The future of phenomics in dairy cattle breeding

Abstract

Increasingly complex dairy cattle production systems require that all aspects of animal performance are measured across individuals’ lifetimes. Selection emphasis is shifting away from traits related to animal productivity toward those related to efficient resource utilization and improved health and welfare/resilience. The goal of phenomics is to provide information for making decisions related to on-farm management, as well as genetic improvement. Genetic selection has been a very successful tool for the long-term improvement of livestock populations, and the rapid adoption of genomic selection over the last decade has doubled the rate at which some dairy cattle populations are improving (García-Ruiz et al., 2016). While details differ somewhat between livestock species, the general objective of breeding programs is the same: the identification of genetically superior males and females that are used as the parents of the next generation. However, the expression of genetic potential also requires that animals are placed in environments that support such performance. For example, Figure 1 shows the increase since 1970 in milk protein yield in U.S. Holstein cattle, partitioned into gains due to increased genetic potential and those associated with improved environment (housing, feeding, etc.). Improved animal efficiency has also resulted in reduced environmental impacts throughout the production chain, which is of importance to consumers around the world (Capper and Cady, 2019). Improvements in genetic potential and cow management have contributed to a sustained increase in productivity per animal, such as for protein yield of U.S. Holstein cows (Source: Council on Dairy Cattle Breeding). Improved animal efficiency may be in conflict with improved health and resilience of animals because of trade-offs (Poppe et al., 2020). Resilience is the capacity of animals to be minimally affected by environmental perturbations, such as diseases or heat waves, or to rapidly return back to the state it had before the perturbation (Berghof et al., 2019). One example is to have cattle that can adapt to climate change, for example, cows that are heat tolerant (Pryce at al., 2018). Big data offer opportunities to better breed dairy cattle with a balanced emphasis on efficiency and resilience. This work will review the current literature related to deep phenotyping of dairy cattle, identify opportunities and challenges associated with new technology for measuring animal performance, and discuss how promising tools may be applied in practice. The phenotype, a measurement of some property or feature of an individual, is the basis of all genetic improvement programs, although its meaning is often assumed and definitions are sometimes rare (e.g., Lush, 1994). While livestock breeders have long used complex selection indices that combine many traits into a single measurement of performance (Cole and VanRaden, 2017), there is a renewed interest in the collection of high-dimensional data on individual animals driven by various genome-mapping initiatives (Houle et al., 2010), environmental challenges (Grossi et al., 2019), and promising new technologies for low-cost phenotyping (e.g., Halachmi et al., 2019). A recent white paper focused on high-throughput phenotyping in livestock species (Koltes et al., 2019), and the U.S. Department of Agriculture’s latest 10-yr blueprint for animal genomics research emphasizes the importance of closing the genome-to-phenome gap (Rexroad et al., 2019). The ultimate goal of these efforts is to understand in detail how information encoded in the genome is translated into a phenotype to support the production of nutritious food from healthy animals. It is important to recognize that the dairy industry has many constituents (dairy producers, dairy processors, breeding companies, etc.), each of which has a different interest in measures of animal performance than do scientists. In the United States, 42 traits of economic importance (5 yield traits; 8 measures of health, fertility, and longevity; 6 direct measures of health; 5 calving traits; and 18 conformation traits) are currently evaluated in the Holstein breed. Each of these traits is directly related to cow profitability, and selection indices are commonly used to combine information from many traits into a single quantity that can be used for animal ranking and selection (Figure 2). Phenotypes included in 21 total merit indices used to rank dairy cattle for profitability 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, 2019); 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. (Source: after Figure 4 in Cole and VanRaden, 2017). Among researchers, there is interest in many other phenotypes, notably those related to milk composition and manufacturing properties, but there is currently little interest from dairy producers because there is no way for them to be paid directly for those traits. An exception is breeding for A2 beta casein, which is gaining popularity and premium supermarket shelf-space in some countries. When the phenotypic recording of these measures does not directly lead to greater farm income, it is a challenge to incentive dairy producers to contribute data, although they are willing to look to the future (e.g., breeding for kappa casein). In addition, dairy processors feel that technological innovations are a more cost-effective solution than genetic improvement for manufacturing properties. This disconnect can give the impression that dairy producers are not willing to change the direction of their selection programs, but actually reflects a lack of market signals. It is tempting to assume that genomic selection provides an answer to all of the problems of the past. However, while genomics helps improve the rate of genetic gain (García-Ruiz et al., 2016), the emphasis on genotypes has often detracted from the importance of phenotypes. Genomic selection can improve only what is measured. Figure 3 shows the often-discussed example of decreased cow fertility associated with selection for increased milk production in U.S. Holsteins. When fertility was not included in the breeding goal, days open increased by approximately 1 d/yr and has only recently begun to show a favorable genetic trend. Changes in genetic merit of Holstein bulls for the production (milk yield, broken green line) and fertility (daughter pregnancy rate, solid blue line) of their daughters from 1957 to 2017 (Source: Council on Dairy Cattle Breeding). One of the key drivers of recent interest in animal phenotypes is the development of a new generation of electronic sensors that can be used to collect detailed, high-frequency measurements about animal performance and their environments in real time or near real time (e.g., Halachmi et al., 2019). Figure 4 provides a summary of data that can be collected using some of these systems. Phenotypes that can be collected at the cow level using sensor-based recording devices (Source: after Figure 5 of Halachmi et al., 2019). The goal of these efforts should not necessarily be to replace existing phenotypes with new ones but to identify new sources of correlated information that can be collected on a large scale. For example, it would not be desirable to stop collecting somatic cell count (SCC) records that are correlated with udder health just because a new mastitis phenotype becomes available (Martin et al., 2018). There are several national databases that contain millions of observations for SCC that are used to compute high-reliability breeding values. New phenotypes, even those based on low-cost, easily collected observations, will require many years for sufficient data to accumulate to match the reliability that current evaluations based on SCC, longevity, and fertility data already have. There is likely more value in using new technologies to supplement data for existing phenotypes that are difficult or expensive to measure, such as computer vision-based measures of feed intake in place of weight-based intakes (Halachmi et al., 2019; Li et al., 2020). Precision management is needed in order to provide the optimal environment for high-performing dairy cows, as well as to make timely management decisions (e.g., Kaniyamattam and De Vries, 2014). This includes a more frequent sampling of milk components (fat, protein, lactose, milk urea nitrogen, and somatic cell counts) and activity monitoring to identify changes in cow behavior associated with the onset of estrus, lameness, or disease and integration of real-time farm-level information (e.g., feed composition and weather). Alternative information about health comes from data recorded by automatic milking systems and electronic milking systems, such as milk yield per milking (Poppe et al., 2020), udder and teat characteristics (Poppe et al., 2019), or other electronic devices, such as sensors or cameras (Song et al., 2019). A recent, multi-institutional effort to develop a “Virtual Dairy Farm Brain” has been organized by Liang et al. (2018) in order to combine expertise from scientists, dairy producers, and industry professionals for the purpose of improving whole-farm decision-making. The prediction of phenotypic performance using new data has not been studied in as much depth in the animal sciences as in the plant sciences (Mir et al., 2019), perhaps, because it is easier to run a seedling through an instrument for deep phenotyping than it is a calf. However, there is growing interest in this topic and the literature is growing (e.g., Goddard et al., 2016; Ho et al., 2019). The most important question overall might be, “How do we provide real-time feedback to dairy producers so that they can take advantage of opportunities and avoid problems before they become acute?” This is driven in part by the biological variability among animals, exacerbated by genotype-by-environmental effects, and the high value of an individual animal. A key opportunity will be to use data and new analytics to rigorously evaluate current management practices to identify assumptions that either no longer hold or, perhaps, never held at all. That is, when presented in a timely and comprehensible manner, data can supplement intuition and guide herd managers to better decisions. It is also critical that the dairy sector does not ignore growing demands from consumers for greater transparency about their food and how it is produced. For example, real-time monitoring of animal health and welfare may have great value as a marketing tool as is the ability to trace food to farms of origin (this is already possible in some markets, e.g., the Red Tractor logo on UK products is designed to give consumers confidence in quality and traceability of products; https://www.redtractor.org.uk/). Advanced analytics may also be necessary if consumers, and the milk processors that are the intermediate between dairy producers and consumers, demand that reproductive hormones are no longer used for routine management. Negative consumer perceptions of recombinant bovine somatotropin (growth hormone) led milk processors in the United States to ban its use, emphasizing that scientific and technical arguments about the value of a particular technology may be ineffective in the face of public resistance. It is better to be proactive on this front than reactive. Whole-animal measurements of performance are important because they provide detailed information about the physiological state of the animal but have the disadvantage that many different types of data must be integrated into decision support systems. Fine measurements considered to be precise measurements of individual physiological indicators (e.g., β-hydroxybutyrate) also provide valuable information but require individual collection and processing of samples. Unfortunately, labor, equipment, and laboratory expenses often prevent wide-spread collection of data from national herds regardless of the exact technology or assay used. Ideally, new phenotyping technologies would build on existing systems for nationwide collection and analysis of phenotypes for millions of cows without substantial additional expenses for dairy producers. Alternatively, there is great potential for the use of test-day milk samples as the source of correlated phenotypes for many traits related to milk composition and the cow’s physiological status. An absorbance spectrum can be generated by beaming infrared light through a milk sample (Figure 5), and the resulting points may be used to develop predictors of many different phenotypes (e.g., De Marchi et al., 2014; Gengler et al., 2016). While there are differences between the instruments sold by the three major vendors of milk-testing equipment, such as Bentley Instruments, Inc., FOSS, and Perten Instruments (formerly Delta Instruments), the same general approach may be used to develop equations for predicting, for example, lactoferrin, fatty acids, and coagulation properties from spectral data. In principle, it is similar to genomic prediction: phenotypes collected in a reference population are regressed on the wavelengths from the spectrum and the resulting weights used to estimate values in the larger population. This supports wide-scale, low-cost phenotyping because individual mid-infrared spectroscopy (MIR) samples can be collected at normal processing speed in milk-testing laboratories without the need for manipulation of the sample. The appeal with MIR is that it could become a cheap way of getting individual cow records for expensive and hard-to-measure phenotypes, such as methane production and metabolic profiles. For example, in a study that included data from five countries, calibration and cross-validation coefficients of determination of MIR predictions of >0.65 and 0.57 were obtained for prediction of methane emissions (Vanlierde et al., 2018). Creation of a mid-infrared spectral phenotype from a milk sample (Source: “MIR for Profit”; https://datagene.com.au/ct-menu-item-7/projects-industry-initiatives/mir-for-profit). While there is great potential associated with these new data, there also are many challenges that must be addressed by the scientific and farming communities. Many of the new technologies being to dairy producers are and their may lack This is a difficult challenge to because of changes in at that have the of and needed to either in with industry or used for are often or not at all because they are considered It is also for data to be or in vendors provide so that data can into on-farm systems, such as dairy producers for to their data, and some provide no for data of to data often are in that do not and data and how the data may be used. there is a lack of transparency in the dairy or integration in which are and become a part of larger is more For example, A may with the for that data be in the data to and this is not to the dairy producers The of new data in animal recording and genetic improvement programs also is such as the for Data as well as national programs, provide detailed for the collection of data that are used in animal improvement This includes definitions of phenotypes to of data. must and and data collection and systems also must such as those related to animal Data as which when data quality For example, the Council on Dairy Cattle requires that changes to as of a based on by an it is not the data, or has to make decisions those data, they be easily integrated into genetic improvement or However, the challenge of all that of how the dairy is to the of their in these new It is of a data are available to new such as for or in there is no way for dairy producers to paid to collect those data. The lack of to is also a growing because of the need to data to and from There are related problems on the because the commonly used in the United States are with and other that can with data the use of and other This additional on of the of and other systems. more technology is to be on farms to the of consumers and processors, there must also be a way for dairy producers to the of those new if the new technology on farm must be most make work The of data that is by sensors and other on-farm management tools can be and may require more from farm of This the integration of such technology and while it is a challenge to dairy producers to collect new phenotypes, it is an large challenge to the collection of phenotypes through The of cows in the Dairy programs has been which is New on how milk recording is and used could its For example, if MIR predictions of reproductive performance or disease are may about milk recording in a different for the of data from systems in genetic improvement programs also are needed to support It is critical that the dairy that genetic is on the recording of and new phenotypes. This is difficult when those do not contribute data with those do support the but no solution to this has been In the United States, is for in programs, so there is a for those do provide data. is not measured be In it is important that to provide value to their when there is or between programs and that genomic evaluation has been paid recently to the growing of data available in all and the need to those data into information with direct and are all by their to be the solution to these challenges (e.g., Cole et al., et al., 2019; et al., 2019). This is the of information to be the high demand for with analytics and the demand from dairy producers for better tools to their However, these new have their from to et al., and there is a to do not dairy producers, consumers, or as a need to focused on food as as possible for a growing population. New analytics technologies will support that goal, but they are not and their should be in In order to the growing demand for protein, it is that the dairy industry make the most efficient use possible of the cow’s ability to plant et al., 2017; Figure This will require the efficient use of all needed on the of from animal management of cows and their and to consumers that dairy animals are healthy and well technologies based on low-cost sensors and data analysis tools will be necessary to those between dairy producers, scientists, and industry will be to these technologies into Holstein and on Farm Farm in (Source: by Cole is the of the and in which is part of the research also is a of the for on a in animal, and from and is a of the for and the is also an in the Department of at and the Department of at the of research currently on genetic improvement of fertility, health, and traits in dairy use of genomic data to support on-farm and identification of genetic in is the at the Association of in a in of dairy cattle from and is a of the in The supports the U.S. industry on a by the of and the of and other for dairy producers. is also a about in national for dairy producers. is of and breeding in the Department of at research is focused on the genomic improvement of traits in are in the of genomic prediction and for traits. research in is focused on the of genomic selection on long-term and is of animal breeding and genomics at and and The is an in livestock breeding animal and genomic current research on resilience indicators based on data in dairy cattle, and the of in genomic breeding programs, and the of for long-term genomic is of and of research genetic gain and herd improvement in the dairy of interest are genetic improvement of traits in dairy cattle, breeding genomic and development of dairy selection is also of in several that the future of dairy research in and is a of the for on was by Dairy by of Genomic New and Selection of the of the United States Department of The of or products in this is for the purpose of information and does not or by the U.S. Department of The is an opportunity and

Topics & Concepts