Genome-wide comparative analyses of correlated and uncorrelated phenotypes identify major pleiotropic variants in dairy cattle

Abstract

While single nucleotide polymorphisms (SNPs) associated with multiple phenotype have been reported, the knowledge of pleiotropy of uncorrelated phenotype is minimal. Principal components (PCs) and uncorrelated Cholesky transformed traits (CT) were constructed using 25 raw traits (RTs) of 2841 dairy bulls. Multi-trait meta-analyses of single-trait genome-wide association studies for RT, PC and CT in bulls were validated in 6821 cows. Most PCs and CTs had substantial estimates of heritability, suggesting that genes affect phenotype via diverse pathways. Phenotypic orthogonalizations did not eliminate pleiotropy: the meta-analysis achieved an agreement of significant pleiotropic SNPs (p < 1 × 10^-5, n = 368) between RTs (416), PCs (466) and CTs (425). From this overlap we identified 21 lead SNPs with 100% validation rate containing two clusters: one consisted of DGAT1 (chr14:1.8 M+), MGST1 (chr5:93 M+), PAEP (chr11:103 M+) and GPAT4 (chr27:36 M+) affecting protein, milk and fat yield and the other included CSN2 (chr6:87 M+), MUC1 (chr3:15.6 M), GHR (chr20:31.2 M+) and SDC2 (chr14:70 M+) affecting protein and milk yield. Combining beef cattle data identified correlated SNPs representing CAPN1 (chr29:44 M+) and CAST (chr 7:96 M+) loci affecting beef tenderness, showing pleiotropic effects in dairy cattle. Our findings show that SNPs with a large effect on one trait are likely to have small effects on other uncorrelated traits.

Figure 1

Overlap of SNPs detected by single-trait GWAS (p _s < 0.05) within raw traits (RTs, a), principal components (PCs, b) and Cholesky transformed traits (CTs, c). Numbers in cells were rounded log2 count of shared SNPs between single-trait GWAS pairs. Overlap significances were based on Fisher’s exact tests (p _f) accounting for the number of SNPs of each pair of single-trait GWAS and the total number of SNPs analysed. The RT associated with the top factor loading value of each PC was shown in the parentheses (Supplementary Table S2).

Figure 2

The breakdown of shared significant SNPs detected for selected traits with the other traits on each chromosome. As highlighted in red dashed boxes, selected milk yield raw trait (a), principal component 18 (b) and Cholesky transformed fat yield (c) had the largest numbers of significant (p _s) SNPs detected by single-trait GWAS for RT, PC and CT, respectively (Table 1). The number of shared SNPs for the each other trait (non-selected) were determined at the p _s < 0.05 level (see methods). Significances of the enrichment of chromosomes containing the amount of significant SNPs detected for selected traits were based on the Fisher’s exact test (p _f). The RT associated with the top factor loading value of each PC was shown in the parentheses (Supplementary Table S2).

Figure 3

Summary of multi-trait meta-analysis of 25 raw traits (RTs) in the discovery population (dairy bulls). (a) Manhattan plot using SNPs with multi-trait meta-analysis p _m < 0.05. The horizontal blue line was p _m<= 1 × 10⁻⁵. Some reported loci affecting milk traits were highlighted. Equivalent Manhattan plots of principal components (PCs) and Cholesky transformed traits (CTs) were shown in Supplementary Figure S3. (b) Venn gram showing the overlap of numbers of significant (p _m < 1 × 10⁻⁵) SNPs from multi-trait meta-analysis of RTs, PCs and CTs in the discovery population.

Figure 4

Relationship of multi-trait meta-analysis significance (p _m) between odd (e.g., 1, 3, 5…) and even (2, 4, 6…) principle components (PCs). Some known loci affecting milk traits were highlighted.

Figure 5

Clustering of lead SNPs representing major loci affecting dairy raw traits (RTs, a) and principal components (PCs, b) in the discovery population. Loci displaying similar effect clustering patterns across RT, PC and Choleskey transformed traits (Supplementary Figure S3) were highlighted in red boxes. Loci labels on the Y-axis were the same for both left (correlation of SNPs’ effects) and right (SNP’s effects on traits) panels. t values with absolute values >= 1 and validated for consistent effect directions between the discovery and validation populations were coloured. The RT associated with the top factor loading value of each PC was shown in the parentheses on the right panel (Supplementary Table S4).

Figure 6

Clustering of the dairy-beef overlapped SNPs on raw traits (RTs, a) and principle components (PCs, b) of the dairy discovery population. Loci displaying similar effect clustering patterns across RT, PC and Choleskey transformed traits (Supplementary Figure S4) were highlighted in red boxes. Loci labels on the Y-axis were the same for both left (correlation of SNPs’ effects) and right (SNP’s effects on traits) panels. t values with absolute values > = 1 and validated for consistent effect directions between the discovery and validation populations were coloured. The RT associated with the top factor loading value of each PC was shown in the parentheses on the right panel (Supplementary Table S4).