Steps to Implement Animal Breeding for Improved Nutritional Quality of Bovine Milk

1. N. Gengler1,2 and H. Soyeurt1 1 Gembloux Agricultural University, Animal Science Unit, Belgium 2 National Fund for Scientific Research, Belgium Steps to Implement Animal Breeding for Improved Nutritional Quality of Bovine Milk

2. Context • Changing breeding goals over last forty years • From yields only • Over type (morphologie) • Towards functional traits (e.g., fertility, longevity) • Limited interest in milk composition except • Always: fat and protein content • Mostly: somatic cell count (udder health) • Also: urea and lactoses (management) • Recently: nutritional quality

3. Milk Quality Traits • Milk fat composition as example • Important variability (3% to 7%) in milk • Composed mostly of fatty acids (FA) • 3 classes: • Saturated (SAT): 70%, Unsaturated (UNSAT): 30% • Monounsaturated (MONO): 25% • Polyunsaturated (POLY): 5% • However far from optimal (human health) • SAT: 30% • MONO: 60% • POLY: 10%

4. Genetic variability existsfor FA Previous, next speaker But implementing Animal Breedingmore complexe process

5. However ImplementingAnimal Breeding Different Steps • Making data available • Adapting models • Implementing routine computation of breeding values • Updating breeding goals and creating and using adapted selection indices • Continuing this ongoing development process towards most advances methods as genomic selection  Presentation will follow this outline

6. Making Data Available - I • Animal breeding needs phenotypes • Until recently difficult to obtain FA composition easily • Based on gas chromatography • Expensive, not in routine • Recent advances based on use of mid-infrared (MIR) spectrometry data • Calibration to predict FA • Similar to predicting fat and protein content

7. Milk sampling (e.g., milk recording) Making Data Available - II • What is MIR spectral data ? MIR spectrometer Spectral data

8. 3000-2800 cm-1: C-H Making Data Available - III • MIR absorption correlated to vibration of specific chemical bonds • MIR spectral data ‘represents’ global milk composition 1700 – 1500 cm-1: N-H 1200 – 900 cm-1: C-O (Sivakesava and Irudayaraj, 2002) 1450-1200 cm-1: COOH

9. Milk sampling (e.g., milk recording) Making Data Available - IV • Using MIR spectral data MIR spectrometer Predicted milk components - Traditional (e.g., fat, protein) - New (e.g., FA) Spectral data

10. Making Data Available - V • Routine milk recording • Currently certain traits available • Major FA (e.g., SAT, MONO, Omega-9)limitation: minor FA • Lactoferin • Minerals • Others under development • Storing MIR spectral data now • Predicting other traits later

11. 2 2 R C R cv Fatty acids (g/dl) Mean SD SEC SEcv RPDcv C4:0 0.13 0.04 0.01 0.94 0.01 0.86 2.69 C6:0 0.09 0.03 0.01 0.94 0.01 0.91 3.41 C8:0 0.05 0.02 0.01 0.90 0.01 0.87 2.80 C10:0 0.12 0.05 0.01 0.92 0.02 0.84 2.49 C12:0 0.15 0.06 0.01 0.94 0.02 0.84 2.48 C14:0 0.49 0.14 0.03 0.96 0.05 0.90 3.14 C14:1 0.01 0.00 0.00 0.41 0.00 0.36 1.25 C16:0 1.40 0.41 0.14 0.88 0.17 0.83 2.46 C16:1 0.08 0.04 0,02 0.76 0.03 0.32 1.22 C18:0 0.56 0.25 0.06 0.94 0.10 0.85 2.62 C18:1 trans 0.17 0.10 0.02 0.95 0.04 0.88 2.83 C18:1 1.07 0.37 0.08 0.95 0.12 0.90 3.23 C18:2 0.11 0.03 0.02 0.73 0.02 0.59 1.57 C18:3 0.03 0.01 0.01 0.71 0.01 0.53 1.46 CLA 0.04 0.02 0.01 0.80 0.01 0.52 1.44 SAT 3.20 0.85 0.08 0.99 0.14 0.97 6.06 UNSAT 1.61 0.48 0.08 0.97 0.13 0.93 3.75 MONO 1.40 0.43 0.08 0.97 0.12 0.93 3.67 POLY 0.21 0.06 0.03 0.79 0.04 0.67 1.75 FA Short 0.41 0.12 0.03 0.94 0.04 0.92 3.54 FA Medium 2.32 0.63 0.13 0.96 0.19 0.91 3.40 FA Long 2.08 0.70 0.14 0.96 0.18 0.93 3.81 Dosage des AG SD= Standard-deviation; SEC= Standard error of calibration; R²c= Coefficient of determination of calibration; SEcv= Standard error of cross-validation; R²cv= Coefficient of determination of cross-validation; RPDcv= SD/SECV

12. Adapting Models - I • Data specific modeling needs: • Longitudinal data: data at every test-day • Multitrait: many (up to 8 and more) milk quality traits that are correlated • Multilactation: less data, more interest to use all available lactations, also linked to absence of historical data • Absence of historic data for new traits:need to use historic correlated traits,e.g., milk yield, fat and protein contents

13. Adapting Models - II • Data specific modeling needs: • Trait definition: some new spectral traits onlyindicators for chemical traits (low RPDcv) • Trait definition: meta-traits • Ratio SAT/UNSAT: linked positively tonutritional and technological properties • Ratios product / substrate: Δ9 indices (next talk) • Potentially adapting models for new fixed effects • E.g., nutritional influence on FA well-known • Heterogeneous variances • Nature of traits • Intra-herd variability  feeding practices

14. Adapting Models - III • Consequence: more complex situation compared to traditional yield test-day models • Advances computing strategies: • Handling of massive missing values  data augmenting techniques • Handling of highly correlated traits  data transformation techniques • Numerous other issues

15. Adapting Models - IV • Also complex situation to estimate (co)variance components: • Multitrait: many correlated milk quality traits, (co)variances needed • Not even nature of traits: different prediction equations different RPDcv, weighting of records • Some spectral traits only indicators for chemical traits:interest to predict inside the model, needs (co)variance between “chemical” and “spectral” traits • Correlations between milk quality and old traits but also other new traits: e.g., those linked to animal robustness as lactoferine

16. Adapting Models - V Consequence: large research needs !!!

17. Implementing RoutineComputations - I • Integration of acquisition of new traits inside genetic evaluation system data flow • Interest to store spectral data on a large scale • Example (known to us): • Southern Belgium (Walloon Region):70 000 cows • Luxembourg:30 000 cows • Already generates nearly 1 000 000 records a year

18. Implementing RoutineComputations - II • Needed (co)variance componentsfirst results become available • Some daily heritabilities (J. Dairy Sci 91:3611-3626) • Milk (kg/day): 0.27 • Fat (%): 0.37 • Protein (%): 0.45 • FA: • SAT (g/100 g milk): 0.42 • MONO (g/100 g milk): 0.14 • Same publication also some needed (co)variances

19. Implementing RoutineComputations - III • Currently few component evaluations • Most genetic evaluations for yields(few exceptions as France) • Milk quality inside evaluation for milk components • E.g., fat, protein • Those traits also needed • As historical correlated data to avoid as much as possible selection bias

20. Implementing RoutineComputations - IV • Expressing genetic results, various possibilities: • Daily base, lactation base • Individual traits: e.g., SAT, UNSAT, MONO • Meta traits: e.g., ratios • Estimate breeding values for all animals • However results for other effects huge potential for management advice: • Not subject of this talk

21. Updating Breeding Goalsand Selection Indices - I • Determine “economic” weights, not easy task • Economic: better milk price • Some dairy companies start to move on this • Health related: social value of more healthy milk  economic value of more healthy milk, reduction of health costs • Other elements, as reputation of milk ashealthy product?

22. Updating Breeding Goalsand Selection Indices - II • Breeding for improved nutritional quality of bovine milk  not at the expenses of other traits • Therefore: • Need to know correlations to traditional traits • E.g., yields, type and functional traits • Also, correlations to other new traits • In particular to robustness traits • However other specific issues to nutritional quality traits

23. Updating Breeding Goalsand Selection Indices - III • Specific issues of nutritional quality traits • Large number of traits: • Which traits to choose and how to choose? • Potential difference between breeding goal traits and index traits: • Breeding goal traits: “chemical traits” • Index traits: “spectral traits” • Doubts that one index fits all situation: • Differentiated index per market as former cheese merit (CM$) and fluid merit (FM$) in USA

24. Updating Breeding Goalsand Selection Indices - IV • Also still large research needs !!!

25. Near Future:Genomic Selection - I • Genomic selection≠QTL detection (previous talk) • Based on dense marker maps (50 000+ SNP) • Linking phenotypic variability to genomic variability • New idea • However under development in nearly all countries • Current implementations mostly • Training population older reliable sires • Predicted population young untested sires

26. Near Future:Genomic Selection - II • Milk quality traits on first hand interesting for genomic selection (prediction) • However • Current implementation needs reliable breeding values from many animals (sires) for training,but genetic evaluations not able to provide this • Genomic selection multitrait setting not yet clear • Nevertheless interesting idea • Why?

27. Near Future:Genomic Selection - III • Genomic information natural way to avoid some current shortcomings: • Few ancestors recorded, risk of selection bias sires (maternal grand sires) could be genotyped • Only recent data, low reliabilities even for older sires larger interest to improve using genomic information • Therefore nutritional quality traits Ideal candidates for genomic selection • Question: How?

28. Near Future:Genomic Selection - IV • How? • Next generation genomic prediction: single step • Recent advances, idea equivalent model • Genomic relationship matrix G reflecting genomic variability replaces (or augments) pedigree based relationship matrix A • Many details under development, progress on • Computing G, inverting G • Combining G and A, potentially on an inverted scale

29. Thank you for your attention Acknowledgments SPW – DGA-RNE different projects FNRS: 2.4507.02F (2) F.4552.05 FRFC 2.4623.08 Email: gengler.n@fsagx.ac.be