Chapter 7. Heritability of LDL Peak Particle Diameter in the Québec Family Study

Yohan Bossé, Marie-Claude Vohl, Jean-Pierre Després, Benoît Lamarche, Treva Rice, D.C. Rao, Claude Bouchard, Louis Pérusse

L’objectif de cette étude était de vérifier l’existence de facteurs familiaux influençant le diamètre principal des particules LDL (DP-LDL). Le DP-LDL a été mesuré par électrophorèse sur gradient de gel de polyacrylamide chez 681 sujets. Le DP-LDL a été ajusté pour l’âge (DP-LDL1), l’âge et l’indice de masse corporelle (IMC) (DP-LDL2), ou l’âge, l’IMC et les triglycérides (DP-LDL3). Les résultats suggèrent que la cellule familiale explique 47.4, 46.7 et 48.9% de la variance totale de ces phénotypes, respectivement. Le patron de corrélations familiales indique aucune corrélation entre époux alors que des corrélations significatives sont observées entre les parents et les enfants et les frères et sœurs avec une héritabilité maximale de 59%, 58% et 52% pour DP-LDL1, DP-LDL2 et DP-LDL3, respectivement. Ces résultats suggèrent que la taille des particules LDL est fortement similaire à l’intérieur des familles et que la ressemblance familiale semble être principalement attribuable à des facteurs génétiques.

Heritability of LDL Peak Particle Diameter in the Québec Family Study

Yohan Bossé1,2, Marie-Claude Vohl1,2,3, Jean-Pierre Després1,2,4, Benoît Lamarche1,3, Treva Rice5, D.C. Rao5, Claude Bouchard6 and Louis Pérusse7

1- Lipid Research Center, CHUL Research Center; 2- Department of Food Sciences and Nutrition; 3- Institute on Nutraceutical and Functional Food; 4- The Quebec Heart Institute; 5- Division of Biostatistics, Washington University School of Medicine, St. Louis, Missouri; 6- Pennington Biomedical Research Center, Baton Rouge, Louisiana, USA; 7- Division of Kinesiology, Department of Social and Preventive Medicine, Laval University, Québec, Canada.

Short title: Familial resemblance for LDL size

Address all correspondence to:

Louis Pérusse, Ph.D. Physical Activity Sciences Laboratory, Division of Kinesiology Department of Preventive Medicine, Laval University, Ste-Foy, PQ G1K 7P4 Canada. Fax: (418) 656-3044; Tel: (418) 656-2131, extension 7831; E-mail: louis.perusse@kin.msp.ulaval.ca

Abstract

LDL size has been associated with the risk of coronary heart disease. The objective of the present study was to verify whether familial factors influence LDL peak particle diameter (LDL-PPD), a quantitative trait reflecting the size of the major LDL subclass. LDL-PPD was measured by 2-16% polyacrylamide gradient gel electrophoresis in 681 members of 236 nuclear families participating in the Québec Family Study. LDL-PPD was adjusted for either age (LDL-PPD1), age and body mass index (LDL-PPD2) or age, body mass index and plasma triglyceride levels (LDL-PPD3), separately in men and women. The residual scores were used to test for familial aggregation using an ANOVA as well as computing maximum likelihood estimates of familial correlations. The ANOVA revealed that family lines accounted for 47.4, 46.7 and 48.9% of the variance in the LDL-PPD1, LDL-PPD2 and LDL-PPD3 phenotypes, respectively. The pattern of familial correlations revealed no significant spouse correlations but significant parent-offspring and sibling correlations for the three LDL-PPD phenotypes with maximal heritability estimates of 59%, 58% and 52% for LDL-PPD1, LDL-PPD2 and LDL-PPD3, respectively. These results suggest that LDL-PPD strongly aggregates in families and that the familial resemblance appears to be primarily attributable to genetic factors. Genes responsible for this genetic contribution remain to be identified.

Key words : genetics, lipoproteins, LDL size

Introduction

Data from case-control [Austin et al., 1994] and prospective [Gardner et al., 1996; Lamarche et al., 2001; Stampfer et al., 1996] studies suggest that small dense LDL particles are associated with increased risk of coronary heart disease (CHD). Defining the genetic and environmental factors modulating LDL particle size may be helpful in understanding its relationship with CHD. Fisher and collaborators [Fisher et al., 1975] were the first to report a genetic effect on LDL subclass phenotypes. Their finding of significant parent-offspring correlations but no spouse resemblance was interpreted as strong support for a genetic determination. Two large twin studies also provided evidence of heritability for LDL particle size. The first study was based on 109 monozygotic (MZ) and 113 dizygotic (DZ) male twin pairs, aged between 59 and 70 years, participating in the third examination of the National Heart, Lung, and Blood Institute Twin Study [Lamon-Fava et al., 1991]. The heritability estimate was performed on the weighted LDL type measured by gradient gel electrophoresis which takes into account both the major and satellite bands. The weighted LDL type intraclass correlation coefficient was higher in MZ twins (0.58) than in DZ twins (0.32), with a heritability of 52% prior to controlling for covariate effects and 39% after adjusting for BMI, alcohol consumption, cigarette smoking and physical activity. Similar results were obtained when only the major LDL band was used as a variable. The second study was based on 203 MZ and 145 DZ pairs of adult female twins, participating in the second examination of the Kaiser Permanente Women Twins Study [Austin et al., 1993b]. In this study, the heritability estimate was performed using the LDL Peak Particle Diameter (LDL-PPD) measured by gradient gel electrophoresis. This phenotype is a continuous variable reflecting the size of the major LDL subclass. Again the intraclass correlation coefficient was higher in MZ twins (0.71) than in DZ twins (0.44) and the heritability coefficient was estimated to reach 54%.

Twin studies represent a powerful design to detect the presence of a genetic effect. However, heritability estimates derived from twin studies should always be interpreted with caution as they assume that the difference in the correlations between MZ and DZ twins is entirely ascribed to genetic factors. This is analogous to saying that both types of twins have been and are exposed to similar environmental conditions. However, failure to meet this assumption typically leads to an overestimation of the heritability. Consequently, estimates derived from twin studies may represent the upper bound estimates of heritability, and other study designs should be used to verify these estimates. Thus far, only one study provide heritability estimates of LDL-PPD based on a family study design [Edwards et al., 1999]. The cohort was based on two family studies, the first ascertained through hyperlipidemic proband surviving a myocardial infarction and the second ascertained through hypertriglyceridemic proband. The heritability coefficient was 34% for the LDL-PPD adjusted for age and gender effects. However, heritability estimates for a multifactorial phenotype derived from such high-risk families may not properly quantify the strength of the familial resemblance for the majority of the population. Thus, the purpose of the present study was to assess the heritability of LDL-PPD based on subjects participating in the Québec Family Study (QFS).

Methods

Population

The QFS is a prospective study monitoring several phenotypes among French-Canadian families with the aim of investigating the genetics of obesity and its comorbidities [Bouchard, 1996]. For the present study, a total of 681 individuals (aged 41.1 ± 17.7 years) including 285 men and 396 women from 236 nuclear families were available. These families included a subsample of 100 families randomly ascertained with regards to obesity, while the remaining families were ascertained through one or more obese probands. Table 1 presents the characteristics of subjects in each of the sex and generation groups. The majority of the families consisted of families with both parents and at least one child (43%), families composed exclusively of siblings (25%) and families with data on mothers and offspring (19%). The study was approved by the Laval University Medical Ethics Commitee, and all subjects provided written informed consent. All the procedures followed were in accordance with institutional guidelines.

Phenotypes measurements

LDL-PPD were measured by electrophoresis with 8X8-cm nondenaturing 2-16% polyacrylamide gradient gels as described in details eslewhere [St-Pierre et al., 2001]. Triglyceride levels were assayed from blood samples collected in the morning after a 12-hour overnight fast. Total triglyceride concentrations were determined enzymatically with commercial kits as previously described [Perusse et al., 1989]. Body mass index (BMI) was calculated as weight (kg)/height (m2).

Data Adjustments

Before computing familial correlations, the LDL-PPD was adjusted for the effects of age and other covariates in both the mean and variance, as explained elsewhere [Perusse et al., 1997]. Briefly, LDL-PPD was regressed on up to a cubic polynomial in age (age, age2 and age3) and other covariates using a stepwise multiple regression procedure (mean regression), performed separately in each age- (<30, 30-50, and ≥50 years) by-sex (male vs. female) group and retaining only terms that were significant at the 5% level. To generate regression equations that were not affected by extreme scores, individuals with phenotypic values beyond ± 3 SD from the mean were identified and temporarily set aside. After estimation of the regression parameters for every group, these outliers were added back for computation of residual scores. One female subject with a residual score above 4 SD and with more then 1 SD from the previous highest score was excluded from the analysis because she was a distinct outlier. The phenotype used to estimate familial correlations was the residual from the mean regression standardized to a zero mean and unit variance. Three LDL-PPD phenotypes based on three different adjustment procedures were computed: LDL-PPD1 adjusted for age effects, LDL-PPD2 adjusted for age and BMI effects and LDL-PPD3 adjusted for age, BMI and plasma triglyceride effects. Table 3 presents the significant covariates for each phenotype within each age-by-sex group. Descriptive statistics and phenotype adjustments were performed using SAS (version 8.02).

Familial Correlation Model

The presence of familial resemblance was first tested using an ANOVA comparing between- versus within-family variance. This test was performed with the general linear model including the LDL-PPD phenotype as the dependent variable and the family line (family number) as the independent variable. The familial correlation model was based on four groups of individuals (fathers=F, mothers=M, sons=S, and daughters=D) leading to 8 correlations: 1 spouse (FM), 4 parent-offspring (FS, FD, MS, and MD), and 3 sibling (SS, DD, and SD) coefficients. Correlations were estimated using maximum likelihood methods of the computer program SEGPATH [Province & Rao, 1995]. A general (Model 1) and eight reduced models (Models 2 through 9) testing specific null hypotheses were fitted to the data. Null hypotheses were tested using the likelihood ratio test, which is the difference in minus twice the log-likelihoods (-2 ln L) between the general and a reduced model. The likelihood ratio is approximately distributed as a χ2, with the degrees of freedom being the difference in the number of parameters estimated in the two models being contrasted. Nonrejected models (P > 0.05) were combined into a single test with the aim of finding the most parsimonious model. This model is the one that best fit the data with the fewest parameters. The most parsimonious model was chosen from among all non-rejected alternatives using the Akaike's Information Criterion [Akaike, 1974] (AIC), which is -2 ln L plus twice the number of estimated parameters. The most parsimonious model is the one with the smallest AIC value.

The specific hypotheses tested in each model were the following. Sex and generation differences in correlations were first considered by testing the hypotheses of no sex differences in offspring in Model 2 (FS = FD, MS = MD, SS = DD = SD, df = 4), no sex differences in offspring or parents in Model 3 (FS = FD = MS = MD, SD = SS = DD, df = 5), and no sex nor generation differences in Model 4 (FS = FD = MS = MD = SD = SS = DD, df = 6). In Model 5, all eight correlations were equated to yield to the so-called environmental model (FM = FS = FD = MS = MD = SD = SS = DD, df = 7). The remaining models tested the strength of the familial resemblance, including the hypotheses of no sibling correlation in Model 6 (SS = DD = SD = 0, df = 3), no parent-offspring correlations in Model 7 (FS = FD = MS = MD = 0, df = 4), no spouse correlation in Model 8 (FM = 0, df = 1) and finally no familial resemblance at all in Model 9 (FM = FS = FD = MS = MD = SD = SS = DD = 0, df = 8). The maximal heritability (h2) was computed using the correlations from the most parsimonious model according to the following equation :

h2 = [(rsibling + rparent-offspring)(1 + rspouse)] / [(1 + rspouse) + 2 (rspouse)(rparent-offspring)]

This maximal heritability is defined as the percent of variance due to all additive familial effects (including both genetic and nongenetic) and is adjusted for the degree of spouse resemblance. The 95% confidence intervals associated with the heritability coefficient was also calculated using the same equation as above by substituting the standard errors obtained from the estimates of the familial correlation.

Results

Prior to familial correlation estimation, LDL-PPD phenotypes were adjusted in the six age-by-sex groups (see Methods) for the effects of age, BMI and triglyceride. There were no significant age effects for LDL-PPD phenotypes, while BMI accounted for less than 10% of the variance. Triglyceride levels had significant effects in each group accounting for between 8 to 48% of the variance in LDL-PPD. The ANOVA (results not shown) revealed that there were about two times more variance between families than within families and that family lines accounted for 47 to 49% of the variance in the LDL-PPD phenotypes. A summary of the correlation model results is presented in Table 2. For each model, the P values and the AIC values are shown. For all three phenotypes the hypotheses of no sibling (Model 6), no parent-offspring (Model 7) and no familial (Model 9) correlations are strongly rejected. On the other hand, the no spouse (Model 8) correlation is accepted for each LDL-PPD phenotype. This pattern of correlations is consistent with the hypothesis that the familial resemblance in LDL size is primarily attributable to genetic factors. For LDL-PPD1 and LDL-PPD2 all tests, except the no spouse correlation, are rejected leaving model 8 as the most parsimonious one. Concerning the LDL-PPD3 phenotype, four models in addition to the no spouse correlation model are accepted: Model 2 for no sex differences in offspring; Model 3 for no sex differences in offspring or parents; Model 4: for no sex nor generation differences; and Model 5: for the environmental model. To determine the most parsimonious model in such case, we combined all nonrejected null hypotheses. This was done by combining the best sex/generation models (Models 2 through 4) with the best model for the level of familial correlations (Models 6 through 9). The best sex/generation model was model 4 (FS = FM = MS = MD = SD = SS = DD) and the best model for the strength of familial correlations was model 8 (FM = 0). As shown in Table 2, the combined test of no sex nor generation differences (Model 4) and no spouse correlation (Model 8) did fit by likelihood ratio test (P = 0.366) and also provided the smallest AIC value (9.63). This combination of models 4 and 8 (i.e., no sex or generation differences and no spouse resemblance) was then chosen as the most parsimonious hypothesis for the LDL-PPD3.

Maximum likelihood estimates of the familial correlations under the general and most parsimonious models are presented in Table 3 for the three LDL-PPD phenotypes. Table 3 also presents the maximal heritability coefficients calculated from the most parsimonious models. For the LDL-PPD1, LDL-PPD2 and LDL-PPD3 the heritabilities are 59%, 58% and 52%, respectively. For all phenotypes, heritability estimates are based on models in which there is no spouse correlation, suggesting that only genetic factors account for these estimates. These estimates of heritability are expressed as a percentage of the residual variance (ie, after removing effects associated with covariates). Heritability can also be expressed as a percent of total variance. This is done by multiplying the heritability estimates by the residual variance after adjustment for the covariates. Table 4 presents the percentage of the total variance explained by the heritability component and the total variance explained by both heritability and covariates. The residual heritability (h2) is fairly similar among the three phenotypes. However, the total variance explained by heritability (h2*) is about 20% lower for LDL-PPD3 as compared to the other phenotypes. This decrease is attributed to removing the variability in LDL-PPD that is shared with triglyceride (i.e., covariance). On the other hand, the total variance explained by both heritability factor and covariates is 5% greater for LDL-PPD3 as compared to LDL-PPD1 and LDL-PPD2.

Discussion

The present study is the first to provide heritability estimates of LDL peak particle diameter (LDL-PPD) using families randomly ascertained with regards to their lipid and lipoprotein profile. Three LDL-PPD phenotypes derived by adjusting variously for the effects of age (LDL-PPD1), age and BMI (LDL-PPD2), and age, BMI and triglycerides (LDL-PPD3) were considered. The results suggest that these phenotypes strongly aggregate in families and are characterized by significant maximal heritability estimates of 59%, 58% and 52%, respectively. In addition, the lack of significant spouse correlation, combined with significant parent-offspring and sibling correlations, suggests that genetic factors are likely the major determinants of the familial aggregation.

Twin studies have demonstrated that approximately 50% of the variability in LDL size is attributed to genetic factors [Austin et al ., 1993b; Lamon-Fava et al ., 1991]. The coefficient was lower when it was derived from 85 families ascertained through hyperlipidemic proband participating in the Genetic Epidemiology of Hypertriglyceridemia (GET) study. This high-risk CHD family study design suggested that approximately one third of the residual variance in LDL-PPD (h2=34%) was attributable to additive genetic effects [Edwards et al ., 1999]. The results of the present study are similar to those reported from twin studies.

Others studies have attempted to uncover the genetics architecture underlying the small dense LDL phenotype. A number of studies have investigated the inheritance of the trait using complex segregation analysis [Austin et al., 1990; Austin et al., 1993a; Austin et al., 1988; Bredie et al., 1996; de Graaf et al., 1992; Friedlander et al., 1999; Vakkilainen et al., 2002]. Despite using different types of family structures, different criteria for proband ascertainment and the use of different techniques to characterize LDL heterogeneity, these studies were consistent in finding a major gene effect influencing the phenotype. Additionally, numerous candidate gene studies have been tested for their potential association or linkage with the small dense LDL phenotype. Unfortunately, due to the inability to replicate positive findings, results derived from these candidate gene studies are inconclusive so far.

LDL-PPD and triglyceride levels are traits with large genetic and environmental correlations [Edwards et al ., 1999]. In the present study, the total variance explained by heritability was lower when LDL-PPD was adjusted for triglycerides. In fact, the heritability factors explained 39% of the total variance in LDL-PPD3 compared to 59% and 57% in LDL-PPD1 and LDL-PPD2, respectively. We assume this reduction is caused by the removal of some of the shared additive genetic and environmental contributions of triglycerides to the variance in LDL-PPD. Such adjustments have the potential to eliminate pleiotropic effects of genes and to narrow the contribution of heritable factors to that specific to the phenotype of interest. This type of adjustment can also be useful in finding genetic loci contributing to LDL size. Indeed, it has been proposed that removing the effects accounted for by covariates and using the residual trait in linkage analysis may increase the likelihood of detecting genes unique to that trait [Comuzzie et al., 1997].

Overall, the results presented in this study indicate a strong familial aggregation with maximal heritability estimates above 50% for LDL-PPD. The pattern of familial correlations suggests that this effect is primarily attributable to genetic factors. Molecular studies are warranted to identify genes responsible for this large genetic contribution.

Acknowledgments

This study was supported by the Canadian Institutes of Health Research (MT-13960, MOP-14014 and GR-15187). The authors would like to express their gratitude to the subjects for their excellent collaboration and to the staff of the Physical Activity Sciences Laboratory for their contribution to the study. Y. Bossé is the recipient of a studentship from the "Fonds pour la formation de chercheurs et l'aide à la recherche (FCAR) et le Fonds de la recherche en santé du Québec (FRSQ)". M.C. Vohl is a research scholar from the FRSQ. J.P. Després is chair professor of human nutrition, lipidology and prevention of cardiovascular disease supported by Provigo and Pfizer Canada. Benoît Lamarche is the recipient of the Canada Research Chair in Nutrition, Functional Foods and Cardiovascular Health. C. Bouchard is partially supported by the George A. Bray Chair in Nutrition.

References

Akaike H. 1974. A new look at the statistical model identification. IEEE Trans Automat Control 19: 716-23.

Austin MA, Brunzell JD, Fitch WL, Krauss RM. 1990. Inheritance of low density lipoprotein subclass patterns in familial combined hyperlipidemia. Arteriosclerosis 10: 520-30.

Austin MA, Hokanson JE, Brunzell JD. 1994. Characterization of low-density lipoprotein subclasses: methodologic approaches and clinical relevance. Curr Opin Lipidol 5: 395-403.

Austin MA, Jarvik GP, Hokanson JE, Edwards K. 1993a. Complex segregation analysis of LDL peak particle diameter. Genet Epidemiol 10: 599-604.

Austin MA, King MC, Vranizan KM, Newman B, Krauss RM. 1988. Inheritance of low-density lipoprotein subclass patterns: results of complex segregation analysis. Am J Hum Genet 43: 838-46.

Austin MA, Newman B, Selby JV, Edwards K, Mayer EJ, Krauss RM. 1993b. Genetics of LDL subclass phenotypes in women twins. Concordance, heritability, and commingling analysis. Arterioscler Thromb 13: 687-95.

Bouchard C. 1996. Genetic epidemiology, association, and sib-pair linkage: Results from the Québec Family Study. In: Bray GA, Ryan DH, eds. Molecular and genetic aspects of obesity. Baton Rouge, LA: Louisiana State University Press. p 470-81.

Bredie SJ, Kiemeney LA, de Haan AF, Demacker PN, Stalenhoef AF. 1996. Inherited susceptibility determines the distribution of dense low- density lipoprotein subfraction profiles in familial combined hyperlipidemia. Am J Hum Genet 58: 812-22.

Comuzzie AG, Mahaney MC, Almasy L, Dyer TD, Blangero J. 1997. Exploiting pleiotropy to map genes for oligogenic phenotypes using extended pedigree data. Genet Epidemiol 14: 975-80.

de Graaf J, Swinkels DW, de Haan AF, Demacker PN, Stalenhoef AF. 1992. Both inherited susceptibility and environmental exposure determine the low-density lipoprotein-subfraction pattern distribution in healthy Dutch families. Am J Hum Genet 51: 1295-310.

Edwards KL, Mahaney MC, Motulsky AG, Austin MA. 1999. Pleiotropic genetic effects on LDL size, plasma triglyceride, and HDL cholesterol in families. Arterioscler Thromb Vasc Biol 19: 2456-64.

Fisher WR, Hammond MG, Mengel MC, Warmke GL. 1975. A genetic determinant of the phenotypic variance of the molecular weight of low density lipoprotein. Proc Natl Acad Sci U S A 72: 2347-51.

Friedlander Y, Kark JD, Sinnreich R, Edwards KL, Austin MA. 1999. Inheritance of LDL peak particle diameter: results from a segregation analysis in Israeli families. Genet Epidemiol 16: 382-96.

Gardner CD, Fortmann SP, Krauss RM. 1996. Association of small low-density lipoprotein particles with the incidence of coronary artery disease in men and women. Jama 276: 875-81.

Lamarche B, St-Pierre AC, Ruel IL, Cantin B, Dagenais GR, Després JP. 2001. A prospective, population-based study of low density lipoprotein particle size as a risk factor for ischemic heart disease in men. Can J Cardiol 17: 859-65.

Lamon-Fava S, Jimenez D, Christian JC, Fabsitz RR, Reed T, Carmelli D, Castelli WP, Ordovas JM, Wilson PW, Schaefer EJ. 1991. The NHLBI Twin Study: heritability of apolipoprotein A-I, B, and low density lipoprotein subclasses and concordance for lipoprotein(a). Atherosclerosis 91: 97-106.

Perusse L, Després JP, Tremblay A, Leblanc C, Talbot J, Allard C, Bouchard C. 1989. Genetic and environmental determinants of serum lipids and lipoproteins in French Canadian families. Arteriosclerosis 9: 308-18.

Perusse L, Rice T, Després JP, Rao DC, Bouchard C. 1997. Cross-trait familial resemblance for body fat and blood lipids: familial correlations in the Quebec Family Study. Arterioscler Thromb Vasc Biol 17: 3270-7.

Province MA, Rao DC. 1995. General purpose model and a computer program for combined segregation and path analysis (SEGPATH): automatically creating computer programs from symbolic language model specifications. Genet Epidemiol 12: 203-19.

Stampfer MJ, Krauss RM, Ma J, Blanche PJ, Holl LG, Sacks FM, Hennekens CH. 1996. A prospective study of triglyceride level, low-density lipoprotein particle diameter, and risk of myocardial infarction. Jama 276: 882-8.

St-Pierre AC, Ruel IL, Cantin B, Dagenais GR, Bernard PM, Després JP, Lamarche B. 2001. Comparison of various electrophoretic characteristics of LDL particles and their relationship to the risk of ischemic heart disease. Circulation 104: 2295-9.

Vakkilainen J, Pajukanta P, Cantor RM, Nuotio IO, Lahdenpera S, Ylitalo K, Pihlajamaki J, Kovanen PT, Laakso M, Viikari JS, Peltonen L, Taskinen MR. 2002. Genetic influences contributing to LDL particle size in familial combined hyperlipidaemia. Eur J Hum Genet 10: 547-52.

Table I. Descriptive Statistics of LDL Peak Particle Diameter and Covariates in Each of the Sex and Generation Groups.

Table II. Summary of Goodness of Fit Tests for LDL Peak Particle Diameter Phenotypes.

Table III. Familial Correlations (±SE) and Maximal Heritability Under the General and the Most Parsimonious Models.

Table IV. Percentage of Variance Explain by Covariates and Heritability.