis placed on trait groups differs, most indices
`include yield, fertility, health, and type traits. Addition
`of milk composition, feed intake, and other traits is
`possible, but they are more costly to collect and many
`are not yet directly rewarded in the marketplace, such
`as with incentives from milk processing plants. As the
`number of traits grows, custom selection indices can
`more closely match genotypes to the environments in
`which they will perform. Traditional selection required
`recording lots of cows across many farms, but genomic
`selection favors collecting more detailed information
`
`Received June 14, 2017.
`Accepted August 22, 2017.
`1 Presented as part of the ADSA Multidisciplinary and International
`Leadership Keynote (MILK) Symposium at the ADSA Annual
`Meeting, Pittsburgh, Pennsylvania, June 2017.
`2 Corresponding author: john.cole@ars.usda.gov
`
`from cooperating farms. A similar strategy may be
`useful in less developed countries. Recording important
`new traits on a fraction of cows can quickly benefit the
`whole population through genomics.
`Key words: breeding program, genetic improvement,
`selection index
`
`INTRODUCTION
`
`Breeding indices are important tools in modern dairy
`cattle breeding. They provide a way to combine infor-
`mation about many traits into a single number that can
`be used to rank animals and make breeding decisions.
`The need for such a tool was recognized very early in
`the history of modern animal breeding, when Hazel and
`Lush (1942) applied the method of Smith (1934) to
`the improvement of economically important traits of
`livestock. The ideal breeding objective for dairy cattle
`remains a popular topic and has been reviewed periodi-
`cally (e.g., Hazel et al., 1994; Philipsson et al., 1994;
`VanRaden, 2004; Miglior et al., 2005; Shook, 2006), but
`there is no single selection objective that is best for all
`populations or all herds within a population.
`Historically, selection indices in the United States
`were developed by the USDA and purebred dairy cattle
`associations, frequently with input from scientists at
`land-grant universities, using data available through
`the national milk recording system and breed type
`appraisal programs. Proposed indices were typically
`reviewed by groups of experts and information about
`the derivation of the indices was published in techni-
`cal and trade publications, ensuring confidence in the
`values because of that review process. Recently, genetic
`evaluations for novel traits and new selection indices
`have been computed and distributed by companies such
`as CRV (Arnhem, the Netherlands), Genex (Shawano,
`WI), and Zoetis (Parsippany-Troy Hills, NJ). This pro-
`vides farmers with new tools and may drive demand for
`new phenotypes, but transparent review processes may
`be lacking. The purpose of this paper is to present a
`brief overview of how selection indices are constructed,
`describe traits included in current indices, review desir-
`able properties of new traits, discuss traits that may
`be included in selection indices in the future, and dem-
`
`3686
`
`
`
`Exhibit 1022
`Select Sires, et al. v. ABS Global
`
`

`

`ADSA MILK SYMPOSIUM: THE DAIRY COW IN 50 YEARS
`
`3687
`
`Figure 1. Changes in fat yield for US Holsteins, 1957 to 2015. The black (red) area represents average production in 1957, the light gray
`(blue) area shows changes due to improved feeding and management, and the dark gray (green) area shows gains from increased genetic merit.
`Color version available online.
`
`onstrate that selection indices are robust to incorrect
`assumptions about model parameters.
`
`SELECTION INDICES
`
`Improving Animal Performance
`
`Animal performance is a function of both genetic and
`environmental factors and interactions among the two.
`Predictions of genetic merit are based on a quantita-
`tive model that assumes that traits are controlled by
`many genes, each of which has a small effect on the
`phenotype (Falconer and MacKay, 1996). This model
`has been found to accurately describe many traits of
`economic importance in dairy cattle (Cole et al., 2009).
`Environmental influences include all sources of pheno-
`typic variation that cannot be attributed to genetics,
`such as nutrition, climate, disease exposure, error in
`measurement, and other unknown factors. These fac-
`tors vary from farm to farm and between individual
`animals on the same farm and may change over time
`(e.g., Windig et al., 2005).
`Figure 1 shows the change in fat yield for US Hol-
`steins between 1957 and 2015. Production in 1957 is
`used as a baseline, and gains over time were found to be
`evenly divided between increased genetic potential and
`improvements in feeding and management. Gains in
`genetics and management each represent 28% of 2015
`production, whereas the 1957 base represents 44% of
`current yield. The proportion of gains from improved
`
`genetics versus improved environment differs from trait
`to trait and is a function of the heritability of a trait.
`Fat yield has a heritability of 20% (VanRaden, 2017),
`whereas daughter pregnancy rate has a heritability of
`only 4% (VanRaden et al., 2004). When the proportion
`of variance in a trait due to genetics is low, it is often
`easier to make gains by improving the environment in
`which the cow is performing, and gains from genetic
`improvement may not be visible to producers for a long
`time.
`
`Construction of Selection Indices
`
`The following discussion focuses on the simplest for-
`mulation of a selection index; greater detail, including
`derivations, may be found in the literature (e.g., Lin,
`1978; Cameron, 1997). When using a selection index,
`the goal is to improve one or more traits, referred to as
`the selection objective, by ranking and choosing mates
`using a combination of one or more traits, known as
`the selection criterion. In modern breeding programs,
`the selection objective is typically a measure of lifetime
`profitability, whereas the selection criterion usually
`comprises traits that are included in national milk re-
`cording programs. In the mathematical terms of Hazel
`and Lush (1942), an index including m terms in the
`selection criterion for an animal takes the form
`
`
`
`I = b1X1 + b2X2 + … + bmXm,
`
`Journal of Dairy Science Vol. 101 No. 4, 2018
`
`Exhibit 1022
`Select Sires, et al. v. ABS Global
`
`

`

`3688
`
`COLE AND VANRADEN
`
`where I is the selection criterion, bi is the emphasis
`placed on the ith trait, and Xi is the animal’s pheno-
`type for the ith trait in the index. Index weights are cal-
`culated as a function of (co)variances among the traits
`in the objective and the criterion and the economic
`weights of the individual traits:
`
`
`
`b = P−1Ga,
`
`where b is a vector of index weights, P is the pheno-
`typic (co)variance matrix for the traits in the selection
`criterion, G is a matrix of genetic (co)variances among
`the traits in the criterion and the objective, and a is a
`vector of economic weights associated with the traits in
`the criterion. If all of the parameters used to compute
`the index are correct, then it is the most efficient way
`of improving all of the traits in the selection objective.
`However, in modern breeding programs, mixed model
`equations include P and G to first obtain multitrait
`evaluations (û), and those are combined directly by
`their economic values as a`û.
`When the traits in the selection criterion and selec-
`tion objective differ, as is often the case, an additional
`calculation is necessary to determine the correlated
`response to selection of the traits in the objective in re-
`sponse to selection on the traits in the criterion. This is
`a straightforward extension of the well-known breeder’s
`equation (Cameron, 1997)
`
`
`
`∆ =gj
`
`b'G
`j
`b'Pb
`
`
`,
`
`where ∆gj is the correlated response of trait j in the
`selection objective in response to selection on the selec-
`tion criterion, and Gj is the correlation between trait j
`and the traits in the selection criterion. This equation
`shows that the correlated response is a function of the
`genetic correlations among the traits in the objective
`and the criterion and the index weights.
`The literature on selection index methodology is
`quite extensive, and many special cases can be accom-
`modated. For example, one trait can be held at a con-
`stant level while others are changed (Kempthorne and
`Nordskog, 1959), economic value can have nonlinear
`relationships with the traits in the index (Goddard,
`1983), selection can proceed in stages where objectives
`change over time (Cunningham, 1975), and quota sys-
`tems can drive the economic value of yield traits (Gib-
`son, 1989). Selection index methodology also has been
`used to determine rates of genetic and economic gain
`under genomic selection programs in a deterministic
`fashion (Dekkers, 2007; König et al., 2009). Readers are
`
`directed to more comprehensive works on selection in-
`dex methodology for additional details (e.g., Van Vleck,
`1993; Weller, 1994; Cameron, 1997).
`
`Contribution of Genomic Information
`
`Genomic selection allows breeders to make decisions
`more quickly by using dense DNA marker informa-
`tion to compute high-reliability predictions of genetic
`merit early in an animal’s life (Nejati-Javaremi et al.,
`1997; Meuwissen et al., 2001). From the perspective of
`the breeding objective, the principal effect of genomic
`selection is on the reliabilities of the breeding values
`used in the index (VanRaden et al., 2009), but the
`technology provides several other advantages, including
`lower costs of proving bulls (Schaeffer, 2006), greater
`rates of genetic gain from shorter generation intervals
`(García-Ruiz et al., 2016), detection of previously
`unknown genetic disorders (VanRaden et al., 2011),
`and identification of genes that influence economically
`important traits (Cole et al., 2011). A trait with a low
`heritability, such as daughter pregnancy rate (h2 =
`0.04), requires more daughter phenotypes to produce a
`breeding value with the same reliability as a trait with
`higher heritability, such as fat yield (h2 = 0.30), and
`genotypes provide more information for low-heritability
`traits. Pedigree information alone is equivalent to ap-
`proximately 7 daughter records, whereas a genotype
`is worth 34 daughter records for fat or 131 daughter
`records for daughter pregnancy rate. Genomics allows
`us to publish useable evaluations much sooner than in
`the past and make more profitable management deci-
`sions on the farm (e.g., Pryce and Hayes, 2012; Van
`Eenennaam et al., 2014).
`
`Selection for Many Traits
`
`The number of traits included in a typical selection
`criterion has grown over time, from 1 or 2 yield traits
`to many nonyield traits, including fertility, health, and
`fitness traits. This allows farmers to make use of more
`information than in the past and takes advantage of
`correlations among traits (important traits rarely have
`correlations of 0 with other important traits). Many
`traits may have direct economic value; for example, milk
`plants often pay premiums for low SCS in addition to
`payments for high protein and fat components. Traits
`can also have indirect value; for example, SCS can pre-
`dict mastitis losses if mastitis is not recorded directly.
`Substantial losses can occur when indirect values are
`ignored—for example, the well-documented negative
`correlation of fertility with milk yield (Figure 2; Lucy,
`2001). Balanced selection improves traits according to
`
`Journal of Dairy Science Vol. 101 No. 4, 2018
`
`Exhibit 1022
`Select Sires, et al. v. ABS Global
`
`

`

`ADSA MILK SYMPOSIUM: THE DAIRY COW IN 50 YEARS
`
`3689
`
`Figure 2. Changes in daughter pregnancy rate (DPR) for US Holsteins, 1957 to 2015. The black (red) area represents average production
`in 1957, the light gray (blue) area shows changes due to improved feeding and management, and the dark gray (green) area shows gains from
`increased genetic merit. Color version available online.
`
`their economic values, and selection indices should be
`periodically updated to include new traits and reflect
`changing economic conditions as well as changing ge-
`netic parameters between and among traits. However,
`as traits are added to an index it becomes increasingly
`difficult to predict a priori whether the new index will
`have greater or reduced response compared with the
`index with fewer traits (Sivanadian and Smith, 1997).
`
`Derivation of Economic Values
`
`The vector of economic values (a) included in the
`calculation of index weights is used to assign values
`to traits based on their importance to the selection
`objective. Two general approaches may be used to
`derive those weights. The first, which might be called
`the empirical approach, uses data from scientific stud-
`ies and field reports to quantify incomes and expenses
`associated with the traits in the selection objective
`and criterion. The goal of this approach is to allow
`the best available economic information to drive the
`formulation of the index, and it is used in the calcula-
`tion of the USDA’s Lifetime Net Merit Index (NM$)
`and some breed-specific indices, such as the American
`Jersey Cattle Association’s (2017) Jersey Performance
`Index. The second, which might be called the subjec-
`tive approach, has been used to construct indices such
`as Holstein Association USA Inc.’s (2017) Total Perfor-
`mance Index (TPI), assigns values to traits based on
`
`the cow that breeders would like to see in the future.
`Those targets for breed improvement are developed
`by groups of breeders and experts and are driven by
`both quantitative and qualitative factors. Quantitative
`factors include incomes and expenses associated with
`costs of raising animals and the value of products sold,
`whereas qualitative factors include such things as the
`desirable conformation for cows of a particular breed.
`Direct economic values for some traits, most notably
`conformation traits, often are difficult to calculate but
`may be very important to farmers who breed and own
`registered cattle. Both approaches to placing values on
`individual traits produce broadly similar results (2010
`NM$ and TPI had a correlation of 0.88), but the dif-
`ferences between the indices reflect important economic
`factors affecting the users. Customized indices at the
`farm level were first delivered by McGilliard and Clay
`(1983) and proposed in Australia (Bowman et al., 1996)
`but were not widely used in the United States. As herds
`continue to grow larger, managers may have an incen-
`tive to customize their own indices (Dickrell, 2017).
`
`Subindices
`
`One way to make indices easier to understand is to
`construct them from a series of subindices. For ex-
`ample, NM$ includes 3 type composites that combine
`information from several traits, and the calving ability
`dollars (CA$) subindex combines sire and daughter
`
`Journal of Dairy Science Vol. 101 No. 4, 2018
`
`Exhibit 1022
`Select Sires, et al. v. ABS Global
`
`

`

`3690
`
`COLE AND VANRADEN
`
`Figure 3. An example of lifetime net merit (NM$) constructed from production (PROD$), longevity (LONG$), fertility (FERT$), conforma-
`tion (TYPE$), and calving ability (CA$) subindices. Panel (a) shows April 2017 NM$, whereas panel (b) shows a hypothetical revision to NM$
`that includes a new health subindex (HEALTH$) and additional traits in some subindices. Color version available online.
`
`calving ease and sire and daughter stillbirth into a
`single quantity. All the breeder will see when the index
`is revised are the changes in emphasis on each of the
`subindices rather than changes to each of the individual
`traits (Figure 3). Farmers need to understand only the
`function of each subindex instead of dozens of traits.
`The Ideal Commercial Cow Index (ICC$; Genex,
`2006) is constructed in this way: ICC$ is the sum of the
`production efficiency (PREF$), health (HLTH$), fertil-
`ity and fitness (FYFT$), milking ability (MABL$), and
`calving ability (CABL$) subindices. The advantages of
`this approach are small when indices contain only a few
`traits but increase rapidly as the number of traits in-
`cluded grows. Composite traits have a similar purpose
`but often are unitless instead of having monetary value.
`The Irish EBI Index (ICBF, 2017) comprises 7 sub-
`indices: milk production, fertility, calving performance,
`beef carcass, cow maintenance, cow management, and
`health. The calving performance subindex receives 10%
`of the total emphasis and includes PTA for direct and
`maternal dystocia, gestation length, and stillbirth. The
`health subindex, with 4% of the emphasis, includes
`direct (clinical mastitis) and indirect (SCC) measures
`of udder health as well as lameness. These examples
`demonstrate the use of direct (e.g., dystocia, clinical
`mastitis) traits in combination with indirect (e.g., ges-
`tation length, SCC) indirect (indicator) traits.
`
`PHENOTYPES IN SELECTION INDICES
`
`What Traits Are Included in Current
`Selection Indices?
`
`The traits included in USDA selection indices over
`time, and weights placed on each, are shown in Table
`1. The first USDA index, Predicted Difference Dollars
`(PD$), included only milk and fat yield in the selec-
`tion criterion, whereas the 2017 revision of NM$ (Van-
`Raden, 2017) includes information about 33 different
`traits when subindices are considered. Selection indices
`differ within and across countries because economic
`conditions, traits recorded, and breeds used are not
`the same everywhere. Figure 4 shows traits included
`in total merit indices from 15 different countries. Trait
`definitions may differ slightly from one country to an-
`other, but common trait groups include yield (e.g., milk
`volume, fat and protein yield), longevity (e.g., produc-
`tive life), fertility (e.g., nonreturn rate, days open), ud-
`der health (e.g., SCS, clinical mastitis), calving traits
`(e.g., dystocia, stillbirth), milking traits (e.g., milking
`speed), and conformation (e.g., udder conformation,
`feet and leg score). Although some broad similarities
`exist among indices—most include direct emphasis
`on protein yield—no two are the same, even within
`a country. For example, NM$ includes more emphasis
`
`Journal of Dairy Science Vol. 101 No. 4, 2018
`
`Exhibit 1022
`Select Sires, et al. v. ABS Global
`
`

`

`ADSA MILK SYMPOSIUM: THE DAIRY COW IN 50 YEARS
`
`3691
`
`Table 1. Traits included in USDA selection indices and the relative emphasis placed on each, 1971 to 2017
`
`Relative emphasis on traits2 (%)
`
`Trait1
`
`PD$,
`1971
`
`MFP$,
`1976
`
`CY$,
`1984
`
`NM$,
`1994
`
`NM$,
`2000
`
`NM$,
`2003
`
`NM$,
`2006
`
`NM$,
`2010
`
`NM$,
`2014
`
`NM$,
`2017
`
`52
`48
`
`27
`46
`27
`
`−2
`45
`53
`
`6
`25
`43
`20
`−6
`
`5
`21
`36
`14
`−9
`7
`4
`−4
`
`Milk
`Fat
`Protein
`PL
`SCS
`UC
`FLC
`BWC
`DPR
`SCE
`DCE
`5
`CA$
`1
`HCR
`2
`CCR
`7
`LIV
`1PL = productive life; UC = udder composite; FLC = foot and leg composite; BWC = BW composite; DPR = daughter pregnancy rate; SCE
`= sire calving ease; DCE = daughter calving ease; CA$ = calving ability dollars; HCR = heifer conception rate; CCR = cow conception rate;
`LIV = cow livability.
`2PD$ = predicted difference dollars; MFP$ = milk, fat, and protein dollars; CY$ = cheese yield dollars; NM$ = net merit dollars.
`
`0
`22
`33
`11
`−9
`7
`4
`−3
`7
`−2
`−2
`
`0
`23
`23
`17
`−9
`6
`3
`−4
`9
`
`6
`
`0
`19
`16
`22
`−10
`7
`4
`−6
`11
`
`5
`
`−1
`24
`18
`13
`−7
`7
`3
`−6
`7
`
`−1
`22
`20
`19
`−7
`8
`3
`−5
`7
`
`5
`1
`2
`
`Figure 4. Traits included in 21 total merit indices of the United States and 16 other countries. Data were collected from genetic evaluation
`centers and purebred cattle associations for Australia (ADHIS, 2014); Canada (CDN, 2017); Denmark, Finland, and Sweden (NAV, 2017);
`France (Genes Diffusion, 2014); Germany (VIT, 2017); Great Britain (AHDB Dairy, 2017); Ireland (ICBF, 2017); Israel (SION, 2015); Italy
`(ANAFI, 2016); Japan (Holstein Cattle Association of Japan, 2010); New Zealand (DairyNZ, 2017); Spain (CONAFE, 2016); Switzerland
`(Holstein Association of Switzerland, 2013); the Netherlands (CRV, 2017); and the United States (Holstein Association USA Inc., 2017;
`VanRaden, 2017). Index abbreviations are HWI = health weighted index; TWI = type weighted index; BPI = balanced performance index; LPI
`= lifetime profit index; NTM = Nordic total merit; GDM = genes diffusion merit; RZG = Relativ Zuchtwert Gesamt (total merit index); £PLI
`= profitable lifetime index; EBI = economic breeding index; PD11 = Israeli 2011 breeding index; PFT = production, functionality, and type
`index; NTP = Nippon total profit; BW = breeding worth; ICO = Índice de Mérito Genético Total (total genetic merit index); ISEL = Index de
`Sélection Totale (total selection index); NVI = Netherlands cattle improvement index; TPI = total performance index; GM$ = grazing merit;
`FM$ = fluid merit; CM$ = cheese merit; NM$ = net merit. Color version available online.
`
`Journal of Dairy Science Vol. 101 No. 4, 2018
`
`Exhibit 1022
`Select Sires, et al. v. ABS Global
`
`

`

`3692
`
`COLE AND VANRADEN
`
`on longevity and less on conformation than TPI. Some
`countries were more far-sighted than others and added
`health traits to their selection programs decades ago,
`providing them with a head start over other countries
`(e.g., Philipsson and Lindhe, 2003).
`
`What Traits Should Be Included in Future
`Selection Indices?
`
`New traits are added to selection indices for many
`reasons. Production economics change over time, such
`as the introduction of incentive payments for milk
`quality or the elimination of quota systems, with a
`corresponding need for adjustments to selection objec-
`tives. Our understanding of biology improves over time,
`which can lead to the adoption of new traits (e.g., Shook
`and Schutz, 1994). Technology also evolves, permitting
`the collection of information that was previously im-
`possible or prohibitively expensive to record (e.g., De
`Marchi et al., 2014). The widespread adoption of ge-
`nomic selection is complementary to those technologies
`because new traits can be predicted on all genotyped
`animals without the need to collect progeny records,
`and phenotyping costs are shared among millions of
`animals. The following discussion will briefly consider
`some traits that are of growing interest to dairy farm-
`ers. Recent comprehensive discussions of new traits and
`phenotyping strategies are provided by Boichard and
`Brochard (2012), Egger-Danner et al. (2015), Gengler
`et al. (2016), and (Pryce et al., 2016).
`Health and Fitness. Some countries have included
`health traits in their selection indices for decades (Mi-
`glior et al., 2005; Heringstad and Østerås, 2013), but
`many have not, and there is growing interest in the use
`of genetic selection to improve cow health and welfare
`(Pryce et al., 2016). There also is increasing pressure
`from consumers and regulatory agencies to reduce the
`use of drugs and increase the perceived welfare of food
`animals (Jensen, 2016; Saitone and Sexton, 2017). Sick
`cows are less profitable than healthy cows due to lower
`production, decreased fertility, and increased labor and
`veterinary costs. They are also more likely to die on the
`farm, which results in lost revenue from beef sales and
`incurs disposal costs that can be evaluated separately
`(Wright and VanRaden et al., 2016).
`Several studies have shown that producer-recorded
`health events from on-farm computer systems are a rich
`source of data for genetic improvement (Zwald et al.,
`2004; Parker Gaddis et al., 2012; Wenz and Giebel,
`2012), and genomic information produces evaluations
`with sufficient reliability for routine use (Parker Gad-
`dis et al., 2014). Direct measures of cow health have
`recently been added to some dairy improvement pro-
`
`grams (Fuerst et al., 2011; Beavers and VanDoormal,
`2016; Vukasinovic et al., 2017), and others are planning
`to introduce evaluations soon (Parker Gaddis et al.,
`2017b). Breeding values for direct measures of immune
`function also have been proposed to improve overall
`animal health (Thompson-Crispi et al., 2012). Although
`heritabilities of these traits are generally low, the aggre-
`gate value of the traits may be large if treatment costs
`related to health and disease are high. However, the
`losses from reduced yield, fertility, and longevity are
`already directly accounted for by those traits.
`Feed Intake. Feed costs represent the largest single
`cost of milk production (e.g., Laughton, 2016), so in-
`creases in the efficiency with which the dairy converts
`feed into milk and milk solids represents a large poten-
`tial economic gain to the producer. At the same level of
`production, a small cow is more efficient than a large
`cow, and NM$ and New Zealand’s Breeding Worth In-
`dex (Livestock Improvement International, 2017) both
`place negative weight on body size as a proxy for ef-
`ficiency. Residual feed intake (RFI), the difference in
`actual intake and intake predicted based on body size
`and level of production (e.g., Koch et al., 1963; Crews,
`2005; Connor, 2015), has been proposed as a selection
`criterion in both dairy and beef cattle. However, RFI
`requires the collection of actual feed intake and BW
`data, which requires that farms install special equip-
`ment, making it an expensive phenotype to collect.
`Genomic selection has reduced the cost of develop-
`ing genetic evaluations for RFI because phenotypes
`can be collected for a relatively small group of animals
`and phenotypes predicted for all animals (Calus et al.,
`2013). Recently, genetic evaluations were introduced in
`the Netherlands for feed intake and in Australia for
`feed saved, which combines genomic predictions of RFI
`with BW (Pryce et al., 2015). Preliminary genomic
`evaluations of feed saved also are available for US Hol-
`steins, although reliabilities were lower than expected
`(VanRaden et al., 2017). Even modest rates of genetic
`improvement for a trait with a large economic value
`result in substantial cumulative gains over time. There
`may be additional benefits associated with RFI because
`efficient cows also emit fewer greenhouse gases, notably
`methane (Hegarty et al., 2007). However, long-term
`strategies with a focus on data consolidation across
`countries, such as the Efficient Dairy Genome Project
`(De Pauw, 2017) and the global Dry Matter Initiative
`project (de Haas et al., 2014), are needed to ensure
`the continued production of new RFI phenotypes to
`support continuing genetic evaluations.
`Fertility. The downward genetic trend in fertility ex-
`perienced by the Holstein breed has stopped, and days
`open are now decreasing (fertility is improving) for US
`
`Journal of Dairy Science Vol. 101 No. 4, 2018
`
`Exhibit 1022
`Select Sires, et al. v. ABS Global
`
`

`

`ADSA MILK SYMPOSIUM: THE DAIRY COW IN 50 YEARS
`
`3693
`
`cattle (García-Ruiz et al., 2016). Genomic evaluation
`has been used to increase the accuracy of genetic evalu-
`ations of fertility as well as identify genomic regions
`associated with variation in days open and pregnancy
`rate (Ortega et al., 2016; Parker Gaddis et al., 2016).
`Fertility continues to be of great economic importance
`to dairy farmers, and there is a need for more precise
`measures of fertility as well as phenotypes that relate
`to new reproductive practices on dairies. Hutchison et
`al. (2017) recently showed that a decrease in age at first
`calving for US Brown Swiss, Holstein, and Jersey cattle
`would result in greater lifetime production of actual
`milk, fat, and protein, although stillbirth rates need
`to be carefully monitored. Progesterone levels may be
`used to define new fertility traits that more accurately
`reflect the physiological status of the cow (Sorg et al.,
`2017). Several recent studies have documented genetic
`variability in response to superovulation and embryo
`transfer protocols (Jaton et al., 2016; Parker Gaddis et
`al., 2017), which are becoming more common, particu-
`larly for matings among elite animals. Greater diversity
`in measures of reproductive performance will help farm-
`ers ensure that they can get cows pregnant when they
`would like, using a variety of available technologies.
`Genetic Diversity. Although not a trait per se,
`genetic diversity remains of concern to animal breeders
`(Howard et al., 2017). Proper use of mating programs
`(e.g., Pryce et al., 2012; Clark et al., 2013; Sun et al.,
`2013) can prevent many immediate problems result-
`ing from excessive inbreeding, and other strategies can
`be used in combination with mating strategies. The
`United States is the only country that adjusts its ge-
`netic evaluations to account for the effects of inbreed-
`ing depression on PTA (VanRaden, 2005), but other
`countries may implement similar adjustments if rates
`of inbreeding continue to increase rapidly. Maps of re-
`combination sites in the bovine genome have recently
`become available (Weng et al., 2014; Ma et al., 2015),
`and simulation suggests that standing genetic variation
`can be manipulated by selecting for increased recombi-
`nation rates (Gonen et al., 2017). However, structural
`changes in the dairy industry leading to an embryo-
`based system of nucleus and multiplier herds may occur
`before selection on recombination rates is adopted.
`Milk Composition. Milk is an important source of
`nutrients in human diets (Pereira, 2014), and it may be
`possible to produce milk with fatty acid profiles and
`protein composition that improve health. However,
`detailed analyses of milk composition are expensive
`and time consuming, limiting the potential number of
`observations available for evaluation. As in the case
`of RFI, genomic selection appears to offer a partial
`solution to the phenotype problem, and recent research
`
`suggests that mid-infrared (MIR) spectral analysis of
`milk samples can provide low-cost, large-scale predic-
`tions of these phenotypes (e.g., Soyeurt et al., 2006; De
`Marchi et al., 2009). Manufacturing properties, such
`as coagulation time and curd firmness in cheeses, also
`can be assessed using MIR (De Marchi et al., 2014),
`enabling selection for those traits. There is growing in-
`terest in milk that is positively associated with human
`health (e.g., Pereira, 2014), such as having a desirable
`fatty acid profile, and consumers are willing to pay
`higher prices for organic or “natural” foods (McFadden
`and Huffman, 2017). However, unlike health and RFI,
`there will be clear economic incentives for dairy farmers
`to select for altered milk composition or manufacturing
`properties only when milk processors, not just consum-
`ers, pay premiums for those traits.
`Omics Data. In addition to the direct and indirect
`measurements of animal performance discussed above,
`there is a growing body of data collected from stud-
`ies of functional biology (e.g., Andersson et al., 2015;
`Suravajhala et al., 2016). Information about what
`genes are expressed in specific tissues at various stages
`of development, detailed knowledge of protein structure
`(including posttranscriptional changes), methylation
`status, and interactions with regulatory elements may
`support better predictions of phenotypic performance.
`Improved reference genomes with better functional an-
`notation are needed to make