Genetic Architecture of Adiposity and Organ Weight Using Combined Generation QTL Analysis




We present here a detailed study of the genetic contributions to adult body size and adiposity in the LG, SM advanced intercross line (AIL), an obesity model. This study represents a first step in fine-mapping obesity quantitative trait loci (QTLs) in an AIL. QTLs for adiposity in this model were previously isolated to chromosomes 1, 6, 7, 8, 9, 12, 13, and 18. This study focuses on heritable contributions and the genetic architecture of fatpad and organ weights. We analyzed both the F2 and F3 generations of the LG, SM AIL population single-nucleotide polymorphism (SNP) genotyped with a marker density of ∼4 cM. We replicate 88% of the previously identified obesity QTLs and identify 13 new obesity QTLs. Nearly half of the single-trait QTLs were sex-specific. Several broad QTL regions were resolved into multiple, narrower peaks. The 113 single-trait QTLs for organs and body weight clustered into 27 pleiotropic loci. A large number of epistatic interactions are described which begin to elucidate potential interacting molecular networks. We present a relatively rapid means to obtain fine-mapping details from AILs using dense marker maps and consecutive generations. Analysis of the complex genetic architecture underlying fatpad and organ weights in this model may eventually help to elucidate not only heritable contributions to obesity but also common gene sets for obesity and its comorbidities.


Obesity is influenced by a combination of environmental and genetic factors. Although some cases of monogenic obesity have been characterized (1,2,3), heritable contributions to obesity are typically due to multiple genes of small effect (4,5,6). A number of different crosses have identified quantitative trait loci (QTLs) contributing to body size and obesity with more than 300 QTLs identified in murine crosses (6).

Overall body size can be conceived of in a modular fashion, with contributions from skeletal structure, organ, muscle, and fat mass (7). These diverse components are only weakly correlated (8) indicating that they are affected by largely nonoverlapping sets of genes. In previous studies of the LG/J and SM/J intercross populations, body weight, fatpad weight, and organ weights were moderately heritable and phenotypic and genetic correlations among body composition traits were low, particularly when controlled for their common correlations with body weight. This quantitative genetic result was reinforced by QTL results showing mostly independent loci for different aspects of body composition (4,8,9). Similar results were observed in a study of organ weights and limb bone lengths in mice from the CAST/Ei by M16i cross (7,10). Understanding the genetic architecture and molecular networks underlying various body size components will aid in our understanding of the physiology of obesity and in understanding and treating its various comorbidities.

Epistatic interactions are expected to contribute significantly to the genetic variation of complex traits. In Cheverud et al. (4), eight adiposity QTLs interacted epistatically with each other and accounted for a significant increase in the amount of heritable variance explained when combined with the direct effects. Adip8 epistatically interacted with all seven of the other adiposity loci. Only epistatic interactions between previously identified direct-effect QTLs were characterized in this study, and it is likely that many two-way epistatic interactions remain unidentified.

Fine-mapping of the relatively large F2-based QTL regions is required to narrow the support intervals and identify causative sequence changes. Several strategies are available to narrow confidence intervals, including the use of congenic lines (11,12,13), recombinant inbred lines (14,15), and advanced intercross lines (AILs) (16,17). Here, we characterize genetic contributions to variation in body weight, organ weights, tail length, and reproductive fatpad weight in the LG, SM AIL F2 and F3 combined mouse population as a first step in fine-mapping our adiposity QTLs. Combining the F2 and F3 generations increases the sample size from 500 to 2,100 animals. Previous studies in the Cheverud lab used a relatively sparse microsatellite marker map (∼25-cM intermarker intervals) whereas our new genotype set is based on a dense, 4 cM, single-nucleotide polymorphism (SNP) map which allows further resolution of mapped QTLs when combined with the increased sample size. We also characterize the epistatic contributions to variation in body composition traits with a full two-way genome scan.

Methods and Procedures

Animal husbandry

Details of animal husbandry and breeding of the LG, SM AIL can be found in Kramer et al. (18). Animals were killed at 17.9 weeks ± 3.9 days. Weights were obtained for the body at necropsy, reproductive fatpad (epididymal in males and perimetrial in females), liver, kidney, heart, and spleen to the nearest 0.01 g. Tail lengths were measured to 0.01 mm using digital calipers. Before analysis, the variance effects of litter size, date of birth, age at necropsy, generation, and sex were removed (19) from the measurements and the residuals used in the mapping analysis.


DNA was extracted using the Qiagen DNeasy 96-well DNA extraction kit (Qiagen, Valencia, CA) from 503 animals from the second LG/J by SM/J F2 intercross population and 1,595 of their F3 progeny. These animals were SNP genotyped for 370 polymorphic SNPs (Supplementary Table S2 online) using the Illumina GoldenGate platform (Illumina, San Diego, CA). The average intermarker interval across the genome was 4 cM. Map distances were generated using the Haldane's algorithm in R/qtl (20).

One-QTL mapping analysis

We performed regression interval mapping (21) as described previously (22)


where Yij is the phenotype of interest, a is the additive genotypic value, d is the dominance genotypic value, Xai is the additive genotype score and Xdi is the dominance genotype score for individual i, eij is the residual, and μ is the constant. Genotype scores were imputed every 1 cM across the entire genome, with markers interspersed at their mapped locations. Genotypes between markers were estimated from the information from the flanking markers and estimated recombination frequencies in each interval between markers (21). We regressed trait values onto the genotypic scores at each marker and imputed location using the SETCOR function in Systat v11.0 (Systat, San Jose, CA). Probabilities were transformed into logarithm of probability (LPR) scores (LOD-equivalent):


Chromosome-wide and genome-wide significance levels were generated via simulation in R. Briefly, we programmed a randomly generated F1 population from the LG/J and SM/J parental populations to provide parental haplotypes. Retaining the experimentally determined family structure (number of progeny for each set of parents), we randomly assigned recombined F2 parental haplotypes to generate the genotypes of the progeny. Thus the genotypic structure of family relationships was maintained. In order to generate the distribution of LPR thresholds under the null hypothesis of no QTL effect, we performed each analysis 1,000 times using the computationally randomized genotypes, retaining the original phenotype sample structure. The genome-wide multivariate 5% significance threshold is an LPR of 7.02. This threshold accounts for multiple comparisons and familial autocorrelation in the sample. A two-QTL model was run for each chromosome that tested significant in the one-QTL test, and was accepted if the fit was significantly (P < 0.05) better than the one-QTL model.

Sex-interaction QTLs were identified using the one-QTL model controlling for locus effects and sex effects. At locations where significant (P < 0.05) sex-interaction QTLs were identified, the sex affected was determined by separate sex analysis. Pleiotropy was tested for as in Ehrich et al. (23). Briefly, the most likely positions for the univariate (linkage) model QTLs were identified as were the pleiotropic, or multiple trait, model QTL locations. The residual sums of squares matrices were then compared for the pleiotropy vs. linkage test as in Knott and Haley (24). Significantly different results indicate that the pleiotropic model can be rejected and that members of the trait set are affected by different, linked genes.

Pleiotropic QTLs were named based upon the traits that were affected by that QTL as well as the chromosome and position. Replicated QTLs used previously published names (4,8). The Adip nomenclature was used for all traits affecting fatpad weight. Wtn loci affect body weight and any other trait except for fatpad or tail length. Bod loci affect overall body size as indicated by effects on soft tissue traits as well as tail length. Org loci affect only organ weights whereas Skl loci affect only tail length. If a potential name was used in a previous publication for a particular chromosome and set of traits, but the QTL location was not replicated, the digit after the period was increased by one.

Epistatic mapping analysis

We performed a two-way epistasis analysis interchromosomally every 1 cM along each chromosome (25). The dependent variables were regressed onto the four interaction components (Xa 1 × Xa 2, Xa 1 × Xd 2, Xa 2 × Xd 1, and Xd 1 × Xd 2) holding the additive and dominance (“Xa” and “Xd,” respectively) genotypic scores at the two loci constant. Therefore, the model for the epistasis scan was as follows:


where Yijk is the dependent variable and μ is the constant.

Significance thresholds for epistasis were generated for several different categories of tests based on prior knowledge of main effects: QTL by QTL, QTL by chromosome, chromosome by chromosome, QTL by genome, and genome by genome. We accounted for multiple tests and family structure when generating adjusted Bonferroni 5% thresholds. Because of computational limitations, we estimated the thresholds as follows. We calculated the effective number of markers, or the M eff (25), for each category. The QTL by QTL tests had one effective comparison. The number of independent comparisons for the QTL by chromosome tests was taken to be the effective number of markers on the chromosome in question whereas the number of independent comparisons for the chromosome by chromosome tests was the product of the effective number of markers on the pair of chromosomes considered. Thresholds were then estimated using the following empirically based equations using the heritability obtained from the correlation among full-sibs in the population:


where h 2 is the heritability calculated using the among-litter variance approach (Ehrich et al. (25)). QTL by QTL thresholds were as follows by trait: fatpad (1.590), body weight (1.731), tail length (1.787), heart (1.450), kidney (1.597), spleen (1.436), and liver (1.576) weights. QTL by chromosome thresholds ranged from 4.018 to 5.204. Typical chromosome by chromosome thresholds ranged from 3.27 to 5.61. Genome by genome thresholds were as follows: fatpad (7.146), body weight (7.885), tail length (8.180), heart (6.406), kidney (7.183), spleen (6.333), and liver (7.072) weights.

Loci identified during the epistasis scan that were not within the confidence interval of a direct-effect QTL were given unique names. Each of these loci was named with “Ep” to denote its identification in the epistasis scan, followed by the chromosome number, a period, and then a number to designate which epistatic QTL on that chromosome was referred to specifically.


Single-trait QTLs

We identified 111 single-trait QTLs, 92 of which were significant at the genome-wide level (Supplementary Table S1 online). We identified the following number of QTLs per trait: fatpad (20), body weight (19), tail length (19), heart (12), kidney (18), spleen (12), and liver (13). The two-QTL model was a better fit than the one-QTL model for at least one trait on chromosomes 1, 4, 6, 7, 8, 10, 11, and 12.

Sexual dimorphism

We identified sex-interactions on chromosomes 1–9, 11–13, 16, and 17. Forty percent of all the QTLs were sex-specific and the vast majority of these were male-specific (72.5%). Sixty-four percent of these sex-interaction QTLs have a significant QTL peak in one sex but no significant peak in the other sex. In 36% of the sex-specific loci, we observed significant peaks for both males and females, but with at least a 1.5-fold higher LOD score for one of the sexes (usually males). Seventeen of the sex-specific loci showed significant opposite effects in the two sexes.

Overall, there was more phenotypic variance explained for males than for females, consistent with the high proportion of male-specific QTLs identified. For each trait, male and female variance explained (respectively) by the QTLs identified in this study tended to be ∼10–40%: fatpad (22.3%, 19.9%), body weight (34.5%, 26.3%), tail length (38.0%, 37.2%), heart (9.5%, 9.5%), kidney (27.7%, 19.5%), spleen (12.6%, 22.5%), and liver (22.2%, 14.5%). The only trait in which more variance was explained in females was spleen weight.

Gene effects

Most single-trait QTLs are additive in their effects. As expected from earlier work on these strains, the additive genotypic values were relatively small (Figure 1, Supplementary Table S1 online). Maximum gene effects were <0.5 s.d. units. Most often, the LG/J allele results in a larger phenotypic value but there were exceptions with larger phenotypic values for the SM/J allele at the following SNPs: rs13477230 (fatpad), rs4224277 (kidney), rs13477996 (heart, kidney), rs3667067 (fatpad), gnf05.069.163 (kidney), rs3669563 (fatpad, body weight, tail length, kidney, liver), rs8254378 (spleen), rs13459120 (body weight, heart, kidney, spleen, liver), rs13481075 (fatpad, body weight, spleen), rs6335879 (fatpad), rs13481361 (tail length), rs3679276 (fatpad), and rs13483409 (heart, kidney). Adip5 (rs3669563) and Adip6 (rs6335879) were both previously determined to exhibit smaller fatpad size with the LG/J allele (4).

Figure 1.

Gene effects histograms. Normalized additive and dominance histograms demonstrate that additive effects are small and dominance effects are most commonly codominant. The graphs represent all phenotypes analyzed in this study. (a) Absolute values of additive gene effects normalized to s.d. (a/s.d.). Absolute values are presented because of the relatively low prevalence of negative additive gene effects scores. (b) Dominance gene effects normalized to additive gene effects (d/a) are displayed. Very low nonsignificant additive values may falsely inflate dominance scores, and so dominance relationships meeting these criteria were excluded from this analysis.

Approximately 29% of the single-trait loci in this study showed significant dominance effects. Overdominant and underdominant loci were defined as being greater than 1.5 or less than −1.5, respectively. Codominance occurs when −0.5 < d /a < 0.5. Dominance of the LG/J allele was observed more commonly than dominance of the SM/J allele by approximately twofold. Five QTLs exhibited overdominance (1.5 < d/a). LG/J allele dominance affected 11 loci contributing to all studied traits. Codominance was observed for seven single-trait QTLs with significant dominance effects affecting all traits except for heart and spleen weights. SM/J allele dominance in six single-trait QTLs affected all traits except for liver weight. Underdominance was present in only three QTLs affecting heart weight, kidney weight, and tail length.

Pleiotropic QTLs

Pleiotropy is defined as a single locus contributing to variation for more than one trait. We tested all of our single-trait QTLs that resided within 30 cM of another single-trait QTL to determine whether pleiotropy could be rejected. When the separate-trait QTL model failed to be a significantly better fit to the data than a pleiotropic locus model, the multivariate pleiotropic location was chosen as the QTL location for the multiple traits considered. QTLs were named based upon the traits that were affected at each locus and also in keeping with previously identified loci (see Methods and Procedures) (4,8,9). We identified 33 separate QTLs (27 pleiotropic and 6 single-trait QTLs) in this study based on the original 111 single-trait QTLs (Figure 2, Table 1, Supplementary Table S1 online). Twenty-five of these thirty-three QTLs were significant at the genome-wide level with eight suggestive QTLs (Adip12, Adip3A, Bod10.2, Adip17, Bod15.1, Skl16.1, Adip19, and Org16.1) significant at the chromosome-wide level. Nineteen of the pleiotropic QTLs exhibit sex-specific effects on at least one trait at that locus. Confidence intervals for pleiotropic loci were narrow, with a median size of 8.5 cM and the QTLs were distributed throughout the entire genome (Figure 2).

Figure 2.

Chromosomal quantitative trait locus (QTL) schematic. Chromosomes 1 through 19 are presented with single-nucleotide polymorphism locations indicated by cross-hatches. Centromeres are all oriented to the top of the figure. Pleiotropic QTLs are displayed in their chromosomal locations. Sex specificity is indicated by either a male sign or a female sign to the right of the QTL exhibiting the sex specificity.

Table 1.  Pleiotropic QTL table
inline image

Traits within pleiotropic QTLs tended to segregate into groups representing overall body size (7) and adiposity (20). Most sex-specific single-trait QTLs were identified as pleiotropic with sex-specific QTLs for other traits. QTLs affecting fatpad typically also affected at least one of the other six traits, most often body weight (75%). A couple of QTLs (Skl8.1 and Skl16.1) affected only skeletal size as represented by tail length. Four QTLs contributed only to a single organ weight (Org4.2, Org6.2, Org7.2, and Org16.1). Typically, any QTLs that affected body weight also affected either organ weights or fatpad weight. Most pleiotropic QTLs affecting body weight affected both skeletal and soft tissue aspects of body size with limited effects on heart and spleen weights. Exceptions to this pattern were QTLs affecting only soft tissue traits in addition to body weight (Adip2, Adip3A, Adip14, Wtn11.1, Bod15.1, Adip19, and Adip8).

Sex-specific effects in the pleiotropic QTLs were not always consistent in the sex affected. Of the QTLs with at least one trait exhibiting sex specificity, 31.6% displayed both male- and female-specific effects on different traits. We also observed 68.4% of pleiotropic QTLs where only a fraction of the traits affected by the QTL showed sex specificity. In only a few of the sex-specific pleiotropic QTLs (15.8%) did all traits affected by that QTL exhibit sex effects on the same sex.

Replication of direct effects

This study validated a large number of direct-effect QTLs originally detected in the F2 generation, contributing to each of the seven traits that we examined. Approximately 70% of the QTLs identified in this study are replicates of QTLs identified previously in F2 and recombinant inbred lines studies (Table 1) (4,8,9). Seven of eight of the Adip QTLs replicated from Cheverud et al. (4). Replication rates were lower for QTLs identified from the recombinant inbred lines (9), where 24 of the 34 single-trait QTLs replicated. In the combination of the two LG, SM AIL F2 populations (8), 60.0% of the body weight QTLs and 66.7% of the organ weight QTLs replicated. Primarily, we were interested in the replication of the Adip loci (4) affecting fatpad weight and body weight. We observed verification of previous F2 data (Intercross II) suggesting that Adip3 consisted of two flanking QTLs rather than one single centrally located QTL (19). Overall, many more significant effects were identified here relative to earlier studies with fewer animals and markers.


We identified a total of 78 epistatic interactions between 88 epistatic loci, each contributing to at least one obesity-related trait (Figure 3, Supplementary Table S3 online). Supplementary Table S3 displays all of the epistatic loci. Only one epistatic interaction, affecting tail length, was identified that exceeded the full genome-by-genome threshold (LPR = 8.18). We identified epistatic interactions for each of our seven traits. The number of epistatic interactions per trait is: 14 (fatpad), 14 (body weight), 16 (tail length), 7 (heart), 9 (kidney), 15 (spleen), and 3 (liver). Only four epistatic interactions affected more than a single trait. None of the epistatic interactions affected more than two traits. Our stringent Bonferroni-adjusted significance thresholds permit identification of only a fraction of the probable existing epistatic interactions.

Figure 3.

Epistatic interaction networks. (a) Epistatic interactions between direct-effects loci are displayed for fatpad weight. All interactions are significant (see Supplementary Table S3 online). (b) Suggestive epistatic interactions. The displayed interaction locus peaks reside with 5 cM of the peak within the previous F2 confidence interval of the loci, but do not pass Bonferroni thresholds for significance.

Sixty-one of the seventy-eight epistatic interactions exhibited significant interaction effects. Standard deviation-normalized epistatic gene effect values are presented in Supplementary Table S3 online. The 17 epistatic interactions that did not display at least one significant epistatic term were typically affected by more than one term, whose combined effects underlie the significant overall epistatic interaction. Thirty-four of the epistatic interactions were additive-by-additive with values in the range of 0.08–0.23 s.d. units. Thirty-seven epistatic interactions displayed additive-by-dominance or dominance-by-additive epistasis with values ranging from 0.12 to 0.31 s.d. units. Twenty-four interactions displayed dominance-by-dominance epistasis with values between 0.16 and 0.40 s.d. units. The greater than expected dominance-by-dominance epistasis that we observed was mostly due to spleen weight. These genotypic values are similar in size to those observed for the direct effects (Supplementary Table S3 online). Intrachromosomal epistatic interactions were not examined. The relative contributions to genetic variance of the epistatic effects for each trait were: fatpad (3.8%), body weight (5.0%), tail length (4.2%), heart (2.4%), kidney (3.7%), spleen (4.2%), and liver (1.4%). As expected, the total epistatic contributions to the phenotypic variance for each of these traits were smaller than the direct effects. On average, the epistatic contribution to variance is ∼15.7% of the total variance explained (Supplementary Table S3 online).

Epistatic interactions involving direct-effect loci affecting fatpad weight are displayed in Figure 3. Two of the loci (Adip11 and Adip8) display two or more interactions, suggesting that these loci may function as regulatory or transcriptional modulators in the molecular networks underlying the development of obesity. Four of the interactions existed only between two loci. The direct-effects data combined with the epistatic interaction data demonstrate that there are a large number of genes contributing to obesity with a complex genetic architecture.

Replication of epistasis results

The strict Bonferroni-corrected significance thresholds that we used in this study inhibited our ability to observe replication of epistatic effects from previous studies (4). Several of the Adip epistatic interactions previously reported resided within 4.1 cM of the peak of our direct effects replicated loci. The interactions where this was observed included: Adip1Adip8, Adip2Adip8, and Adip3Adip8 (Figure 3). Although these results do not meet Bonferroni criteria for replication, they may be suggestive of true interactions that do not pass statistical significance due to lack of power.


We identified a large number of single-trait QTLs (>100) contributing to obesity-related traits measured at necropsy. Ample evidence of complex genetic architecture was observed contributing to fatpad weight, body weight, and organ weights. Both the nature of the gene effects and the relationships of the traits and QTLs were multifaceted. The majority of gene effects were additive, but significant dominance effects were observed in ∼30% of the QTLs. Overall, the LG/J allele contributed to an increased phenotypic value relative to the SM/J allele but the SM/J allele contributed to a larger phenotype in ∼25% of the single-trait QTLs. Most of the single-trait QTLs could be condensed into 27 pleiotropic QTLs, with only 6 single-trait QTLs. About half of the single-trait QTLs exhibited epistatic interactions either with other direct-effect QTLs or with previously unidentified genomic regions. Overall, these results display a complexity reminiscent of the suspected molecular complexity underlying obesity and the metabolic syndrome.

A large proportion (40%) of these QTLs exhibited sex specificity. The QTLs were mostly male-specific, which is unsurprising given previous results indicating sexual dimorphism in body size (23). Males also responded more strongly than females to a high-fat diet as indicated by worsening obesity and diabetes-related measures (18,23,26). About half of the QTLs for organ weights, body weight, and tail length that replicated positions between the current study and the study by Kenney-Hunt et al. (8) reflect the same pattern of sex specificity. The differences between the studies for body weight and organ weight sexual dimorphism are likely to be the result of our larger sample size and enhanced resolution.

Our pleiotropy findings reflected the intertrait correlations observed in Kramer et al. (18). Correlations were highest between body weight and fatpad weight (0.77) and significantly lower for organ weights with any other trait (median correlation is 0.32). Our results matched expectations from phenotypic correlations reported previously (18) that fatpad (74.1%) and body weight (70.4%) would be most commonly pleiotropic and that kidney weight and liver weight would exhibit more pleiotropic relationships than either heart or spleen weight. Both kidney weight and liver weight are indicative of known obesity comorbidities, so the apparent overlap of heritable contributions to these traits with fatpad weight or body weight is intriguing. We anticipate that a number of these pleiotropic QTLs may segregate into multiple QTLs in future fine-mapping studies.

The current analysis method confers a number of advantages over a traditional F2 QTL mapping study. First, we identified many more QTLs per trait than were observed using solely an F2 population with the same parental strains. Second, we observed an increase in the statistical support for each of these QTLs, with an increase in mean LPR for adiposity-related QTLs of 9.7 over Cheverud et al. (4). Average LPR values increased as a result of increased sample size, increased marker density, and confounding familial autocorrelation. The mean LPR for all traits (11.46) far exceeded the genome-wide significance threshold for all traits (7.02), which is a larger proportion of LPR exceeding the threshold than is typically observed in F2 studies of these traits (4,8). We obtained a greater proportion of genome-wide significant results, with 79.6% of single-trait QTLs significant at the genome-wide level compared to the same traits in previous F2 studies (4,8). Due to typical F2 confidence interval sizes of ∼10–40 cM, it is often difficult to resolve multiple QTLs per chromosome or to identify true pleiotropy. Our results represent a three- to fourfold increase in mapping precision relative to previous F2 work in LG/J and SM/J (8.5 cM vs. 29 cM (4), 30 cM (8)). We also observed narrower confidence intervals compared to the LGXSM RIL (22 cM (9)).

We replicated seven of the eight original direct-effects Adip loci from an earlier F2 study in the Cheverud lab (4), but were unable to replicate all of the epistatic interactions between these loci. Increased resolution in this study resulted in enhanced identification of multiple QTLs for many chromosomes for each trait, and may have shifted the locations of the QTL peaks enough such that the epistatic interactions detected originally reside within the confidence interval of the Adip QTLs, but not at the peaks. This would inhibit replication due to the relatively small effects of epistatic contributions to genetic variation combined with the higher thresholds required to reach significance in the face of the extremely large number of comparisons performed. It is likely that future fine-mapping results will identify multiple QTLs per each peak in this study, each with differing direct and epistatic effects that considered together reflect the results in the current study and in previous F2 studies.

Recently, more studies have examined obesity and obesity-related traits in light of QTL identification and characterization of genetic architecture (27). The current study identified by far the largest number of QTLs with the narrowest confidence intervals contributing to fatpad weight, body weight, and organ weights. A suggestive QTL on chromosome 4 in the LXS recombinant inbred lines (27) overlaps with Adip11, which not only exhibits strong direct effects, but also functions as a “hub” within the adiposity epistatic interaction network. QTLs identified in an NMRI8 by DBA/2J cross (28) may correlate with our results. Many of those QTLs reside within the same general regions on the same chromosomes as our QTLs (2,3,12,13) but genomic positions were not provided for exact comparison purposes. We identified more epistatic interactions contributing to organ weights than other studies have produced (7,10). There was meager overlap between studies, likely due to the fact that different strains were used between the published literature and our work. We expect that only a subset of loci identified in studies using other strains will replicate as the genetic background differs significantly between studies and epistatic influences are likely to affect results. Our data appears to match observations from studies of diabetic nephropathy (29), cardiovascular disease (30), and nonalcoholic fatty liver disease (31) suggesting that comorbidities of obesity are mediated by overlapping but somewhat different sets of genes from those identified for obesity and diabetes. Further analysis of the heritable variation affecting organ weights and related phenotypes are necessary for elucidation of obesity-related comorbidities.

Like other QTL studies of obesity and obesity-related traits, we anticipate that these QTLs will eventually map to novel genes as well as genes in known obesity-related pathways. Candidate genes have been discussed previously (4,8,9) for many of the QTLs in this study. Examples include CPN10 for Adip1, Ucp2 for Adip3B, and Mc4r for Adip8. Table 2 lists a number of candidate genes for each of the pleiotropic QTLs identified in this study, although this is not an exhaustive list.

Table 2.  Candidate gene table
inline image

This study is an initial genome-wide approach to fine-mapping. Future studies are required to narrow confidence intervals to regions small enough to reasonably assess gene expression, allelic discrimination, and potential causal sequence changes. Also, statistically identified QTL regions will need to be biologically verified via creation of congenic and subcongenic lines and differential complementation tests. We are likely to find that many of our QTL regions correlate to coding or expression changes in genes associated with the PPAR γ interaction network affecting adipogenesis (32,33), MCR signaling cascades (34), circadian clock networks (35), mitochondrial energy homeostasis (36,37), and inflammation cascades (38).

Supplementary Material

Supplementary material is linked to the online version of the paper at http:www.nature.comoby


We gratefully acknowledge the funding support of the National Institutes of Health grant DK055736 and Biotechnology and Biological Sciences Research Council (BBSRC) grant BB/C/516936. The authors thank Doug Falk for efforts in SNP data generation. The authors also thank Elizabeth A. Norgard, Elizabeth Ann Carson, and Mihaela Pavličev for critical reading of this manuscript.


The authors declared no conflict of interest.