Defining QTL Boundaries from GWAS Results

Although it is necessary to identify candidate genes, there is actually no well spread method to define QTL boundaries from genome wide association studies (GWAS). Indeed, GWAS results are mostly shown using simple Manhattan plots because in an association panel the link between linkage disequilibrium (LD) and genetic or physical distances is much more complex than in a biparental population.  Thus, authors fixed a more or less arbitrary window around associated markers, often based on mean LD decay. But massive variation of LD exists along the chromosomes.

Moreover, authors control the error Type I (false positive) using a p-value threshold or through method based on p-value distribution to define significant marker-trait association. But the way QTL boundaries are designed can also lead to false positive. Thus, the real error Type I is composed of two errors, (1) a QTL is defined although the initial marker-trait association is not real, (2) the marker trait association is real but the QTL boundaries do not contain the causal mutation.

We used genotyping data (~24,000 SNP) from a 214-variety elite wheat panel defined for association mapping. We assessed by simulation the power of detection and false discovery rate of several QTL construction methods depending on traits heritabilities, locus explained variance, and p-value threshold. Results revealed that real error Type I is much higher than expected and mainly controlled by the ability to well-define QTL boundaries. Methods are discussed regarding QTL coverage, real error Type I, and power of detection balanced altogether.