The genetic background of bovine milk infrared spectra

Milk infrared (IR) spectroscopy is a cheap, quick and high-throughput technique that has been widely used to determine milk components. It has been used as the standard method for routine quantification of fat, protein and lactose content of milk, and it is a promising technique to obtain information about milk composition. The aim of this thesis was to explore the genetic background of bovine milk IR spectra, identify the environmental factors affecting milk IR spectra, and combined use genotypic information and milk IR spectra in predicting dairy cattle phenotypes. Two studies were conducted to explore the genetic background of milk IR spectra of Holstein Friesian dairy cows in the Netherlands. Studies were focused on individual IR wavenumbers, and results showed that for many of them 20 to 60% of variation can be attributed to genetic factors. Polymorphisms of individual gene diacylglycerol O-acyltransferase 1 (DGAT1), k-casein (CSN3) and b-lactoglobulin (LGB), as well as lactation stage of dairy cows and the different dates of IR analysis have significant effect on the values of milk IR spectra. Genome wide association study (GWAS) identified the associated genomic regions. In addition to the regions that related to milk fat, protein and lactose content, this thesis detected 3 new regions related to phosphorus, orotic acid and citric acid content in milk. Knowledge of the genetic background of milk IR spectra could enhance the prediction for dairy cattle phenotypes. This thesis investigated if combined use of genotypes of dairy cows can improve the prediction for milk fat composition. Results suggest that prediction accuracy of unsaturated fatty acids can be considerably improved by adding stearoyl-CoA desaturase (SCD1) genotypes of dairy cows. Predicting methane (CH4) emission based on milk IR spectra is of great interest due the environmental impact of dairy production. This thesis showed the importance of validation strategy in interpreting the results of predicting CH4 emission. This result has general value in milk IR prediction for dairy cattle phenotypes that a block cross validation with farms as block could reflect the true predicative ability for independent observations. This thesis also suggested to predict based on IR wavenumbers from water absorption regions of the spectra as a negative control, to detect potential problem due to dependency structure in the data.