Most variation between individuals in behavior, physiology, morphology, disease susceptibility and reproductive fitness can be attributed to the segregation of multiple quantitative trait loci (QTLs) with individually small effects, whose expression is conditional on the environment1. Understanding the genetic and environmental factors that cause this variation is of fundamental importance for medicine, agriculture, evolution and the emerging discipline of functional genomics. Genome scans for QTLs have become a cottage industry in the past 15 years, facilitated by the discovery of abundant, highly polymorphic molecular markers in many species and the development of sophisticated statistical methods of analysis2,3. Identifying the genes underlying QTL peaks by positional cloning has proven elusive in all but a handful of cases where the QTLs had larger than usual effects4. On page 1197 of this issue, Binnaz Yalcin and colleagues5 describe a combined genetic and functional strategy for identifying genes corresponding to QTLs with small effects.

The elusive QTLs

There are three main problems facing high-resolution QTL mapping. First, an increasingly large number of recombination events in each QTL interval are required to map the QTLs to decreasingly smaller intervals. Second, individual QTLs have small effects that are sensitive to the environment; therefore, the phenotype of a single individual is not a reliable indicator of QTL genotype. Third, even after whittling down the QTL interval to a region containing several positional candidate genes, one must determine which of them corresponds to the QTL. Guessing (the 'candidate gene' approach) has worked in cases where the genetic basis underlying the trait phenotype is well understood. But the availability of complete genome sequences for many organisms means it is possible to peruse the gene list in a candidate QTL region, inevitably finding many predicted genes of unknown function and a handful of known genes with no a priori relationship to the trait. This is particularly true for variation in complex behaviors, which have been shown in Drosophila melanogaster to be attributable to subtle variants in pleiotropic genes6,7. Alternative strategies such as scanning genes in the region for potentially causal sequence polymorphisms or differences in transcript abundance are not optimal. Even if only two lines are used to map the QTLs, there will be many sequence differences between them, and all will be associated with variation in the trait. It is not yet certain to what extent differences in transcript abundance will correlate with differences in gene expression at the level of trait phenotype, and small differences in expression confined to a few cells or tissues may be relevant but not detectable with current technology.

Historical haplotypes

Anxiety-like behavior is a classic quantitative trait in mice. When placed in a brightly lit open-field arena, reactive or anxious mice will freeze in place and defecate, whereas nonreactive mice will explore their environment and defecate less. Thus, measures of locomotion and numbers of fecal pellets in the open-field environment are two negatively genetically correlated traits that can be combined into a single index of 'emotionality', EMO8.

Previously, a QTL that affects EMO was mapped to a 0.8-cM region on chromosome 1, using a heterogeneous population of mice derived from a cross of eight standard inbred lines8. Yalcin et al.5 argued that the increased historical recombination in an outbred population would enable even finer mapping of this QTL. They scored anxiety phenotypes for 729 individuals from the commercially available MF1 outbred strain and determined single-nucleotide polymorphism (SNP) genotypes for 42 markers spanning the 3.5-Mb region to which the QTL mapped. Tests for association of each SNP genotype with variation in anxiety identified a single weakly significant marker, after adjusting the experiment-wide significance level to account for multiple tests. But consideration of associations of inferred progenitor haplotypes with differences in trait phenotypes can be more powerful than single-marker associations9 (Fig. 1). This multipoint mapping method was applied to the haplotypes of all 729 individuals, inferred by considering them a fine-grained mosaic of either four or eight inbred strains. The single QTL fractioned into three closely linked QTLs, each of which contributed ∼5% of the total phenotypic variance and which did not interact epistatically. The first and third QTLs contained no known genes. The 95% confidence interval for the second QTL contained only two genes, regulator of G-protein signaling 2 (Rgs2) and Rgs13.

Figure 1: The power of haplotype mapping compared with single-marker analysis.
figure 1

The haplotypes of four inbred strains are depicted for two SNPs, which flank a causal QTL that is associated with variation in high (H) and low (L) EMO between the strains but which has not been genotyped. There is no association between mean anxiety levels between the two SNPs considered individually. Reconstructing the haplotypes partitions the strains into two groups with divergent phenotypes, indicating there is a QTL between the typed SNPs that affects EMO.

Full size image

An anxious gene

Quantitative complementation tests to mutations at positional candidate genes have been used in D. melanogaster to identify genes that functionally interact with, and possibly correspond to, the QTL alleles3. The test requires strains containing a mutated (m) and a wild-type (+) allele of the candidate gene, preferably in the same coisogenic background. These strains are then crossed to the strains containing the alternative QTL alleles (high or low) and the four F1 progeny genotypes (high/m, low/m, high/+, low/+) are scored for the trait. Quantitative failure of the mutation to complement the QTL alleles is inferred if the difference in mean phenotype between the high/m and low/m genotypes is greater than that between the high/+ and low/+ genotypes (Fig. 2). This is detected as a statistical 'Cross' (m or +) by 'Line' (high or low) interaction in a two-way analysis of variance.

Figure 2: The quantitative complementation test.
figure 2

Anxiety-like behavior is assessed by an index (EMO) combining ambulatory behavior and defecation in the open-field environment for four genotypes: m/high, m/low, +/high and +/low. (a) Quantitative complementation. There is a difference in mean performance of the two parental strains when crossed to both mutant and wild-type; and, if the mutation is not completely recessive, there is a difference between mutant and wild-type strains when crossed to both high and low strains. The magnitude of the difference in phenotype between the two parent strains is the same in the background of the mutation and the wild-type allele, indicating the mutated allele does not interact with the QTL alleles. (b) Quantitative failure to complement. Here the magnitude of the difference in phenotype between the two parent strains is greater in the background of the mutation than the background of the wild-type allele, indicative of an interaction between the mutated allele and the QTL alleles.

Full size image

In the first application of this method to mammals, Yalcin et al.5 showed that a recessive knockout mutation of Rgs2 failed to complement the parental high- and low-anxiety QTL alleles. Rgs2 is widely expressed in the brain, and individuals homozygous with respect to the mutation are more anxious than wild-type mice. Failure to complement, whether qualitative or quantitative, can arise from an allelic interaction between the mutation at the candidate gene and the homologous QTL alleles, or from an epistatic interaction between the candidate gene mutation and nonhomologous segregating QTLs. In either case, however, Rgs2 is implicated as modulating natural variation in anxiety.

From QTLs to genes

Because the outbred strain of mice used in this study captures much of the allelic diversity among common inbred strains and is commercially available, it should be possible to use the method of probabilistic ancestral haplotype reconstruction to map QTLs affecting any quantitative trait in mice with high resolution. Such studies might show that single QTLs fractionate into multiple linked QTLs, as is common in D. melanogaster10,11. Finally, despite the emerging complexity of the genetic architecture of quantitative traits, it is possible to use quantitative complementation tests to identify genes that functionally interact with QTLs and thus modulate variation in the trait. The complementary efforts to generate mutations in all mouse genes will be an invaluable resource for systematically testing all positional candidate genes for interactions with QTLs, not just the few for which mutations are currently available12,13. This strategy is not restricted to mice but can be applied to any organism with the requisite genetic and genomic resources. This bodes well for eventually describing quantitative genetic variation in terms of complex genetics rather than complex statistics.