Phenomics

Prediction of nitrogen use efficiency and nitrogen losses from milk mid-infrared spectra, and their genetic parameter estimates in Norwegian Red cows

---

Abstract

The objectives of this study were (1) to evaluate the potential of using milk mid-infrared (MIR) spectra for prediction of nitrogen use efficiency (NUE) and nitrogen losses (NL) for individual dairy cows, and (2) to estimate genetic parameters for both observed (ONUE, ONL) and MIR spectra predicted NUE (PNUE) and NL (PNL). The study used data from 1,933 Norwegian Red cows from 16 herds, collected from 2022 to 2025. NUE was defined as ratio of gram milk nitrogen output to gram nitrogen intake and expressed as percentages, while NL was defined as gram nitrogen intake minus gram milk nitrogen output. Milk spectra-based prediction models were developed using 2,253 observations (1,287 cows) and partial least square regression, followed by genetic analyses. Model performance was evaluated using (i) 5-fold grouped cross-validation by cow and (ii) leave-out-one-farm cross-validation, with and without herd effect correction on phenotypes. Grouped cross-validation yielded coefficients of determination (R²) of 0.39-0.69 for NUE and 0.41-0.60 for NL, with root mean square error (RMSE) of 2.9-4.0 and 49.6-60.6, respectively. In contrast, leave-out-one-farm cross-validation showed lower predictive accuracy (R² = 0.19-0.50 for NUE and 0.17-0.42 for NL) and higher RMSE (3.8-4.9 for NUE and 75.2-60.2 for NL), indicating limited model generalizability. Inclusion of milk yield, days in milk, parity, and/or herd as predictors alongside milk spectra improved the calibration. Best performing predictive model was used to generate PNUE and PNL phenotypes from large milk spectral data for genetic analysis. Genetic parameters were estimated using single- and multi-trait repeatability animal models fitted to 12,229 test-day observations and a pedigree of 43,299 animals. Heritability estimates (± standard error) for ONUE and PNUE ranged from 0.12± 0.022 to 0.14±0.023 and 0.14±0.026 to 0.15±0.026, respectively, whereas ONL and PNL showed lower heritability of 0.07±0.014 to 0.09±0.019. Genetic correlations (± standard error) between ONUE and PNUE were positive and high (0.85±0.044 to 0.87±0.042), suggesting that PNUE is a reliable proxy trait for indirect selection on NUE. ONUE showed a negative and moderate genetic correlation with ONL (−0.42±0.127 to −0.51±0.120), but a low positive genetic correlation with PNL (0.15±0.133). Genetic correlations between PNUE and NL traits were low with high standard error (−0.02±0.154 to 0.09±0.137). ONL and PNL showed a moderate positive genetic correlation (0.46±0.130), reflecting limited predictive accuracy of NL from milk spectra and other biological covariates. In conclusion, milk MIR spectra, combined with biological covariates, can predict NUE and NL with moderate accuracy. Both NUE and NL traits are heritable. There are medium to high favorable genetic correlations between the corresponding observed and predicted NUE and NL traits. Therefore, our results suggest that PNUE and PNL can be useful indicator traits for improving nitrogen utilization and losses through indirect selection.

Keywords: 2026

How to Cite:

Belay, T., Heringstad, B., Karlsson, L., Kidane, A., Meuwissen, T., Prestløkken, E. & Sommerseth, J., (2026) “Prediction of nitrogen use efficiency and nitrogen losses from milk mid-infrared spectra, and their genetic parameter estimates in Norwegian Red cows”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286322. doi: https://doi.org/10.31274/wcgalp.23926

Rights: 1