PSIII-7 Relationship of Age and Genetics with the Methylation Profile of Beef Cattle

Abstract This study aimed to compare models for the prediction of cow age from DNA methylation profiles and estimate the heritability of the proportion of methylated sites (PM) and methylation status at each site (MS). Methylation data from blood samples of cows (n=136) were generated from the HorvathMammalMethylChip40 array that consists of 34,324 CpG sites that mapped to the bovine genome. Methylation status was determined by the distribution of the methylation values, with values above, within and below 2 standard deviations classified as methylated (2), intermediately methylated (1) and unmethylated (0), respectively. Principal component analysis (PCA) was applied to a (co)variance methylation status matrix. The first and second PC accounted for 25.65% and 9% of the total variance, respectively. Five Bayesian models (Bayesian ridge regression, BayesA, BayesB, BayesCπ and Bayesian LASSO) were implemented with the BGLR package in R. Bootstrapping validation (n=400) was used to evaluate the tested models, with 102 and 34 individuals in the training and validation sets, respectively. The correlation between the predicted and true age was high (r = 0.97 to 0.99). A BayesA model performed the best (r = 0.99, MSE = 0.11 and slope = 0.93), while Bayesian LASSO was the least accurate (r = 0.97, MSE = 0.26 and slope = 0.88). Heritability was estimated using GBLUP implemented in the BGLR package. The mean (SD) heritability estimate for PM was 0.46 ± 0.10 and the heritability of MS ranged from 0.18 to 0.73 (mean = 0.33). The 10% of sites with the highest heritability (343 sites; mean = 0.62) were located in exon (91), intron (84), intergenic (152), and promoter (16) regions. The largest number of these top sites (31) were located on chromosome 3 in genetic or intergenic regions close to transcription factor binding sites (i.e., FOXO6, ELAV4 and LMO4).