Comparison of single and multiple trait statistical methods for QTL detection in dairy cattle

The principle focus of this thesis was the study of the effect of application of multivariate methods on power of QTL detection and accuracy of QTL position estimates in dairy cattle granddaughter design mapping studies. To investigate the QTL associations with correlated traits and traits not extensively studied, a genome scan for QTL detection was carried out using a granddaughter design, which included 433 bulls distributed in six sire families. Trait values used were estimated breeding values (EBV) derived from a test day model or animal model for 47 traits (production, functional, and type). Bulls were genotyped for 69 microsatillite markers. The QTL were analyzed by across-sire and within-sire linear regression of EBV on the probability that the son receives one or the other paternal QTL allele, given the marker information. QTL were detected for all traits, including those with a low heritability. For across-sire analysis, 71 QTL were detected with a false discovery rate less than 10%. Within-sire analysis detected 177 QTL at the same level of significance. Possible reasons for detecting a larger number of QTL in within-sire analysis were the probability of identical QTL detected in multiple families and the increase in power of detection for QTL segregating in a single family due to reduction of error variance. Most substitution effects ranged from 0.6 to 1.0 genetic standard deviation. This genome scan study confirmed several already published QTL but most QTL were original, particularly for non-production traits. A pattern of QTL detected included identical or proximal markers associated with several correlated traits, which indicated the possibility of the presence of pleiotropic QTL. Fitting a multivariate model in the presence of pleiotropic QTL was expected to yield residuals that minimize violation of the assumptions underlying single QTL single trait models. Data were simulated with the same granddaughter design and marker structure of the genome scan study. EBV for two traits were simulated as the sum of 20 pleiotropic QTL with a genetic correlation range of +/-0.5 among allelic effects. Multivariate maximum likelihood and least squares methods including composite interval techniques were compared to the regression method of the QTL scan study. Comparison was based on number of QTL detected, distance from the true QTL, and residuals derived from fitting predicted genotypes. Based on a false discovery rate of less than 5%, the only significant difference was noted for a larger distance from true QTL with composite interval techniques. Overall, multivariate methods did not show an advantage in QTL detection in granddaughter design data when trait values used were high accuracy EBV with residual correlation due to shared genetic sources.