A lipidomic-based biological clock model indicates greater metabolic age in housed dairy cows compared to pasture-based dairy cows

Metabolic clock models use biomarkers to assign a metabolic age that can provide insight into relative health. In dairy cows, plasma glycerophospholipids containing omega-3 fatty acids are lower with age and lower in cows from total mixed ration, housed systems (HOUSED) compared with extensive, pasture-based systems (PASTURE). We hypothesized that plasma lipid metabolic clock models trained exclusively on either HOUSED or PASTURE cows would produce systematically biased predictions when applied to cows from the alternate system, potentially indicating different metabolic aging between systems. Data were obtained from a previous observational study of 794 peak-milk cows (~58 days in milk), from 29 Australian farms (14 PASTURE, 15 HOUSED) and stratified by parity. A total of 186 lipid species, including glycerophospholipids, sphingomyelins, and triacylglycerols, were quantified. A prediction-oriented Cubist machine-learning algorithm used training: test data splits of HOUSED: PASTURE and PASTURE: HOUSED. Models were fitted using either the full lipid dataset or 10 pre-screened predictor lipids associated with age in both systems. The direction and magnitude of systematic bias were consistent across all approaches. HOUSED-trained models had near-zero training bias but predicted PASTURE cows as 616 d younger than their true age. Conversely, PASTURE-trained models predicted HOUSED cows as 326 d older than their true age. Results indicated that, based on their lipid profiles, HOUSED cows had greater metabolic age than PASTURE cows of the same chronological age. A secondary conclusion reinforced the need for caution when extrapolating machine-learning models to external data. The system-level difference in metabolic age may have implications for cow health and longevity, though this remains to be explored.