Use of multivariate statistical analyses to preselect SNP markers for GWAS on residual feed intake in dairy cattle

• Published in: Journal of animal science Vol. 94; p. 155
• Main Authors: Dimauro, C, Manca, E, Rossoni, A, Santus, E, Cellesi, M, Gaspa, G
• Format: Journal Article
• Language: English
• Published: Champaign Oxford University Press 01.10.2016
• Subjects: Animals; Bovidae; Calves; Cattle; Cluster analysis; Dairy cattle; Data processing; Discriminant analysis; Feed efficiency; Feeds; Genomes; Heritability; Livestock breeding; Markers; Multivariate analysis; Phenotypes; Regression analysis; Single-nucleotide polymorphism; Statistical analysis
• ISSN: 0021-8812; 1525-3163
• DOI: 10.2527/jam2016-0323

An index currently used to evaluate feed efficiency in cattle is the residual feed intake (RFI) whose heritability is around 0.20-0.40. Genome wide association studies (GWAS) can contribute to breeding programs aimed at improving RFI by detecting genomic regions and candidate genes that regulate it. However, the detection of significant SNP in GWAS with high density SNP platforms is often hampered by the severity of Bonferroni's p-value correction for multiple testing, due to huge number of tests. The pre-selection of markers could be an option to mitigate this problem. In the present research, a multivariate approach was used to select a pool of markers that could have any chances to be associated with RFI. Data consisted of 1092 Brown Swiss young bulls genotyped with the Illumina's 50K BeadChip. Animals were divided into two groups, according to RFI: high RFI (HRFI) for RFI > 0.5 standard deviations from the mean RFI; low RFI (LRFI) for animals with RFI < - 0.5 standard deviations from the mean. The two groups consisted of 266 and 280 animals, for LRFI and HRFI, respectively. Individuals that did not belong to the two groups were discarded.Three multivariate discriminant techniques were applied to data. The stepwise discriminant analysis was used to select 152 genome-wide most discriminant markers that were retained for the further analyses. The canonical discriminant analysis significantly separated the LRFI from the HRFI group, and the extracted canonical function was able to correctly assign 92% of animals to the correct group. Canonical coefficients associated to the 152 SNP in the canonical function were useful to rank markers according to their discriminant power. The ability of the selected SNP in depicting the RFI profile of calves was tested by developing a k-means cluster analysis that correctly classified 84% of individuals. For instance, a GWAS was also developed by regressing RFI phenotypes on SNP covariates. After p-values were corrected for multiple testing, no significant marker was obtained by using all original variables (41,183). When only the selected 152 SNP were used, 5 significant markers were obtained.