Modelling QTL effect on Bos taurus autosome 6 using random regression test day models

This PhD project is realized by Tomasz Suchocki
Supervisor Joanna Szyda

The genetic variation of quantitative traits is attributed to the combined effect of several loci with large effects (QTL – Quantitative Trait Locus/Loci) and many loci with a small effect each – jointly treated as a polygenic effect. The QTL effect has been incorporated into statistical models, either as a fixed term (e.g. Knott and Haley, 1992) or as a random term (e.g. Fernando and Grossman, 1989), but up to now it has been mainly considered as a time-independent variable. However, for many traits recorded repeatedly it is very interesting to investigate the behaviour of QTL and polygenes across time.

For dairy cattle, one of the first models for test day records enabling the incorporation of time-dependent variation was proposed by Ptak and Schaeffer (1993). In this model, a so-called fixed regression model, an additive genetic variance of polygenes is still assumed to be constant across time, but some fixed effects are modelled as variable across lactation stages. A further development comprises a so-called random regression model, which additionally allows for time-dependent modelling of random effects by assuming that variances of additive genetic and permanent environmental effects are variable across lactation (e.g. Schaeffer, 2004). However, none of these models considers the effects of quantitative trait loci (QTL).

Consequently, the major goal of this study is to estimate the position and effect of QTL for milk, fat and protein yields, as well as for somatic cell score based on test day records, and to test whether the effects of QTL and polygenes are constant or variable throughout lactation. The analysed data set originates from the Chinese Holstein-Friesian dairy cattle population, and consists of 23 paternal half-sib families (716 daughters of 23 sires). For interval QTL mapping a sequence of three models is used. The first one is a simple lactation model and the next two are based on random regression models. In the first random regression model a QTL effect is incorporated as constant in time and in the second as variable in time.

The analysis shows that for each production trait at least one QTL significant after correction for multiple testing (Bonferroni) is detected. For milk and protein yield only the model with QTL effect variable in time is able to report significant QTL, while for fat yield each of the models discovers significant QTL positions. The QTL for milk yield is located at the 52nd cM, near the marker ILSTS035. For protein yield there exist two significant QTL positions at the 33rd and the 56th cM, in the vicinity of markers BMS470 and BMS2460, respectively. For fat yield the QTL is located at the 18th cM, near BM413. Additionally, the investigation for milk and protein yield confirms that effects of these QTL are variable in time.

Likelihood ratio test profiles for model with longitudinal QTL effect and all analysed traits.

The novel QTL detection method based on a longitudinal model is an appropriate and powerful tool for the detection of QTL, which effect vary in time. When a QTL is incorporated into a model as constant in time, the effect of this QTL is averaged over the whole lactation stages and may thereby be difficult to detect. On the other hand, in longitudinal studies, variance components are estimated across the whole length of the time period and thereby use the information included in each observation across lactation.