Overview of commonly recorded traits

At present, 30 to 40 traits are commonly recorded in dairy cattle and used in selection schemes implemented in many developed countries (Banos, Reference Banos2010). The first category is related to milk recording. Milk production is recorded throughout lactation and milk is sampled to be analyzed in laboratories using mid-infrared (MIR) spectrometry to provide milk composition (mainly fat, protein, somatic cells). Other optional measures are proposed; for example, for lactose or urea. These measures are performed on a regular basis generally through milk recording organizations according to several possible designs combining the following options: recording by a technician or by the farmer; recording of one or all milkings on each test day; time interval from 3 to 6 weeks between test days.

In many countries, longevity and survival are assessed through the animal traceability process: each animal, alive or dead, is identified at birth and traced on its passport throughout its whole life up to its death (natural death, slaughter), as well as its movements across farms and its possible exportation. Conditions of calving, sex, twinning are recorded at calving during the identification declaration.

Fertility-related phenotypes (conception rate, intervals, number of services, gestation length, etc.) are obtained mainly through insemination and calving data. Complementary natural mating data require a declaration by the farmer. A special information system is required for embryo transfer.

Type data are obtained through a specialized recording system. Specially trained technicians visit farms and score the animals for a number of conformation traits and survey the farmer for some additional traits such as milking speed or temperament. Additional traits are recorded by milk-recording agencies, such as clinical mastitis.

In France, all the above-mentioned traits are routinely recorded on a large scale. For example, 100% dairy cows are identified, 100% AIs are recorded, ∼70% dairy cows are milk recorded and ∼30% are classified for type (http://www.inst-elevage.asso.fr/spip.php?page=article_espace&id_espace=935&id_article=19072).

Many new traits could be recorded with marginal extra effort

Making better use of milk-recording data is the first opportunity and a priority. Developments based on improving the use of milk test-day records have made it possible to better exploit lactation curves and especially persistency (Druet et al., Reference Druet, Jaffrezic, Boichard and Ducrocq2003). Eding et al. (Reference Eding, de Haas and de Jong2009) showed that mastitis occurrences could be predicted, to some extent, by somatic cell count data.

However, the most appealing development is a broader use of MIR spectra. At present, they are used to predict a limited number of milk components, especially total fat and protein percentages, but could be used to predict many more traits, as soon as prediction equations are available. After the pioneering work in Gembloux University (Soyeurt et al., Reference Soyeurt, Dardenne, Dehareng, Lognay, Veselko, Marlier, Bertozzi, Mayeres and Gengler2006), this approach is under study in other countries, such as in France with the ‘PhenoFinlait’ project (Brochard et al., Reference Brochard, Faucon, Barillet, Bolard, Brunschwig, Duhem, Eggen, Esvan, Ferrand, Fritz, Gastinel, Guerin, Journaux, Krychowski, Lagriffoul, Larroque, Lecomte, Leray, Leroux, Leverrier, Martin, Mattalia, Miranda, Palhiere, Peyraud and Boichard2009). The principle is the following. Prediction equations need to be established by comparing MIR spectra with reference measures of a given trait for several hundreds of samples presenting a large variability. These prediction equations could then be used to predict this new phenotype on all the samples analyzed in milk laboratories (e.g. ∼25 million in France) at virtually no additional cost.

Several conditions, however, need to be met in order to make this approach successful. The first one is to collect MIR spectra on a routine basis. The corresponding database is very large, as each spectrum is a plot of ∼1000 (x, y) points. MIR spectra are an example of high-throughput data with regard to both the number of measures and the number of variables.

A second condition is the calibration to adjust for the drift of the analyzer and for differences across analyzers. The recommended method is to regularly analyze reference milk samples to assess these differences and adjust for them (Leray et al., Reference Leray, Ferrand, Larroque, Astruc, Douguet, Brochard and Duhem2011).

The first series of candidate traits predicted by MIR are milk fatty acids (Soyeurt et al., Reference Soyeurt, Dardenne, Dehareng, Lognay, Veselko, Marlier, Bertozzi, Mayeres and Gengler2006; Ferrand et al., Reference Ferrand, Huquet, Barbey, Barillet, Faucon, Larroque, Leray, Trommenschlager and Brochard2011). Accurate predictions could be obtained for 15 to 25 different individual fatty acids or groups of fatty acids. The first results on genetic parameters show a negative correlation between saturated and unsaturated fatty acids, a positive correlation between saturated fatty acids and total fat content and a medium-to-high heritability for all components (Soyeurt et al., Reference Soyeurt, Gillon, Vanderick, Mayeres, Bertozzi and Gengler2007a and Reference Soyeurt, Dardenne, Dehareng, Bastin and Gengler2008). These estimates are very similar to those obtained with the classical methods (e.g. Stoop et al., Reference Stoop, Van Arendonk, Heck, Van Valenberg and Bovenhuis2008) and validate the approach. Fatty acid desaturation activity could also be measured by some fatty acid ratios (Soyeurt et al., Reference Soyeurt, Dehareng, Mayeres, Bertozzi and Gengler2009a). This approach provides many potential new traits and it is not yet clear how all these results should be included in the breeding goal. Therefore, additional work is needed to define what should be the objective in terms of milk composition but it is likely that fatty acids will be included very soon in the breeding goal. Encouraging results, although less accurate, were also obtained in the prediction of individual proteins (e.g. lactoferrin, Soyeurt et al., Reference Soyeurt, Colinet, Arnould, Dardenne, Bertozzi, Renaville, Portetelle and Gengler2007b) or minerals (Soyeurt et al., Reference Soyeurt, Bruwier, Romnee, Gengler, Bertozzi, Veselko and Dardenne2009b).

MIR spectral data are also accurate predictors for parameters related to milk processing properties such as coagulation, acidity and pH (Dal Zotto et al., Reference Dal Zotto, De Marchi, Cecchinato, Penasa, Cassandro, Carnier, Gallo and Bittante2008; De Marchi et al., Reference De Marchi, Fagan, O'Donnell, Cecchinato, Dal Zotto, Cassandro, Penasaand and Bittante2009). Cecchinato et al. (Reference Cecchinato, De Marchi, Gallo, Bittante and Carnier2009) demonstrated that selection for milk with better cheese processing ability predicted from MIR data is feasible. Finally, it should be emphasized that any trait correlated with milk composition could be indirectly predicted by MIR spectra and this opens a completely new way to implement performance recording, not only for genetic purposes but also for traits related to health or reproduction. For instance, body mobilization is correlated with unsaturated fatty acids and body condition score could be predicted by MIR (McParland et al., Reference McParland, Banos, Wall, Coffey, Soyeurt, Veerkamp and Berry2011). The prediction of the reproductive status (pregnant or not) is still a very partially solved question and current methods (progesterone test, echography, rectal palpation) still lack accuracy. Its prediction by MIR in milk is under study in different laboratories. If successful, this method would provide a very useful tool to characterize fertility.

Many other traits could be recorded at a low cost. Foot infections are among the main reasons for involuntary culling in dairy cattle (Fourichon, Reference Fourichon, Seegers, Bareille and Beaudeau2001; Eriksson, Reference Eriksson2006). Hoof trimming is a regular service provided by specialized technicians. This care is an excellent opportunity to score the health of the feet of the animals, as already implemented in some countries (Cramer et al., Reference Cramer, Lissemore, Guard, Leslie and Kelton2008; Stoop et al., Reference Stoop, de Jong, van Pelt and van der Linde2010). A standardized scoring table has been established in France and a pilot study has been launched to collect these data on a large scale.

More generally, health traits are the major challenge in the coming years. Health is the major component of sustainable production and selection. With the notable exception of the Nordic countries, which have been recording health traits for many years, large-scale recording is limited to somatic cell counts in most countries. However, many efforts have been done recently in different countries; for example, Austria with the Gesundheitsmonitoring Rind Projekt (Egger-Danner et al., Reference Egger-Danner, Obritzhauser, Fuerst-Waltl, Grassauer, Janacek, Schallerl, Litzllachner, Koeck, Mayerhofer, Miesenberger, Schoder, Sturmlechner, Wagner and Zottl2010) and Canada with the National Health Project (Van Doormaal, Reference Van Doormaal2009). Indeed, this health information already exists at the farm level with the sanitary notebook, an on-farm recording mandatory in many countries. Collecting this disease occurrence information at a large scale and in a centralized and normalized way would make it possible to use it in selection. Although all information is of interest, a major focus could be placed on metabolic diseases such as ketosis, milk fever, acidosis, on frequent infectious diseases such as mastitis (when not already done), lameness or metritis and on reproductive disorders (retained placenta, cystic ovarian disease).

Specialized recording systems are needed for other health traits

In future, we will need more robust dairy cows. We will also have to reduce the use of chemicals to limit environment contamination and to meet the demand of the citizens. Finally, global warming may increase the pathogen pressure, particularly for parasites and insect-borne diseases. Selection could contribute to an integrated management of health for many diseases with a complex determinism. Digestive and respiratory diseases are good targets for young animals as they are responsible for very important economic losses. However, they need a specialized recording system, in the farms and, possibly, in the slaughterhouses (Assie et al., Reference Assie, Seegers, Makoschey, Desire-Bousquie and Bareille2009). Another interesting target is paratuberculosis. This disease is very difficult to eradicate as the pathogen can remain for a very long time on the farm and the diagnosis, based on Elisa and PCR tests, is still quite difficult (Collins and Manning, Reference Collins and Manning2005). An accurate phenotyping requires different tools and a longitudinal analysis over several years. In spite of the difficulty to generate large samples, the first estimates of genetic parameters (Gonda et al., Reference Gonda, Chang, Shook, Collins and Kirkpatrick2006) have been published, as well as the first Quantitative Trait Loci detection results (Minozzi et al., Reference Minozzi, Buggiotti, Stella, Strozzi, Luini and Williams2010). According to these results and the knowledge in man and other species for similar diseases, it is likely that paratuberculosis has a large genetic determinism. If confirmed, genomic selection could be a useful tool to fight against this very costly pathology at the expense of an important phenotyping effort.

Genetic defects

All dairy cattle breeds are genetically small populations. Because of the initial bottleneck at breed creation in the last century and to the very limited number of artificial inseminated bulls widely spread, French breeds have an effective size ranging from 20 to 50 (Boichard et al., Reference Boichard, Maignel and Verrier1996) and the situation is not different in the other countries. Even the Holstein breed has a limited genetic size at the worldwide level because of the extensive spreading of the North American germplasm in the 1980s (Miglior, Reference Miglior2000). This low size generates an inbreeding increase rate of ∼1% per generation and is the cause of recurrent emergences of recessive defects. Each breed has its own defects and their frequency, initially at a low level, may rapidly increase when a popular ancestor is responsible for a large proportion of the population genes and is found to be a carrier.

Recent molecular tools (single nucleotide polymorphism genotyping and sequencing) allow rapid mapping and identification of the mutation (Charlier et al., Reference Charlier, Coppieters, Rollin, Desmecht, Agerholm, Cambisano, Carta, Dardano, Dive, Fasquelle, Frennet, Hanset, Hubin, Jorgensen, Karim, Kent, Harvey, Pearce, Simon, Tama, Nie, Vandeputte, Lien, Longeri, Fredholm, Harvey and Georges2008). Homozygosity mapping is a method based on the search of a chromosomal region always homozygous in the affected animals, whereas rarely homozygous in the population. When used on several individuals, it can map the defect in a region of one or several megabases. New sequencing technologies, when applied to captured regions or even to the whole genome make it possible to identify the causative mutation. With a test for the causative polymorphism, the defect could be eradicated at a limited cost, by selecting only noncarrier young males as future AI bulls, even when its frequency in the population is already quite high.

This strategy requires that affected calves are detected early, accurately characterized at the clinical level by veterinarians and that tissues are sampled for DNA analysis. An observatory should be set up in order to systematically detect early emerging abnormalities, at birth or later in life. Such an observatory should involve all the organizations in close contact with the farmers, particularly veterinarians and AI technicians, as well as geneticists for genomic analyses. The Danish observatory (Agerholm, Reference Agerholm2007) proved to be very efficient to detect emerging defects and to characterize them. The French National Observatory for Congenital Abnormalities was set up in 2001 after the emergence of Complex Vertebral Malformation syndrome in Holstein and is still very active. It has been shown to be successful to detect different syndromes such as caprine-like generalized hypoplasia syndrome, atresia coli or different ataxia (Capel and Duchesne, Reference Capel and Duchesne2010).

The situation is quite similar for the trait ‘presence/absence of horns’. In modern dairy production systems, horns are not desirable and dehorning is practiced almost systematically at a young age. However, this practice is criticized by many welfare organizations and may be forbidden in the future. A genetic alternative is well known with the polled gene mapped on chromosome 1 but not yet fully characterized. It is a dominant gene with no known adverse pleiotropic effect in contrast with other genes with comparable effects (Capitan et al., Reference Capitan, Grohs, Weiss, Rossignol, Reversé and Eggen2011). The polled gene, however, is not observed in all breeds. When genetic introgression with successive crossbreeding generations is not accepted, the only alternative is to carefully search for naturally occurring polled calves without any associated disorder.

Carcass and meat quality traits

In most European countries, a complete traceability of animals and meat has been set up after the mad cow disease crisis. Increasingly, cuts are maturated and meat pieces are prepared in the slaughterhouses and delivered ready for sale in the supermarkets. In theory, the combination of these two characteristics makes it possible to accurately know not only the carcass weight and its classification, but also the meat yield of each animal, the percentage of the different cuts and possibly the maturation time required per animal, a very important parameter for meat tenderness. This strategic information already exists in some slaughterhouses; however, to our knowledge, its use in selection has not yet been studied. Alternatively, digital images can be used to predict carcass cut yields (Pabiou et al., Reference Pabiou, Fikse, Cromie, Keane, Nasholm and Berry2011). Tenderness could also be predicted by biochemical markers and protein chips are presently under development to quantify several hundreds of proteins simultaneously (Guillemin et al., Reference Guillemin, Jurie, Cassar-Malek, Hocquette, Renand and Picard2011).

Three dimensional (3D) imaging

New imaging technologies are expected to provide interesting solutions to old and only partially solved questions. Tomography is developed for smaller animals but is not adapted to the large body size of cattle. 3D imaging could provide a very accurate description of the exterior body surface at a reasonable cost. It could be used to predict many morphological indicators. As already mentioned, it could be used to predict carcass cut yields. In live animals, a trait of special interest is body condition score (Bewley et al., Reference Bewley, Peacock, Lewis, Boyce, Roberts, Coffey, Kenyon and Schutz2008; Halachmi et al., Reference Halachmi, Polak, Roberts and Klopcic2008) because of its association with productivity, health, longevity and welfare (Roche et al., Reference Roche, Friggens, Kay, Fisher, Stafford and Berry2009). Imaging can accurately detect and measure angularity, and intensive work in this field is currently being carried out.

Precision farming

Electronic devices are more and more present on the farm. The 2010 North American Conference on Precision Farming of Toronto (http://www.precisiondairy2010.com/pdm2010conference) provided a complete picture of this rapidly evolving field. In most cases, they include a recording instrument (physical, chemical, optical, etc.), a system for data recovery and an interpretative and diagnostic software. They could be set up nearly anywhere, at key locations on the farms, on the cow and even in the cow. Of course, all devices require electronic identification of the animals.

The earliest devices have been set up in the milking parlor or in the milking robot. They analyzed different milk characteristics and produced indicators based on temperature, conductivity, milk quantity, milk composition, production balance across quarters or milking flow. Typical phenotypes produced were relative to milk yield and composition, mastitis, milking speed, as well as indicators of their evolution over successive days. Some devices such as Herd Navigator take samples and include an on-farm analyzer for a more accurate diagnosis with biological markers such as lactate dehydrogenase for mastitis (Friggens et al., Reference Friggens, Chagunda, Bjerring, Ridder, Hojsgaard and Larsen2007; Hojsgaard and Friggens, Reference Hojsgaard and Friggens2010), beta-hydroxyburate for ketosis and progesterone for reproduction.

In many farms, concentrate is distributed individually by feeding machines, and concentrate consumption and feeding times are recorded. Roughage consumption is still very demanding to record, in terms of investments and labor, and is reserved to experimental facilities.

Automatic weighing several times a day is a promising way to predict the daily weight change, the random variations due to feeding and drinking being controlled by the accumulation of measures.

Cows can be equipped with other devices. Efficient heat-detection systems are already in the market and other systems can inform the farmer about calving, cardiac rhythm, chewing activity or cow activity. Some devices are encapsulated in bolus and put into the rumen. When the cow is close to a receptor, the bolus can transmit information about past temperature.

Potential applications are virtually unlimited to characterize health, behavior, reproduction, production or feeding. Their development depends either on their economical interest for the farmer or the interest of the breeding company to contract with the farmer to collect information. An important challenge is the access to these data, as most of these devices have their own software and do not always allow data exports. Another major point is data quality and standardization, and the International Committee for Animal Recording is an appropriate and independent organization to define the guidelines.

Environmental footprint

Since 2006 and the ‘Livestock's Long Shadow’ report (Food and Agriculture Organization, 2006), cattle are criticized for contributing largely to overall green house gas production, particularly through the methane produced by their rumen fermentations. Although many of these arguments have been discussed (see for instance the analyses of Pfimlin, Reference Pfimlin2008 or Mitloehner, Reference Mitloehner2009) and sometimes revoked, the image of cattle production has been deteriorated and methane production is likely to become and stay as a trait of large interest in the next decade. Consequently, although green house gas has presently no market value and, therefore, no economic weight in the breeding objective, decreasing its environmental footprint is a major challenge for the cattle industry. Of course, there are many ways to decrease the environmental impact of dairy cattle and a complete diagnosis at the production system is necessary to define the best action plan (Martin et al., Reference Martin, Morgavi and Doreau2010; Bertrand and Barnett, Reference Bertrand and Barnett2011). In developed countries, the most important one is probably to increase longevity in order to decrease the replacement rate and the number of nonproductive animals. When production level is low, increasing productivity could reduce environmental footprint per production unit. However, a direct action on methane production should not be excluded. Indeed, there is an important individual variability in methane production, even for a given feeding system. At a given production level, this variability is related to feed efficiency: methane production corresponds to a loss of energy and is negatively correlated to feed conversion. Limited data in the literature, mostly in beef cattle, show that methane production is genetically determined and is related to feed efficiency (Hegarty et al., Reference Hegarty, Goopy, Herd and McCorkell2007). The first step is to define the most relevant traits and the way to measure them. Methane production could be directly recorded by analyzing the gas produced by the cow. At least two approaches have been used: in the first one, the gas produced is collected in a bottle and methane is measured by comparison with a reference gas (SF6) delivered at a constant rate by a bolus put in the rumen (Pinares-Patiño et al., Reference Pinares-Patiño, Baumont and Martin2003); in the second approach, the cow is placed in a respiratory chamber. The principle of the respiratory chamber was recently adapted to the milking robot. Although the robot is not fully confined and the cow is present only for a short time, Madsen et al. (Reference Madsen, Bjerg, Hvelplund, Weisbjerg and Lund2010) showed that methane production could be measured by MIR by comparing methane and CO₂ concentrations. This approach could be routinely implemented and provide repeated measures for each cow at each milking, as shown by Lassen et al. (Reference Lassen, Løvendahl and Madsen2010). Measuring feed consumption and residual feed intake provides more information but is much more demanding in equipment and labor than methane recording. Even if these equipments are improved and simplified, these measurements will be reserved to experimental facilities. In the future, metagenomics is expected to provide a direct measure of rumen flora composition and their control by the cow and to predict fermentation processes. In any case, this type of phenotype will remain very demanding. If a large reference population is required for genomic selection, the most appealing way would be to produce such a population at the international level through a large collaboration.

Toward a new organization of selection and data recording?

Genomic selection is a revolution and the major changes will affect data recording. Selection will need more diverse phenotypes, which is individually more expensive and is recorded on fewer animals than now, and fewer farms will be involved in the production of these new phenotypes. Breeding organizations will need reference populations of 20 000 to 30 000 recorded animals and will contribute to data collection more than now. Because selection and phenotype production will be more tightly linked, future breeding organizations may incorporate these two activities. International collaboration will be necessary for traits difficult and expensive to measure and for sharing reference populations but it may be hampered by the strategic importance of the phenotypes. For a breeding company, having access to some rare phenotypes is a competitive advantage. However, as soon as these new phenotypes are frequently recorded, there is a clear advantage to share them in order to increase the overall accuracy of genomic estimated breeding values and decrease the phenotyping cost. Breeding objectives are evolving everywhere throughout the world with an increasing weight of functional traits. This trend will continue and additional traits may be included in the breeding goals. The role of the dairy factories is not clear but they will influence selection at least through the price of the products and, therefore, through the definition of the breeding objective and especially for fine milk composition. The large potential genetic gain will favor customized objectives according to farms or environments. A critical question concerns genotype × environment (G × E) interactions. With a limited reference population in a given environment, selection would be more sensitive to G × E than traditional progeny tests with daughters distributed in a large number of herds. Therefore, it is recommended to produce phenotypes over a range of environments.