Evidence for a Major Gene Elevating Serum Bilirubin Concentration in Utah Pedigrees
In case-control studies, lower serum bilirubin levels have been associated with increased risk of developing coronary heart disease (CHD). We have also previously shown that serum bilirubin has a significant polygenic component. The purpose of the present investigation was to determine whether there was statistical evidence for a major gene explaining a significant portion of individual variation in serum total bilirubin levels and whether this gene might alter the risk of CHD. Serum bilirubin measurements were obtained from 1240 adults in 84 Utah pedigrees screened twice: from 1980 to 1983 and again from 1983 to 1986. Bivariate maximum-likelihood segregation analysis of serum bilirubin levels obtained from the two clinic visits indicated that a major gene was responsible for elevated levels in 11.5% of the persons in these pedigrees. Phenotypic variations in visit 1 bilirubin arising from polygenes were highly correlated with the phenotypic variation due to polygenes in visit 2 bilirubin, indicating a stable genetic contribution to bilirubin over 2.5 years of follow-up. The major gene explained 27% and 28% of the variance in bilirubin levels at visit 1 and visit 2, respectively. There were no correlations of unmeasured environmental factors influencing bilirubin between the two clinic visits. At both visits, persons with early CHD had lower levels of bilirubin than unaffected persons (P<.01). The odds ratio for the risk of CHD in the high-homozygote group was 0.31, P=.09. We conclude that there is a major gene modestly raising serum bilirubin levels. Since cross-sectional data indicate that low serum bilirubin levels increase the risk of CHD, this major gene may be protective against CHD for about 12% of the persons in this study.
- Received May 19, 1995.
- Revision received February 22, 1996.
Two recent case-control studies have implicated lower serum total bilirubin levels as a strong risk factor for coronary heart disease (CHD).1 2 Total bilirubin levels were independently associated with CHD after inclusion of standard lipid, lipoprotein, smoking, and hypertension risk factors in a logistic model predicting CHD. The odds ratio for severe CHD, as determined by coronary angiography, in the earlier study of total bilirubin was 0.68.1 In the later study, in which CHD was defined as early myocardial infarction (age <55 for males and <65 for females), coronary bypass surgery or coronary angioplasty, the odds ratio was 0.23.2 The strength of the protective effect of higher bilirubin levels was similar to the protective effect of higher HDL-C levels. A recent prospective study found a U-shaped CHD risk distribution for quintiles of bilirubin.3 The middle quintile had a relative risk of 0.68 (0.51 to 0.89 95% confidence interval) compared with the lowest or highest quintile.
Since CHD itself4 and risk factors for CHD5 6 have significant heritable components, the present study was initiated to test whether there was a genetic component to total bilirubin levels. Polygenic heritability of bilirubin has been estimated as 58%.7 However, major gene influences were not modeled in this study. A bimodal distribution for bilirubin levels has been shown for both males and females.8 A recent study has found that homozygosity for a TATAA element with two extra bases was associated with elevated bilirubin levels and may be associated with mild cases of Gilbert's syndrome.9 We now present evidence from Utah pedigrees for both a major gene with large effects and for polygenic effects for total bilirubin levels using measurements from two separate clinic visits in a bivariate segregation analysis. Elevation of total bilirubin levels by this inferred gene indicates that the gene may be protective against CHD.
The Utah pedigrees studied have been extensively described.10 11 12 The first clinical screening took place from 1980 to 1983 and the second screening took place from 1983 to 1986. Mean length of follow-up was 2.5 years (95% had 2- or 3-year follow-up). There were 84 pedigrees containing 1240 adults ages 20 and older who had serum bilirubin measured at both clinic visits. Individuals were not excluded because of any disease or medication status, although none had evidence of liver or blood diseases that might affect total bilirubin levels. There were no significant differences in total bilirubin between those taking any specific medication versus unmedicated subjects. Since CHD has been related to bilirubin, excluding these persons may cause one to miss finding a major gene. Persons with CHD were also not excluded because a mechanism is not known that would reduce bilirubin after the occurrence of CHD.
Pedigree sizes were 40 pedigrees with 1 to 4 members, 13 pedigrees with 5 to 9 members, 16 pedigrees with 10 to 29 members, and 15 pedigrees with 30 to 86 members. There were 299 descendants in 8 pedigrees ascertained for 2 or more stroke deaths before age 75; 572 descendants in 18 pedigrees ascertained for 2 or more coronary heart deaths before age 55; 328 relatives of 49 probands selected from the Salt Lake City component of the Hypertension, Detection and Follow-up Program; and 41 relatives of 9 probands referred to us from miscellaneous sources. This study followed procedures approved by an institutional review committee, and informed consent was obtained from all subjects.
Blood was drawn with the subject in a recumbent position. Persons were requested to fast overnight and not to take medications, including over-the-counter drugs, during that period. Clinical chemistries, including bilirubin, were measured using an autoanalyzer (SMAC II Analyzer, Technicon Instruments Corp). Plasma cholesterol,13 triglycerides,14 and HDL-C15 were also measured. Sitting blood pressures were measured using an automated blood pressure device (Infrasonde SR-2, Sphygmetrics, Inc) after sitting for 5 minutes and four such measurements taken at least 2 minutes apart during the same clinic visit were averaged. Alcohol consumption was derived by adding up the number of 12-oz beers, 4-oz glasses of wine, and shots of liquor consumed per week. CHD was defined as a history of myocardial infarction, coronary bypass graft, or percutaneous transluminal angioplasty.
Statistical and Genetic Analyses
Bilirubin was measured in milligrams per deciliter and converted to SI units. Before genetic analysis, bilirubin levels were standardized to a mean of zero and standard deviation of one. Since the distribution was significantly skewed, maximum likelihood power transformations using the formula y=6/p[(x/6+1)p −1] were performed to reduce skewness.16 The power transformation estimate for visit 1 total bilirubin, with a skewness of 2.74, was p=−2.24. Skewness was reduced to −1.66. For visit 2, skewness was reduced from 2.77 to −1.15, with a transformation estimate of p=−2.76.
Major locus inheritance of total bilirubin levels was modeled using maximum likelihood segregation analysis.17 18 Maximum likelihoods were obtained using PAP and GEMINI software.19 20 To test for segregation of a major gene, a bivariate segregation model was fit to the data. Prior to fitting that model, univariate models were fit to provide starting estimates for the large number of parameters in the bivariate model. The bivariate model assumed that total bilirubin resulted from a major genetic locus with a large effect, an additive polygenic effect, covariate effects, and random, unmeasured environmental effects, each acting independently. Serum bilirubin measurements from both the first and second clinic visits were used in the model as a bivariate phenotype arising from a single major gene locus. The effect of this locus could differ at each time point, as could the effects of polygenes and covariates. Correlations between the polygenic effects at each clinic visit and between unmeasured environmental effects at each visit were modeled. This allows one to examine whether new sources of polygenic or environmental variation arise (or decrease) over time.
Parameters of the bivariate model included the gene frequency (p) for the common allele for low (L) bilirubin levels, means for each of the three genotypes (μiLL, μiLH, μiHH), i=1,2 for visit i measurements, a common standard deviation for each genotype (σi), which is allowed to differ between clinic visits i=1,2; and polygenic heritability (hi2). Age, gender, and body mass index (BMI) were included in the model as covariates such that the bilirubin level within each genotype was determined by the equation yi=αij + βij gender + γij age + δij BMI; j=LL,LH,HH genotypes; i=1,2 clinic visits; age and BMI are visit 1 values, and gender=1 for males and 2 for females. The bivariate bilirubin distribution also involves the correlation, ρ, between bilirubin measurements at visit 1 and visit 2. This correlation is partitioned into the genetic correlation, ρg, and the environmental correlation, ρe. The genetic correlation reflects the correlation of the same polygenes on the two longitudinal bilirubin levels, whereas the environmental correlation reflects the correlation of unmeasured environmental factors between the two visits. The general model includes transmission parameters (τ1, τ2, τ3), which are tested for departure from the expected mendelian values of 1, 1/2, and 0, respectively. The transmission parameters are also tested for departure from all three τ's being equal to the gene frequency (environmental model). No ascertainment correction was used in the analysis.
Significance tests compared log-likelihood differences of a submodel to a more general model, with the degrees of freedom determined by the difference in the number of estimated parameters. Minus 2 times the log-likelihood difference is approximately distributed as χ2. Hypothesis testing proceeded by first fitting a general transmission model with common standard deviations across genotypes but differing between clinic visits and common covariate slopes across genotypes but differing between clinic visits. Common slopes were used because there were no significant differences in slopes in the univariate models. Unequal slopes were tested in the bivariate model after the most parsimonious model with equal slopes was identified, to confirm the univariate results. To test whether the same major locus was detected at both clinic visits, the means of each distribution at visit 1 were fixed equal to one another (polygenic model), but the means at visit 2 were not fixed. Rejection of this model when compared with the model where two or three means at each visit are estimated indicates a major effect of the single locus on bilirubin levels at both visits.
Means of study variables for each genotype or CHD category were calculated by regressing total bilirubin on gender and BMI. Residual values from this regression were used to test for group differences using a two-sample t test. Because of the nonindependence of the samples used in the t test, the standard deviations are too small and significance levels are somewhat more significant than if an independent sample were used.
Table 1⇓ shows the means of the study variables for all subjects combined. Fifty percent of the sample were females who had a lower visit 1 total bilirubin mean and standard deviation (8.8±4.3 μmol/L) than males (11.2±5.8 μmol/L, P<.0001 for the mean and P<.0001 for the standard deviation). There were mild correlations of total bilirubin with age (r=−.08, P=.007) and body mass index (r=−.09, P=.002). Total bilirubin ranged from 2 to 60 μmol/L, although the second highest value was only 44 μmol/L. The distribution of total bilirubin is shown in the Figure⇓. Direct bilirubin ranged from 0 to 8.6 μmol/L. Over the 2.5 years between clinic visits, total bilirubin decreased by 1.6±4.2 μmol/L with a correlation of .65 (P<.0001) between bilirubin levels from the two visits.
Table 2⇓ shows the genetic model estimates for total bilirubin levels measured during the two clinic visits. The estimate of τ1 hit the boundary at 1.0 and was fixed at that value for subsequent analyses. A major gene hypothesis could not be rejected when comparing the mixed codominant model to the general model (χ2=5.12, df=2, P=.08). The general model indicated that there was an excess of offspring who had the allele for high bilirubin levels but whose parents were classified as having the homozygous low bilirubin genotype (τ3=.19). A nontransmissable, environmental model was rejected (χ2=8.81, df=2, P=.01). Under the codominant model, the correlation between the polygenic sources of variation for visit 1 and visit 2 bilirubin was .94, whereas the environmental correlation was .0007. Environmental correlation was not significantly different from 0 (χ2=0.00, df=1, P=.99). This indicates that the polygenic variance was constant over the 2.5 years of follow-up and that there was no correlation between unmeasured environmental effects at visit 1 and visit 2 in these pedigrees.
Under the recessive model, the genetic correlation estimate hit the upper bound of .99 and was fixed at that value. So that this recessive model would be a submodel of the codominant model, the codominant model was rerun fixing the genetic correlation to .99 and obtaining a −2 ln likelihood of 4837.924. The recessive model could not be rejected when compared with this codominant model (χ2=2.60, df=2, P=.27). The dominant model was rejected, however (χ2=30.6, df=2, P<.00001). The covariates were tested for significance in the recessive model, and age was not significantly related to bilirubin (χ2=3.90, df=2, P=.14). The effect of BMI on total bilirubin was significantly different from zero (χ2=15.23, df=2, P=.0005). Therefore the most parsimonious model was a recessive bivariate model with gender and BMI covariates, polygenic inheritance, no significant correlation of unmeasured environmental effects over time, and very high correlation of the polygenic heritability over time.
To test whether there was any interaction of gender or BMI with bilirubin genotype, the slopes of each covariate were allowed to differ across genotypes. There were no significant differences in slopes between genotypes for either covariate (χ2=2.18, df=4, P=.70). To test for a single major locus effect on bilirubin levels at both time points, the two genotype means for visit 1 were forced to equal each other, whereas the two means at visit 2 were both estimated. This model was rejected when compared with the most parsimonious recessive model in Table 2⇑ (χ2=77.2, df=1, P<.00001).
Under the recessive model, the separation of distribution means was 1.88 standard deviations for visit 1 bilirubin and 1.95 standard deviations for visit 2 bilirubin, indicating considerable overlap between distributions (see Figure). Table 3⇓ shows the percent of the variance explained by the major gene, polygenes and unmeasured effects at visit 1 and 2. The percent of persons recessive for high bilirubin levels was estimated as 12%. The mean total bilirubin of the entire sample after standardization and power transformation was −0.20. Since the mean of the homozygous high genotype at visit 1 was 0.88 (1.91 -.46×1.5 -.013×25.8) compared with −0.33 (.69 -.46×1.5 -.013×25.78) for the other two genotypes, this indicates that the major gene is associated with elevated total bilirubin levels.
With a recessive mixed model used to estimate the genotypic probabilities of each person, means of other risk factors for CHD for the two genotypes are given in Table 4⇓. Individuals were assigned to a genotype using the maximum probability over the three genotypes. Persons with the high bilirubin genotype were younger, had lower triglycerides, and drank less. Other variables were similar between the two genotypes, supporting the claim that bilirubin acts as an independent risk factor for CHD. Direct bilirubin, as a percentage of total bilirubin, was lower in those with the high genotype (13.8% in the high genotype versus 17.9% in the low genotype, P=.0001). There was no clear relationship between amount of alcohol consumed and bilirubin levels. The percent of males and females in each genotype was not significantly different (P=.12).
A comparison of persons with and without CHD at each clinic visit (Table 5⇓) confirms the cross-sectional relationship of lower bilirubin levels in persons with CHD found in the other two studies.1 2 Total bilirubin levels decreased over time to a similar extent (1.5 and 1.6 μmol/L) in both cases and controls, preserving the 1.4 μmol/L difference between cases and controls at both visits (P<.01). Serum bilirubin levels were lower in persons with earlier age at onset of CHD (before age 60) compared to persons with CHD onset at age 60 or later. Only 4 of the 56 CHD cases at visit 2 had a probability greater than 10% of being high homozygotes (27%, 37%, 42%, and 46% for the 4 persons) and only two of the four were assigned to the high-homozygote genotype. The odds ratio for the risk of CHD in the high-homozygote group was 0.31, P=.09.
In addition to the .65 correlation for bilirubin levels between the two visits, it is of interest to examine the consistency of bilirubin levels within genotype. Since there is substantial overlap between genotype distributions, a crude measure of stability was calculated by looking at the number of persons in the high genotype who had visit 1 bilirubin levels less than and more than the mean of the low genotype at visit 1 compared with the number of persons in the high genotype who had visit 2 bilirubin levels less than and more than the mean of the low genotype mean at visit 2. There were 129 persons whose genotypic probability placed them in the high genotype. Of these, only 2 persons had visit 1 bilirubin levels below the visit 1 mean of the low genotype and only 1 person had a visit 2 bilirubin level below the visit 2 mean of the low genotype. Thus, there is crude concordance of bilirubin level and genotype assignment, as indicated by the genetic model testing for pleiotropy.
Persons who had diabetes, kidney disease, or were smokers or were nonfasting had nonsignificantly lower total bilirubin levels compared with persons without disease, without medications, and who were fasting and nonsmokers. All differences tended toward lower bilirubin levels, which would not artificially induce evidence for a major gene for high bilirubin levels because these persons were included in the analysis.
This study shows that serum total bilirubin concentrations are influenced by a major gene. The effect of the underlying gene is much larger than the difference in bilirubin levels between persons with and without CHD. Even so, elevations in bilirubin are mild (19 μmol/L versus 9 μmol/L) and not in the range usually seen in Gilbert's syndrome, which results in elevated indirect (and total) bilirubin due to glucuronyl transferase deficiency. Except for the rare Crigler-Najjar syndrome, which occurs in neonates and also affects indirect bilirubin, elevated bilirubin levels appear to have no adverse effects.
Beneficial effects of elevated bilirubin have been proposed, however, such as reducing oxidation of lipids21 22 and increasing cholesterol clearance.23 Therefore, the mildly elevated levels seen in persons inferred to have the underlying gene described in this study, may provide protection against CHD through one of these mechanisms. Since there were no differences in lipid or lipoprotein levels in persons with and without the inferred gene, the antioxidative effects may be the most likely explanation for protection from CHD. Only 2 of the 56 persons with CHD were assigned to the high bilirubin genotype distribution, and these 2 had probabilities less than 50% of being that genotype, suggesting that most persons with CHD lacked the protective effect from this gene associated with bilirubin levels. Total bilirubin was not different between hypertensive and normotensive subjects or between diabetic and nondiabetic subjects. Smoking and alcohol had nonsignificant relationships with bilirubin levels.
Since 88% of the study sample did not have elevated bilirubin levels, they would be expected to have the same risk of CHD as the general population through any variety of CHD risk factors. Those with the protective genotype (odds ratio for CHD of 0.31) may have additional resistance to one or more of these risk factors, depending on the mechanism that bilirubin reduces risk. Persons with early-onset CHD had lower bilirubin levels than persons with late-onset CHD. The decrease in total bilirubin in CHD patients compared with healthy patients replicated the results found in two other cross-sectional studies.1 2 A prospective study also found lower bilirubin in patients with CHD, although there was evidence for another group of CHD patients with elevated bilirubin levels (a U-shaped risk function). Additional physiological studies in persons with high and low bilirubin levels need to address whether the antioxidant or cholesterol clearance mechanisms differ between groups. The molecular basis of this purported major gene for bilirubin also needs to be defined. If it can be shown that the promoter region defect in the UDP-glucuronosyltransferase 1 gene9 is linked to elevated bilirubin levels segregating in these pedigrees, more specific characterization of gene carriers may be done. Homozygosity for an extra TA in the TATAA element in a sample of 55 normal subjects was estimated at 16%, which is close to the 12% estimated in our study. The mean bilirubin level of the three homozygous normal controls was 17.1 μmol/L compared with means of 19.4 and 17.5 μmol/L at visit 1 and visit 2 in the homozygous high-genotype group of our study. The similarities of these estimates strongly suggest that the promotor abnormality in the UDP-glucuronosyltransferase 1 gene may be found to segregate with elevated bilirubin levels in these pedigrees. If linkage of the promotor region is not found, other genes must be identified.
The longitudinal nature of this study provides information on how stable different causes of phenotypic variation are over time. The estimate of the correlation between the variation due to polygenes at the two visits ranged between .93 and .99, with the most parsimonious model having an estimate of .99 (at the boundary). This shows that variation from this source remains constant over the 2.5 years of follow-up. This suggests that over this short time period aging does not influence polygenic variation despite a significant drop in mean total bilirubin levels. Although we did not model specific common environmental factors in this analysis, there was no correlation over time between unmeasured environmental factors. This may suggest that unmeasured environmental factors acted acutely and changed before the second screening cycle began. The mean separation between genotype means for the two clinic visits remained nearly the same (1.88 versus 1.95 standard deviations) despite the drop in bilirubin levels over time.
The genetic estimates found in this study may not represent the Utah population from which the pedigrees were ascertained. Since ascertainment correction was not used, the estimates, particularly the gene frequency, may be biased from population estimates. The probands used to select the CHD and stroke pedigrees were deceased, since computerized Utah death certificates were used to identify the probands. Therefore it was not possible to adjust for the bilirubin levels of the probands. However, selection for CHD and stroke would tend to select families with lower total bilirubin levels. Since the major locus is responsible for elevated bilirubin levels, the gene frequency estimate may be too low. Blood pressure and hypertensive status were clearly not related to bilirubin levels in this study, and the hypertension-ascertained pedigrees, which contained very few cases of CHD, would be expected to require minimal ascertainment correction. Therefore, we believe that lack of appropriate ascertainment provides conservative estimates of the effects of the proposed major locus in the Utah population. Replication of this study using data sets that are population based will help determine how biased our estimates are.
If 12% or more of the population have a strong protective effect against CHD because of a major gene for bilirubin, this would have significant population implications in further reducing CHD incidence. This would be especially true if it is confirmed to have as large a protective effect as HDL-C.2 Clinical intervention that mimics the effect of the major gene, once it is identified and understood, might allow an effective preventive treatment in susceptible individuals. Clearly, much more needs to be done to reach this point. Fortunately, there are many existing data sets that already have the required data to answer many of the questions raised in this article.
This study was supported by National Institutes of Health grants HL-24855 and HL-21088.
Schwertner HA, Jackson WG, Tolan G. Association of low serum concentration of bilirubin with increased risk of coronary artery disease. Clin Chem. 1994;40:18-23.
Hopkins PN, Wu LL, Hunt SC, James BC, Vincent GM, Williams RR. Higher serum bilirubin is associated with decreased risk for early familial coronary artery disease. Arterioscler Thromb Vasc Biol. 1996;16:250-255.
Breimer LH, Wannamethee G, Ebrahim S, Shaper AG. Serum bilirubin and risk of ischemic heart disease in middle-aged British men. Clin Chem. 1995;41:1504-1508.
Williams RR, Skolnick M, Carmelli D, Maness AT, Hunt SC, Hasstedt SJ, Reiber GE, Jones RK. Utah pedigree studies: design and preliminary data for premature male CHD deaths. In: Sing CF, Skolnick M, eds. The Genetic Analysis of Common Diseases. New York, NY: Alan R. Liss, Inc; 1979:711-729.
Williams RR, Hunt SC, Schumacher MC, Hopkins PN, Wu LL, Barlow GK, Stults BM, Wilson DE, Lalouel JM. Genes, hypertension and coronary heart disease: evidence for shared metabolic pathophysiology. In: Smith U, Bruun NE, Hedner T, Hokfelt B, eds. Hypertension: An Insulin-Resistant Disorder—Genetic Factors and Cellular Mechanisms. Amsterdam, the Netherlands: Elsevier; 1991:89-101.
Williams RR, Hasstedt SJ, Hunt SC, Wu LL, Hopkins PN, Berry TD, Stults BM, Barlow GK, Kuida H. Genetic traits related to hypertension and electrolyte metabolism. Hypertension. 1991;17(suppl I):I-69-I-73.
Owens D, Evans J. Population studies on Gilbert's syndrome. J Med Genet. 1975;12:152-156.
Hunt SC, Hasstedt SJ, Wu LL, Williams RR. A gene-environment interaction between inferred kallikrein genotype and potassium. Hypertension. 1993;22:161-168.
Wahlefeld AW. Triglycerides: determination after enzymatic hydrolysis. In: Bergmeyer HU, ed. Methods of Enzymatic Analysis. Vol 4. New York, NY: Academic Press, Inc; 1974:1831-1835.
Warnick GR, Benderson JM, Albers JJ. Dextran sulfate-Mg2+ precipitation procedure for quantitation of high-density-lipoprotein cholesterol. Clin Chem. 1982;28:1379-1388.
Hasstedt SJ. PAP: Pedigree Analysis Package, revision 4. Salt Lake City, UT: Department of Human Genetics, University of Utah; 1993.
Lalouel JM. GEMINI: a computer program for optimization of a nonlinear function. Department of Medical Biophysics and Computing, University of Utah; 1979.
Stocker R, Yamamoto Y, McDonagh AF, Glazer AN, Ames BN. Bilirubin is an antioxidant of possible physiological importance. Science. 1987;235:1043-1046.
Stringer MD, Gorog PG, Freeman A, Kakkar VV. Lipid peroxides and atherosclerosis. Br Med J. 1989;298:281-284.
Yla-Herttuala S, Palinski W, Rosenfeld ME, Parthasarathy S, Carew TE, Butler S, Witztum JL, Steinberg D. Evidence for the presence of oxidatively modified low density lipoprotein in atherosclerotic lesions of rabbit and man. J Clin Invest. 1989;84:1086-1095.