Implementation of a routine genetic evaluation of milk coagulation properties in Italian Holstein using a mixed reference population of bulls and cows

Cheese production is one of the most important segments of the agrifood sector in Italy and milk coagulation properties (MCP) are a key factor for an efficient cheesemaking process. Milk coagulation properties, referred as rennet coagulation time (RCT), curd firmness (a30) and curd-firming time (k20), are available from mid-infrared spectroscopy (MIRS) prediction models implemented within the official national milk recording system (LEO project, PSRN mis 16.2, AIA, 2023). Aim of this study was to assess the possibility to genetically improve MCP in Italian Holstein population and to develop a routine genetic evaluation for such traits. A multiple-trait repeatability linear animal model was employed, with RCT, a30, k20 and casein percentage (CAS) as outcome variables. Fixed effects were the interaction between year and season of recording, between parity (1,2+), year and age at calving class (7 classes) and between parity, year and days in milk class (10 classes). Random effects were contemporary groups, animal permanent environment, animal additive genetic and residuals. The models for RCT, a30 and k20 accounted also for somatic cell score as covariate. A total of 64,720 records from 150 herds, randomly sampled from the full dataset of 4,001,769 observations after edits, were used for variance components estimation using THRGIBBS1F90. The pedigree was traced back to 4 generations and was composed of 59,124 individuals. Convergence was assessed using R package BOA. The posterior mean (PM) for heritability was 0.33 for CAS, 0.11 for RCT, 0.16 for a30, and 0.15 for k20. The genetic correlation between RCT and a30 was -0.87, highlighting their antagonistic relationship; the same conclusion can be drawn from the correlation between k20 and a30 (-0.98). RCT and K20 were positively correlated (0.77). CAS was negatively genetically correlated to both RCT and k20 (-0.04 and -0.76, respectively), and positively to a30 (0.51). A SNPBLUP model was employed for estimating genomic breeding values (GEBV) using two distinct training populations: solely bulls and both bulls and cows (mixed reference population). The validation of GEBVs, conducted with complete and partial datasets (with a three-year back cutoff date for phenotypes), consistently demonstrated that employing a mixed training population results in reduced dispersion and heightened reliability for these traits. These results showed the feasibility of selecting for MCP improvement within the Italian Holstein population. Furthermore, they establish the foundation for implementing a routine genetic evaluation aimed at enhancing cheese production, utilizing a mixed reference population for SNP effects estimation.