Large-scale monitoring of milk infrared-predicted blood biomarkers of stress resilience: effects of breed and dairy systems and their interactions

: Intensified dairy production has increased cows' metabolic stress, which is often masked by good coping mechanisms that impede early detection of subclinical disorders and fertility decline. In particular, inflammation and oxidative stress compromise immune function, increasing susceptibility to disease. Noninvasive milk Fourier-transform mid-infrared (MIR) spectroscopy offers a promising alternative to laborious blood profiling for monitoring metabolic health. This study investigated the associations between milk infrared-predicted blood biomarkers of stress resilience and the sources of variation, focusing particularly on breed-specific responses in relation to DIM, parity, and dairy system. Previously developed calibration equations for 16 blood metabolites related to hepatic function and damage, oxidative stress, inflammation, and innate immunity were applied to a large population database using a gradient boosting machine approach. This database comprised 2,175,148 test-day records that included Fourier-transform MIR milk spectral data, obtained from 250,071 dairy cows from 1,509 multibreed farms belonging to the Parmigiano Reggiano Protected Designation of Origin Consortium. Records were collected by the Breeders Association of Emilia-Romagna (Reggio Emilia, Italy) during official milk recording, and released by the Italian Breeders Association (Rome, Italy). This data set was merged with a field data set provided by the Parmigiano Reggiano Consortium, comprising information on feeding and farm characteristics, allowing us to classify the farms into 4 dairy systems using a non-hierarchical k-means clustering procedure: 2 traditional systems (one located in the Apennines, one in the Po Plain), and 2 modern systems, one using and one not using total mixed rations. The 16 predicted biomarkers were analyzed individually using mixed models that included fixed effects of DIM, parity, month of sampling, year, breed, and dairy system, and the interactions breed × DIM, breed × parity, and breed × dairy system, and random effects of herd and animal. Overall, population-level MIR-predicted blood biomarkers were consistent with expected physiological patterns of DIM and parity, capturing postpartum hepatic overload and innate immune activation followed by gradual immune-metabolic recovery. Breed-specific patterns were evident: Brown Swiss cows exhibited lower hepatic recovery postpartum, maintaining higher concentrations of aspartate aminotransferase (AST) and alkaline phosphatase (ALP) throughout lactation (P < 0.01). We found breed- and farm system-specific variations in oxidative stress biomarkers, total reactive oxygen metabolites (ROMt), and advanced oxidation protein products (AOPP), with Reggiana cows in modern systems exhibiting the highest levels of ROMt (P < 0.001). Antioxidant biomarkers, ferric-reducing antioxidant power (FRAP) and total thiol groups (SHp), were elevated in traditional systems (P < 0.01), especially in the Reggiana. Inflammatory biomarkers, such as haptoglobin and ceruloplasmin, varied according to breed and parity. Holsteins had the lowest ceruloplasmin levels, Brown Swiss and Reggiana cows the highest, regardless of system. Notably, Reggiana exhibited increased haptoglobin in late lactation (P < 0.05). Since all patterns were consistent with physiological responses previously observed on a small-scale using gold standard measures, these findings support the use of MIR-predicted biomarkers for routine population-level monitoring, potentially integrated with existing milk quality recording systems, and highlight opportunities for breed-specific health strategies tailored to different dairy systems. The predicted biomarkers could also serve as novel phenotypes for selective breeding.