Genomic assessment of reproduction traits in Holstein dairy cattle across 3 lactations using additive genetic models and post hoc random forest analysis

The typical objective of genomic analyses is to assess additive genetic variance in traits. However, the nonadditive component of genetic variation is often disregarded. Consequently, genomic analyses may not directly elucidate the complex genomic structures or other potential underlying mechanisms, such as pleiotropy, dominance, or epistatic effects. Furthermore, polygenic traits are likely to be subject to nonadditive interactions. Specifically regarding traits pertaining to fitness, including fertility, genomic regions exhibiting nonadditive genetic effects, potentially resulting from directional dominance or epistatic effects, have been identified and require further investigation. In this study, data from more than 7,400 German Holsteins dairy cows with continuous observations of their reproduction performance across the first 3 lactations were analyzed. In the first instance, variance component estimations for 12 observations, distributed across 4 different traits across the 3 lactations, were conducted. The results obtained confirmed low h² for all traits, with the lowest value, h² = 0.016, observed for stillbirth maternal in the third lactation and the highest being h² = 0.128 for metritis in the first lactation. Hereafter, GWAS were employed as an initial step to identify chromosomes of interest for each trait and lactation combination. Hereby, a total of 23 genomic regions were identified as significantly associated and subsequently investigated using a machine learning random forest (RF) approach to screen for putative further nonadditive regions of interest. The correlation (r) between repeated RF models exhibited a mean value of r = 0.854 to r = 0.973, while the average proportion of incorrectly predicted animals remained between 0.102 and 0.244. A direct comparison with the 35 significantly associated markers identified by GWAS revealed common markers, as well as the chromosome- and trait-specific architecture, which displayed different patterns of association signals across the complex of reproduction traits. Screening database records confirmed the identified markers in proximity to previously described genes in the context of reproduction as well as dairy cattle genomics. The findings of our study represent a contribution to a better understanding of the complexity of further nonadditive genetics underlying functional traits using GWAS results, with particular attention to regional clustering. Furthermore, they may serve as a foundation for regional in-depth analysis using a broader cohort of animals.