Quantitative Trait Linkage Analysis of Lipid-Related Traits in Familial Type 2 Diabetes: Evidence for Linkage of Triglyceride Levels to Chromosome 19q

Families were ascertained for at least two siblings with known type 2 diabetes and no more than one parent known to have diabetes. All available family members, including parents of the proband sib pair, all available siblings, and offspring of both the proband sib pair and their diabetic siblings were tested if they were >18 years of age. All nondiabetic individuals underwent a standard 75-g glucose tolerance test with diabetes diagnosed by World Health Organization (WHO) criteria (23). Height and weight were obtained for all individuals, and age of diagnosis was ascertained for diabetic subjects. Lipid levels were determined using standardized methods in either the Cardiovascular Genetics Research Laboratory of the University of Utah (3,6) or at Medlantic Institute Penn Medical Laboratory (24). All measures were performed in plasma after a 12-h fast. LDL cholesterol was calculated from total cholesterol, HDL cholesterol, and triglycerides and was considered unknown for individuals with TG >4.52 mmol/l. Lipid levels for families only in the extension set were measured by Medlantic Institute’s Penn Medical Laboratory (24). In the initial 19 families (referred to subsequently as “screening study”), genotypes were available for 407 individuals, including 379 for whom lipid measures were available. Of those individuals, 288 were nondiabetic by 1980 WHO criteria. At least 60 additional spouse controls were included for determination of allele frequencies. The “extended study” of 42 families included four simple sib pairs, but the remaining 19 new families were multigenerational, extended families. In the extended study, we tested a total of 610 individuals; lipid measures were available for 576 individuals, and 414 family members were nondiabetic. The extended study included 100 spouse controls for determination of allele frequencies. The number of relative pairs is shown in Table 1 for all available individuals in the data set from the screening set and the extension family set. Diabetes was diagnosed by glucose tolerance test or by treatment with insulin or oral hypoglycemic agents, as described elsewhere (25). Medication history was not used in the analysis, and all subjects were included regardless of antilipid medications. All subjects provided written informed consent under approval by the University of Utah Institutional Review Board, and all studies were conducted at the University of Utah General Clinical Research Center.

We tested 440 markers at a mean interval by our map of 9.2 cM for each of the 22 autosomes (14). Microsatellite markers were chosen from published maps (26–31) for heterozygosity >0.7. Most markers used in this study were dinucleotide repeats taken largely from Genethon maps (27,29), with additional markers at candidate genes for diabetes and from other maps (30,31). Most markers were amplified with a 5′ ³²P-labeled primer, separated on denaturing polyacrylamide gels, and detected by autoradiography as described previously (25,32). Markers were scored by two individuals, and discrepancies were resolved before analysis. Inheritance errors were resolved by review of original scores; unresolvable errors were scored as unknown. Markers run recently were labeled fluorescently. Infrared dyes were appended to an M13 primer. An M13 sequencing primer was appended to the standard amplification primer, and PCR was conducted with co-amplification of the labeled M13 primer and the amplification primers. Markers were detected on a LICOR GR4200 sequencer and scored semiautomatically with GeneImager software (Scanalytics, Billerica, MA).

We used variance-component analysis implemented in the SOLAR (Sequential Oligogenic Linkage Analysis Routines) computer program (33) to estimate the heritability of each tested trait both in the screening study and from the full phenotypic data available in the extended study. Heritability was estimated by including all available individuals, including diabetic family members, after correction for age and sex. No ascertainment correction was included.

We conducted quantitative trait linkage analysis of TG, total cholesterol, HDL cholesterol, TG-to-HDL ratio (20), and LDL-to-HDL ratio as a measure of cardiovascular risk. Each variable was adjusted for age and sex and transformed to reduce both skew and kurtosis to 0 before analysis (34). We used the multipoint variance-component approach to test linkage of each trait in all family members by maximum likelihood methods. In variance-component analysis, the variance is broken into locus-specific effects determined by the identity by descent relationships, the residual additive genetic effects, and individual-specific random environmental factors (18,33). We used the SOLAR computer program to calculate locus-specific identity-by-descent information for each pair of relatives with the multipoint extension, also implemented in SOLAR (33). Significance was expressed as LOD scores obtained by converting natural logarithm (ln) likelihood values into values of log₁₀. Because TG may be elevated by poor glucose control, we tested TG linkage twice: first in all individuals, then only in nondiabetic family members, while considering the phenotype of diabetic family members as unknown. Distances were derived from multiple sources and confirmed using the ILINK program from actual family data (14). In contrast to our published dichotomous trait analysis of type 2 diabetes, where the first marker was considered to be at 0 cM, the maps for the current study estimate the location of the first marker as 5–10 cM from the telomere. Thus, distances are generally 5–10 cM greater than those published previously (14).

We conducted the screening study on 19 extended families for which full genotypic data were available. Table 1 summarizes the relative pairs, based on TG levels, for both screening and extended studies. We tested four measured traits related to diabetic dyslipidemia: TG in all members, TG in nondiabetic family members, total cholesterol, and HDL cholesterol. Additionally, we tested linkage of three derived measures of cardiovascular risk: LDL cholesterol, LDL-to-HDL ratio, and TG-to-HDL ratio. Heritabilities for these measures ranged from 0.31 for HDL to 0.52 for TG (Table 2). Figure 1 shows LOD score curves using our linkage map (14) for six traits: TG in all family members, total and HDL cholesterol, calculated LDL cholesterol, TG-to-HDL ratio, and TG-to-LDL ratio. In all, 37 contiguous regions had LOD >1.0 for at least one measured or derived trait (Fig. 1), and nine regions had LOD scores exceeding 1.5 (Fig. 1 and Table 3). TG tested in only nondiabetic individuals (data not shown) closely paralleled that in all individuals, albeit at generally lower LOD scores, with three exceptions. Possible linkage of TG in only nondiabetic family members was seen on chromosomes 8 (171–188 cM; LOD 1.53) and 21 (0–4 cM, LOD 1.12), but was not seen in all family members. The LOD on 19q at position 74–80 was 2.56 in all family members but only 1.38 in nondiabetic family members. The lower LOD scores when only nondiabetic individuals were tested were expected given the smaller sample size.

We focused on three regions with initial LOD >1.5: HDL linkage to chromosome 1q, which maps near the apolipoprotein AII (ApoA2) gene and in a region with putative genes for type 2 diabetes (10,14–16); HDL linkage to chromosome 2q in the region of NIDDM1 (35) and calpain 10 (36); and TG linkage to the ApoC2/ApoE region of chromosome 19 (Fig. 2). Other regions were examined for all available data but were not specifically expanded and thus represent incomplete analyses. Of these three regions, analysis of the extended study data (see research design and methods) showed increased evidence for linkage only for TG on chromosome 19q, where the LOD score reached 3.16 at marker D19S178 (Fig. 2). The corresponding LOD score when diabetic individuals were excluded was 2.54, consistent with the loss of data. This peak of linkage is 3 cM from marker ApoC2 according to our map, but only 1.5 cM according the recent comprehensive maps (37). By current physical maps (38), D19S178 and the ApoC2/ApoE/ApoC1/ApoC4 gene cluster are separated by ∼1.5 Mb.

Because errors in genetic distances might alter LOD scores when multiple markers in close proximity are analyzed using multipoint methods (39), we also performed a 2-point variance-component analysis using SOLAR. The peak LOD score was confirmed at 3.15 at marker D19S178, with nearby markers KLK, DM, HRC, and GYS1 showing 2-point scores of 1.99, 2.32, 1.02, and 1.15, respectively. These markers cover a physical distance from 48.6 Mb (D19S178) to 55.5 Mb (KLK1) on 19q13 (38). In contrast, the ApoC2 microsatellite polymorphism (49.8 Mb), which is in close physical proximity to the ApoE, ApoC1, ApoC2, and ApoC4 genes, showed a 2-point LOD score of only 0.45.

Macrovascular disease is the major complication of type 2 diabetes. Considerable previous data have suggested that the metabolic risk factors for macrovascular disease cluster in family members and probably predate the onset of hyperglycemia (4). Indeed, nondiabetic members of 16 of the families in the present study showed a 58% prevalence of lipid abnormalities (3). In the present study, we have used a variance-component approach to confirm that each of the major lipid traits is heritable in families ascertained for at least two diabetic siblings, with heritabilities in the same range as those for insulin resistance–related traits (14). These findings extend and confirm our earlier study (3) and other studies that have demonstrated insulin resistance-related traits in family members of diabetic individuals. Although such findings support the significance of shared genetic predisposition in controlling lipid-related traits, they do not necessarily indicate a major genetic locus controlling these traits. Genome scans offer one method to search for such loci. We used data from a genome scan conducted in our 19 largest families (screening study) to examine five traits that affect risk for macrovascular disease or, in the case of TG-to-HDL ratio, were shown by others (20) to be controlled by major genes. We then extended three of these regions to 19 additional multiplex families and 4 diabetic sibships (extended study) and examined all 42 families together. Within the screening study, no trait reached significance, with the highest LOD scores just over 2.0. Regions previously reported to be linked to lipid-related phenotypes were generally not replicated. These findings were not surprising. Even if a major gene were present, multiple interacting loci are likely to control each of these traits. Both complex disease and quantitative trait linkage studies in humans consequently have somewhat limited power, and replication of previously significant linkage can be difficult to achieve (39,40).

None of the nine regions with LOD scores >1.5 in our screening study corresponds to earlier reports. We found little evidence for linkage of TG to 15q (18), and although we found modest evidence for linkage to TG and TG/HDL ratio on chromosome 7, our peak linkage is at least 50 cM proximal to that reported by Shearman et al. (20). Large differences in map distances (total chromosome lengths of 150 cM for Shearman et al. and 236 cM in our study) between our studies might explain some of this discrepancy, and this region merits further analysis. In the recent study of Pima Indians (41), modest evidence for linkage of TG was shown to chromosomes 2p, 3, and 5q. In contrast, we have no evidence for linkage in these regions. We found three regions with modest evidence for linkage to HDL cholesterol, but none overlap with the regions identified by Almasy et al. (42) or with the regions of suggestive linkage in Pima Indians (chromosomes 3, 7, and 20). We also found no evidence for linkage of any lipid-related trait to the strong candidate regions on 11q (apolipoprotein AI, CIII, or AIV; ApoA1/ApoC3/ApoC4), hepatic lipase (15q21-q23), or cholesterol ester transfer protein (CETP) (16q21) (7). Finally, neither screening nor extended linkage data showed evidence for linkage of total cholesterol to 19p, where significant linkage was reported in the region of the LDL receptor (LDLR) in Pima Indians (41), although we found some evidence for linkage of TG-to-HDL ratio in this region (Fig. 2).

Several studies have recently reported linkage results for FCHL, a complex trait that is characterized by glucose intolerance and insulin resistance, and thus shares features of the insulin resistance/metabolic syndrome (12). One FCHL locus in Finnish families (10) and in a mouse model of FCHL (43) maps to the same region of chromosome 1q as a confirmed type 2 diabetes susceptibility gene (14–16). In contrast to the studies of FCHL as a dichotomous phenotype (affected versus unaffected), we examined lipids as quantitative (continuous) traits. We found no evidence for linkage of lipid-related traits to lipoprotein lipase (LPL), the ApoA1/ApoC3/ApoA4 cluster on 11q, the lecithin-cholesterol acyltransferase (LCAT) gene on 16q (7), or loci on chromosome 10 (12). We did find evidence for linkage of HDL cholesterol in the region of chromosome 1q21–q23 that harbors the FCHL locus in Finnish families (10), but this modest evidence for linkage did not increase in the extended study. The ApoA2 gene is the strongest candidate in this gene-rich region. Transgenic mice overexpressing ApoA2 have elevated free fatty acid levels and increased triglycerides, altered HDL levels, and insulin resistance (44). We have screened the ApoA2 gene for variation and identified only noncoding changes in intronic and 3′ flanking regions (S.C.E., unpublished data), which show only modest association with diabetes. We have also failed to find coding region alterations of the retinoid X receptor γ gene in this region, which is known to form dimers with peroxisome proliferator-activator receptors α and γ. In preliminary analysis, we found little evidence for association of these variants with diabetes (S.C.E., unpublished data).

Our best evidence for linkage of any lipid trait was TG to 19q13.2. Because we have tested six different lipid-related phenotypes, and this value does not quite reach the levels proposed for genome-wide significance using allele-sharing methods (45), we would consider this linkage to fall between suggestive and significant. Two recent studies suggested that under some circumstances, variance-component analysis was subject to excessive error rates, particularly with extreme sampling distributions and skewed trait distributions (46,47). In the study of Hirschhorn et al. (47), empiric P values from simulation ranged from 0.02 to 0.24 for a LOD score of 3.1, albeit for smaller families and a trait with different characteristics. An LOD score of 3.1 may be compatible with failure to reach nominal genome-wide significance (P < 0.05). Thus, even our finding on chromosome 19q may represent false-positive linkage. However, several factors argue for the importance of this finding. In the present study, we normalized all data before analysis, and although we did not test the significance by simulation, extensive simulations by others have suggested that variance component analysis does not lead to inflated P values for normal data. Although similar artifacts (outliers or skewed data) might also alter LOD scores for nondiabetic family members, our finding of a suggestive LOD score for nondiabetic family members in this region of 2.54 close to the peak of linkage in all family members supports the biological significance of this region. This lower LOD score might be expected because diabetic subjects were excluded in this analysis. The close correspondence of these analyses suggests that the genetic predisposition to hypertriglyceridemia affects both diabetic and nondiabetic individuals similarly and is not likely to be due to hyperglycemia itself, nor to an interaction of this susceptibility locus with hyperglycemia, nor to diabetic individuals with very high triglyceride levels. Second, the TG-to-HDL ratio also maps to this location, although triglycerides and TG-to-HDL ratio show a strong correlation and this finding is expected. Third, this linkage is near a plausible cluster of candidate genes. These factors increase the confidence that a locus controlling TG levels is present in this region, despite some uncertainty regarding the statistical significance of the LOD scores derived from the variance-component linkage analysis. ApoE has been implicated in LDL cholesterol levels and in clearance of chylomicron remnants (7), but to our knowledge, no other study has reported linkage of TG to this region.

Despite strong suggestive evidence for linkage of TG to marker D19S178, the two-point LOD score for the ApoC2 microsatellite, which is within the cluster of genes that represent the obvious candidate genes in this region, is <0.5. For complex traits, LOD scores at any single locus depend on the amount of information that marker provides in the informative families in which susceptibility or protective alleles segregate (39). Thus, lack of linkage to ApoC2 does not exclude this gene or adjacent genes as explanations for this linkage. ApoC2 is a cofactor for lipoprotein lipase, and rare mutations cause familial hypertriglyceridemia (7). Unusual mutations in the promoter region, exons, and splice sites of the ApoC2 gene (48) cause hypertriglyceridemia by inactivating the gene, but paradoxically overexpression of ApoC2 in transgenic mice also causes hypertriglyceridemia (49). Among the common variants of ApoC2 are two amino acid changes (50) whose role in hypertriglyceridemia has not been evaluated, to our knowledge. Interaction of rare ApoC2 variants and the ApoE4 allele has also been described (51). Finally, the ApoE2 allele also may cause hypertriglyceridemia (52). The roles of the ApoC1 and ApoC4 genes are uncertain, but variation is a plausible cause of hypertriglyceridemia.

We also found linkage of TG-to-HDL ratio at the same location on chromosome 19 (LOD 2.727), with more modest evidence for linkage on chromosome 19p (LOD 1.9225) where total cholesterol was linked in Pima Indians (41). The TG-to-HDL ratio is an additional cardiovascular risk factor in type 2 diabetes, and an increased TG-to-HDL ratio commonly accompanies the insulin resistance-to-metabolic syndrome (20). Shearman et al. (20) reported heritability estimates for TG and TG-to-HDL ratio similar to ours (0.396 for TG, 0.488 for TG-to-HDL ratio). Interestingly, in their study both TG and TG-to-HDL ratio mapped to the same region of chromosome 7q, albeit with lower LOD scores than we report in the present study. Although both their study and ours might be interpreted to suggest that the same genes may control both TG levels and the TG-to-HDL ratio, these two phenotypes show a correlation of 0.94 in our complete sample. Thus, correspondence of the linkage may reflect the fact that TG and TG-to-HDL ratio in this sample in fact represents the same trait, as is reflected in the nearly invariant inverse relationship between TG and HDL cholesterol.

Replication of linkage findings is important in understanding the role of a locus in a complex disease, particularly when findings do not quite reach genome-wide significance. Such replication has been difficult to achieve and might require a study size at least three times larger than that of the current study (39). Given the strength of the ApoC2-to-ApoE-to-ApoC1-to-ApoC4 gene cluster as candidates for this linkage, and the higher power of association studies over linkage studies to detect variants in complex diseases (53), a direct search for association of common variants with TG and TG-to-HDL ratio appears justified. Such studies might lead to earlier intervention in the management of cardiovascular risk before the onset of hyperglycemia in these high-risk individuals.

These studies were supported by grant DK39311 (S.C.E.) from the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health (NIH), from the Research Service of the Department of Veterans’ Affairs, by a Harold Rifkin Family Acquisition Grant from the American Diabetes Association, and by the National Center for Research Resources, NIH, through the General Clinical Research Center.

We thank Michael Hoffman, Rachel Mayorga, Kelli Barrett, Denise Fuller, Phaedra Yount, and Kui Teng for technical support, and Kimberly Wegner, Teresa Maxwell, and Holly Tuckett for family ascertainment and DNA preparation.