Prediction and genetic analysis of blood β-hydroxybutyrate based on milk mid-infrared spectra in Fleckvieh dairy cows

Hyperketonemia is a common metabolic disorder in early-lactation dairy cows that impairs milk performance and health. The gold-standard indicator for diagnosis is the blood metabolite BHB. However, measuring blood BHB on a large scale is cumbersome and expensive, resulting in a paucity of available phenotypes, therefore precluding hyperketonemia from most genetic evaluation systems. An alternative approach is the prediction of blood BHB concentration from milk mid-infrared (MIR) spectra. Thus, the objectives of the present study were to develop prediction equations for blood BHB concentration using milk MIR spectroscopy and to estimate genetic parameters of and among MIR-spectra-predicted BHB (MIR-BHB), milk production traits, blood BHB, and clinically diagnosed ketosis. For development of prediction equations, a dataset comprising 555 observations of blood BHB concentration measured with a handheld device from 365 early-lactation dairy cows across 25 farms was applied. The prediction equations were derived by comparing different machine learning algorithms (partial least squares, artificial neural network, and support vector machine) using 212 milk MIR spectra and milk yield as predictors. Prediction performance was moderate in both the leave-1-cow-out and the leave-1-farm-out validation scenarios, with coefficients of determination ranging from 0.18 to 0.34. The MIR-BHB phenotypes were subsequently generated for 92,006 test-day records routinely collected within the Austrian milk recording system between 2019 and 2021. The dataset comprised test-day records up to 120 DIM from 21,130 Fleckvieh cows originating from 318 farms, as well as partly measured blood BHB concentrations and clinical ketosis diagnoses. Heritabilities and phenotypic and genetic correlations were estimated using univariate and bivariate linear animal mixed models for 4 DIM periods. The identified heritability of MIR-BHB ranged from 0.13 to 0.18, and MIR-BHB exhibited a negative genetic correlation with protein%, a positive correlation with milk yield, fat:protein ratio, and fat%, and a weak negative to positive correlation with lactose%. Heritability varied only marginally, whereas genetic correlations demonstrated stronger deviations across DIM periods. Strong genetic correlations were found between MIR-BHB and measured blood BHB (±SE, 0.65 ± 0.16) as well as clinical ketosis (0.84 ± 0.46). Results imply the potential of MIR spectra to predict large-scale phenotypes of blood BHB, which may serve as a novel heritable trait for the inclusion of hyperketonemia in genetic evaluation systems of dairy cows.