Search

Additional file 2: Figure S1. Manhattan plot of p values from single-trait GWAS results of heat tolerance milk (A), fat (B), protein (C) yield slope traits for the Holstein cow discovery set (N = 20,623). The dashed line represents p-value cut-off = 0.001. Figure S2. QQ-plot for heat tolerance milk (HTMYslope), fat (HTFYslope), and protein (HTPYslope) from GWAS of the Holstein cow discovery set (N = 20,623). Figure S3. Manhattan plot of p values from single-trait GWAS results of heat tolerance milk (A), fat (B), protein (C) yield slope traits for combined set of Holstein and Jersey cow discovery set (N = 25,766). The dashed line represents p-value cut-off = 0.001. Figure S4. QQ-plot for heat tolerance milk (HTMYslope), fat (HTFYslope), and protein (HTPYslope) from GWAS using a combined set of Holsteins + Jersey cows (N = 25,766). Figure S5. Minor allele frequency (MAF) distribution of the 50k SNP data and the selected ���top SNPs��� (most significant) from the imputed-whole genome sequence variants. Figure S6. Accuracy of genomic predictions in Holsteins (A; N = 1223), Jersey (B; N = 6338), and Holstein���Jersey crossbreds (C; N = 790) validation cows for milk (MYint), fat (FYint) and protein (PYint) yield intercept traits from different SNP sets based on the BayesR and BayesRC methods: (a) standard 50k SNP array (50k; colored grey) (b) 50k + top SNPs selected from single-trait GWAS (colored blue) and multi-trait meta-analysis (colored orange) at a less stringent cut-off threshold of [��� log10(p-value) ��� 2] and a more stringent p-value of [��� log10(p-value) ��� 3]. The top SNPs were selected from GWAS of Holstein cows (N = 20,623). Vertical lines represent standard errors calculated from three (Holsteins) and two (Jersey) random validation subsets. Figure S7. Bias of genomic predictions in Holsteins (A; N = 1223), Jersey (B; N = 6338), and Holstein���Jersey crossbreds (C; N = 790) validation cows for milk (MYint), fat (FYint), and protein (PYint) yield intercept traits from different SNP sets based on the BayesR and BayesRC methods. Figure S8. Bias of genomic predictions in Holsteins (A; N = 1223), Jersey (B; N = 6338), and Holstein-Jersey crossbreds (C; N = 790) validation cows for heat tolerance milk (HTMYslope), fat (HTFYslope), and protein (HTPYslope) yield slope traits from different SNP sets based on the BayesR and BayesRC methods. Figure S9. QTL discovery using the single-breed (Holstein cows; left) and the across-breed (Holsteins + Jersey cows; right) discovery set.

Publication pdf can be found here: http://www.aaabg.org/aaabghome/fullproc24.php

Additional file 1: Table S1. Number (included in brackets), breed, and sex of animals with phenotypes and genotypes used for different study objectives. Table S2. Additive genetic variance (AG) and genomic heritability ( $${h}^{2}$$ h 2 ) estimates for heat tolerance (slope) traits based on 50k SNP data for Holstein cows (N = 29,107), Holstein bulls (N = 3323), Jersey cows (N = 6338) and crossbred cows (N = 790). Table S3. Additive genetic variance (AG) and genomic heritability ( $${h}^{2}$$ h 2 ) estimates for milk intercept traits based on 50k SNP data for Holstein cows (N = 29,107), Holstein bulls (N = 3323), Jersey cows (N = 6338) and crossbred cows (N = 790). Table S4. Number of informative markers defined as ���top SNPs��� selected from single-trait GWAS and multi-trait meta-analyses of intercept traits in the Holstein discovery cow set (N = 20,623). Table S5. Number of SNPs with the same effect direction at different GWAS p-value cut-off and their corresponding false discovery rate (FDR) between Holstein bulls (N = 3323) versus Holstein cows (N = 1223) and Holstein bulls (N = 3323) versus Jersey cows (N = 6338). The false discovery rate effect direction (FDR ED) was computed following [39] and the conventional false discovery rate (FDR) was calculated following [47]. Table S6. Accuracy and the dispersion bias of predictions (in brackets) from non-linear Bayesian methods (BayesR and BayesRC) versus GBLUP models. Table S7. Number of overlapping selected ���top SNPs��� between intercept and slope traits and the proportion of SNPs with the sample effect direction detected from the Holstein cow discovery set at a GWAS p-value cut-off of 0.001. Table S8. Correlation between estimated breeding values for heat tolerance (based on BayesR and BayesRC) versus Australian selection index and balanced performance index values for the Holstein validation cows (N = 1122) used in this study. Table S9. Correlation of GBV between whole ( $${GBVs}_{W}$$ GBVs W ) versus ���partial��� data ( $${GBVs}_{P}$$ GBVs P ), and the dispersion bias of predictions ( $${\mathrm{b}}_{\mathrm{w},\mathrm{ p}}$$ b w , p ) in the Holstein validation cows (N = 1223) using a linear regression method [45] based on GBLUP and BayesR models.

Background Sharing individual phenotype and genotype data between countries is complex and fraught with potential errors, while sharing summary statistics of genome-wide association studies (GWAS) is relatively straightforward, and thus would be especially useful for traits that are expensive or difficult-to-measure, such as feed efficiency. Here we examined: (1) the sharing of individual cow data from international partners; and (2) the use of sequence variants selected from GWAS of international cow data to evaluate the accuracy of genomic estimated breeding values (GEBV) for residual feed intake (RFI) in Australian cows. Results GEBV for RFI were estimated using genomic best linear unbiased prediction (GBLUP) with 50k or high-density single nucleotide polymorphisms (SNPs), from a training population of 3797 individuals in univariate to trivariate analyses where the three traits were RFI phenotypes calculated using 584 Australian lactating cows (AUSc), 824 growing heifers (AUSh), and 2526 international lactating cows (OVE). Accuracies of GEBV in AUSc were evaluated by either cohort-by-birth-year or fourfold random cross-validations. GEBV of AUSc were also predicted using only the AUS training population with a weighted genomic relationship matrix constructed with SNPs from the 50k array and sequence variants selected from a meta-GWAS that included only international datasets. The genomic heritabilities estimated using the AUSc, OVE and AUSh datasets were moderate, ranging from 0.20 to 0.36. The genetic correlations (rg) of traits between heifers and cows ranged from 0.30 to 0.95 but were associated with large standard errors. The mean accuracies of GEBV in Australian cows were up to 0.32 and almost doubled when either overseas cows, or both overseas cows and AUS heifers were included in the training population. They also increased when selected sequence variants were combined with 50k SNPs, but with a smaller relative increase. Conclusions The accuracy of RFI GEBV increased when international data were used or when selected sequence variants were combined with 50k SNP array data. This suggests that if direct sharing of data is not feasible, a meta-analysis of summary GWAS statistics could provide selected SNPs for custom panels to use in genomic selection programs. However, since this finding is based on a small cross-validation study, confirmation through a larger study is recommended.

Additional file 4. Analyses of variants located around previously reported wool and hair genes. A set of variants in and close to 453 previously reported genes associated with wool/fur/hair traits across mammals (Published-Gene set) were identified including 1619 coding variants and 11,712 variants up- and down-stream of the genes. A BayesR fine-mapping analysis identified which of these variants were most associated with wool traits and a GBLUP tested the proportion of genetic variance that the sequence variant set explained for each of the wool traits.

Additional file 2: Table S1. STAR parameters used for alignment.

Additional file 4: Table S2. RNA-seq library information. Mean, minimum, maximum and median values of the number of read pairs that pass trimming and filtering (clean reads), and are uniquely mapped reads, and the proportion of clean reads that map uniquely to the genome in liver and muscle samples. The distribution of the proportion of clean reads that are uniquely mapped in liver and muscle samples is also shown.