Relevance of mid-infrared spectroscopy predictions of milk fine composition and technological properties for selective breeding

Abstract Text:

In order to evaluate the potential use of novel milk infrared predictions as indicator traits in selective breeding, genetic variation in 92 traits, describing the fine composition and technological properties of milk, and their predictions was assessed in the Italian Simmental (IS) population. The genetic relationship between measured traits (MT) and infrared predictions (IP) was investigated. Fatty acid and protein composition, lactoferrin and mineral contents were available for 1,040, 3,337, 558 and 689 individual milk samples, respectively. Measures of pH and milk coagulation properties were available for 3,438, and 3,266 samples, respectively. Curd yield and curd composition were obtained by laboratory micro-cheese making techniques for 1,177 samples. Infrared calibration models were developed for all traits and IP were obtained for 143,198 spectra of 17,619 IS cows. (Co)variance components for MT and IP were estimated in a set of bivariate analyses, each including one MT and its IP. There was a positive relationship between the R² in cross-validation (R²_CV) of calibration models and the decrease in both the phenotypic variance (r = 0.78; P < 0.01) and the additive genetic variance (r = 0.61; P < 0.01) of IP compared to the estimates for MT. For the 92 traits, the average decrease in total variance of IP compared to the variance of MT was approximately 35%. The decrease in genetic variance was on average 64%. As a consequence, 88 traits exhibited lower h² estimates for IP than for MT. The R²_CV exhibited a positive relationship (r = 0.57; P < 0.01) with the estimated genetic correlation (r_a) between MT and IP. For calibration models having R²_CV > 0.75, r_a between IP and MT was greater than 0.9. The variability in the estimated r_a values increased when R²_CV decreased and, for calibration models having moderate predictive ability, estimates of r_a ranged from 0.2 to 1.

Calibration equations showing high predictive accuracy would lead to a faster genetic progress compared to calibration models having moderate or low prediction accuracy. Nevertheless, the estimated r_a between IP and MT was generally very high, even when calibration models had moderate R²_CV. Hence, even calibrations showing low predictive accuracy may be successfully used in selective breeding, particularly when multiple predictions per animal are available from the routine application of calibration models.

Keywords:

infrared spectroscopy, animal breeding, fine milk composition.