|Home | About | Journals | Submit | Contact Us | Français|
Bilirubin, resulting largely from the turnover of hemoglobin, is found in the plasma in two main forms: unconjugated or conjugated with glucuronic acid. Unconjugated bilirubin is transported into hepatocytes. There, it is glucuronidated by UGT1A1 and secreted into the bile canaliculi. We report a genome wide association scan in 4300 Sardinian individuals for total serum bilirubin levels. In addition to the two known loci previously involved in the regulation of bilirubin levels, UGT1A1 (P = 6.2 × 10−62) and G6PD (P = 2.5 × 10−8), we observed a strong association on chromosome 12 within the SLCO1B3 gene (P = 3.9 × 10−9). Our findings were replicated in an independent sample of 1860 Sardinians and in 832 subjects from the Old Order Amish (combined P < 5 × 10−14). We also show that SLC01B3 variants contribute to idiopathic mild unconjugated hyperbilirubinemia. Thus, SLC01B3 appears to be involved in the regulation of serum bilirubin levels in healthy individuals and in some bilirubin-related disorders that are only partially explained by other known gene variants.
Bilirubin results from hemoglobin catabolism (and to a lesser extent from other heme-containing proteins) within the reticulo-endothelial system. Bilirubin is transported into the liver, where specific carrier proteins belonging to the Organic Anion Protein Transporter family (1) are thought to mediate its uptake. In hepatocytes, bilirubin is glucuronidated by (UDP)-glucuronosyltransferase and thereby transformed into a water soluble conjugated molecule (2). Conjugated bilirubin is actively secreted into the bile canaliculi by a membrane ATP-dependent transporter, designated multidrug resistance-associated protein 2 (MRP2) (3). In plasma, bilirubin occurs primarily unconjugated tightly bound to albumin, or conjugated as a water-soluble glucuronide. High levels of unconjugated bilirubin in plasma may indicate increased production by hemolysis, ineffective erythropoiesis or decreased conjugation, whereas conjugated hyperbilirubinemia reflects decreased excretion by damaged biliary duct or hepatic parenchymal cells.
Within the SardiNIA study we conducted a genome-wide association (GWA) analysis (4,5) in 4300 Sardinians to identify genetic factors affecting a series of quantitative traits, including serum bilirubin levels (see Material and Methods). The strongest associations were observed on chromosome 2, in the promoter of UGT1A1 (uridine diphosphoglucuronyl-transferase) gene (rs887829, P = 6.21 × 10−62, Fig. 1 and Supplementary Material, Table S1), a major and well-known modifier of bilirubin levels (6), and on the X chromosome near the G6PD (glucose-6-phosphatedehydrogenase) gene (rs766420, P = 9.40 × 10−09, Fig. 1 and Supplementary Material, Table S1).
In addition, a third locus on chromosome 12p12.2 reached genome-wide significance (Fig. 1). The strongest P-value was observed at SNP rs2117032 (P = 4.74 × 10−08, Supplementary Material, Table S1) downstream of the 3′-UTR of the SLCO1B3 (solute carrier organic anion transporter family, member 1B3) gene (7) (Fig. 2); each copy of the minor allele C was responsible, on average, for an increase of 0.048 mg/dl in bilirubin levels. The second ranked marker in this region was rs17680137 (P = 2.3 × 10−7), located in intron 7 of SLCO1B3, in modest LD with rs2117032 (r2 = 0.29), and responsible for a slightly larger effect (0.067 mg/dl per copy of the minor allele G). Our genotype strategy was designed to take advantage of the relatedness of the sample. In fact we genotyped first a small number of markers (10K GeneChip) in largest families to identify shared stretches of chromosome between individuals, and then genotyped with a denser map (the 500K chip) a subset of individuals, selected to be dispersed along the known pedigrees with an overlap of 436 individuals genotyped with both platforms. Next a within-family imputation method was used to predict missing genotypes (see Material and Methods). The two markers, rs17680137 and rs2117032, belong to the 500 K SNP array and 10 K SNP array, respectively, and thus were imputed the first in the majority of the samples, the second in only one-fourth. To evaluate the effect of the different number of genotyped and imputed samples, we directly genotyped both markers in individuals for whom only imputed genotypes were available. Using all available genotypes, the association test showed SNP rs17680137 as more representative of total bilirubin variation (P = 3.9 × 10−9, see Table 1 and Supplementary Material, Table S1). Interestingly, when including this marker in the model, SNP rs2117032 still showed some evidence for association (P = 0.00050). Furthermore, we observed a significant interaction between the two markers (P = 1.28 × 10−04). Using all available genotypes, we were able to infer the complete haplotypes of 3431 individuals for which the wet-genotype for both markers were available and we observed that only haplotypes carrying both positive or both negative alleles were significant (HGC P = 1.8 × 10−07 and HCT P = 2.1 × 10−07, see Table 2). Notably, we observed that the haplotype frequencies were similar to the minor allele frequency of the two variants (HGC freq = 0.26 and HCT freq = 0.43), and that each haplotype was highly correlated with one variant (HGC and rs17680137 cor = 0.97, HCT and rs2117032 cor = 0.98) (Table 2). Those results were also replicated in the Amish population, where the only two haplotypes associated with bilirubin levels were those carrying both positives or both negatives alleles. Each haplotype was highly correlated with one of the two markers (Table 2). Thus the two SNPs, rather than having independent effects, are likely imperfectly tagging a third common variant of which the allele responsible for an increase in bilirubin levels would be probably associated with the HGC haplotype, and the other allele with the HCT haplotype.
We also evaluated the effect of possible confounding factors related to the particular hematological features of the Sardinian population, and repeated the association tests including as covariates Mean Corpuscular Hemoglobin, Mean Corpuscular Volume, total Hemoglobin, β-thalassemia carrier status and G6PD levels (8,9). None of those factors affected the association at the three loci. As expected, adjustment for G6PD levels reduced the association at the G6PD locus, somewhat strengthening the UGT1A1 signal (Supplementary Material, Table S2). In addition, we observed that the association of UGT1A1 and SLCO1B3 appeared stronger in women than in men, although we found no significant difference in the magnitude of the effect between sexes. In contrast, the association with the X-linked G6PD locus was stronger in men, most likely because of the major impact of G6PD deficiency in males. In fact, there was no difference in association between G6PD genotype and bilirubin when G6PD level was taken into account (Supplementary Material, Table S3). Finally, none of the tests for epistasis between the top SNPs at the three loci was significant after correction for multiple testing (Supplementary Material, Table S4). Additional studies are required to confirm this observation.
Our findings at the SLCO1B3 locus were successfully replicated in an independent sample of 1860 Sardinians, previously described and reported as SardiNIA stage 2 (10), with an effect in the same direction as that observed in the GWA sample (P < 2 × 10−04, Table 1). Moreover, we also replicated the association in 832 individuals enrolled in the Amish Heredity and Phenotype Intervention (HAPI) Heart study (11) (P < 7 × 10−03, Table 1), suggesting that variants in the SLCO1B3 influence bilirubin levels not only in Sardinians but also in other populations of Caucasian origin. Combined, the evidence for association at this locus was P < 5 × 10−14 (Table 1).
As shown in Fig. 2, although our results point to the SLCO1B3 gene, weaker association encompasses a genomic region of 700 Kb harboring, in addition to SLCO1B3, four more genes in the same family of membrane transport proteins. Among them, SLCO1B1 (12) has been previously implicated in bilirubin liver uptake (13) and associated with bilirubin levels in Asian populations. In this gene two coding variants, namely Val174Ala/rs4149056 and Asp130Asn/rs2306283, have been described as major determinants of bilirubin variation (14,15), and they weakly correlated with rs17680137 (r2 = 0.008 and r2 = 0.359, respectively) and with rs2117032 (r2 = 0.031 and r2 = 0.302, respectively). To compare their effects with polymorphisms at SLCO1B3 we used inferred genotypes based on HapMap (16) CEU (Utah residents with ancestry from northern and western Europe) haplotypes available from previous studies (17), and tested for association with total, conjugated and unconjugated bilirubin. Only nominal association with total bilirubin was detected at these coding variants (P = 0.079 and P = 3.6 × 10−03, Table 3). In addition, when including rs4149056 and rs2306283 as covariates in the model, SNPs at the SLCO1B3 locus still remain highly significant (Supplementary Material, Table S5), supporting the hypothesis that the association signal observed in the SLCO1B3 is not modulated by an indirect effect of these coding variants. Furthermore, since the SNPs tested in the GWAS covered 66% of the common variants in the SLCO1B1 at r2 > 0.5, it is unlikely that other SNPs in this gene may have a stronger effect than those in the SLCO1B3 locus. The differential contribution of the two transporters may be owing to population-specific genetic features. In Asians, for example, SNP rs17680137 is monomorphic.
Interestingly, we observed that SNPs in SLCO1B1 were mostly associated with conjugated bilirubin (P = 1.7 × 10−04 for rs4149056), whereas no association was observed with unconjugated bilirubin (P = 0.26, Table 3). In contrast, SNP rs17680137 in SLCO1B3 showed a stronger association with unconjugated bilirubin (P = 5.2 × 10−09) compared with conjugated bilirubin (P = 0.0004). A similar pattern of association was observed in the SardiNIA stage 2 sample (Table 3), where SNP rs4149056 was directly genotyped. Again, the association with unconjugated bilirubin at SNPs rs17680137 and rs2117032 remained significant after accounting for the two SNPs in the SLCO1B1 gene (Supplementary Material, Table S5).
Conjugated bilirubin was not available in the Amish cohort, and thus a parallel analysis could not be performed.
We next tested for any effect of the UGT1A1 gene on the association of bilirubin levels with the SLCO1B3 gene. Including SNP rs887829, the top marker on the UGT1A1 locus in the model, the association at rs2117032 and rs17680137 did not disappear but rather increased (data not shown), suggesting that SLCO1B3 explains part of the variability not accounted for by UGT1A1 polymorphisms. To explore this possibility further, we evaluated whether this locus may be responsible for idiopathic mild hyperbilirubinemia in patients lacking mutations in the UGT1A1 gene. Hence, we genotyped 1004 healthy individuals and 261 unrelated patients with hyperbilirubinemia (defined as serum total bilirubin >1.2 mg/dl) (2) from the SardiNIA sample for the mutation in the TATAA element of the 5′ promoter region of the UGT1A1 gene (2). We observed a 5% enrichment in the frequency of the G allele of SNP rs17680137 in the hyperbilirubinemic patients compared with the healthy subgroup (P = 0.029; Table 4). Notably, most of the difference was attributable to patients homozygous for the wild-type allele, A(TA)6TAA at the UGT1A locus (N = 95, P = 0.031), rather than to patients carrying at least one copy of the mutated allele A(TA)7TAA (N = 166, P = 0.053) (2). Thus one can infer that SLCO1B3 not only acts in the general population, but also affects mild hyperbilirubinemias.
GWA analysis for serum bilirubin levels on 4300 Sardinian revealed two striking signals in the UGT1A1 and the G6PD genes. The association with UGT1A1 is well known. In fact, mutations in this gene are responsible for the Crigler–Najjar syndrome as well as the more common mild unconjugated hyperbilirubinemia of Gilbert syndrome (2). The association of G6PD with bilirubin levels was also expected, given the high frequency of deficiency in G6PD in the Sardinian population; this condition is indeed associated with a shorter life span of red blood cells, resulting in a modest increase in bilirubin levels (18). Furthermore, the results revealed an additional locus on chromosome 2p12.2. The strongest associated marker in this region, an intronic SNP of the SLCO1B3 gene, was successfully replicated in an independent Sardinian sample and in a cohort of Amish individuals. Notably, this gene is highly expressed in the basolateral membrane of hepatocytes and is a member of the OATP family encoding the Organic Anion Transporter Polypeptide Na+-independent, OATP1B3 (7). The OATP proteins are expressed in a variety of tissues, including gut, brain, kidney and liver where they play important roles in drug absorption and distribution and excretion, with a wide spectrum of substrates that include both endogenous and xenobiotic compounds (1). Thus SLCO1B3 is a plausible candidate gene responsible for changes in bilirubin levels. Interestingly, another solute carrier organic anion transporter, SLCO1B1, previously implicated in liver bilirubin uptake (13), maps in the same region, but only weaker association was observed at this locus. In particular, we found that SNPs at SLCO1B1 were mostly associated with conjugated bilirubin, although no association was detected with unconjugated bilirubin. Although we cannot exclude the possibility of a differential contribution of the two transporters in a population-specific fashion, the results produced in this study indicate that although the SLCO1B3 gene regulates unconjugated serum bilirubin, SLCO1B1 appears to act on conjugated bilirubin. SLCO1B1, with an high affinity for glucuronated bilirubin (19), may contribute (together with MPR3) to reverse transport from hepatocytes to plasma in normal conditions, and very likely also in cholestasis when the major export pump MPR2 is downregulated (20).
Further studies are necessary to identify the SLCO1B3-specific causative variants and then to clarify the functional role of both OATP genes on conjugated and unconjugated bilirubin. Finally, the SLCO1B3 SNPs identified here were also found to be associated with unconjugated hyperbilirubinemia of unknown origin. Speculatively, they may also act as modifiers of the clinical course of some phenotypes, as already shown for UGT1A1 polymorphisms in hemolytic anemia, and anemia resulting from ineffective erythropoiesis, and neonatal jaundice (8,9,21). Thus, our findings not only reveal an additional regulator of bilirubin levels in the general population but may also contribute to the identification of genetic variants necessary to predict and prevent severe clinical manifestations in hematological and hepatic disorders linked to hyperbilirubinemia.
We recruited and phenotyped 6148 individuals, males and females, ages 14–102 years, from a cluster of four towns in the Ogliastra province of Sardinia (4). During physical examination, a blood sample was collected from each individual, and divided into two aliquots. One aliquot was used for genomic DNA extraction and the second (7 ml peripheral blood without anticoagulant) to characterize several blood phenotypes, including evaluation of serum total and conjugated bilirubin levels, measured with Express Plus (Bayer) equipment, according to manufacturer's instruction.
The normality range was 0.2–1.0 and 0–0.2 mg/dl for total and conjugated bilirubin, respectively. Hyperbilirubinemia was defined when total bilirubin serum concentration exceeded 1.2 mg/dl (2).
We genotyped from the whole sample 4305 individuals selected to represent the largest available families, regardless of their phenotypic values (5). Specifically, 1412 were genotyped with the 500K Affymetrix Mapping Array Set and 3,329 with the 10K Mapping Array Set, with 436 individuals genotyped with both arrays. This genotyping strategy allowed us to examine the majority of our cohort in a cost-effective manner because genotypes for the SNPs that passed quality-control checks could be propagated through the pedigree via imputation (22,23). Bilirubin measurements were available for 4300 individuals among the 4305 genotyped. A total of 362 129 SNPs passed initial quality-control checks (5) and were tested for association. Although the statistical analysis was ongoing, an additional 2 million autosomal markers from HapMap imputed data became available. Repeating the GWA analysis using this larger data set of markers detected no additional loci, and the top markers at chromosomes 2 and 12 loci were confirmed. Further details of the strategy for imputation and data analysis are given in the Statistical Analysis section below.
We designed a ParAllele Custom Chip (from Affymetrix) to replicate the regions associated with bilirubin levels as well as other traits studied in the SardiNIA Project. Genotyping was performed in 1862 Sardinian, selected within the SardiNIA sample and independent from the individuals included in the GWA scan (kinship coefficient = 0 with those directly typed with the 500K chip), and indicated here as SardiNIA stage 2 data set (17). SNPs rs17680137 and rs2117032, were included in the ParAllele Custom chip, and also genotyped in additional individuals using TaqMan SNP genotyping assay (Applied Biosystems) according to the protocol. The concordance rate between inferred and wet genotypes was 74.7 and 93.4% for rs17680137 and rs2117032 (over N = 1460 and N = 202 samples), respectively. SNP rs4149056, not included in the custom chip was genotyped in the SardiNIA stage 2 data set using TaqMan technology. Mutations in the TATAA element of the 5′ promoter region of the UGT1A1 gene were analyzed by direct sequencing (ABI-PRISM 377; Applied Biosystems, USA) of genomic DNA.
For individuals genotyped with a sparse map, we used a modified version of the Lander–Green algorithm (23) to estimate IBD sharing at the location of the SNPs being tested and identify stretches of haplotype shared with close relatives who were genotyped at higher density and probabilistically infer missing genotypes (22). Our inference approach allowed us to account for uncertainty in genotype assignment, estimating, instead of the most likely genotype, an expected genotype score, representing the expected number of copies of a reference allele (a fractional number between 0 and 2). The genotype scores were then used in the family-based association test for evaluating the additive effect of each marker, as described elsewhere (22,24). Owing to computational constraints, we divided large pedigrees into sub-units with ‘bit-complexity’ of 19 or less (typically, 20–25 individuals) before all the analyses. The association test fits a simple regression model and uses a variance component approach to account for correlation between different observed phenotypes within each family. To evaluate association on the X chromosome, we modeled a polygenic variance component shared according to an X-linked kinship coefficient in addition to the usual autosomal polygenic variance component (4,24). Prior to all analyses, traits were normalized using quantile normalization to avoid inflation of type I error rates; gender, age and age squared were included as covariates.
In the SardiNIA stage 2 the association with bilirubin levels was tested as in the SardiNIA GWA sample. To infer haplotypes we used all available genotypes and Merlin software. We created a variable for each haplotype coded as 0, 1 or 2, representing the number of copies carried by an individual. A linear regression model was fitted for each haplotype. Epistasis and haplotype-specific tests were performed in the GWA sample using the R kinship package.
The HAPI Heart Study was initiated in 2002. Participants of the HAPI Heart Study comprised adults from the Old Order Amish community of Lancaster County, PA, who were recruited over a 3-year period. The study aims and recruitment details, including ascertainment criteria, have been described previously (11). Physical examinations were conducted at the Amish Research Clinic in Strasburg, PA. A blood sample was obtained for measurement of bilirubin levels. Biochemical assays were performed by Quest Diagnostics (Horsham, PA, USA).
Genotypes for SNPs rs17680137, rs10770763 (proxy of rs2117032; r2 = 0.96), rs11045879 (proxy of rs4149056; r2 = 0.86) and rs2291075 (proxy of rs2306283; r2 = 0.86), were obtained from a previous GWAS performed using the 500K Affymetrix Mapping Array Set.
Statistical analysis was performed using in house developed software. In brief, we performed a measured genotype approach utilizing a t-test of the beta coefficient for the SNP variable. We included sex, age and age squared as fixed covariates in the model and a polygenic component modeled as a random effect to account for the full 14 generation pedigree of the Amish. The bilirubin trait was normalized using quantile normalization to avoid inflation of type I error rates; gender, age and age squared were included as covariates, as in the SardiNIA GWA sample.
The work was supported by the following National Institutes of Health grants: HL084729 to G.R.A., HG002651 to G.R.A., National Institute of Aging contracts NO1-AG-1-2109 to the SardiNIA (‘ProgeNIA’) team and 263-MA-410953 to the University of Michigan (G.R.A).The Old Order Amish study was supported by National Institutes of Health research grant U01 HL72515, the University of Maryland General Clinical Research Center, grant M01 RR 16500, the Clinical Nutrition Research Unit of Maryland P30 DK072488, and the Baltimore Veterans Administration Medical Center Geriatrics Research.
G.R.A, D.S., A.C., A.R.S. and M.U. designed the study; R.G., A.S., D.S. and M.U. provided material and reagents; F.B., A.M., G.U., M.D., S.L. and A.M. performed the genotyping; M.G.P., L.P., M.M., M.M., A.R S. performed characterization of phenotype and data collection; S.S., P.F.M., analyzed the data; S.N., L.C., F.B., A.R.S., P.F.M. contributed to draft the manuscript; S.S., D.S., A.C. and M.U. wrote the paper.
We thank the many individuals who generously participated in this study, Monsignore Piseddu, Bishop of Ogliastra, the Mayors and citizens of the Sardinian towns (Lanusei, Ilbono, Arzana and Elini), and the head of the Public Health Unit ASL4, the province of Ogliastra for their volunteerism and cooperation. In addition, we are grateful to the Mayor and the administration in Lanusei for providing and furnishing the clinic site. We thank the team of physicians Angelo Scuteri, Marco Orrù, Maria Grazia Pilia, Maria Cristina Spada, Danilo Fois, Liana Ferreli, Francesco Loi, and nurses Paola Loi, Monica Lai and Anna Cau who carried out participant physical exams, and Monica Balloi who recruited the volunteers.
Conflict of Interest statement. None declared.