Genome wide association study associated with milk protein composition

Abstract
Introduction: Genomic selection has provided the dairy industry with a powerful tool to increase genetic gains on economically important traits such as milk production (Taylor et al. 2016). One way to identify new loci and confirm existing QTL is through genome-wide association analysis (GWAA). In addition identifying of genes loci with large effects on economically important traits, has been one of the important goal to dairy cattle breeding. QTL assisted selection and genomic regions affecting the production traits have been considered to increase the efficiency of selection and improve production performance. Genome wide association studies typically focus on genetic markers with the strongest evidence of association. However, single markers often explain only a small component of the genetic variance and hence offer a limited understanding of the trait under study. A solution to tackle the aforementioned problems, and deepen the understanding of the genetic background of complex traits, is to move up the analysis from the SNP to the gene and gene-set levels. In a gene-set analysis, a group of related genes that harbor significant SNP previously identified in GWAS, is tested for over-representation in a specific pathway.
Material and methods: The present study aimed to conduct a genome wide association studies (GWAS) based on Gene-set enrichment analysis for identifying the loci associated with milk protein composition traits. For each cow, a total of eight traits including protein yield, protein percentage, αs1-casein, αs2-casein, β-casein, κ-casein, α-lactalbumin and β-lactoglobulin were recorded using plink software and no any correction to adjust the error rate. The gene set analysis consists basically in three different steps: (1) the assignment of SNPs to genes, (2) the assignment of genes to functional categories, and finally (3) the association analysis between each functional category and the phenotype of interest.
In brief, for each trait, nominal P-values < 0.05 from the GWAS analyses were used to identify significant SNP. Using the biomaRt R package the SNP were assigned to genes if they were within the genomic sequence of the gene or within a flanking region of 15 kb up- and downstream of the gene, to include SNP located in regulatory regions. For the assignment of the genes to functional categories, the Gene Ontology and Kyoto Encyclopedia of Genes and Genomes pathway databases were used. The GO database designates biological descriptors to genes based on attributes of their encoded products and it is further partitioned into 3 components: biological process, molecular function, and cellular component. The KEGG pathway database contains metabolic and regulatory pathways, representing the actual knowledge on molecular interactions and reaction networks. Finally, a Fisher’s exact test was performed to test for overrepresentation of the significant genes for each gene-set. The gene enrichment analysis was performed with the goseq R package. In the next step, a bioinformatics analysis was implemented to identify the biological pathways performed in BioMart, Panther, DAVID and GeneCards databases.
Results and discussion:
Gene set enrichment analysis has proven to be a great complement of genome-wide association analysis (Gambra et al., 2013; Abdalla et al., 2016). Among available gene set databases, GO is probably the most popular, whereas KEGG is a relatively new tool that is gaining ground in livestock genomics (Morota et al., 2015, 2016). We had hypothesized that the use of gene set information could improve prediction. However, neither of the gene set SNP classes outperformed the standard whole-genome approach. Gene sets have been primarily developed using data from model organisms, such as mice and flies, so it is possible that some of the genes included in these terms are irrelevant for milk production. It is likely that a better understanding of the biology underlying milk production specifically, plus an advance in the annotation of the bovine genome, can provide new opportunities for predicting production using gene set information. According to gene set enrichment analysis, 20 categories from gene ontology and KEGG pathway were associated with the related to traits (P˂0.05). Among those categories, the Oxytocin signaling pathway, Glycerolipid metabolism, Response to progesterone, detection of calcium ion, Complement and coagulation cascades and amino acid binding including candidate genes CDH13, P4HTM, SPP1, CSN1S1, CSN2, SERPINA1, SLC35A3, ODC1 and PAEP have significant association with protein yield and content, phosphorylation of glycoproteins, coagulation and curd firming milk and lactose synthesis. Some of these regulatory regions, such as enhancers, are located far from the genes. Therefore, although the gene might be part of the analysis, the relevant variant would probably not be included in the gene set SNP class. Finally, linkage disequilibrium interferes with the use of biological information in prediction because irrelevant regions (regions without any biological role) capture part of the information encoded in relevant regions, causing both regions to exhibit similar predictive abilities. The use of very high density SNP data or even whole genome sequence data could alleviate some of these issues.
Finally, it is worth noting that our gene-set enrichment analysis was conducted using a panel of SNP obtained from a single marker regression GWAS, which relies on a simplified theory of the genomic background of traits, without considering for instance the joint effect of SNP. Hence, other approaches (e.g., GWAS exploring SNP by SNP interactions) might provide a better basis for biological pathway analysis.
Conclusion: Our finding showed a genetic selection potential for improving milk quality to milk protein composition per animal. Since genetic improvements are heritable, cumulative, and permanent, this improvement would be permanent and beneficial.

Keywords: pathway analysis, casein, α-lactalbumin, β-lactoglobulin, candidate gene