Low-Density Lipoprotein Particle Size Loci in Familial Combined Hyperlipidemia

Michael D. Badzioch; Robert P. Igo, Jr; France Gagnon; John D. Brunzell; Ronald M. Krauss; Arno G. Motulsky; Ellen M. Wijsman; Gail P. Jarvik

From the Department of Medicine, Divisions of Medical Genetics (M.D.B., R.P.I., F.G., A.G.M., E.M.W., G.P.J.) and Metabolism, Endocrinology, and Nutrition (J.D.B.), Departments of Genome Science (A.G.M., E.M.W., G.P.J.) and Biostatistics (R.P.I., E.M.W.), University of Washington, Seattle; Department of Epidemiology and Community Medicine (F.G.), University of Ottawa, Ontario, Canada; and Life Sciences Division (R.M.K.), Ernest Orlando Lawrence Berkeley National Laboratory, University of California, Berkeley.

ABSTRACT

Objective— Low-density lipoprotein (LDL) size is associated with vascular disease and with familial combined hyperlipidemia (FCHL).

Methods and Results— We used logarithm of odds (lod) score and Bayesian Markov chain Monte Carlo (MCMC) linkage analysis methods to perform a 10-cM genome scan of LDL size, measured as peak particle diameter (PPD) and adjusted for age, sex, body mass index, and triglycerides in 4 large families with FCHL (n=185). We identified significant evidence of linkage to a chromosome 9p locus (multipoint lodmax=3.70; MCMC intensity ratio =21) in a single family, and across all 4 families to chromosomes 16q23 (lodmax=3.00; IR=43) near cholesteryl ester transfer protein (CETP) and to 11q22 (lodmax=3.71; IR=120). Chromosome 14q24-31, a region with previous suggestive LDL PPD linkage evidence, yielded an IR of 71 but an lodmax=1.79 in the combined families.

Conclusions— These results of significant evidence of linkage to 3 regions (9p, 16q, and 11q) and confirmatory support of previous reported linkage to 14q in large FCHL pedigrees demonstrate that LDL size is a trait influenced by multiple loci and illustrate the complementary use of lod score and MCMC methods in analysis of a complex trait.

We used lod score and Markov chain Monte Carlo linkage analysis to perform a genome scan of adjusted LDL size in 4 large families with familial combined hyperlipidemia. We identified significant evidence of linkage to chromosomes 9p, 16q23, and 11q22 and confirmatory evidence for 14q24-31. LDL size is influenced by multiple loci.

Key Words: LDL size ? linkage scan ? familial combined hyperlipidemia

Introduction

Cardiovascular disease (CVD) remains the major cause of death in Western societies. Low-density lipoprotein (LDL) size, quantitatively measured as the peak particle diameter (PPD), has been shown to predict CVD.1–3 A related trait, small, dense LDL is associated with myocardial infarction,4 as well as with elevated apolipoprotein B (apoB), triglycerides (TGs), and decreased high-density lipoprotein cholesterol (HDL-C) levels.4,5 LDL size is highly correlated with HDL-C and TG levels, and examination of whether its predictive power in CVD is independent of these correlations has given contradictory results.1–3 It has been hypothesized that small, dense LDL causes vascular disease directly through increased ability to permeate the vascular epithelium, or through increased susceptibility to oxidation,6,7 in addition to effects on other atherogenic traits.

The previous studies demonstrating that variation in LDL size is heritable were reviewed recently by Bosse et al.8 Genetic heterogeneity is suggested by the linkage and association studies. Linkage of LDL size to cholesteryl ester transfer protein (CETP) and APOA1/C3/A5 has been described in families with familial combined hyperlipidemia (FCHL).9 Only the CETP locus is well established to cause common variation in LDL size.8–11 Allayee et al also found weak evidence of linkage of LDL size to the manganese superoxide dismutase (MnSOD; chromosome 6q25), a region suggested in a study of familial hypertriglyceridemia.12

Several strategies can be used to improve the power to detect evidence for linkage to a genetically complex trait such as LDL size.13 First, large pedigrees are advantageous because they tend to have more power for linkage detection than do small pedigrees for an equivalent total number of individuals,13,14 and they also can be analyzed individually, which can reduce the problem of genetic heterogeneity. Second, ascertainment of pedigrees through an extreme phenotype that is related to the complex trait of interest can reduce the genetic heterogeneity and increase the potential for segregation at trait loci. Successful examples of these strategies include the mapping of BRCA1 and Alzheimer disease loci.15,16 Third, it can be helpful to take covariates into account, including correlated traits that also have a genetic basis,13,17 as underscored by the finding of significant evidence of lipoprotein lipase linkage to LDL size only when adjusted for TG.18 The only previous genomic scan of LDL PPD that obtained significant results and used a TG covariate was of French-Canadians19 that identified a 17q21 locus. Finally, it is useful to use methods of analysis that effectively efficiently accommodate the possibility of multilocus inheritance, such as Markov chain Monte Carlo (MCMC).20

Families with FCHL, the most common genetic cause of hyperlipidemia,21,22 may represent a relatively homogeneous sample with respect to LDL density. The major lipid abnormalities seen in FCHL are an increase in TG or cholesterol caused by an increase in small very low-density lipoprotein, an increase in intermediate density lipoprotein, and an increase in small dense LDL.9,23,24 Indeed, the small LDL and high apoB seen in FCHL are found whether the patient is hypercholesterolemic or hypertriglyceridemic.25 Additionally, small LDL appeared to segregate and predict FCHL independently of a putative apoB level locus.24

The relationship between LDL size and FCHL and the understanding of the need for proper covariate adjustments motivated the genomic scan for LDL size loci in large families presented here. The use of FCHL families may reduce genetic heterogeneity, as may the use of fewer, large families, improving power to detect linkage. Further, our segregation analyses determined that adjustment of TG was an important factor in ability to map LDL PPD loci. Only 1 genome scan of LDL PPD adjusted for TG with significant results has been reported,19 and there has been no such genomic scan of LDL PPD in families with FCHL. The use of novel Bayesian MCMC analytic methods allows for: (1) consideration of multilocus trait models that are not prespecified, (2) multipoint linkage analyses using all chromosome markers simultaneously, and (3) comparison of these results to those of logarithm of odds (lod) score methods. Identification of LDL size loci will aid in diagnosis, treatment, and perhaps prevention of vascular disease, as well as help unravel the complex genetics of FCHL.

Methods

Families

Analyses were performed on members of 4 large white pedigrees with well-characterized FCHL. These pedigrees each include 3 generations of sampled adults and multiple large sibships and were described in detail previously.24,26 Pedigree 1 was identified through a survivor of a premature myocardial infarction with hyperlipidemia. The probands in pedigrees 2 to 4 were markedly hypertriglyceridemic and originally were thought to represent bilineal FCHL families.27 These FCHL families were collected because of their large size. Pedigrees 1 to 4 have also been referred to previously as pedigrees 41, 428, 2100, and 2103, respectively.14,26 Seven subjects were excluded from phenotype analyses: 3 had type II diabetes, and 4 were taking lipid-lowering medications. There were 185 remaining subjects with phenotype and genotype data; 45, 38, 64, and 38 subjects each in families 1 to 4, respectively. Body mass index (BMI) was computed using self-reported height and weight. This study was approved by the University of Washington Institutional Review Board.

Laboratory Methods

LDL PPD (size in nanometers) was determined by gradient gel electrophoresis of whole plasma.28–31 Standard enzymatic methods were used to determine levels of TG (UV method) and HDL-C on an Abbott Spectrum analyzer.32,33

Genotyping and Error Resolution

Genotyping for the 10-cM genome scan was provided by the National Heart, Lung, and Blood Institute (NHLBI) Mammalian Genotyping Service (James Weber, Marshfield, Wis; NHLBI NO1-HV-48141). We used the Weber 9 panel of 387 markers, with an average heterozygosity across markers in our data set of 0.76 (SD=0.06). The average percentage of markers for which genotypes were obtained in an individual was 98.4%. The genotyping error, determined by blind duplicate samples, was 0.7%. Family structure errors, sample switches, and genotyping errors were identified and resolved through a combination of methods, including (1) evaluation of allele sharing in pairs of individuals under the assumption of a 0% genotyping error rate, as computed by RELPAIR;34 (2) new methods to evaluate more complicated relationships among trios of individuals, with a model that allows for genotyping error, implemented in ECLIPSE;35 and (3) computation of single-locus likelihoods on complete pedigrees for specific marker data. When pedigree structure problems could not be resolved, relevant individuals were removed from further analysis. Genotypes of individual markers with evidence of genotyping error were removed from any individuals for whom the presence of a genotype would be informative in a linkage analysis.

Covariate Adjustments

The primary trait was PPD adjusted for variability attributable to age, sex, BMI, and TG. Adjustment for TG was undertaken to map quantitative trait loci (QTLs) for LDL size rather than TG levels because these 2 variables are highly correlated, and our previous work demonstrated that adjustment for TG was crucial to identifying the strong linkage of LDL size to the lipoprotein lipase gene in heterozygous lipoprotein lipase deficiency.18 Segregation analyses discussed below further supported the increased power of mapping LDL size adjusted for PPD. Natural log TG (lnTG) was used because of the marked skewing of the TG distribution; 2 significantly predicted PPD (P<0.05) and was used for PPD adjustment. The linear term lnTG did not predict further variation in PPD (P>0.05).

For lod score analyses, linear regression across all 4 families was used to adjust for age, sex, BMI, and 2. Each family had regression residuals for covariates adjusted to a mean of zero, eliminating differences in mean family LDL size. The resulting adjusted trait for the individual and combined family lod score analyses is PPDasbt. Secondary lod score analyses were done without 2 adjustment (PPDasb) to determine whether TG adjustment influenced the linkage signals.

For MCMC analyses, the same covariates were considered as part of the joint linkage and segregation analyses. The MCMC trait was termed PPDasbt2 as a reminder that the lod score and MCMC analyses include the same covariates, but they are used differently. The MCMC approach accounts for the relationship among individuals, whereas the regression approach assumes that individuals are unrelated. Nevertheless, the coefficient differences between the resulting estimates were small (data not shown). For MCMC analyses, but not for the lod score analyses, covariate adjustments differ among individual family analyses.

Segregation Analyses

Complex segregation analysis (CSA) was performed on PPDasbt to estimate parameter values for Mendelian transmission models used in the lod score linkage analyses. Separate models were generated for individual families and for all families combined. PAP version 5 was used to compute likelihoods for the models considered.36,37 PAP approximations are fairly good even for very large pedigrees.38,39 No ascertainment correction was undertaken as a result of the lack of information on the relationship of the FCHL-based ascertainment to the PPD trait. Likelihood surfaces were interrogated repeatedly and systematically to avoid acceptance of a local maximum.

Lod Score Linkage Analyses

The lod score approach to linkage analysis40 as implemented by FASTLINK41,42 was used. This model-based approach is expected to have highest power to detect linkage if the Mendelian models considered are approximately correct. The basic genome scan used 2-point (1-marker) and 3-point (2-marker) linkage analysis with PPDasbt. Two-point methods for linkage detection are less sensitive than multipoint methods to model misspecifications,43,44 and are not affected by map misspecification, whereas multipoint methods provide more linkage information. Multipoint methods provide a valid test of linkage even when model misspecification may lead to biased estimates of location. Allele frequencies were estimated from our larger set of 15 FCHL families, the remaining 11 of which did not have PPD measure available, to avoid false-positive results attributable to low allele frequencies.45 Regions with 2-point maximum lod scores (zmax) 1 for all families, zmax2 for single families, 3-point zmax>2, or intensity ratios (IRs) 20 from the MCMC analyses were followed by multipoint lod score analyses with sets of 3 adjacent markers and adjusted PPD. Multipoint analyses used sex-specific map distances obtained from the Marshfield Medical Research Foundation web site converted to Haldane distances. In addition to gaining comparability with the MCMC analyses (below) that currently only use the Haldane map function, this map function is somewhat more robust to trait model misspecification for the lod score analyses because of the longer resulting genetic map. Sex-specific maps were used because sex-averaged map results can inflate the lod score when the male and female maps differ in length.46 Chromosomal locations are referred to here using the sex-averaged Haldane map. Sensitivity of analyses to marker allele frequencies were performed for regions with a zmax2 by increasing all rare marker frequencies to 0.025. This did not appreciably change the results. Multipoint lod score analyses were limited to 3 markers and a small number of regions because of the excessive computing time needed to carry out these analyses. The large size of the pedigrees allowed us to evaluate the possibility of genetic heterogeneity by examining the 4 families together and independently.

MCMC Linkage Methods

A Bayesian MCMC approach, as implemented in the computer program Loki,20,47 was also used for linkage analysis. Unlike the lod score method, these MCMC analyses allow for multiple QTLs, with the number of QTLs treated as a random variable and estimated in a joint linkage and segregation analyses. Thus, the genetic model is oligogenic and not prespecified. Loki version 2.4.5 was used for analysis. Details for carrying out the steps of the analyses are described previously.47,48 The information on unobserved multilocus marker states obtained with this program was shown recently to agree well with other estimates and to be superior to estimates obtained with multipoint approximation methods.49,50 In the analyses presented here, the MCMC sampler was run for 100 000 iterations, with every second iteration retained for estimating posterior distributions.

These MCMC methods require specification of previous distributions for analysis. For the number of QTLs in the model, a Poisson distribution with a mean of 2 was assumed, based on the mean number of loci contributing 10% of the total variance in a preliminary MCMC segregation analyses. The value for the variance in the genotype effect (?) was similarly obtained by varying ? in short segregation analysis–only runs and using the value that maximized the genetic variance. Allele frequencies were assumed uniform on 0 to 1. Marker allele frequencies were estimated within each MCMC run. Loki uses the Haldane map function for multipoint analysis, and for these analyses, sex-average maps were used.

As for the lod score analyses, the MCMC analyses were performed on all families combined and on each family separately. Each MCMC genome scan consisted first of an analysis of each chromosome using all markers on the chromosome in a single multipoint analysis. Second, all regions showing positive evidence of linkage were reanalyzed with single-marker analyses to ensure that there were no false-positive results from map distance misspecification or use of sex-average maps.46 All model parameters and covariate adjustments were simultaneously estimated within each joint MCMC linkage and segregation iteration. Thus, each MCMC iteration potentially considers a different model. Although MCMC methods are more robust than lod score methods to genetic heterogeneity,51,52 it was not known a priori whether a combined analysis of all 4 families would detect a locus that segregated in only 1 family, nor was it clear that enough data existed within 1 family to estimate all parameters simultaneously in the MCMC methods because the oligogenic model requires a large number of parameters.

There is no standard score that can be related easily to the lod score or to a P value to evaluate the strength of evidence for linkage. A commonly used measure of support in Bayesian analyses is the Bayes factor (BF),53 which is the ratio of the odds under 2 hypotheses compared. In an MCMC linkage analysis, the hypotheses that can be compared are that of linkage versus that of uniform QTL locations on the genome. A statistic that approximates the BF is the ratio of posterior to previous probabilities of linkage, and 2 estimates of this statistic are the L-score and the IR.47,48 The L-score represents the expectation of the probability ratio over the MCMC run, and the IR represents the ratio of the expectations of the same probabilities. L-score and IR are conservative estimates of the BF. We use the IR here because it is somewhat less conservative, where evidence of linkage is reasonably strong.48

Results

Models From CSA

When a single Mendelian model for PPDasbt transmission was generated for the combined families, it was most similar to the family 1–specific model (Table 1). The proportion of the total trait variance attributable to the codominant major locus (Vg/Vt) estimated by the combined family model was 0.60, and the polygenic variance was near zero. The correlation between LDL PPD and BMI was –0.36 and between LDL PPD and TG was –0.64. When PPDasb was considered, the model was substantially different (Table 1), with a rarer allele and larger genotype mean deviations. For PPDasb, the Vg/Vt of 0.36 was substantially lower than the 0.60 found for PPDasbt. The larger contribution of the major locus component to the variance supports using PPDasbt as the trait of interest.

TABLE 1. Models for Linkage of LDL PPD Adjusted for Age, Sex, BMI, and 2

Linkage Analyses

Four regions gave results that strongly supported linkage. These results are summarized here. All regions with significant or suggestive linkage analysis results are summarized in Table 2. The single and 2-marker lod scores for all families and each family individually for the entire genome scan, as well as the parametric multipoint plots, are available online at http://www.atvb.ahajournals.org.

TABLE 2. Candidate LDL Size Loci or Related Traits Mapped to Regions Detected (lod3) or Suggested by This Study (lod2.0 or IR >20)

The strongest evidence for linkage found by analysis of individual families was on chromosome 9 (Figure 1; Table 2). These analyses were based on the family-specific CSA models for PPDasbt. In family 1, a multipoint zmax=3.70 was found at D9S921 using the 3 telomeric 9p markers in family 1. The 2-point zmax=2.89 at marker D9S921 (24 cM; 9p23) and 3-point zmax=3.66 (D9S2169/D9S921). The regional zmax for the remaining individual families were low or zero. Using the MCMC method, chromosome 9 yielded an IR of 21 at 24 cM (9p23-ter) in family 1. To determine whether the differing covariate adjustments resulted in differences between the MCMC and lod score method analyses in the single family analyses, PPDasbt was used as the trait in the MCMC analyses, rather than estimating the covariate adjustments within the joint linkage and segregation analysis. In family 1, the IR for PPDasbt2 of 21 increased to 52 at 17 cM when the preadjusted PPDasbt values were used as input.

Figure 1. Genome scan summary of maximum 3-point lod scores and multipoint MCMC logIR in all families combined and in each family analyzed separately for linkage of LDL PPD using covariates of age, sex, BMI, and 2. Lod scores are plotted at the midpoint of the 2 markers used. MCMC results have been truncated to a low value of zero and plotted in a log10 scale for consistency with the lod score results.

Considering the 4 families combined, there was significant evidence for linkage to regions of chromosomes 11, 14, and 16 (Figure 1; Table 2). These lod score analyses considered the model from the joint segregation analysis of all 4 families. On chromosome 11, zmax=3.72 with 3-point analysis using markers D11S2371 and D11S2002 (95 cM). The 2-point zmax=1.52 with D11S2371 (85 cM; 11q13). The MCMC-combined family scan yielded an IR of 120 at 93 cM (chromosome 11q22). Considering the families individually, the 3-point zmax was 0.60, 1.30, 1.36, and 1.01 under the all-family model and 0.70, 0.44, 1.37, and 0.11 under the family-specific models for families 1 to 4, respectively. The maximum single family IR was 13.4 in family 2. IRs were <3 in the other families.

MCMC analyses yielded a signal of IR=71 at 85 cM on chromosome 14q. This region did not provide a lod score >3. The 2-marker zmax=1.77 with markers D14S592 (74 cM) and D14S588 (84 cM); the 2-point zmax=1.51 with D14S588. This signal was largely attributable to family 4, which had a 3-marker zmax=2.37. The remaining families had zmax ranging from 0.3 to 0.4. The single-family IR was 24.3 in family 4 and 6.7 in family 2.

For chromosome 16, the 3-marker (D16S3253-GATA67G11-D16S2624) multipoint zmax for the combined families was 3.00 near D16S516 (110 cM; 16q23), with a second peak of zmax=2.22 near D16S753. The 2-point zmax=1.66 at GATA67G11 (89 cM; 16q21). The 2-marker zmax=2.83 (D16S3253-GATA67G11) for the 4 families combined with 2.29, 0.37, 0.91, and 1.18 for families 1 to 4, respectively, using D16S3253-GATA67G11 or adjacent markers GATA67G11-D16S2624 and the combined family model. Family 1 has a very wide region with a 2-marker zmax>1.5, but under the family-specific model, the peak shifted from D16S3253-GATA67G11 to D16S753-D16S3396. This shift, some 20 cM toward the p-terminus, places the peak closer to the CETP locus (Figure 2). The MCMC IR=43, with a wide peak centered at 105 cM on chromosome 16q23. Using MCMC methods, all 4 families individually provided some support for linkage of PPDasbt2 to chromosome 16, with the highest IRs being 9.4 in family 1 and 7.6 in family 4. Families 3 and 4 had bimodal peaks on chromosome 16 (Figure 1), which may account for the wide peak seen in the analysis of the combined families (Figure 2).

Figure 2. Multipoint MCMC IR for linkage of LDL PPD using covariates of age, sex, BMI, and TG levels on chromosomes 9 (family 1), 11, 14, and 16. Chromosome numbers are indicated on each graph with the genome scan markers, and their positions are indicated on the top axis. The dark horizontal bar on the chromosome 16 figure indicates the estimated location of CETP.

Analyses Without TG Adjustment

As expected, without the TG adjustment (PPDasb), chromosome 9, 11, 14, and 16 linkage results substantially changed. The chromosome 9p family 1 2-point zmax dropped from 2.89 to 0.83. The combined family chromosome 14 2-point zmax dropped from 1.31 to 0.46, the chromosome 16 from 1.66 to 0.55, and chromosome 11 (85 cM) zmax from 1.52 to 0.45. Considering all 4 families, the strongest signal for PPDasb was suggestive evidence for linkage to the APOA1/C3/A5 gene cluster. In family 3, the ordered markers D11S1986 (118 cM), APOC3 (123 cM), and D11S1998 (126 cM) suggested linkage to PPDasb with zmax=1.95, 0.37, and 1.81, respectively. The APOC3 marker had a high rate of homozygosity in key individuals14 and thus lower informativeness for linkage analysis in this data set. In this region, the 2-marker (D11S2000-D11S1986) multipoint zmax=2.59 in family 3. All families had positive lod scores in this region, and the combined families yielded zmax=1.87 at D11S1998 and a 2-marker zmax=2.08. The APOC3 region gave no evidence of linkage when LDL size was adjusted for TG levels.

Other regions with zmax >1.5 for PPDasb across all 4 families combined included 1.52 at marker D1S1609 and 1.85 at D14S599. Each family had a single zmax>2: family 1 had zmax=2.29 at D19S433; family 2 had 2.17 at D21S1446; family 3 had zmax=2.09 at D9S1838; and family 4 had zmax=3.17 at D20S480. None of these regions overlapped with regions of interest for PPDasbt.

Discussion

We present evidence for the existence of 4 loci that influence TG-adjusted LDL PPD on chromosomes 9p, 11q, 14q, and 16q in a sample of 4 large FCHL pedigrees. LDL PPD has all the hallmarks of a complex trait, yet we were able to obtain strong and consistent evidence of linkage with statistically efficient model-based analyses by using a combination of 2 methods: a classic lod score and also a MCMC joint oligogenic segregation and linkage analysis approach. With both methods, we obtained strong evidence of linkage to chromosome 9p in a single family, and to chromosomes 11q and 16q in a joint analysis of all 4 families. In addition, we found consistent evidence for linkage at chromosome 14q24, which was stronger for the MCMC than the lod score method.

Our results, along with highly suggestive linkage to this region reported by Bosse et al,19 are highly consistent with a locus in this 14q region detected in 2 very different populations. Despite the lack of a standard criterion for significance of the IR, comparison of IR signals that are known to be significant by lod methods should suggest the range of IR consistent with linkage within a data set. The IR of 71 for the chromosome 14 region was intermediate to the IRs of chromosomes 11 and 16, regions that were significant across all families by the lod score method. Similarly, the single family IR of 24 was equivalent to chromosome 9 IR of 21, which also corresponded to a significant lod score. IRs for individual family analyses are not comparable to the IRs from the larger data set of all families combined. A second line of evidence supporting a 14q QTL is that this was a highly suggestive region (lod score=2.8) in the only other reported scan of TG-adjusted LDL PPD, which used a French-Canadian population.19 At sequential markers D14S592 (74 cM), D14S588 (84 cM; our peak), and D14S53 (97 cM), Bosse et al reported a zmax=1.73, 1.44, and 2.79. The dip at D14S588 may be attributable to less information or genotyping errors at the marker. We did not genotype D14S53, the marker at which the Bosse group reported their peak; instead, we genotyped D14S606 at 103 cM. The stronger results by MCMC than the lod score method may suggest that this locus is less consistent with the prespecified model used by the lod score analyses than chromosome 9, 11, and 16 loci. It is unclear to what extent the loci detected here are involved in the etiology of FCHL in these families, but there is growing evidence that FCHL is a multifactorial trait, and these may be contributing loci.

Our results support genetic heterogeneity in LDL size loci across and within families. MCMC methods originally estimated between 2 and 3 QTLs per family, and in these analyses, each family had 2 to 3 regions, with IRs >10, consistent with the original predictions. In the presence of locus heterogeneity, loci detected by individual studies are expected to vary with the population sampled and the method of ascertainment of the families.54,55 The detection of LDL size loci in families with FCHL is consistent with the excess of small, dense LDL seen in subjects with FCHL.9,23,24 That we detect multiple loci here suggests that even within FCHL, multiple loci contribute to variation in LDL size variation. This genetic heterogeneity is a major obstacle for the mapping and cloning of novel LDL size loci and will require strategies to minimize heterogeneity within a sample, such as the use of large families with similar features, as well as analytic methods that are more robust to heterogeneity and can provide relatively accurate gene localization such as MCMC.

Essential to this mapping success was the demonstration that PPD adjusted for TG, in addition to age, sex, and BMI, had a larger genetic contribution from a codominant single locus than did PPD not adjusted for TG. Indeed, linkage results on LDL PPD adjusted for age, sex, and BMI but not TG failed to detect any significant evidence of linkage, reducing the linkage evidence at the regions identified with TG adjustment and giving suggestive evidence (lod scores 2.1 to 2.6) in the APOA1/C3/A5 region. The stronger linkage results for TG adjusted LDL PPD suggests that the loci detected are not TG loci that are detected because of the relationship of TG and LDL size. Alternatively, the APOA1/C3/A5 cluster is known to affect TG56–58 and has been suggested as an FCHL susceptibility locus.59–61 Polymorphisms in this cluster have also been linked to HDL-C levels in these FCHL families.14

Comparison of results of the MCMC and lod score results was illuminating. The computational requirements differed. The single marker lod score analyses were faster than the MCMC analyses. In contrast, the MCMC multipoint analyses, incorporating all markers on a chromosome, were much faster than the 2- or 3-marker lod score analyses. Indeed, 3-marker lod score analyses were the limit of what was computationally practical in these pedigrees. The 2 methods gave generally consistent results, but there were also some differences. Both methods detected linkage to chromosomes 11 and 16 across all families combined. Although evidence from the lod score methods for a chromosome 14 QTL was only suggestive, evidence from the MCMC analysis was stronger for chromosome 14 than for 16. This may reflect differences in the number of markers that can be accommodated, the trait model that is estimated, or some other difference. For both methods, the chromosome 9 signal was only detected when family 1 was analyzed separately. Given the strength of this signal using lod score methods, it was initially surprising that there was no signal in the combined family set when analyzed by MCMC methods. This, as well as other strong signals seen in the MCMC single family analyses but not in the combined analyses, suggests that when a family is large enough, single family analyses by MCMC may yield information not found in the combined analyses. This may be a particular issue when ascertainment procedures differ among pedigrees.

Several other observations on the MCMC methods were made. First, single-family analyses supported locus heterogeneity between families, suggesting the specific families that were responsible for the linkage peaks in the combined family MCMC analysis. When families are not large, this might be approached by analyzing all but 1 family at a time to see which families contribute to the signal.62 Second, the large improvement in the IR for chromosome 9p linkage when preadjusted PPDasbt data were analyzed instead of PPDasbt2 being estimated within the MCMC analyses supports the concern that the available data in single families may be insufficient for accurate simultaneous estimation of trait models and covariate adjustments. Conversely, the IR for chromosome 16 remained 40 across the combined families when the preadjusted data were analyzed instead of the raw data, which may be expected given the larger data set used to estimate the covariate effects within the MCMC analysis.

Strong candidate genes that occur at regions we identify as potentially harboring LDL size loci are summarized in Table 2. The CETP locus on chromosome 16q13 lies between markers D16S3253 and GATA67G11 and, although not located at the signal peak, may contribute to this linkage signal, particularly if undetected genotyping errors or mutations exist or if there are multiple QTLs across the families. As discussed above, CETP has established effects on LDL size. Lecithin-cholesterol acyltransferase is further telomeric at 16q22.1 and has shown linkage to FCHL in 1 study.63 Without the TG adjustment of PPD, the chromosome 11 lod score peak was at the APOA1/C3/A5 gene cluster region, (120 cM) 35 cM from the peak for PPD with the TG adjustment (85 cM). The peak at the APOA1/C3/A5 gene cluster without but not with TG adjustment suggests that the linkage signal for PPDasbt at 85 cM is a separate locus from that at the APOA1/C3/A5 cluster. Bosse et al19 also detected a zmax=2.4 for PPD adjusted for age at the chromosome 5q region, near 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMGCR), where we detected a multipoint zmax=1.42 across families and zmax=1.56 in family 2.

In summary, lod score and MCMC results yielded significant linkage evidence for a chromosome 9p locus for LDL size in a single family with FCHL as well as significant evidence in all families combined for chromosome 11q and 16q loci. Our results also support previous evidence of linkage to chromosome 14q. Use of MCMC and lod score analyses in this complex trait were complementary. Identification of LDL size loci will aid in diagnosis, treatment, and perhaps prevention of vascular disease, as well as help unravel the complex genetics of FCHL.

Acknowledgments

This research was supported by National Institutes of Health (NIH) grant HL 30086 (G.J.), with additional support from NIH grants HL 18574, GM 46255 (E.W.), and HL 33577 (R.K.), a Pew Biomedical Scholar Award (G.J.), and a postdoctoral fellowship from the Canadian Institutes of Health Research (F.G.). Portions of this study were conducted at the Lawrence Berkeley Laboratory, supported by US Department of Energy contract DE-AC03-76F00098 and the Clinical Research Center of the University of Washington, supported by NIH grant RR00037. Genotypes were provided by the NHLBI Mammalian Genotyping Service grant NHLBI NO1-HV-48141. Computer systems support was provided by Hiep D. Nguyen, and the database was constructed and maintained by Ted Holzman.

References

Gardner C, Fortmann SP, Krauss RM. Association of small low-density lipoprotein particles with the incidence of coronary artery disease in men and women. J Am Med Assoc. 1996; 276: 875–881.

Stampfer M, Krauss RM, Ma J, Blanche PJ, Holl LG, Sacks FM, Hennekens CH. A prospective study of triglyceride level, low-density lipoprotein particle diameter, and risk of myocardial infarction. J Am Med Assoc. 1996; 276: 882–888.

Lamarche B, Tchernof A, Moorjani S, Cantin B, Dagenais GR, Lupien PJ, Despres JP. Small, dense low-density lipoprotein particles as a predictor of the risk of ischemic heart disease in men. Prospective results from the Quebec Cardiovascular Study. Circulation. 1997; 95: 69–75.

Austin M, Breslow JL, Hennekens CH, Buring JE, Willett WC, Krauss RM. Low-density lipoprotein subclass patterns and risk of myocardial infarction. J Am Med Assoc. 1988; 269: 1917–1921.

Austin M, King M, Vranizan K, Krauss R. Atherogenic lipoprotein phenotype: a proposed genetic marker for coronary heart disease risk. Circulation. 1990; 82: 495–506.

Chait A, Brazg RL, Tribble DL, Krauss RM. Susceptibility of small, dense, low-density lipoproteins to oxidative modification in subjects with the atherogenic lipoprotein phenotype, pattern B. Am J Med. 1993; 94: 350–356.

de Graaf J, Hendriks JC, Demacker PN, Stalenhoef AF. Identification of multiple dense LDL subfractions with enhanced susceptibility to in vitro oxidation among hypertriglyceridemic subjects. Normalization after clofibrate treatment. Arterioscler Thromb. 1993; 13: 712–719.

Bosse Y, Perusse L, Vohl MC. Genetics of LDL particle heterogeneity: from genetic epidemiology to DNA-based variations. J Lipid Res. 2004; 45: 1008–1026.

Allayee H, Aouizerat BE, Cantor RM, Dallinga-Thie GM, Krauss RM, Lanning CD, Rotter JI, Lusis AJ, de Bruin TW. Families with familial combined hyperlipidemia and families enriched for coronary artery disease share genetic determinants for the atherogenic lipoprotein phenotype. Am J Hum Genet. 1998; 63: 577–585.

Rotter J, Bu X, Cantor RM, Warden CH, Brown J, Gray RJ, Blanche PJ, Krauss RM, Lusis AJ. Multilocus genetic determinants of LDL particle size in coronary artery disease families. Am J Hum Genet. 1996; 58: 585–594.

Barzilai N, Atzmon G, Schechter C, Schaefer EJ, Cupples AL, Lipton R, Cheng S, Shuldiner AR. Unique lipoprotein phenotype and genotype associated with exceptional longevity. J Am Med Assoc. 2003; 290: 2030–2040.

Austin M, Edwards KL, Monks SA, Koprowicz KM, Brunzell JD, Motulsky AG, Mahaney MC, Hixson JE. Genomewide scan for quantitative trait loci influencing low-density lipoprotein size and plasma triglyceride in familial hypertriglyceridemia. J Lipid Res. 2003; 44: 2161–2168.

Wijsman E, Amos C. Genetic analysis of simulated oligogenic traits in nuclear and extended pedigrees: summary of GAW10 contributions. Genet Epidemiol. 1997; 14: 719–735.

Gagnon F, Jarvik GP, Motulsky AG, Deeb SS, Brunzell JD, Wijsman EM. Evidence of linkage of HDL level variation to APOC3 in two samples with different ascertainment. Hum Genet. 2003; 113: 522–533.

Hall J, Lee MK, Newman B, Morrow JE, Anderson LA, Huey B, King MC. Linkage of early-onset familial breast cancer to chromosome 17q21. Science. 1990; 250: 1684–1689.

Schellenberg G, Bird TD, Wijsman EM, Orr H T, Anderson LL, Nemens E, White JA, Bonnycastle L, Weber JL, Alonso ME, Potter H, Heston LL, Martin GM. Genetic linkage evidence for a familial Alzheimer’s disease locus on chromosome 14. Science. 1992; 258: 668–671.

Goddard K, Jarvik GP, Grahm J, McNeney B, Hsu L, Siegmund K, Grossner S, Olson J, Wijsman EM. Analysis of quantitative risk factors for a common oligogenic disease. Genet Epidem. 1995; 12: 759–764.

Hokanson J, Brunzell JD, Jarvik GP, Wijsman EM, Austin MA. Linkage of low-density lipoprotein size to the lipoprotein lipase gene in heterozygous lipoprotein lipase deficiency. Am J Hum Genet. 1999; 64: 608–618.

Bosse Y, Perusse L, Despres JP, Lamarche B, Chagnon YC, Rice T, Rao DC, Bouchard C, Vohl MC. Evidence for a major quantitative trait locus on chromosome 17q21 affecting low-density lipoprotein peak particle diameter. Circulation. 2003; 107: 2361–2368.

Heath S. Markov Chain Monte Carlo segregation and linkage analysis for oligogenic models. Am J Hum Genet. 1997b; 61: 748–760.

Goldstein J, Schrott HG, Hazzard WR, Bierman EL, Motulsky AG. Hyperlipidemia in coronary heart disease. II. Genetic analysis of lipid levels in 176 families and delineation of a new inherited disorder, combined hyperlipidemia. J Clin Invest. 1973; 52: 1544–1568.

Brunzell J, Schrott HG, Motulsky AG, Bierman EL. Myocardial infarction in the familial forms of hypertriglyceridemia. Metabolism. 1976; 25: 313–320.

Hokanson J, Austin MA, Zambon A, Brunzell JD. Plasma triglyceride and LDL heterogeneity in familial combined hyperlipidemia. Arterioscler Thromb. 1993; 13: 427–434.

Jarvik G, Brunzell JD, Austin MA, Krauss RM, Motulsky AG, Wijsman EM. Genetic predictors of FCHL in 4 large pedigrees. Influence of ApoB level major locus predicted genotype and LDL subclass phenotype. Arterioscler Thromb. 1994; 14: 1687–1694.

Ayyobi AF, Brunzell JD. Lipoprotein distribution in the metabolic syndrome, type 2 diabetes mellitus, and familial combined hyperlipidemia. Am J Cardiol. 2003; 92: 27J–33J.

Wijsman E, Brunzell J, Jarvik GP, Austin M, Motulsky A, Deeb S. Evidence against linkage of familial combined hyperlipidemia to the apolipoprotein AI-CIII-AIV gene complex. Arterioscler Thromb Vasc Biol. 1998; 18: 215–226.

Chait A, Brunzell JD. Severe hypertriglyceridemia: role of familial and acquired disorders. Metabolism. 1983; 32: 209–214.

Krauss R, Burke DJ. Identification of multiple subclasses of plasma low-density lipoproteins in normal humans. J Lipid Res. 1982; 23: 97–104.

Nichols A, Krauss RM, Musliner TA. Nondenaturing polyacrylamide gradient gel electrophoresis. In: Segest JP, Albers JJ, ed. Methods in Enzymology: Plasma Lipoproteins, Part A, Preparation, Structure and Molecular Biology. Orlando, Fla: Academic Press; 1986.

Austin M, Mykkanen L, Kuusisto J, Edwards KL, Nelson C, Haffner SM, Pyorala K, Laakso M. Prospective study of small LDLs as a risk factor for non insulin dependent diabetes mellitus in elderly men and women. Circulation. 1995; 92: 1770–1778.

Hokanson J, Austin MA, Brunzell JD. Measurement and clinical significance of low-density lipoprotein subclasses. In: Rifai N, Warnick GR, Dominiczak MH, eds. Handbook of Lipoprotein Testing. Washington, DC: American Association of Clinical Chemistry; 1997.

Warnick G. Enzymatic methods for quantification of lipoprotein lipids. Methods Enzymol. 1986; 129: 101–123.

Bachorik P, Albers, JJ. Precipitation methods for quantitation of lipoproteins. Methods Enzymol. 1986; 129: 78–100.

Boehnke M, Cox NJ. Accurate inference of relationships in sib-pair linkage studies. Am J Hum Genet. 1997; 61: 423–429.

Sieberts S, Wijsman EM, Thompson EA. Relationship inference from trios of individuals, in the presence of typing error. Am J Hum Genet. 2002; 70: 170–180.

Hasstedt S, Cartwright P. PAP: Pedigree Analysis Package. Salt Lake City, UT: Department of Medical Biophysics and Computing, University of Utah; 1981.

Hasstedt S. A mixed-model likelihood approximation on large pedigrees. Comput Biomed Res. 1982; 15: 295–307.

Olshen A, Wijsman EM. Pedigree analysis package vs. MIXD: fitting the mixed model on a large pedigree. Genet Epidemiol. 1996; 13: 91–106.

Snow G, Wijsman EM. Pedigree analysis package (PAP) vs. MORGAN: model selection and hypothesis testing on a large pedigree. Genet Epidemiol. 1998; 15: 355–369.

Morton N. Sequential tests for the detection of linkage. Am J Hum Genet. 1955; 7: 277–318.

Cottingham R, Idury, RM, Schaffer, AA. Faster sequential genetic linkage computations. Am J Hum Genet. 1993; 53: 252–263.

Schaffer A, Gupta SK, Shriram K, Cottingham RW Jr. Avoiding recomputation in linkage analysis. Hum Hered. 1994; 44: 225–237.

Clerget-Darpoux F. Bias of the estimated recombination fraction and lod score due to an association between a disease gene and a marker gene. Ann Hum Genet. 1982; 46: 363–372.

Risch N, Giuffra, L. Model misspecification and multipoint linkage analysis. Hum Hered. 1992; 42: 77–92.

Ott J. Strategies for characterizing highly polymorphic markers in human gene mapping. Am J Hum Genet. 1992; 51: 283–290.

Daw E, Thompson EA, Wijsman EM. Bias in multipoint linkage analysis arising from map misspecification. Genet Epidemiol. 2000; 19: 336–380.

Wijsman EM. Joint segregation and linkage analysis using Markov chain Monte Carlo methods. In: Camp N, Cox A, eds. Quantitative Trait Loci: Methods and Protocols. Totowa, NJ: Humana Press; 2002.

Wijsman E, Yu D. Joint oligogenic segregation and linkage analysis using Bayesian Markov chain Monte Carlo Methods. Mol Biotechnol. In press.

Wijsman E. Summary of Group 8: development and extension of linkage methods. Genet Epidemiol. 2003; 25 (suppl 1): S64–71.

Goring H, Williams JT, Dyer TD, Blangero J. On different approximations to multilocus identity-by-descent calculations and the resulting power of variance component-based linkage analysis. BMC Genet. 2003; 4 (suppl 1): S72.

Heath S, Snow GL, Thompson EA, Tseng C, Wijsman EM. MCMC segregation and linkage analysis. Genet Epidemiol. 1997; 14: 1011–1016.

Daw E, Heath SC, Wijsman EM. Multipoint oligogenic analysis of age of onset data with application to large Alzheimer’s disease pedigrees. Am J Hum Genet. 1999; 64: 839–851.

Kass R, Raftery AE. Bayes factors. J Am Stat Assoc. 1995; 90: 773–795.

Goldgar D, Easton DF. Optimal strategies for mapping complex diseases in the presence of multiple loci. Am J Hum Genet. 1997; 60: 1222–1232.

McCarthy M, Kruglyak L, Lander ES. Sib-pair collection strategies for complex diseases. Genet Epidemiol. 1998; 15: 317–340.

Pennacchio L, Olivier M, Hubacek JA, Cohen JC, Cox DR, Fruchart JC, Krauss RM, Rubin EM. An apolipoprotein influencing triglycerides in humans and mice revealed by comparative sequencing. Science. 2001; 294: 169–173.

Talmud PJ, Hawe E, Martin S, Olivier M, Miller GJ, Rubin EM, Pennacchio LA, Humphries SE. Relative contribution of variation within the APOC3/A4/A5 gene cluster in determining plasma triglycerides. Hum Mol Genet. 2002; 11: 3039–3046.

Lai C, Tai ES, Tan CE, Cutter J, Chew SK, Zhu YP, Adiconis X, Ordovas JM. The apolipoprotein A5 locus is a strong determinant of plasma triglyceride concentrations across ethnic groups in Singapore. J Lipid Res. 2003; 44: 2365–2373.

Xu C-F, Talmud P, Schuster H, Houlston R, Miller G, Humphries S. Association between genetic variation at the APO AI-CIII-AIV gene cluster and familial combined hyperlipidaemia. Clin Genet. 1994; 46: 385–397.

Groenendijk M, Cantor RM, Funke H, Dallinga-Thie GM. Two newly identified SNPs in the APO AI-CIII intergenic region are strongly associated with familial combined hyperlipidaemia. Eur J Clin Invest. 2001; 31: 852–859.

Eichenbaum-Voline S, Olivier M, Jones EL, Naoumova RP, Jones B, Gau B, Patel HN, Seed M, Betteridge DJ, Galton DJ, Rubin EM, Scott J, Shoulders CC, Pennacchio LA. Linkage and association between distinct variants of the APOA1/C3/A4/A5 gene cluster and familial combined hyperlipidemia. Arterioscler Thromb Vasc Biol. 2004; 24: 167–174.

Neuman RJ, Yuan B, Gerhard DS, Liu KY, Yue P, Duan S, Averna M, Schonfeld G. Replication of linkage of familial hypobetalipoproteinemia to chromosome 3p in six kindreds. J Lipid Res. 2002; 43: 407–415.

Aouizerat B, Allayee H, Cantor RM, Dallinga-Thie GM, Lanning CD, de Bruin TW, Lusis AJ, Rotter JI. Linkage of a candidate gene locus to familial combined hyperlipidemia:lecithin:cholesterol acyltransferase on 16q. Arterioscler Thromb Vasc Biol. 1999; 19: 2730–2736.

Ghosh S, Watanabe RM, Valle TT, Hauser ER, Magnuson VL, Langefeld CD, Ally DS, Mohlke KL, Silander K, Kohtamaki K, Chines P, Balow Jr J, Birznieks G, Chang J, Eldridge W, Erdos MR, Karanjawala ZE, Knapp JI, Kudelko K, Martin C, Morales-Mena A, Musick A, Musick T, Pfahl C, Porter R, Rayman JB. The Finland-United States investigation of noninsulin-dependent diabetes mellitus genetics (FUSION) study. I. An autosomal genome scan for genes that predispose to type 2 diabetes. Am J Hum Genet. 2000; 67: 1174–1185.

Watanabe RM, Ghosh S, Langefeld CD, Valle TT, Hauser ER, Magnuson VL, Mohlke KL, Silander K, Ally DS, Chines P, Blaschak-Harvan J, Douglas JA, Duren WL, Epstein MP, Fingerlin TE, Kaleta HS, Lange EM, Li C, McEachin RC, Stringham HM, Trager E, White PP, Balow J Jr, Birznieks G, Chang J, Eldridge W. The Finland-United States investigation of noninsulin-dependent diabetes mellitus genetics (FUSION) study. II. An autosomal genome scan for diabetes-related quantitative-trait loci. Am J Hum Genet. 2000; 67: 1186–1200.

Pajukanta P, Allayee H, Krass KL, Kuraishy A, Soro A, Lilja HE, Mar R, Taskinen MR, Nuotio I, Laakso M, Rotter JI, De Bruin TW, Cantor RM, Lusis AJ, Peltonen L. Combined analysis of genome scans of Dutch and Finnish families reveals a susceptibility locus for high-density lipoprotein cholesterol on chromosome 16q. Am J Hum Genet. 2003; 72: 903–917.

Chiodini B, Lewis CM. Meta-analysis of 4 coronary heart disease genome-wide linkage studies confirms a susceptibility locus on chromosome 3q. Arterioscler Thromb Vasc Biol. 2003; 23: 1863–1868.

Horikawa Y, Oda N, Cox NJ, Li XQ, Orho-Melander M, Hara M, Hinokio Y, Lindner TH, Mashima H, Schwarz PEH, del Bosque-Plata L, Horikawa Y, Oda Y, Yoshiuchi I, Colilla S, Polonsky KS, Wei S, Concannon P, Iwasaki N, Schulze T, Baier LJ, Bogardus C, Groop L, Boerwinkle E, Hanis CL, Bell GI. Genetic variation in the gene encoding calpain-10 is associated with type 2 diabetes mellitus. Nat Genet. 2000; 26: 163–175.

Hanis C, Boerwinkle E, Chakraborty R, Ellsworth DL, Concannon P, Stirling B, Morrison VA, Wapelhorst B, Spielman RS, Gogolin-Ewens KJ, Shepard JM, Williams SR, Risch N, Hinds D, Iwasaki N, Ogata M, Omori Y, Petzold C, Rietzch H, Schroder HE, Schulze J, Cox NJ, Menzel S, Boriraj VV, Chen X, Lim LR, Lindner T, Mereu LE, Wang YQ, Xiang K, Yamagata K, Yang Y, Bell GI. A genome-wide search for human noninsulin-dependent (type 2) diabetes genes reveals a major susceptibility locus on chromosome 2. Nat Genet. 1996; 13: 161–166.

Kovacs P, Hanson RL, Lee YH, Yang X, Kobes S, Permana PA, Bogardus C, Baier LJ. The role of insulin receptor substrate-1 gene (IRS1) in type 2 diabetes in Pima Indians. Diabetes. 2003; 52: 3005–3009.

Chen S, Tan KC, Lam KS. Effect of the microsomal triglyceride transfer protein -493 G/T polymorphism and type 2 diabetes mellitus on LDL subfractions. Atherosclerosis. 2003; 167: 287–292.

Rainwater DL, Almasy L, Blangero J, Cole SA, VandeBerg JL, MacCluer JW, Hixson JE. A genome search identifies major quantitative trait loci on human chromosomes 3 and 4 that influence cholesterol concentrations in small LDL particles. Arterioscler Thromb Vasc Biol. 1999; 19: 777–783.

Pullinger CR, Eng C, Salen G, Shefer S, Batta AK, Erickson SK, Verhagen A, Rivera CR, Mulvihill SJ, Malloy MJ, Kane JP. Human cholesterol 7alpha-hydroxylase (CYP7A1) deficiency has a hypercholesterolemic phenotype. J Clin Invest. 2002; 110: 109–117.

Couture P, Otvos JD, Cupples LA, Wilson PW, Schaefer EJ, Ordovas JM. Association of the A-204C polymorphism in the cholesterol 7alpha-hydroxylase gene with variations in plasma low density lipoprotein cholesterol levels in the Framingham Offspring Study. J Lipid Res. 1999; 40: 1883–1889.

Hui D, Howles PN. Carboxyl ester lipase: structure-function relationship and physiological role in lipoprotein metabolism and atherosclerosis. J Lipid Res. 2002; 43: 2017–2030.

Chen Q, Reis SE, Kammerer C, Craig WY, LaPierre SE, Zimmer EL, McNamara DM, Pauly DF, Sharaf B, Holubkov R, Bairey Merz CN, Sopko G, Bontempo F, Kamboh MI. Genetic variation in lectin-like oxidized low-density lipoprotein receptor 1 (LOX1) gene and the risk of coronary artery disease. Circulation. 2003; 107: 3146–3151.