Triiodothyronine Exerts a Major Pleiotropic Effect on Reverse Cholesterol Transport Phenotypes
Abstract The thyroid hormone triiodothyronine (T3) is known to be a potent mediator of APOA1 gene expression. With the use of multivariate quantitative genetic analysis, we have assessed the magnitude of shared effects of T3 on plasma concentrations of apolipoprotein AI (apo AI) and three related phenotypes: HDL-C, apo AII, and LpAI (which is a concentration of apo AI that contains HDL particles). Maximum likelihood techniques were used to simultaneously estimate mean effects and variance components in large, extended Mexican American families living in San Antonio, Tex. We found that T3 accounted for 16%, 23%, 21%, and 37% of the additive genetic variance in HDL-C, apo AI, apo AII, and LpAI, respectively, while explaining virtually none of the random environmental variance in these phenotypes. T3 also has a pronounced effect on the pairwise genetic correlations among the four phenotypes: After the pleiotropic effects of T3 concentrations are controlled for, the genetic correlations are reduced by 6% in the case of HDL-C and apo AI and 97% for apo AII and LpAI. Thus, genes that influence T3 have a significant effect on HDL-C, apo AI, apo AII, and LpAI and also on the correlations among these phenotypes.
- Received June 7, 1995.
- Accepted November 22, 1995.
High density lipoprotein cholesterol (HDL-C), apolipoproteins AI (apo AI) and AII (apo AII), and LpAI are all involved in reverse cholesterol transport, and their plasma concentrations are considered to be a strong indicator of risk for the development of atherosclerosis. Given their combined role in reverse cholesterol transport (ie, the removal of exchangeable cholesterol from extrahepatic tissues to the liver for excretion), it is not surprising that these HDL measures should exhibit considerable intercorrelation phenotypically. Furthermore, it can be hypothesized that much of the phenotypic correlation among these variables is in fact due to an underlying structure of genetic correlations, indicative of significant pleiotropy (ie, shared genes).
Because of their role in the regulation of metabolic levels and gene expression (via stimulation of messenger RNA production1 ), thyroid hormones are involved to some extent with practically all aspects of normal physiological activities. Specifically with regard to the apolipoproteins involved in reverse cholesterol transport, effects of thyroid hormones on apo AI levels have been noted in numerous studies. In women, plasma apo AI levels are decreased with hypothyroidism and increased with hyperthyroidism.2 3 4 Perhaps most important, triiodothyronine (T3) has recently been found to be a potent mediator of APOA1 gene expression.5 It has been shown that the regulation of APOA1 gene expression by T3 is the result of a thyroid hormone response element located at the 5′ end of APOA1.5
Because of the functional relationship that has been reported among this set of traits, we have formally tested hypotheses concerning the extent to which the expected phenotypic correlations among these traits can be attributed either to shared genetic effects (ie, pleiotropy) or to shared random environmental effects. Here we have decomposed the phenotypic correlations among these reverse cholesterol transport phenotypes and T3 into their constituent additive genetic and random environmental correlations by use of variance decomposition techniques. From these analyses it was also possible to examine the degree to which the intercorrelations among HDL-C, apo AI, apo AII, and LpAI are accounted for in terms of each of their common additive genetic and random environmental correlations with T3.
In this study data were obtained from a large sample of Mexican American families living in San Antonio, Tex, who are participants in the San Antonio Family Heart Study (SAFHS), a broader project examining the genetics of atherosclerosis and its risk factors. Ascertainment of families for the SAFHS was made without regard to preexisting medical conditions. For this project, data were available for 445 individuals (182 men; 273 women). These individuals ranged in age from 16 to 87 years with an average age of 39.7 years. Of the 445 individuals, 10 are unrelated to any other individuals in the sample, whereas the remaining individuals are distributed among 25 pedigrees. The majority of these pedigrees contain at least three generations of relatives. They range in size from 3 to 71 individuals with an approximate mean size of 12 to 15 individuals. An Institutional Review Board (University of Texas Health Science Center at San Antonio) approved the procedures, and all subjects gave informed consent.
The phenotypes used in this analysis were measured in serum samples collected after a 12-hour fast and include total T3, HDL-C, apo AI, apo AII, and LpAI. Serum concentration of total T3 was measured using a commercial radioimmunoassay kit (Diagnostic Products Co). The radioimmunoassay for total T3 had an intra-assay coefficient of variation (CV) of 5.9% and an interassay CV of 8.7%. Mass measurements of apo AI, apo AII, and LpAI were made at Medical Research Laboratories, Highland Heights, Ky (Dr Evan Stein, Director). Plasma concentrations of apo AI were measured by nephelometry,6 apo AII by a competitive immunoassay,7 and LpAI by differential electroimmunoassay with a commercial kit (Sebia, 23 rue Maximilien-Robespierre). HDL-C was measured according to the protocol of Warnick et al.8
With the use of established quantitative genetic theory, it is possible to extend a univariate genetic analysis to encompass the multivariate state.9 10 11 12 Variance decomposition techniques,10 11 12 which use maximum likelihood methods and are implemented in a modified version of the computer program PAP (Pedigree Analysis Package, University of Utah),13 were used to simultaneously estimate both the genetic and environmental correlations among all pairs of traits as well as individual trait means, phenotypic standard deviations, heritabilities, sex and age effects, and selected environmental covariates for total T3 and four reverse cholesterol transport phenotypes (HDL-C, apo AI, apo AII, and LpAI). The phenotypic correlation (ρP) between a pair of traits can be expressed in terms of the underlying genetic (ρG) and environmental (ρE) correlations, correcting for the use of related individuals, by the equation
where h21 is the heritability of trait 1 and h22 is the heritability of trait 2. The heritability (h2) represents the proportion 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 due solely to genetic factors. In other words, the phenotypic correlation between a pair of traits can be expressed as a function of the shared genetic and environmental effects (expressed as the genetic and environmental correlations) between a pair of traits conditional on the portion of their respective phenotypic variances accounted for by the effect of genes (ie, their heritabilities). It is possible to model the multivariate phenotype of an individual as a linear function of the measurements on the basis of the individual’s traits. From such a model we can obtain the phenotypic variance-covariance matrix from which the additive genetic and random environmental components are estimated, given the relationships (kinship coefficients) obtained from the pedigree and with the use of standard quantitative genetic theory. From the genetic and environmental variance-covariance matrices, it is then possible to directly estimate the additive genetic correlation (ρG) (ie, the quantitative expression of pleiotropy) and the environmental correlation (ρE) between pairs of traits. Simply stated, a significant nonzero genetic correlation is a direct measure of the extent of shared genes 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.9 14 15
To examine the genetic effects of T3 on the reverse cholesterol transport phenotypes, we calculated the conditional phenotypic, genetic, and environmental covariance matrices with standard matrix formulas that use the observed maximum likelihood parameter estimates of the full model in which T3 was included as a separate phenotype, with these conditional estimates themselves being maximum likelihood estimates. We then obtained estimates of residual heritabilities (ie, heritabilities of reverse cholesterol transport phenotypes after removing the separate genetic and environmental effects of T3) and residual genetic correlations (ie, genetic correlations between reverse cholesterol transport phenotypes after correcting for the genetic effects of T3).16 In other words, the heritabilities as well as the genetic and environmental correlations among the four reverse cholesterol transport phenotypes were reestimated while accounting for their common interactions with T3.
Despite the fact that the multivariate model used in this study allows only for additive genetic effects, it is possible that major genes could actually be involved in determining the common variation shared among these five phenotypes. The maximum likelihood methods used here, however, are robust to deviations from multivariate normality in the underlying distribution. Therefore, valid maximum likelihood estimates for the parameters of the genetic model can be obtained.17 Also, as part of a general screening process we have not detected significant household effects for any of the traits presented here.
ln Likelihood Ratio Test
The significance of both the genetic and environmental correlations between any pair of traits and the significance of heritability and covariate effects were tested by comparing the likelihoods from restricted models, in which each of these parameters was in turn constrained to equal 0.0, to the 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 distributed asymptotically as a χ2 with degrees of freedom equal to the difference in the number of parameters estimated in the two models being compared and is calculated as
In this case, the comparisons of the restricted models to the general model have one degree of freedom. An ln likelihood score for any of the restricted models that was significantly worse than that of the general model (P≤.05) was considered evidence of a significant, non-zero, correlation, heritability, or covariate effect.
Table 1⇓ provides the male and female means, and the phenotypic standard deviations, along with their associated standard errors for total T3, HDL-C, apo AI, apo AII, and LpAI as estimated in the general model (ie, that is the model in which all parameters were simultaneously estimated). Maximum likelihood estimates were also obtained for age and sex effects as well as for a variety of other covariates (eg, smoking, diabetic status, use of diabetic medications, dietary cholesterol). The selection of covariates considered was based on the results of previous analyses by our group (unpublished data). None of the covariates considered were found to have a significant effect on all of the traits, and several were found to have no significant effect on any of the traits. Maximum likelihood estimates of the significant covariate effects, along with their associated standard errors and probability values, are given in Table 2⇓. It must be remembered that no model is a true reflection of the natural world and undoubtedly there exist as yet unidentified covariates that could have a significant effect. However, the inclusion of such covariates would serve only to strengthen the results (ie, increase the relative genetic contribution).
Maximum likelihood estimates and standard errors obtained from the general model for the heritabilities (h2) of each trait, as well as the genetic (ρG) and environmental (ρE) correlations among them, are provided in Table 3⇓. All five traits show moderate to pronounced levels of heritability (0.33 to 0.52); and on the basis of likelihood ratio tests, all are significantly different from zero at P≤.001 (Table 3⇓). Significant (.001≤P≤.05) to highly significant (P<.001) genetic correlations (ρG=0.48 to 0.88) were detected among all pairs of traits except for T3:HDL-C and apo AII:LpAI (Table 3⇓). With respect to the environmental correlations only four pairwise comparisons were found to be significant (ie, P≤.05); these were HDL-C:apo AI, HDL-C:LpAI, apo AI:apo AII, and apo AI:LpAI (Table 3⇓).
When the estimates of heritabilities and genetic and environmental correlations provided in Table 3⇑ are inserted into Equation 1, it is possible to calculate the phenotypic correlations among these traits corrected for the use of related individuals. The pairwise phenotypic correlations between T3 and HDL-C, apo AI, apo AII, and LpAI are .096, .199, .257, and .098, respectively. With respect to the phenotypic correlations among the reverse cholesterol transport phenotypes, the values range from .230 for apo AII:LpAI to .776 for HDL-C:apo AI.
Partitioning of the variance for each of the reverse cholesterol transport phenotypes so as to account for the portion attributable to the effect of T3 revealed a reduction of 16%, 23%, 21%, and 37% of the additive genetic variance in HDL-C, apo AI, apo AII, and LpAI, respectively (Table 4⇓). With respect to the effects of T3 on the random environmental variances of these four reverse cholesterol transport phenotypes, the impact was small to nonexistent, accounting for 0.6%, 0.0%, 1.4%, and 4.4%, respectively, in HDL-C, apo AI, apo AII, and LpAI (Table 4⇓). Likewise, the portion of the overall phenotypic variance in these four phenotypes accounted for by T3 was negligible, with reductions of 0.9%, 4.0%, 6.6%, and 1.0% for HDL-C, apo AI, apo AII, and LpAI, respectively. Also, as can be seen from Table 5⇓, the removal of the common pleiotropic effect of T3 on the pairwise correlations among the remaining four phenotypes produced reductions in ς2G (ie, the portion of the covariance between a pair of traits accounted for by the genetic correlation) ranging from 6% to 97%. The first column of Table 5⇓ shows the uncorrected genetic correlation (ρG) between the four reverse cholesterol transport phenotypes (which are the same values as presented in Table 3⇑). Column 2 of Table 5⇓ provides the residual genetic correlation (ρG) between the four reverse cholesterol transport phenotypes after correcting for the common correlation with T3. Column 3 presents the percent reduction in the genetic variance explained (which is the genetic analogue of the more familiar r2) and is obtained by dividing the squared value of column 2 (which is the genetic correlation between the traits corrected for their common correlation with T3) by the squared value of column 1 (which is the genetic correlation between the traits uncorrected for their common correlation with T3) and subtracting this number from 1 and then multiplying by 100. The largest impact of the removal of the common effect of T3 was seen on the genetic correlation between apo AII and LpAI (a reduction of 97.5%), whereas the smallest was observed between HDL-C and apo AI (a reduction of 5.6%).
The results of this analysis indicate that a substantial degree of pleiotropy exists among serum levels of T3, apo AI, apo AII, LpAI, and HDL-C, indicating that they all have some genetic determinants in common, while at the same time exhibiting little evidence of an extensive pattern of shared random environmental effects. Specifically, it appears that the correlation between T3 and these reverse cholesterol transport phenotypes is entirely due to common genetic effects, and in fact the corresponding small environmental correlations combined with the heritabilities of these traits (using Equation 1) yield phenotypic correlations that would not be considered significantly different from zero in a statistical analysis in which these genetic effects were ignored. The finding that all the phenotypes examined had significant heritabilities along with positive genetic correlations among this set of phenotypes implies that the expression of these levels is to some degree under genetic control and that those genes responsible for elevated serum levels for one of these traits also produce elevated levels in the others. That is, the phenotypic intercorrelation observed among this group of variables is primarily driven by the influence of a set of shared genes (ie, pleiotropy) as opposed to common environmental factors. Although the existence of extensive pleiotropy among the four reverse cholesterol transport phenotypes is in itself not surprising, the rather consistent and significant genetic correlation of T3 with these variables is intriguing. This pattern of genetic correlation between T3 and the reverse cholesterol transport phenotypes becomes even more interesting when the pairwise genetic correlations among the four lipoprotein phenotypes are reconsidered after their correction for the effects of their common correlation with T3. Specifically, it was found that the removal of the common pleiotropic effect of T3 resulted in reductions of 6% to 97% in the portion of the genetic variance (ρ2G) explained by the pairwise genetic correlations among these four phenotypes. In effect this demonstrates that a substantial portion of the apparent pleiotropy among these reverse cholesterol transport phenotypes actually comes from genes that they all have in common, to varying degrees, with T3. Common pleiotropy with T3 accounts for a relatively small portion of the genetic correlation between HDL-C and apo AI (≈6%), thereby suggesting that most of the genes shared between these two traits are unique to this pair of traits and not shared with T3. At the other extreme, however, virtually all of the pleiotropy between apo AII and LpAI (≈98%) is due to the common pleiotropic effects of T3, indicating that they have virtually no genes unique to themselves.
The magnitude of the genetic impact of T3 on this set of reverse cholesterol transport phenotypes is also demonstrated when its contribution to the additive genetic variance of each of the remaining four phenotypes is considered. Partitioning the additive genetic variance of each phenotype into the portion accounted for by T3 resulted in the reduction of the overall additive genetic variance in these traits by 16% to 37%. Thus, it can be concluded that T3 contributes greatly to the additive genetic variance in each of these traits and as a result has a strong influence on their heritability. These findings lend further support to the conclusion that T3 exerts a significant pleiotropic effect on this group of lipoprotein phenotypes. It would appear, therefore, that serum T3 levels are an important factor to consider in the genetic analysis of any of these reverse cholesterol transport variables as well as perhaps other lipoprotein phenotypes. In fact, the recognition, and ultimate disentanglement, of such genetic interactions could open new avenues to understanding this group of lipoprotein variables and the normal metabolic as well as pathogenic processes in which they are involved.
This study was supported by National Institutes of Health grants HL45522, GM15803, and DK44297.
Samuels HH, Forman BM, Horowitz ZD, Zheng-Sheng Y. Regulation of gene expression by thyroid hormone. J Clin Invest. 1988;81:957-967.
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;29:1221-1229.
Stein EA, DiPersio L, Pesce AJ, Kashyap M, Kao JT, Srivastava L, McNerney C. Enzyme-linked immunoabsorbant assay of apolipoprotein AII in plasma, with use of a monoclonal antibody. Clin Chem. 1986;32:967-971.
Warnick GR, Benderson J, Albers JJ. Dextran sulfate-Mg2+ precipitation procedure for quantitation of high-density-lipoprotein cholesterol. Clin Chem. 1982;28:1379-1388.
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: Department of Genetics, University of Utah; 1989.
Graybill FA. Theory and Application of the Linear Model. Belmont, Mass: Duxbury Press; 1976.
Beaty TH, Self SG, Liang KY, Connolly MA, Chase GA, Kwiterovich PO. Use of robust variance components models to analyse triglyceride data in families. Ann Hum Genet. 1985;49(pt 4):315-328.