Abstract Nitrogen losses from dairy systems, such as nitrate leaching and nitrous oxide emissions, can disturb the nitrogen cycle. An effective mitigation strategy involves selecting cows with improved nitrogen-use efficiency and reduced nitrogen excretion. Accurate genetic evaluations are dependent on routine access to large quantities of individual cow phenotypic data on the trait of interest, a feat that is not simple for nitrogen utilization, especially in grazing dairy cows. A workaround is to incorporate information from a proxy trait into a multi-trait selection index; for example, predictions from mid-infrared (MIR) spectroscopy of milk, a widely used method globally for predicting major milk components. This study utilized nitrogen utilization data and accompanying milk mid-infrared spectra from 1,186 Irish grazing cows across 4 experimental farms between the years 2007 and 2018; the 2 nitrogen utilization metrics explored were nitrogen use efficiency (NUE) and nitrogen balance (NBAL). The accuracy of (indirect) genetic selection for nitrogen utilization using milk MIR spectra-based predictions of nitrogen utilization traits was quantified; neural networks or partial least squares regression were used in the phenotypic prediction. The MIR predictions used were those derived from cow-level cross-validation or leave-one-farm-out. Irrespective of prediction method or validation approach, the partial phenotypic correlations between the observed nitrogen utilization values and their respective predictions were stronger for NUE (0.46 to 0.51) than for NBAL (0.23 to 0.28). Based on cow-level cross-validation, the genetic correlations between the observed nitrogen utilization values and those predicted from the milk MIR ranged from 0.69 (SE = 0.06) to 0.84 (SE = 0.06) for NUE and from 0.57 (0.07) to 0.71 (SE = 0.10) for NBAL. Using the leave-one-farm-out validation, the genetic correlations between the observed nitrogen utilization values and those predicted from the milk MIR ranged from 0.64 (SE = 0.07) to 0.75 (SE = 0.06) for NUE and from 0.43 (SE = 0.08) to 0.73 (SE = 0.09) for NBAL. Thus, even though the accuracy of phenotypic prediction is low, milk mid-infrared spectral predictions serve as a reliable proxy for genetic selection aimed at improving nitrogen utilization.
Tavernier et al. (Sun,) studied this question.