An Obesity-Related Locus in Chromosome Region 12q23-24

Two hundred sixty European-American families (1,297 subjects) were chosen, as previously described (15,16). Briefly, all family probands (BMI ≥40 kg/m²) had at least one obese sibling (≥30 kg/m²) and at least one parent and one sibling who were of normal weight (<27 kg/m²). All subjects gave informed consent, and the protocol was approved by the Committee on Studies Involving Human Beings at the University of Pennsylvania.

BMI was calculated based on measured height and weight: BMI = weight (in kilograms)/height (in meters)². Percentage of fat was measured by bioelectric impedance (Tanita TBF310 Pro Body Composition Analyzer; Tanita, Arlington Heights, IL). Waist circumference was measured when blood samples were collected. All quantitative variables were adjusted for linear effects of age within generation and sex using SPSS version 11.0.

DNA was extracted using a high-salt method (17) and diluted to 10 ng/μl for genotyping. Three hundred eighty-two polymorphic Marshfield microsatellite markers from the Marshfield Screening Set 11 were genotyped by the Marshfield Center for Medical Genetics. Map distances were taken from the Marshfield Database (http://research.marshfieldclinic.org/genetics/). One family was duplicated (coded as different family) as an inner control for genotyping. Sex-specific PCR markers were amplified to verify sex. Mendel checks were performed by Merlin (18), and all errors were corrected or dropped.

We performed nonparametric multipoint linkage analyses using Genehunter version 1.3 (19) for BMI. We used dichotomous obesity affection status: BMI ≥27, ≥30, ≥35, and ≥40 kg/m². Gene frequencies were estimated by counting alleles, using all individuals who provided DNA. This approach gives asymptotically unbiased estimates of the allele frequencies (20). All Mendelian inheritance errors were checked and resolved using Merlin and Genehunter 1.3. Any genetically unrelated parents and siblings were excluded, as were all half-siblings.

Quantitative phenotypes (BMI, percentage of fat, and waist circumference) were analyzed using the family regression test (Merlin_regress) (21). We used phenotypes after linear correction for age within generation and sex. Following this standardization, higher-order age effects were not significant. Outliers (more than four times the SD) were dropped in our analyses.

Using sex, age, and the interaction age × sex as covariates, a variance-components approach was performed with Solar 2.6.6 (22). We corrected for ascertainment based on the primary proband (23).

We estimated empirical significance levels for the analyses of qualitative phenotypes. We used the Simulate program (24) to generate 500 replicates of simulated markers genotypes based on real family structures, map distance, and allele frequencies, as we used in the Genehunter analyses. Simulated genotypes were analyzed using Genehunter 1.3. Empirical P values were calculated by dividing the number of replicates that exceeded the observed nonparametric linkage (NPL) score by the number of replicates (500 replicates × 382 markers = 191,000 replicates). Empirical P values (P_step-down) were also given by a step-down method to control for multiple testing (25).

For quantitative traits, we simulated 500 replicates of our datasets for family regression analyses using the “simulate” function in Merlin. As we did for Genehunter, empirical P values were computed by dividing the number of replicates that exceeded the observed LOD score by the number of replicates (i.e., n = 191,000).

After detecting linkage on 12q23-24, three additional microsatellite markers IGF-I (108 cM), D12S1339 (116 cM), and D12S349 (135 cM) were picked to fill the gaps flanking the peak from the genome scan. Genotyping of microsatellite markers was performed as previously described (15).

To verify our previous linkage result on 10p12, two microsatellite markers (D10S197 and D10S193) were manually genotyped after the Marshfield genome scan finished. Data analyses were carried out using Genehunter and Merlin after new genotyping data were merged.

Among 1,297 European-American individuals, the average BMI values (±SD) for probands, fathers, mothers, sisters, and brothers were 48.7 ± 9.1, 29.1 ± 6.2, 32.3 ± 9.1, 33.6 ± 10.0, and 30.7 ± 7.7 kg/m², respectively. Most probands were female (237 of 251), and about two-thirds of all subjects were female (895 of 1,297). We had DNA available from both parents for 143 families, one parent for 111 families, and no parent for 6 families. Sibship size ranged from 2 to 8, and most families (n = 244) had 2–6 sibs (1–15 sibpairs), with a median sibship size of 3. Numbers of affected sib pairs for BMI ≥27, ≥30, ≥35, and ≥40 kg/m² were 658, 472, 282, and 134, respectively. The numbers of families having at least one affected sibpair were 224 (224 of 260, 86.15%) for BMI ≥27, 203 (78.08%) for BMI ≥30, 163 (62.69%) for BMI ≥35, and 100 (38.46%) for BMI ≥40 kg/m², respectively. Clinical characteristics (including BMI, percentage of fat, and waist circumference) of all samples, probands, fathers, mothers, sisters, and brothers are shown in Table 1. To demonstrate distributions of BMI, percentage of fat, and waist circumference, Table 1 also includes skewness and kurtosis for reference.

The 1,297 European-American individuals were genotyped by Marshfield Medical Genetics Center using 382 microsatellite markers across the human genome. Marker heterozygosity was from 0.49 to 0.92 and averaged 0.76 ± 0.06. The average interval between markers was 8.94 cM, with the maximum gap of 17.5 cM on chromosome 18.

Using Genehunter 1.3, we obtained the highest NPL score (2.52, P = 0.0039) at the marker D7S1799 (114 cM, BMI ≥27 kg/m²) on chromosome 7. NPL scores as well as nominal and empirical P values are shown in Table 2. In addition to chromosome 7, chromosomes 3 (D3S2045, 124 cM), 8 (D8S2324, 94 cM), 9 (D9S910, 104 cM), 12 (GATA49D12N and PAH, 18 and 109 cM, respectively), and 13 (D13S894, 33 cM) also had linkage signals (NPL scores >1.85) (Table 2). Simulation tests yielded empirical P values that were all <0.05 for those loci.

We obtained significant results on chromosomes 4 (D4S1644, 143 cM), 12 (D12S2070, 125 cM), 13 (D13S800 and D13S779, 55 and 83 cM, respectively), and 21 (57.8 cM) (Table 3) (Fig. 1). We also found suggestive linkage on chromosome 2 (38 cM, LOD 1.70, P = 0.003) for BMI (Fig. 1). The most significant result was from chromosome 21 (D21S1446, 57.8 cM) for percentage of fat (LOD 4.27, P = 0.00001). Marker D12S2070 (125 cM) revealed significant linkage for BMI (2.98, P = 0.0001), percentage of fat (3.79, P = 0.00001), and waist circumference (2.86, P = 0.00014). Empirical genome-wide P values based on simulations (500 simulations each for 382 markers) ranged from 0.00002 to 0.0003 at the same genomic location, 125 cM from the p-terminus in chromosome region 12q23.

Chromosome 13 also showed positive linkage signals: LOD 2.70 (P = 0.0002) and 2.82 (P = 0.0002) for BMI on 55 cM (D13S800) and 83 cM (D13S779), respectively. We also found LOD 1.80 (P = 0.002) for marker D13S779 for waist circumference. Estimates of trait mean, variance, and heritability had little effect on results from the family regression analyses (data not shown).

Variance components analyses showed suggestive linkages on chromosomes 12, 13, and 21 (Table 3). The most significant result was on chromosome 13 (83 cM, D13S779), with LOD 3.11 for BMI. We obtained LOD scores of 2.33, 2.40, and 2.11 on chromosome 12 (D12S2070, 125 cM) for BMI, percentage of fat, and waist circumference, respectively.

After we added three additional markers to the 12q23-24 region, regression analysis using Merlin_regress gave a LOD score of 4.08 (P = 0.00001) for percentage of fat on the marker D12S1339 (116 cM). D12S2070 yielded LOD scores of 3.57 (P = 0.00003) and 3.05 (P = 0.00009) for BMI and waist circumference, respectively. For qualitative BMI, we obtained an NPL score of 1.60 at the marker D12S1339 (BMI ≥30 kg/m²). The results of the fine mapping on chromosome 12 are shown in Fig. 2.

After five gap-filling markers (D10S197 and D10S193) were added, Genehunter NPL scores improved from 0.84 to 2.58 (D10S197, BMI ≥27 kg/m²).

We have identified strong evidence for linkage of obesity-related phenotypes to markers in chromosome region 12q23-24. The evidence for linkage is consistent across correlated obesity phenotypes and analytic methods, and the peak LOD score meets criteria for genome-wide significance (LOD 4.08, P = 0.00001). The result was based on a large and unique sample segregating extreme obesity and normal weight.

The results for chromosome 12 gave LOD scores based on regression analyses of 3.79, 2.98, and 2.86 for the correlated phenotypes, percentage of fat, BMI, and waist circumference, respectively. After adding three microsatellite markers (IGF-I, D12S1339, and D12S349) to the 12q23-24 region, LOD scores for percentage of fat, BMI, and waist circumference increased from 3.79 to 4.07, 2.98 to 3.56, and 2.86 to 3.05, respectively. The peaks for BMI and waist circumference remained unchanged, whereas the peak of percentage of fat shifted to the adjacent (and newly added) marker, D12S1339.

Empirical genome-wide P values based on simulations (500 simulations each for 382 markers) ranged from 0.00002 to 0.0003 at the same genomic location, 125 cM from the p-terminus in chromosome region 12q23. Qualitative analyses revealed an overlapping peak ∼16 cM centromeric for BMI ≥30 kg/m², with an NPL score of 1.92 and a genome-wide P value of 0.02. Because of a gap in the marker density, this second peak is at the next proximal centromeric marker. Although the multiple phenotypes examined are correlated, we assumed four independent phenotypes in the simulations. Overall, results were consistent across regression analyses for three related quantitative phenotypes and NPL analyses of BMI qualitative thresholds.

Perusse et al. (26) reported a genome scan for abdominal fat based on computed tomography scans for 156 European-American families from Quebec. Approximately one-half of the families were ascertained through an obese individual (BMI >32 kg/m²). They found linkage to markers in chromosome region 12q22-24, with a maximum LOD of 2.88. In our original scan with 92 families (15) we detected a similar peak, but with a Z score of only 1.4 in 12q23-24 (a proximal peak had a somewhat higher score in the earlier scan). Because the 1.4 value did not reach our threshold for suggestive linkage, we did not report the result at that time.

There have been two possibly relevant QTLs that map to the homologous regions in mouse and rat. One was the QTL (Qsbw) for body weight identified in a cross between C57BL/6J and Quackenbush-Swiss mice (27). Another study (28) reported a QTL (Weight1 or bw/gk1) with a strong influence on body weight in a GKxBN rat cross. Both QTLs had large effects, with Weight1 accounting for 24% of the phenotypic variance and the Qsbw locus causing an increase of 0.4 SD units. Both QTLs are located very near the gene for IGF-I, which in humans maps to chromosome 12q23-24. Another rat QTL (Dmo7p) for fat weight is also located in the region (29).

In addition to the human linkage and mouse QTL studies, there have been two reported associations of obesity-related phenotypes to genes located in this region of chromosome 12q22-24. Sun et al. (30) reported that percentage of body fat and other body fat-related variables were associated with polymorphisms in the IGF-I gene. One study (31) reported a possible association between BMI and the lipoprotein receptor SCARB1 (scavenger receptor class B, member 1), which appears to be located distal to IGF-I in 12q24, probably too far to contribute to reported linkage results in humans.

A candidate gene ACACB (acetyl-CoA carboxylase-β) localizes within 0.5 Mb of the marker D12S1339, our peak LOD score for percentage of fat. ACACB controls fatty acid oxidation in skeleton and heart muscle cells. ACACB knockout mice (AccII^−/−) have a higher fatty acid oxidation rate and lower amounts of fat than normal controls (32).

In addition to ACACB, there are several plausible candidate genes located within the linked region on chromosome 12q23-24, including PMCH (promelanin-concentrat-ing hormone) and IGF-I. MCH (melanin-concentrating hormone) knockout mice showed reduced body weight (due to hypophagia) (33). However, there are 39 genes or expressed sequence tags within a 20-Mb region for which we obtained a LOD score >2.0. At present, this locus is insufficiently localized to go beyond hypothesizing about possible causative genes.

Previous linkages have been reported to ∼70 locations in 150 reports from 60 articles, including 29 genome scans (30 including this one) (13). However, most studies were based on small samples, and most results were at most suggestive. Only three previously reported linkages achieved genome-wide significance (LOD >3.6) based on a sample of 1,000 or more subjects. This fact should dramatize the need for large samples in genome scans for complex traits.

We detected linkage at locations previously identified in other studies on earlier subsets of our cohort based on 78–220 families, particularly on chromosome 7q (34–36) and 20q (15). Linkage results from the scan were not as strong as in previous studies, apparently because of lower marker coverage in the linked regions.

We also found suggestive linkage on chromosomes 2, 3, 4, 7, 8, 9, and 13 for both qualitative and quantitative obesity phenotypes. For chromosome 21, we found a high LOD score and significance level. However, the result was based on a single marker that had a relatively low information content, which was compounded by the fact that it is a terminal marker. The linkage results for this location were not consistent across phenotypes or analytic methods, and we suspect the result may not be reliable.

One candidate gene on chromosome 13 is IRS2 (insulin receptor substrate 2, close to marker D13S779 at 83 cM). IRS2-disrupted mice have type 2 diabetes (37), and case-control studies in human subjects also showed associations between IRS2 polymorphisms and type 2 diabetes (38) or obesity (39). Table 3 shows another peak of linkage on chromosome 13 around 55 cM (GATA64F08 and D13S317). Feitosa et al. (40) found an LOD 3.0 for BMI on marker GATA11C08 (8 cM upstream) and the flanking region on 13q14-22. They suggested that HTR2A (5-hydroxytryptamine receptor 2A) could be a candidate gene in this region.

As we reported in our fine mapping study (36), chromosome 7 (114 cM) showed linkage for dichotomous BMI in this study. Because of the lower marker density (the marker GATA23F05 locates 4.5 Mb upstream of our previous peak D7S692), it is not surprising that the LOD score is slightly lower in this study compared with the previous results from fine mapping, especially for quantitative traits. There were multiple linkage peaks found in 7q by different groups of researchers within or outside the leptin (LEP) gene region for obesity and related phenotypes (34,41–44).

Although there are no previously reported human linkage reports near our chromosome 3, 4, 8, and 9 peaks, multiple QTL and association studies suggested that obesity-related genes may lie in those chromosome regions. We found suggestive linkage for multiple phenotypes for a marker (D4S1644, 4q31.21) only 0.25 Mb downstream of uncoupling protein 1 (UCP1). Both animal models (45) and human association studies (46,47) indicate that the key role UCP1 plays in energy balance could influence common obesity. The chromosome 8 linkage peak (D8S2324, 8q21.11) localizes within the mouse QTL Hlq4 (48), and the chromosome 9 peak (D9S910, 9q22.23) overlaps with mouse QTL Qlw4 (49,50).

Importantly, we missed a previously reported linkage peak on chromosome 10p (51,52). Although the Marshfield marker set has an average coverage of 10 cM, there is variability in the density of coverage. This peak linkage falls within a 16-cM gap. To validate the previous linkage result and exclude the possibility that the lack of results in the current scan might be due to failure to replicate in the more recently accrued families, we filled the gap between Marshfield markers on 10p. We then found a maximum NPL score of 2.58, a more than threefold increase from the initial genome scan maximum value of 0.84. The two additional markers increased the information content at the peak from <60 to >90%. The fact that an established linkage result can be missed in a 10-cM (average) scan supports the development and use of finer maps in genome scans of complex traits. Although genotyping costs will be higher, it appears to be the only way to fully extract information from family cohorts already collected at great expense.

Much progress has been made in methods for analyses of quantitative traits in the 5 years since our first genome scan. Three major methods are widely used: Haseman-Elston-based regression, variance components (20,21), and the family regression method by Sham et al. (21). The first two methods are sensitive to departures from normality, particularly variance components. Using simulated, nonnormally distributed data, Sham et al. (21) found that their family regression method was insensitive to distribution properties. Because our families were selected through multiple probands, we used the family regression method to minimize any effects of the ascertainment. We used variance components (Solar) to verify our family regression results. While the estimates of variance accounted for cannot be interpreted based on analyses of our sample, the localization of linkage should be relatively robust. The results matched very well with those from the family regression analyses (Table 3).

Homogeneity is very important in family selection for linkage analyses. To minimize admixture, we only selected non-Hispanic, European-American subjects in this study. It has been shown that European and European-American populations appear to have minimum population structure (53).

Our chromosome 10 results provide a dramatic demonstration of the effect of marker density on linkage detection for complex traits. Risch (54) examined the issue of linkage detection from the standpoint of marker information content. Using scheme 2 in an article by Risch (54), where parental information is used to infer identity-by-descent status, his results (see Fig. 3 of that study) showed that the percentage of maximal expected LOD score as compared with fully informative markers is approximately proportional to the polymorphism information content of the marker under study. Therefore, near the true disease locus, we expect the LOD score to on average increase by a factor of 1.2 to 1.6 when the polymorphism information content is increased from 60–80 to 90–95%. This will also result in the reduction of the CI length, although the extent of this reduction depends on the strength of the true signal. Obviously, when the markers are located farther from the disease locus, the effect can be much more pronounced, threefold in this case.

Two recently published studies have addressed the issue of map density from the standpoint of CIs rather than linkage detection. They examined the question from very different perspectives and found consistent results but drew opposite conclusions. Schulze et al. (55) provided a proof-of-principle demonstration by going from a 10-cM to a 1-cM map in linkage analyses of bipolar disorder. Their LOD score increased from 4.7 to 5.4, and the 1 LOD CI decreased from 12 to 9 cM, i.e., by roughly 25%. Atwood and Heard-Costa (56) examined the effect of map density on location errors for quantitative traits analyzed by variance components methods through a series of simulations. Their largest sample size was one-third smaller than ours, 200 families with five members. For QTL accounting for 20–40% of the variance, they found decreases in location error of 7–40% by going from a 10-cM to a 1-cM map, values well within the range found in the bipolar study. They concluded that fine linkage mapping was of little value. However, their view seems to ignore the effort and expense involved in genotyping. On the other hand, higher marker density and its resulting information increases the power for detection and reduces CIs.

In summary, we completed a genome scan of a large sample of families segregating extreme obesity and normal weight. We obtained consistent support for linkage to markers in chromosome region 12q21.2-23 across obesity phenotypes and analytic methods. The peak LOD score met genome-wide significance based on simulations. This result appears to be one of the strongest yet reported for an obesity trait. Obviously, further study and independent replications are needed.

This work was supported by the National Institutes of Health (R01DK44073, R01DK48095, and R01DK56210 to R.A.P. and R01GM59507 to H.Z.). Except for five markers, genotypes were completed by the National Heart, Lung, and Blood Institute-supported Marshfield Genotyping Service, and we thank James L. Weber, Director, Donna Dorshorst, and Ying Fan.

We acknowledge the cooperation of our subjects. We thank Quan Cao, Jeffrey Hannah, Balasahib Shinde, Elizabeth Joe, Jan Merideth, Cameron Braswell, Cathleen Garrigan, and Kye Yun for technical assistance.