Shared and Unique Genetic Effects Among Seven HDL Phenotypes
Abstract The purpose of this study was to investigate the genetic control of various HDL measures and to determine the proportion of genetic variance explained by shared genes (ie, pleiotropy) and the proportion unique to each trait. The data used were drawn from large, randomly ascertained pedigrees of Mexican Americans participating in the San Antonio Family Heart Study. Data were available for 655 individuals (258 men and 397 women) in 26 families. We performed a multivariate quantitative genetic analysis to simultaneously estimate both the additive genetic and random environmental correlations among seven HDL phenotypes. These seven HDL phenotypes can be divided into two categories: measures of concentration and estimates of particle size. Concentration was measured for apo A-I, apo A-II, esterified cholesterol, and unesterified cholesterol, and particle size was estimated for apo A-I, apo A-II, and esterified cholesterol. The heritabilities (h2) for each of the seven traits were significantly greater than zero (P<.05) and ranged from 0.2 to 0.6. When considered in a pairwise fashion, all combinations of these traits showed marked genetic correlations (ρG=0.33 to 0.87) and all were significantly greater than zero (P<.05), indicative of pleiotropic effects. However, we found substantial unique genetic variance for each of these traits even after accounting for the effects shared in common with all the remaining measures. We conclude that the genetic variation in these HDL phenotypes is a result of the action of common as well as unique genes.
- Received July 9, 1996.
- Accepted October 2, 1996.
High-density lipoprotein is a complex lipoprotein composed of cholesterol (both esterified and unesterified), apolipoproteins (principally apo A-I and apo A-II), triglycerides, and phospholipids, and its plasma concentrations are considered to be a strong indicator of risk for the development of atherosclerosis. Clinically, HDL levels can be assessed by measurement of serum concentrations of any of its several different constituents (eg, cholesterol, apo A-I, apo A-II) or by estimating average particle sizes. Given the direct relationship of the various constituents of HDL to the particle as a whole, as well as the common role of these constituents in reverse cholesterol transport, it is not surprising that all the clinical measures are strongly correlated phenotypically. In other words, to some degree all of these clinical measures are reflections, though perhaps from varying perspectives, of the same larger phenotype.
Given the physical and functional relationships among these HDL measures, we can hypothesize that many of the observed phenotypic correlations are in fact due to an underlying structure of common genetic effects (ie, pleiotropy), with this situation being manifest as a pattern of significant ρG among all of these various HDL measures. To test this hypothesis, we conducted a multivariate quantitative genetic analysis in which we decomposed the phenotypic correlations among a group of seven related HDL measures into the portion due to genes shared in common and that due to shared environment. By use of this approach, it is possible to directly assess the extent of pleiotropy underlying the relationships among HDL measures. At the same time, however, it is also possible to use this information to remove the common genetic influences on each of these measures so as to allow for the identification of the genetic elements unique to each of these measures.
This study uses data obtained from a large sample of Mexican American families living in San Antonio, Tex, who are participants in the San Antonio Family Heart Study, a broader project examining the genetics of atherosclerosis and its risk factors.1 Ascertainment of families for that study was done without regard to preexisting medical conditions. For this project, data were available for 655 individuals (258 men and 397 women) ranging in age from 16 to 87 years, with an average age of 39.7 years. Of the 655 individuals, 10 were unrelated to any other individuals in the sample, and the remaining individuals were distributed among 26 pedigrees. The sizes of these pedigrees ranged from 3 to 71 individuals, with the majority of the pedigrees containing at least three generations of relatives.
At the clinic visit, a blood sample was obtained after an over-night fast. Plasma was divided into aliquots2 and stored at −80°C until used. In addition, information about age, sex, health status, and medications was also collected. An Institutional Review Board (University of Texas Health Science Center at San Antonio) approved the procedures, and all subjects gave informed consent.
HDL Concentration Measures
β-Lipoproteins were precipitated from plasma samples by use of dextran sulfate,3 and HDL cholesterol concentrations were assayed enzymatically in a Gilford SBA-300 Clinical Chemistry Analyzer with commercial reagents supplied by Boehringer-Mannheim Diagnostics. HDL-UC concentrations were measured by use of a similar kit (Wako) except that the enzyme cholesterol esterase was not added; the concentrations of HDL-EC were calculated as the difference between HDL-C and HDL-UC. The interassay coefficients of variation for control products in these assays were 6.2% for HDL-C and 9.0% for HDL-UC. Apolipoprotein concentrations were determined by a commercial laboratory (Medical Research Laboratories). Apo A-I concentrations were determined by nephelometry,4 5 and apo A-II concentrations were determined with competitive immunoassays.6 The interassay coefficients of variation for control products in these assays were 3.5% for apo A-I and 4.4% for apo A-II.
HDL Particle Size Measures
Size distributions of HDL particles were determined by electrophoresis of plasma in nondenaturing 3% to 31% polyacrylamide gradient gels, which were made in the laboratory as previously described.7 After electrophoretic separation, the proteins were transferred to nitrocellulose paper and detected by binding with sheep anti–apo A-I and sheep anti–apo A-II antibodies (both from Boehringer-Mannheim) as previously described.8 9 These antibodies were in turn bound by donkey anti-sheep IgG (Chemicon International, Inc), which was radioiodinated by the chloramine T method,10 and distributions were measured by densitometry of autoradiograms with an LKB Ultroscan Laser Densitometer with GSXL software (Pharmacia). The distribution of cholesteryl esters among HDLs was detected by staining with Sudan black B and densitometry as described.11 12 HDL absorbance profiles were decomposed by a curve-fitting procedure as suggested previously13 ; we developed in house a program to automatically fit curves representing the HDL subclasses and to calculate fraction of absorbance in each of the HDL subclasses. We summed the data to generate fractional absorbance for HDL3 particles and for those larger than HDL3. To avoid the problems of singularity for proportional data (ie, the two fractions must sum to one and are not independent), we used a log ratio transformation as suggested by Aitchison,14 in which the fractional absorbance for larger HDLs was divided by the fractional absorbance for HDL3. Thus, each of the size distributions was summarized into a single variable, appropriate for statistical analyses, which was positively correlated with average size of HDLs; these variables are HDL-Csize, HDL-A-Isize, and HDL-A-IIsize.
In addition to the seven HDL measures, a number of covariate effects were also included in the analysis. Since all of these traits had previously been screened for a variety of covariate effects in univariate analyses, the multivariate analysis included only covariates that had previously been found to be significant for at least one of the seven traits examined. In addition, all seven HDL measures had previously been screened for shared household effects, but none were detected. The covariate effects that were included in the multivariate analysis were sex, sex-specific age, hormone replacement therapy, diabetic status, and use of diabetic medications. It must be remembered, however, that no model is a true reflection of the natural world, and undoubtedly as yet unidentified covariates exist that could have a significant effect. However, the inclusion of such environ-mental covariates would only serve to strengthen the results (ie, increase the relative genetic contribution) by adding further resolution to the model.
By use of established quantitative genetic theory, it is possible to extend a univariate genetic analysis to encompass the multivariate state.15 16 17 18 Variance decomposition techniques16 17 18 using maximum-likelihood methods and implemented in a modified version of the computer program PAP19 were used to simultaneously decompose the phenotypic correlations among seven measures of size and concentration of HDL into their underlying additive ρG and random ρE. This decomposition was accomplished by conditioning the phenotypic correlations on the degree of relationship among all the individuals in the sample by use of extended pedigree information. Specifically, the parameterization of the additive genetic model used here allowed for the simultaneous estimation of mean effects, six covariate effects, phenotypic SDs, and heritabilities for each trait and estimates of ρG and ρE for all pairwise combinations of the traits.
The phenotypic correlation (ρP) between a pair of traits within an individual can be expressed in terms of the underlying ρG and ρE, correcting for the use of related individuals, by the equation (Equation 1) where h12 is the heritability of trait 1 and h22 is the heritability of trait 2. The heritability (h2) represents the portion of the phenotypic variance accounted for by the total additive genetic variance (h2=σ2G/σ2P), or in other words, the portion of the similarity in a given trait between related individuals that is predicted by degree of relatedness. The phenotypic correlation between a pair of traits is expressed as a function of the shared genetic and environmental effects (expressed as the ρG and ρE) between a pair of traits weighted by the portion of their respective phenotypic variances accounted for by the effect of genes (ie, their heritabilities).
To estimate the additive ρG and the random ρE for pairs of traits, the multivariate phenotype of an individual is first modeled as a linear function of the individual's trait measurements, and the matrix of kinship coefficients is generated, expressing relatedness among all pairs of individuals in the pedigree. Using standard quantitative genetic theory, the phenotypic variance-covariance matrix and its additive genetic and random environmental components are then obtained. From these matrices, ρG and ρE are estimated directly. A significant nonzero ρG is a direct measure of the extent of genes shared in common between a pair of traits (ie, pleiotropy). A more detailed explanation of the extension of this methodology to the multivariate state can be found in previously published work.15 20 21
To examine the unique genetic component of each of these seven traits independent of all the others, we calculated the conditional phenotypic, genetic, and environmental covariance matrices using standard matrix formulas employing the observed maximum-likelihood parameter estimates of the full model in which all the traits were included as separate phenotypes, with these conditional estimates themselves being maximum-likelihood estimates. We then obtained estimates of residual heritabilities (ie, heritabilities of each of these traits after the common genetic and shared environmental effects of the remaining six traits were removed). In other words, the heritabilities, along with the variance, for each of the seven traits were reestimated while their common interactions with the six remaining traits were accounted for.
Even though the multivariate model used in this study allows only for additive genetic effects, it is possible that major genes could be involved in determination of the common variation shared among these seven phenotypes, which would lead to deviations from multivariate normality. The maximum-likelihood methods used here, however, are robust to such deviations in the underlying distribution. Therefore, valid maximum-likelihood estimates for the parameters of the genetic model can be obtained.22
Log-Likelihood Ratio Test
The significance of both the ρG and ρE and the estimates of heritability and covariate effects were tested by comparing the ln likelihoods from restricted models, in which each of these parameters was in turn constrained to equal zero, against the ln likelihood for the general model in which all parameters were estimated. The ln likelihood values of the general and the restricted models were compared by use of a likelihood ratio test. This test yields a statistic that is distributed asymptotically as a χ2 with df equal to the difference in the number of parameters estimated in the two models being compared and is calculated as −2(ln likelihoodrestricted model−ln likelihoodgeneral model). In this case, the comparisons of the restricted models to the general model have 1 df. An ln likelihood score for either of the restricted models that was significantly worse than that of the general model (P≤.05) was considered to be evidence of a significant, nonzero, correlation, heritability, or covariate effect.
Meanings and Covariate Effects
Table 1⇓ provides MLEs and associated standard errors for the seven HDL measures and all statistically significant (P≤.05) covariate effects as simultaneously estimated in the full heptavariate model. These parameters include the male mean (μmale), a sex effect (βsex), male age effect (βagemale), female age effect (βagefemale), effect of exogenous hormone use (βhormone), effect of diabetes status (βdiabetes), and the effect of diabetic medication (βdmeds). In the case of apo A-I, apo A-II, and HDL-UC, there are no significant sex effects, so the male mean is in fact the MLE for the mean in both sexes. With respect to measures of HDL-EC, HDL-A-Isize, HDL-A-IIsize, and HDL-Csize, there are significant sex effects (P≤.05), so the female means for these traits are obtained by adding 0.10, 0.20, 0.14, and 0.21, respectively, to the MLE for the male mean for the corresponding trait. Significant effects of age were detected for three measures of HDL concentration: apo A-I, apo A-II, and HDL-EC. Significant effects for hormone replacement therapy were detected for apo A-I, apo A-II, and HDL-UC. A significant effect of diabetic status was detected for HDL-EC, and significant effects of diabetes (βdiabetes) and use of diabetic medications (βdmeds) were detected for HDL-A-Isize.
SDs and Heritabilities
Table 2⇓ provides the maximum-likelihood estimates and associated standard errors for the phenotypic SDs (σ) and heritabilities (h2) for each of the seven HDL measures in the heptavariate analysis. The SD estimates range from 0.17 for apo A-I to 0.47 for HDL-Csize. The heritability estimates range from 0.17 for HDL-A-IIsize to 0.61 for apo A-I, with all seven traits having heritabilities significantly greater than zero at P≤.05.
Table 3⇓ provides the additive ρG among these seven HDL measures along with their associated standard errors. As indicated in the table, the ρG are all positive, and all are significantly greater than zero at P≤.05. These additive ρG range from a moderate value of 0.33 (apo A-II:HDL-A-Isize) to a relatively large value of 0.87 (HDL-UC:HDL-Csize).
Table 4⇓ provides the random ρE among these seven HDL measures. Only the random ρE between apo A-I:HDL-A-Isize and apo A-II:HDL-UC were not significantly different from zero at P≤.05. Unlike the pattern of ρG, the random ρEvary in sign (both positive and negative correlations), and the significant negative random ρE were detected only between apo A-II and the three HDL size measures, with values of −0.33, −0.22, and −0.36 between apo A-II:HDL-A-Isize, apo A-II:HDL-A-IIsize, and apo A-II:HDL-Csize, respectively. The significant positive random ρE ranged from 0.18 (apo A-I:HDL-A-IIsize) to 0.65 (HDL-A-IIsize:HDL-Csize).
SDs and Heritabilities of the Unique Genetic Component
Table 5⇓ provides the maximum-likelihood estimates and associated standard errors for the phenotypic SDs (σ) and heritabilities (h2) for each of the seven HDL measures in the heptavariate analysis after removing the common effects of the remaining six traits. The SDs ranged from 0.13 for apo A-I to 0.29 for HDL-Csize. The heritability estimates range from 0.10 for HDL-UC to 0.31 for apo A-I. For all but two of the traits (HDL-UC and HDL-A-IIsize), residual heritabilities were significantly greater than zero at P<.05.
Percent Variance Remaining After Corrections for Shared Genetic Effects
Table 6⇓ shows the percentage of the additive genetic, random environmental, and phenotypic variance remaining after correction for the shared effects with the other six traits. Between 35% and 71% of the total phenotypic variance remained after each trait was adjusted for the pleiotropic effects shared with the other six traits. Table 6⇓ also gives the percentage of the additive genetic and random environmental variances remaining after adjustment for the shared genetic effects. As would be expected, a substantial amount of random environmental variance remains after this adjustment. However, there is a marked decrease in the additive genetic variance of each of the seven traits, with percentages of additive genetic variance remaining ranging from 13% (HDL-UC) to 36% (HDL-A-IIsize).
On the basis of the significant additive ρG among these seven HDL measures, we can conclude the existence of a substantial degree of pleiotropy (ie, additive genetic effects common to all seven traits). The finding of significant heritabilities for all of the phenotypes examined, together with the pattern of significant positive additive ρG among all pairs of these phenotypes, implies that the expression of these traits to varying degrees is under genetic control and that those genes responsible for elevated values for one of these traits also leads to higher values for the other traits. It should also be emphasized that because a preliminary screening of these seven HDL traits failed to detect any significant common household effect, these heritabilities and the ρG reflect solely the action of genes and are not confounded by nongenetic familial effects. The magnitudes of the heritabilities for each of these traits and the ρG among them indicate that the phenotypic intercorrelation observed among this group of variables is largely driven by the common influence of additive effects of a set of shared genes (ie, pleiotropy). Despite these findings of a high degree of commonality with respect to genetic effects among this set of seven HDL phenotypes, there is no evidence of complete pleiotropy (ie, ρG in each case is significantly less than one). This means that there are additional genes that uniquely exert significant effects on each of the traits (and subsets of them). This point is further supported by the fact that there are relatively pronounced residual genetic variances for each of these traits after adjustment for the effects of pleiotropy. In the case of HDL-UC and HDL-A-IIsize, however, with their nonsignificant heritabilities after correction for pleiotropy, it would appear that perhaps the bulk of their genetic component is derived from the genes common among all seven of the traits.
Although the overall values for the random ρE are not quite as large as those for the ρG, there is still clear evidence of shared random environmental effects among the majority of this set of phenotypically related HDL traits. The significant negative random ρE between apo A-II and the three measures of particle size (HDL-A-Isize, HDL-A-IIsize, and HDL-Csize) indicate that the random environmental factors leading to increased values of apo A-II in turn result in lower values in each of the three size measures. In contrast, all of the significant positive ρE among this set of traits indicate that those random environmental factors leading to increased values in one trait also lead to increased values in each of the others.
We found significant random ρE for most traits. These indicate the existence of other important covariates that were not included in this model or that are still unmeasured; however, the identification of these environmental covariates would not decrease the genetic effects already detected. Likely candidates for these random environmental covariates include diet and endocrine hormones. Age and sex are well established factors associated with lipoprotein variation. Diabetes has been associated with low concentrations of HDL cholesterol, and in this study we found a significant effect for HDL-EC but not for HDL-UC or for apo A-I, the principal apolipoprotein of HDL. Recently, we reported that diabetic individuals had significantly smaller HDL particles when lipoprotiens were stained for apo A-I distributions but not when stained for cholesterol distributions.23 The present study, including a substantially larger data set, confirms these earlier results and reveals a significant counteracting effect of diabetic medications on HDL size phenotypes.
In general, these findings reveal a strong pattern of ρG among all seven of these traits, which is indicative of pronounced pleiotropy and suggests that at some level, each of them is in fact a proxy for a larger and perhaps as yet undefined phenotypic entity of HDL. Given the results of this analysis, it would be ill advised to treat these seven phenotypes as truly independent measures of different aspects of HDL without first accounting for their rather significant shared genetic and environmental components. However, we have also shown that a significant unique genetic component exists for each of these traits after the removal of the shared effects. Therefore, the next task is to identify ways to exploit both the common information inherent in each of these seven measures and those that would allow for the identification of the unique elements that are also inherent in each of these phenotypes. Recognition of such genetic and environmental interactions and their ultimate disentanglement will further the understanding of this group of lipoprotein variables and of the normal metabolic and pathogenic processes in which they are involved.
Selected Abbreviations and Acronyms
|HDL-EC||=||esterified HDL cholesterol|
|HDL-UC||=||unesterified HDL cholesterol|
This study was supported by NIH grants HL-45522, GM-15803, and DK-44297.
Mitchell BD, Kammerer CM, Blangero J, Mahaney MC, Rainwater DL, Dyke B, Hixson JE, Henkel RD, Sharp RM, Comuzzie AG, VandeBerg JL, Stern MP, MacCluer JW. Genetic and environmental contributions to cardiovascular risk factors in Mexican Americans: the San Antonio Family Heart Study. Circulation. 1996;94:2159-2170.
Cheng M-L, Woodford SC, Hilburn JL, VandeBerg JL. A novel system for storage of sera frozen in small aliquots. J Biochem Biophys Methods. 1988;13:47-51.
Warnick GR, Benderson J, Albers JJ. Dextran sulfate-Mg2+ precipitation procedure for quantitation of high-density-lipoprotein cholesterol. Clin Chem. 1982;28:1379-1388.
Hogle DM, Smith RS, Curtiss LK. Quantitation of plasma apolipoprotein A-I using two monoclonal antibodies in an enzyme-linked immunosorbent assay. J Lipid Res. 1988;29:1221.
Smith SJ, Cooper GR, Henderson LO, Hannon WH, Apolipoprotein Standardization Collaborating Group. An international collaborative study on standardization of apolipoproteins A-I and B, I: evaluation of a lyophilized candidate reference and calibration material. Clin Chem. 1987;33:2240.
Stein EA, DiPersio L, Pesce AJ, Kashyap M, Kao J-T, Srivastava L, McNerney C. Enzyme-linked immunoabsorbent assay of apolipoprotein AII in plasma, with use of a monoclonal antibody. Clin Chem. 1986;32:967-971.
Rainwater DL, Andres DW, Ford AL, Lowe WF, Blanche PJ, Krauss RM. Production of polyacrylamide gradient gels for the electrophoretic resolution of lipoproteins. J Lipid Res. 1992;33:1876.
Greenwood FC, Hunter WM, Glover JS. The preparation of 131I-labelled human growth hormone of high specific radioactivity. Biochem J. 1963;89:114.
Rainwater DL, Blangero J, Hixson JE, Birnbaum S, Mott GE, VandeBerg JL. A DNA polymorphism for LCAT is associated with altered LCAT activity and high density lipoprotein size distributions in baboons. Arterioscler Thromb. 1992;12:682.
Verdery RB, Benham DF, Baldwin HL, Goldberg AP, Nichols AV. Measurement of normative HDL subfraction cholesterol levels by gaussian summation analysis of gradient gels. J Lipid Res. 1989;30:1085.
Aitchison J. The Statistical Analysis of Compositional Data. New York, NY: Chapman and Hall; 1986.
Boehnke M, Moll PP, Kottke BA, Weidman WH. Partitioning the variability of fasting plasma glucose levels in pedigrees. Am J Epidemiol. 1987;125:679-689.
Hopper JL, Mathews JD. Extensions to multivariate normal models for pedigree analysis. Ann Hum Genet. 1982;46:373-383.
Hasstedt SJ. Pedigree Analysis Package. Revision 3.0. Salt Lake City, Utah: University of Utah Department of Genetics; 1989.
Singh ATK, Rainwater DL, Haffner SM, VandeBerg JL, Shelledy WR, Moore PH Jr, Dyer TD. Effect of diabetes on lipoprotein size. Arterioscler Thromb Vasc Biol. 1995;15:1805-1811.