Discussion
In this study, we described a large, densely mapped, segregating F2 mouse population designed to study the genetic regulation of several traits associated with the so-called metabolic syndrome. Several groups, including ours, have reported the advantage of combining traditional genetics with genome-wide gene expression analysis for the dissection of complex traits. This study improved on past models by including over 300 animals (three times the size of previous studies) of both sexes, allowing for the incorporation of sex-specific effects on underlying genetic regulation.

Significant Sex Bias in the Regulation of Both Complex Traits and Gene Expression
Given the known dichotomy between females and males in the susceptibility and control of obesity, this study was designed to sufficiently power the detection of significant QTLs for this and other traits with sex-dependent effects. Note, however, that these effects can extend to traits without overall mean differences between the sexes. Previous studies have described the advantages of performing QTL analysis both with and without sex as an interactive covariate [22–25]. Analyzing the sexes separately is suboptimal since it reduces sample size in both groups, thus reducing power to detect main QTL effects, as demonstrated by our genome scan of Chromosome 5 (Figure 1B). Furthermore, separate analyses would not allow for the detection of QTLs that have opposing, or sex-antagonistic, effects in females and males and would hinder the detection of QTLs specific to one sex.
Accordingly, we detected five cQTLs for the gonadal fat mass trait on Chromosomes 1, 3, 5, 11, and 19. The detection of all five cQTLs was “driven” by the larger effect in females, with significant improvement by the incorporation of sex*additive and sex*dominant parameters. QTLs associated with obesity, gonadal fat, and abdominal fat have been reported before overlapping with cQTLs on Chromosomes 1 [26–28], 5 [26,29], and 11 [19,29] reported here, whereas the cQTL on Chromosome 3 represents a novel QTL for this trait. The Chromosome 19 cQTL for fat mass was recently reported by us [5] in the BXD intercross F2 progeny from the strains B6 and DBA (which shares the same haplotype at this region as the C3H strain used in this study). Interestingly, significant heritability and genetic regulation was seen in this F2 population despite the hyperlipidemic, proinflammatory ApoE−/− background and the high-fat Western diet. This background possesses several advantages, such as allowing the modeling of human-like disease states. The predominantly female-driven effects of the five cQTLs likely reflect the significant effect of differential gonadal hormone secretions on the genetic regulation of this complex trait.
The identification of genes underlying cQTLs remains a challenge. The widespread availability of genome-wide expression analysis has begun to address this by providing a snapshot of transcription in relevant organs and thus providing initial information for which genes can differentiate a given trait. Furthermore, by treating transcript levels as quantitative traits, we can map the genetic regulation underlying differential gene expression (eQTLs). Those eQTLs that have cis-acting variations affecting their transcription are potential candidate genes for the trait. At a single trait, genome-wide significance level of 0.05, we detected 6,676 eQTLs representing 4,998 genes, of which 2,118 were cis-acting. At increased thresholds, the proportion of cis-eQTLs increased, which is in good agreement with previous studies [5,15] and likely reflects the increased power to detect cis-acting variations affecting transcription. Of all 6,676 significant eQTLs, 1,166 possessed significant sex interactions. Of these, 304 were cis and 852 were trans, suggesting that only a minority of the sex-specific effects on the regulation of gene expression occur through polymorphisms within the gene itself. Rather, underlying genetic regulation of most transcripts is the result of interactions between trans loci and sex-specific factors (e.g., hormones). As with cQTLs, sex bias in the predominantly trans genetic regulation of gene expression is likely secondary to different sex hormone profiles.
Recently, using a similar dataset, our group demonstrated that significant cis-eQTLs (p < 5 × 10−5) largely represent true positives [30] and are enriched for highly polymorphic regions over the mouse genome. The cis-eQTLs presented in Table 5 overlap with one of the gonadal fat mass cQTLs and should be considered potential candidates. Given the sex effects in the gonadal fat mass cQTLs, we reasoned that the cis-eQTLs with significant sex*additive and sex*dominant effects should receive priority consideration. The use of eQTLs to dissect cQTLs is a method still in its infancy, with uncertain efficacy and applicability. Nevertheless, application of this analysis to this dataset provides some tantalizingly attractive candidate genes.
One shortcoming of this approach, however, is that candidate genes are limited to those whose transcript expression levels vary in association with a nearby polymorphism that differs between the parental strains—in other words, genes with significant and detectable cis-eQTLs. However, it is not strictly necessary for candidate genes to have evidence of such linkage: polymorphisms underlying a trait cQTL can affect gene function or posttranslational modifications. Nevertheless, several phenotypes are known to be regulated, at least partly, at the level of transcription or mRNA stability, which is exactly what our methods are designed to detect. A separate problem is that organ-specific gene expression differences may preclude one from detecting the relevant causative gene if the tissue arrayed is not the tissue where the control is exerted. This is particularly relevant for a trait such as adipose tissue mass, which is controlled by multiple tissues. We propose that analysis of correlated genes can provide guidance as discussed below.

Genes Correlated with Gonadal Fat Mass Illustrate Tissue-Specific Regulation of the Trait
In an effort to identify genes associated with the fat mass trait, but not necessarily candidate genes underlying the trait cQTLs, we fitted linear models to assess the degree of association between transcripts and gonadal fat mass. As with QTLs, sex-specific correlations were modeled. At an FDR of 1%, 4,613 genes were found to be significantly correlated with gonadal fat mass, of which 4,254 (98%) showed sex-biased correlation. As indicated in Tables 4 and 5, several genes with detectable cis-eQTLs are also significantly correlated with the trait and are even further prioritized as candidate genes.
Thus far, studies that have examined the “genetics of gene expression” are in good agreement regarding the increased power to detect cis-eQTLs relative to trans [5,9,15,16,31]. It is unclear at this time, however, what exactly is the significance of trans-eQTLs and the nature of the underlying polymorphisms associated with them. Furthermore, the eQTL hotspots reported in this and previous studies [5,9,10] largely represent trans-eQTLs. This localization suggests some functional significance to these regions. Of the 4,613 genes correlated with gonadal fat mass, 1,130 generate 1,478 significant eQTLs, of which 1,023 (69%) are trans-acting. These eQTLs are significantly enriched at one locus (Chromosome 19). Interestingly, this hotspot was coincident with a cQTL associated with fat mass reported in this study. Since these transcripts represent those significantly correlated with gonadal fat mass, the localization of their eQTLs to these regions strongly supports the notion that the genes with trans-eQTLs represent downstream targets of candidate regulatory genes located at the position of significant linkage. This means that the genes may be causal but downstream of the gene responsible for the cQTLs, or they may be reacting to the increased gonadal fat mass and associated metabolic changes. These data also suggest that identifying such loci that show overrepresentation of highly correlated genes is a means to identify which of the trait cQTLs are more likely controlled by the tissue arrayed. As expected, the Chromosome 19 locus was enriched for trans-eQTLs with substantially greater effects in females. Functional and promoter analysis of genes with a common trans-eQTL may prove enlightening. Furthermore, gene expression network construction and analysis may be improved by the incorporation of experimentally demonstrated cis versus trans regulation.

Conclusion
The integration of traditional genetics with genome-wide expression analysis was first proposed by Jansen and Nap 4 y ago [32]. Advances in genomic technology and bioinformatic resources since then have vastly improved the applicability of these methods to the dissection of complex traits. Taking into account sex-specific effects will similarly improve the sensitivity to detect underlying genetic regulation, especially for phenotypes known to be affected by sex. Furthermore, network analyses are increasingly being applied to complex phenotypes [11,33]. Regardless of which variables are used in the construction of these networks, whether they measure gene expression or protein interactions, accounting for sex specificity, hormonal status, or construction of different networks for females and males would likely more accurately represent the complexity associated with these phenotypes.
We reported here on the initial genetic and genomic analysis of an F2 intercross population designed to recapitulate several traits associated with human metabolic syndrome. Using 334 mice of both sexes genotyped at high density, this is the largest study of its kind to date designed, and it is strongly powered to detect subtle effects of genetic regulation and sex specificity. We identified five cQTLs for the gonadal fat mass trait, all with greater effects in females. We also detected several thousand significant liver eQTLs, a significant fraction of which are sex-biased, demonstrating how meaningful effects of sex on gene expression extend beyond overall mean differences. We demonstrated the application of linkage and correlation methods to identify candidate genes. Finally, we showed that localization of a subset of liver genes linked in trans to a cQTL region can identify relative tissue contributions to the genetic regulation of a complex trait. We anticipate that the application of these and similar methods would significantly improve the elucidation of the genetic regulation underlying complex phenotypes.