Local properties of the genome can bias QTL analyses

Abstract

Given the ever-increasing availability of molecular markers, the explosive growth of quantitative trait loci (QTL) mapping studies is not surprising. For the difficult-to-study but important quantitative traits, QTL mapping assists in our understanding of their genetic architecture: the numbers of loci that contribute to the trait, their locations, their relative effect sizes, and the interactions among these genes. QTL mapping studies are also the first step towards the cloning and molecular characterization of the genes that underlie variation in quantitative traits. In recent years, the genetics community has become more aware of caveats and limitations of QTL analyses. For instance, they tend to overestimate the effect sizes of those QTLs with the largest effects, particularly when the sample size is modest (<1000 genotypes assayed). However, most of the known biases are ones of methodology and not of biology. Now, Noor et al.1xSee all References1 argue that properties of the genome can bias QTL analysis. Specifically, they warn that variation in recombination rates and gene density among different regions of the genome can bias substantially the results of mapping studies.This warning is based upon simulations they performed to assess the ability of QTL mapping programs to detect QTLs placed at random in Drosophila melanogaster. They placed QTLs in two different ways. In one, they placed QTLs at random with respect to recombination position (RR). This reflects the assumption that the density of QTLs per recombination unit is constant across the genome. They also placed QTLs at random with respect to the 14 000 coding sequences identified in the complete genome of D. melanogaster (RC). This placement reflects the assumption that the probability that a QTL exists is constant across coding regions, a more reasonable assumption than that reflected in RR. Yet, the tacit assumption of most simulated QTL mapping studies is that QTLs are distributed randomly based on the recombination map. Using single-marker linear regression and composite interval mapping, Noor et al. detect fewer QTLs in RC simulations than in RR simulations. They also find that in the RC simulations, QTLs are detected most often in regions where the ratio of gene density to recombination is high. These results show that, at least for D. melanogaster, the variation in the distribution of the number of coding regions per recombination unit is sufficient to distort the detection of QTLs significantly. These results are robust to changes in several parameters including marker density, heritability and QTL effect sizes.One possible example of this biasing effect is the ‘small X effect’ seen in genetic studies of behavioral isolation between different species of the D. melanogaster group. Differential selection pressures affecting the two sexes has been invoked as a possible explanation for this phenomenon. Noor et al. argue that this phenomenon might just be an artifact of the X chromosome having a lower density of genes per recombination unit than the autosomes. Supporting their contention, they find substantially more X-linked QTLs are detected in the RR than in the RC simulations.This bias can easily be corrected in D. melanogaster studies and the bias seen in D. melanogaster could be a ‘worst case scenario’, because this species has only three major pairs of chromosomes. Biases can be lower in species where the ratio of chromosomes to genes is higher. Still, this study should caution researchers in their interpretations of QTL analyses. As the authors state, ‘results of QTL mapping should be taken as hypotheses to be tested by additional genetic methods, particularly in species for which detailed genetic and physical genome maps are not available’. These results also highlight the continuing importance studying the natural history of genomes.