Population-Based Genome-wide Association Studies Reveal Six Loci Influencing Plasma Levels of Liver Enzymes

Abstract

Plasma liver-enzyme tests are widely used in the clinic for the diagnosis of liver diseases and for monitoring the response to drug treatment. There is considerable evidence that human genetic variation influences plasma levels of liver enzymes. However, such genetic variation has not been systematically assessed. In the present study, we performed a genome-wide association study of plasma liver-enzyme levels in three populations (total n = 7715) with replication in three additional cohorts (total n = 4704). We identified two loci influencing plasma levels of alanine-aminotransferase (ALT) (CPN1-ERLIN1-CHUK on chromosome 10 and PNPLA3-SAMM50 on chromosome 22), one locus influencing gamma-glutamyl transferase (GGT) levels (HNF1A on chromosome 12), and three loci for alkaline phosphatase (ALP) levels (ALPL on chromosome 1, GPLD1 on chromosome 6, and JMJD1C-REEP3 on chromosome 10). In addition, we confirmed the associations between the GGT1 locus and GGT levels and between the ABO locus and ALP levels. None of the ALP-associated SNPs were associated with other liver tests, suggesting intestine and/or bone specificity. The mechanisms underlying the associations may involve cis- or trans-transcriptional effects (some of the identified variants were associated with mRNA transcription in human liver or lymphoblastoid cells), dysfunction of the encoded proteins (caused by missense variations at the functional domains), or other unknown pathways. These findings may help in the interpretation of liver-enzyme tests and provide candidate genes for liver diseases of viral, metabolic, autoimmune, or toxic origin. The specific associations with ALP levels may point to genes for bone or intestinal diseases.

Main Text

Plasma liver-enzyme tests are widely used in the clinic to identify patients with liver diseases, to monitor the course and severity of these diseases and the effect of therapies, and to detect drug-induced liver injury.^1,2 These tests also have substantial epidemiologic significance that extends beyond the liver, given that they have been shown to be prospective risk factors for type 2 diabetes, cardiovascular disease, and all-cause mortality in multiple large studies.^3–6 Therefore, it is of interest to identify genes or loci affecting these markers in order to establish whether such loci are also associated with these clinical endpoints.

Plasma liver-enzyme levels are influenced by environmental and genetic factors. The estimated heritabilities range from 33% for alanine-aminotransferase (ALT) to 61% for gamma-glutamyl transferase (GGT).^7,8 So far, only a limited number of genes that influence liver-enzyme levels have been identified, mostly those responsible for Mendelian liver diseases such as mutations in the HFE gene in hemochromatosis (MIM 235200).^9,10 A thorough understanding of the genetic determinants of plasma liver enzymes is important for proper interpretation of these tests. Indeed, such information could assist in our understanding of the interindividual differences in the propensity for development of liver dysfunction in the presence of toxins or conditions such as metabolic syndrome. As such, identification of genes associated with liver-enzyme levels could reveal previously unsuspected candidate genes for liver diseases of viral, metabolic, autoimmune, or toxic origin.

The goal of the present study was to identify, by using a genome-wide association (GWA) approach, genes influencing plasma levels of ALT and aspartate-aminotransferase (AST), two markers of hepatocyte injury and liver fat accumulation, and of alkaline phosphatase (ALP) and GGT, used primarily as indicators of biliary or cholestatic diseases and heavy alcohol consumption.

In the discovery phase, we carried out independent GWA studies in three population-based cohorts, the CoLaus Study from Lausanne Switzerland^11,12 (n = 5636), the InCHIANTI Study from Tuscany Italy¹³ (n = 1200), and a subset of the LOLIPOP Study from West London UK^14,15 (n = 879) (Table 1). The clinical characteristics of the participants are described in Table 2. The mean levels of liver-enzyme tests varied somewhat between populations, presumably because of slight differences in the demographics of the populations under study and methodological differences in the assays (Table S1 available online). Accordingly, we used study-specific criteria for GWA-genotyping quality control^11,14 and liver-enzyme-level analyses (Appendix A). Additional genotypes were imputed on the basis of the HapMap Phase II data with the software IMPUTE.¹⁶ For these imputed data, we performed association analysis with SNPTEST, using the full posterior probability genotype distribution for each study separately (adding in relevant covariables). Only SNPs with a posterior probability score >0.90, high genotype information content (proper_info >0.5), and minor allele frequency >0.01 were considered for these imputed association analyses.

Table 1.

Description of Six Collections Used in GWAS on Plasma Liver-Enzyme Levels

	CoLaus	InCHIANTI	LOLIPOP
Sample size (n)	5636	1200	879	1006	1005	2693
Location	Lausanne, Switzerland	Tuscany, Italy	London, UK	London, UK	London, UK	London, UK
Ethnicity	European white	European white	European white	Indian Asian	European white	Indian Asian
Study Design	Population-based	Population-based	Population-based enriched with coronary artery disease	Nested metabolic syndrome case and control	Nested metabolic syndrome case and control	Nested metabolic syndrome case and control
Data Usage	GWA Discovery	GWA Discovery	GWA Discovery	Replication	Replication	Replication
Liver enzymes	AST,a ALT,b GGT,c and ALPd	AST, ALT, GGT, ALP	ALT, GGT, ALP	ALT, GGT, ALP	ALT, GGT, ALP	ALT, GGT, ALP
Genotyping Platform	Affymetrix 500k	Illumina HumanHap550	Affymetrix 500k	Perlegen, custom array	Perlegen, custom array	Illumina HumanHap300
Number of SNPs genotyped	370,697	496,032	387,549	248,537	266,722	317,968

^aAST, aspartate aminotransferase.

^bALT, alanine aminotransferase.

^cGGT, gamma-glutamyltransferase.

^dALP, alkaline phosphatase.

Table 2.

Clinical Characteristics of the Participants in the Discovery Data Sets from CoLaus, InCHIANTI, and LOLIPOP Population-Based Studies

	CoLausa	InCHIANTIa	LOLIPOPa
n	5636	1200	879b
Sex (% F)	52.8	55.2	23.0
Age (years)	53.2 ± 10.8	68.4 ± 15.9	52.8 ± 10.4
Fasting glucose (mmol/L)	5.56 ± 1.16	5.24 ± 1.36	5.50 ± 1.54
Insulin (μIU/mL)	8.76 ± 6.25	10.9 ± 6.15	10.3 ± 10.7
BMI (kg/m2)	25.8 ± 4.6	27.1 ± 4.1	27.9 ± 4.8
Waist (cm)	89.3 ± 13.4	NA	96.7 ± 13.0
Hip (cm)	101.0 ± 9.3	NA	103.0 ± 9.0
Total cholesterol (mmol/L)	5.60 ± 1.04	NA	5.32 ± 1.07
LDL-cholesterol (mmol/L)	3.34 ± 0.92	3.43 ± 0.90	3.25 ± 0.93
Triglycerides (mmol/L)	1.40 ± 1.20	1.42 ± 0.88	1.61 ± 1.16
HDL-cholesterol (mmol/L)	1.64 ± 0.44	1.43 ± 0.39	1.36 ± 0.35
Albumin (g/L)	44.2 ± 2.5	42.3 ± 3.2	43.6 ± 2.9
Smoking status (% smokers)	11.8	18.8	64.4
Alcohol consumption (%)c	77.3	NA	75.7
Type 2 Diabetes (%)	6.6	11.1	11.3
AST (U/L)	29.7 ± 13.5	20.2 ± 0.4	NA
ALT (U/L)	27.6 ± 19.5	18.1 ± 0.5	29.1 ± 16.4
ALP (U/L)	63.4 ± 20.4	204.2 ± 71.5	80.9 ± 38.5
GGT (U/L)	32.1 ± 39.9	26.2 ± 28.7	40.2 ± 44.8

^aResults are expressed as mean ± SD.

^bSubset from the LOLIPOP population-based study enriched with CAD cases.

^cAlcohol consumption was defined as alcohol intake ≥1 unit per week.

Quantile-quantile plots (Figure 1) revealed the presence of a substantial number of SNPs associated with ALT, GGT, and ALP levels at a genome-wide significance level (p value < 10⁻⁷). No SNP with plasma levels of AST was associated with genome-wide significance. It is not clear why AST was uninformative in this case. Highly significantly associated SNPs were located within discrete regions of the genome (Figure 2), including two loci for ALT (10q24.2 and 22q13.31), two loci for GGT (12q24.31 and 22q11.23), and five loci for ALP (1p36.12, 6p22.2, 9q34.13, 10q21.2-21.3, and 19q13.33).

Figure 1.

Quantile-Quantile Plot for GWA for ALT, GGT, ALP, and AST in a Meta-analysis from CoLaus, InCHIANTI, and LOLIPOP Discovery Study Data Sets with Imputed Autosomal SNPs

The blue and red dotted lines correspond to 95% CI and 99% CI, respectively. Each panel shows data for the enzymes as follows: (A), ALT; (B), GGT; (C), ALP; and (D), AST.

Figure 2.

GWA Analysis of ALT, GGT, and ALP

Represented here are three Manhattan plots showing the degree of association (−log10 p value) between imputed SNPs included in the meta-analysis of three discovery collections totaling 7715 individuals. Red dots correspond to the SNPs with p value ≤ 10⁻⁸.

A total of 74 Affymetrix-genotyped SNPs (24 for ALT, 8 for GGT, and 42 for ALP, Table S2) with a p value ≤ 10⁻⁵ in the discovery phase was subsequently examined in 3699 Indian Asian and 1005 European white participants from the LOLIPOP study (Table 1). After joint analysis of discovery and replication data, 32 SNPs (10 for ALT, 6 for GGT, and 16 for ALP, Table S3 and Figures S1–S8) reached genome-wide significance with directionally consistent signals (Figures S9–S16 for the leading SNPs) among the discovery and replication cohorts. The remaining 42 SNPs were not considered as replicated because of either lack of high-quality imputation (Appendix A) of the presence of inconsistent direction of the effect across studies or because they did not reach genome-wide significance level (p ≤ 10⁻⁸). It is also conceivable that absence of replication was due to LD differences between the European white and Indian Asian cohorts. Out of the 32 replicated SNPs, 19 SNPs were considered to be independent (r² < 0.4) (Table 3). These SNPs were located within two loci for ALT, two loci for GGT, and four loci for ALP.

Table 3.

Independent SNPs Associated with Liver-Enzyme Levels with Genome-wide Significance in Combined GWAS Analysis of Discovery and Replication Data Sets

SNPa	CHR	Positionb	Nearest Gene	Region	Minor Allelec	MAFd	Three Discovery Collections (n = 7,715)		Three Replication Collections (n = 4,704)		Six Collections (n = 12,419)
							Pooled Beta-Coefficient (standard error)	Combined p Value	Pooled Beta-Coefficient (standard error)	Combined p Value	Pooled Beta-Coefficient (standard error)	Combined p Value
ALT
rs11597390	10	101851425	CPN1	intergenic	A	0.36	−0.044 (0.008)	2.9 × 10−8	−0.023 (0.013)	0.08	−0.039 (0.007)	1.5 × 10−8
rs11597086	10	101943695	CHUK	Intronic	C	0.43	−0.038 (0.007)	3.6 × 10−7	−0.030 (0.012)	0.01	−0.036 (0.006)	1.8 × 10−8
rs11591741	10	101966491	CHUK	intronic	C	0.43	−0.038 (0.007)	4.5 × 10−7	−0.029 (0.012)	0.02	−0.035 (0.006)	3.0 × 10−8
rs2281135	22	42657471	PNPLA3	intronic	T	0.18	0.065 (0.010)	8.2 × 10−12	0.051 (0.012)	1.3 × 10−5	0.060 (0.007)	8.4 × 10−16
rs2143571	22	42716587	SAMM50	intronic	A	0.19	0.046 (0.009)	9.4 × 10−7	0.038 (0.012)	1.8 × 10−3	0.043 (0.007)	7.2 × 10−9
GGT
rs1169313	12	119905390	HNF1A	intronic	C	0.38	−0.007 (0.001)	3.2 × 10−8	−0.003 (0.001)	7.0 × 10−5	−0.005 (0.001)	1.8 × 10−10
rs4820599	22	23314767	GGT1	intronic	G	0.31	0.006 (0.001)	3.9 × 10−6	0.008 (0.002)	1.8 × 10−6	0.007 (0.001)	4.0 × 10−11
ALP
rs1780324	1	21567063	NBPF3-ALPL	intergenic	T	0.43	0.032 (0.005)	1.4 × 10−10	0.031 (0.007)	1.0 × 10−5	0.031 (0.004)	7.0 × 10−15
rs9461011	6	24548330	GPLD1	intronic	G	0.23	0.033 (0.006)	4.4 × 10−8	0.030 (0.009)	0.001	0.032 (0.005)	1.9 × 10−10
rs9467160	6	24549725	GPLD1	intronic	A	0.24	0.033 (0.006)	2.2 × 10−8	0.035 (0.009)	1.3 × 10−4	0.034 (0.005)	1.2 × 10−11
rs8176720	9	133162427	ABO	synonymous	G	0.32	0.031 (0.005)	1.4 × 10−8	0.031 (0.008)	7.0 × 10−5	0.031 (0.004)	4.3 × 10−12
rs641959	9	133163253	ABO	intronic	G	0.26	0.035 (0.006)	6.8 × 10−10	0.024 (0.008)	0.004	0.031 (0.005)	2.1 × 10−11
rs514708	9	133163297	ABO	intronic	T	0.26	0.035 (0.006)	6.2 × 10−10	0.024 (0.008)	0.004	0.031 (0.005)	1.9 × 10−11
rs672316	9	133167679	ABO	intronic	C	0.27	0.032 (0.006)	1.4 × 10−8	0.023 (0.008)	0.006	0.029 (0.005)	4.4 × 10−10
rs657152	9	133168819	ABO	intronic	T	0.39	−0.057 (0.005)	4.6 × 10−29	−0.029 (0.007)	4.9 × 10−5	−0.047 (0.004)	1.7 × 10−30
rs474279	9	133169171	ABO	intronic	T	0.24	0.030 (0.006)	3.4 × 10−7	0.023 (0.008)	0.006	0.028 (0.005)	8.3 × 10−9
rs552148	9	133183035	ABO	upstream	T	0.24	0.032 (0.006)	3.1 × 10−8	0.022 (0.008)	0.008	0.029 (0.005)	1.2 × 10−9
rs12355784	10	64791571	JMJD1C	intronic	A	0.48	0.026 (0.005)	4.7 × 10−7	0.025 (0.007)	2.7 × 10−4	0.025 (0.004)	5.0 × 10−10
rs10761779	10	64944933	REEP3	intergenic	G	0.49	0.025 (0.005)	3.9 × 10−7	0.024 (0.007)	4.5 × 10−4	0.025 (0.004)	6.9 × 10−10

^aIndependent SNPs (r² < 0.4) were selected from each loci.

^bPosition is of NCBI Build 35.

^cMinor allele corresponds to forward strand and is of NCBI Build 35.

^dMAF, Minor allele frequency. It is based on CoLaus study.

The ALT-associated 10q24.2 region (Figure S1) spans three genes. CHUK (IKK-α) is a ubiquitously expressed serine threonine protein kinase that modulates the NF-κB-transcription-factor-dependent activation of several genes implicated in insulin resistance. Its intronic SNP rs11591741 has been reported to be associated with expression of its upstream gene CWF19L1 (CWF19-like 1) in human liver (p = 1.47 × 10⁻³⁹),¹⁷ probably through its transcriptional regulatory effect. CPN1 encodes arginine carboxypeptidase-1, a liver-expressed plasma metallo-protease that protects the body from potent vasoactive and inflammatory peptides containing C-terminal arginine or lysine (such as kinins or anaphylatoxins), which are released into the circulation.^18,19 Defects in CPN1 are the cause of carboxypeptidase N deficiency (MIM 212070). ERLIN1 encodes a member of the prohibitin family of proteins that define lipid-raft-like domains of the endoplasmic reticulum.²⁰ The second association peak for ALT on chromosome 22q13.31 (Figure S2) encompasses a 57 kb fragmented linkage disequilibrium (LD) region encompassing SAMM50 and PNPLA3. SAMM50 is a subunit of mitochondrial SAM translocase complex for importation of proteins such as metabolite-exchange anion-selective channel precursors,^21,22 whose N-terminal domain is essential for the biogenesis of mitochondria. The variation of imputed SNP rs3761472 (p = 2.7 × 10⁻⁹) causes N-terminal Asp110Glu substitution in SAMM50, which may lead to mitochondrial dysfunction and impaired cell growth. PNPLA3 (ADPN) is a liver-expressed transmembrane protein with phospholipase activity.²³ It has been shown to be significantly upregulated during adipocyte differentiation, and in response to fasting and feeding, indicating its role in facilitating both energy mobilization and lipid storage in adipose and liver.²⁴ PNPLA3 mRNA expression was elevated in subcutaneous and visceral adipose tissue of obese subjects.²⁵ The lead SNP rs2281135 is in complete LD (r² = 1) with intronic SNPs rs1010022 and rs2072907, two obesity-associated tagSNPs that showed significant differences in adipose PNPLA3 mRNA expression and adipocyte lipolysis from experimental data.²⁵ Two imputed nonsynonymous SNPs within PNPLA3 (rs738409 [Ile148Met], p = 3.7 × 10⁻¹⁰, and rs2294918 [Lys434Glu], p = 6.0 × 10⁻⁴), represent putative exonic splicing silencer elements and might play a role in gene regulation.²⁶ Homozygous carriers of the GG genotype for rs2281135 had a 34% greater risk (OR 1.34 [1.13–1.60]) of having ALT levels above upper limits of normal (defined as 36 U/L in females, and 60 U/L in males) in the CoLaus study. The effect of the lead SNPs at both CPN1 and PNPLA3 loci was not specific for ALT because these SNPs were also associated with plasma levels of AST (Table 4), suggesting that these genes are generally predisposed to hepatocyte dysfunction.

Table 4.

Association of the Lead SNPs with Plasma Levels of Liver Enzymes in the CoLaus Study

SNP	Nearest Gene	ALT		AST		ALP		GGT
		Beta (% 95 CI)	p Value	Beta (% 95 CI)	p Value	Beta (% 95 CI)	p Value	Beta (% 95 CI)	p Value
rs11597390	CPN1	−0.043 (−0.062−0.023)	1.4 × 10−5	−0.022 (−0.035−0.009)	0.0009	−0.011 (−0.023−0.001)	0.08	−0.004 (−0.008−0.0001)	0.04
rs2281135	PNPLA3	0.064 (0.041–0.087)	2.6 × 10−8	0.035 (0.020–0.050)	5.7 × 10−6	−0.005 (−0.019−0.010)	0.51	−0.001 (−0.006−0.003)	0.61
rs1169313	HNF1A	0.002 (−0.016–0.019)	0.86	0.001 (−0.011–0.012)	0.92	0.013 (0.002–0.024)	0.03	−0.006 (−0.009−0.002)	0.0009
rs4820599	GGT1	−0.002 (−0.021−0.017)	0.82	0.003 (−0.009–0.016)	0.6	0.006 (−0.006–0.018)	0.36	0.003 (−0.001–0.007)	0.09
rs1780324	NBPF3-ALPL	0.01 (−0.007–0.027)	0.26	0.015 (0.003–0.026)	0.01	0.031 (0.020–0.042)	5.6 × 10−8	0.003 (−0.001–0.006)	0.14
rs9467160	GPLD1	0.003 (−0.018–0.024)	0.76	−0.003 (−0.017−0.011)	0.66	0.031 (0.018–0.044)	5.6 × 10−6	0.001 (−0.003–0.005)	0.65
rs657152	ABO	0.012 (−0.007–0.030)	0.21	−0.001 (−0.013−0.011)	0.88	−0.061 (−0.072−0.050)	2.5 × 10−25	0.001 (−0.002–0.005)	0.57
rs12355784	JMJD1C	−0.003 (−0.021−0.015)	0.72	0.001 (−0.011–0.013)	0.83	0.026 (0.014–0.037)	1.6 × 10−5	0.003 (−0.001–0.007)	0.1

When examining the association with GGT levels, we identified a series of SNPs within 300 kb on either side of the gene encoding GGT1 on chromosomal region 22q11.23 (Figure S3). As such, these SNPs met the definition of cis-acting SNPs.²⁷ The typed rs4820599 (p = 4.0 × 10⁻¹¹) is in high LD with previously reported variant rs5751901 (r² = 0.71) associated with serum GGT¹³ and with rs6519519 (r² = 0.96), a SNP associated with transcript abundance in lymphoblastoid cells;²⁸ such a finding suggests that the genetic association observed here is due to differential expression of the gene. Another association peak was observed on chromosomal region 12q24.31 (Figure S4) encompassing HNF1A (TCF-1), the hepatic nuclear factor predominantly expressed in the human liver. This transcriptional regulatory protein plays a prominent role in the activation of a large family of hepatocyte-specific genes involved in hepatocyte differentiation and liver development.^29,30 Mutations within this gene are associated with type III form of maturity-onset diabetes of the young (MODY3 [MIM 600496])³¹ and hepatic adenomas, which are frequently accompanied by an elevation in plasma GGT level (MIM 142330).³² SNP rs7953249 (p = 5.1 × 10⁻⁸) was also closely associated in the CoLaus study with plasma C-reactive protein levels (p = 4.4 × 10⁻¹⁰), consistent with recently reported observations.³³ It is conceivable that variants within this gene, including imputed rs2464196 (Ser487Asn, p = 3.2 × 10⁻¹²) located within the C-terminal transactivation domain of HNF1A, may broadly affect the transcriptional effect of this nuclear factor.

The strongest association with ALP levels was observed within the ABO locus on chromosome 9q34.13 (Figure S5). Association between ALP levels and the ABO blood group has been reported previously.³⁴ This association was specific to plasma ALP (Table 4) with SNP rs657152 (p = 1.7 × 10⁻³⁰) accounting for 2% of the total variance of ALP levels in the CoLaus study (Table S4). The mechanism underlying this association remains unclear and may be due to genetically determined variations in the proportion of isoenzymes among different blood types because it has been shown that the appearance of intestinal ALP in the plasma, particularly after fatty meals, is associated with ABO blood group and secretor status.³⁵ In addition, we detected a strong association with the ALPL locus (Figure S6), which encodes nonspecific alkaline phosphatase found in liver, bone, kidney, and other tissues and is responsible for hypophosphatasia.³⁶ Association of orthologous ALPL gene with serum ALP levels has been reported in mice³⁷ but not in humans. One cis-acting SNP rs1780324 within ALPL had been shown to markedly affect the gene expression in lymphoblastoid cells.²⁸ One trans-acting signal was identified within GPLD1 (glycosylphosphatidylinositol specific phospholipase D1) gene (Figure S7). GPLD1 hydrolyzes the inositol phosphate linkage in proteins anchored by phosphatidylinositol glycans (GPI-anchor), thus releasing these proteins from the membrane. Elevated serum levels of GPLD1 and hepatic mRNA expression have been reported in nonalcoholic fatty liver disease (NAFLD).³⁸ Finally, one additional trans-acting SNP was localized in the JMJD1C (TRIP8)-REEP3 (receptor accessory protein 3) region. JMJD1C encodes thyroid-hormone-receptor interactor 8, a hormone-dependent transcription factor that regulates expression of a variety of specific target genes³⁹ (Figure S8). The observation that none of the ALP-associated genes were crossassociated with other liver enzymes (Table 4) may indicate that the association signal was generated by variations in bone- or intestine- rather than liver-metabolic pathways.⁴⁰

The findings reported here were generated on a limited number of genome-wide scans, and it is anticipated that adding more scans with enriched liver phenotypes within the discovery and replication phase will further expand the number of genes and loci associated with plasma liver-enzyme levels in the population.

At this stage, the discovery of five trans-acting loci and one cis-acting locus influencing plasma levels of liver enzymes may point to mechanisms regulating these enzymes and could assist in the interpretation of liver tests in the clinic. Most importantly, these genes represent candidates for susceptibility to liver diseases. Analysis between variants within these genes and NAFLD, alcohol-, viral, autoimmune, or toxin-induced liver injury is warranted.

Appendix A

Material and Methods

Study Populations. The CoLaus study is a population-based sample consisting of 5694 Lausanne residents aged 35 to 75 years. The study design and protocol have been described previously.^11,12,42

The London Life Sciences Population (LOLIPOP) study is an ongoing population-based cohort study of ∼30,000 Indian Asian and European white men and women, aged 35–75 years, recruited from the lists of 58 general practitioners in West London, United Kingdom.^14,15 The participants included in the present study are a subset of the LOLIPOP cohort study. These subsets were assembled for specific experiments for identification of genetic variants underlying metabolic syndrome and coronary artery disease (CAD). The samples were genotyped on separate platforms (Table 1).

The InCHIANTI study is a population-based sample that includes 1200 individuals of <65 age and 1155 individuals of age ≥65 years. Details of the study design and protocol have been described previously.¹³ A total of 1200 subjects who had both WGS data and liver-enzyme levels measured were included in this study.

Genome-wide Association Statistics. Linear-regression analyses were done on natural log-transformed ALT, ALP, and AST and power-transformed GGT independently in each study. Regression analyses were done with the PLINK software package⁴³ adjusted by age, gender, and geographical principle components analyzed by EIGENSOFT software,⁴⁴ plus smoking and alcohol intake if they were significant covariates for the trait.

Imputation. For CoLaus, InCHIANTI, and LOLIPOP discovery set genotypes were imputed separately on the basis of the HapMap Phase II build35 data with the software IMPUTE.¹⁶ For these imputed data, we performed association analysis with SNPTEST, using the full posterior-probability genotype distribution for each study separately (adding in relevant covariables). Only SNPs with a posterior probability score >0.90, high genotype information content (proper_info >0.5), and MAF >0.01 were considered for the imputed association analyses. SNP imputation for LOLIPOP substudies used the MACH program (v1.0). For European data sets, CEPH haplotypes in HapMap were used as reference haplotypes. Indian Asian data sets were imputed on the basis of a combination (mixed) of HapMap populations, given that this showed greater concordance with real genotypes, compared with use of any one HapMap population. Imputation analyses were based on HapMap build35, dbSNP build 125. Only SNPs with RSQR ≥0.3 were considered for the meta-analysis.

Meta-analysis. For the initial GWA screen, analyses were performed within study, with study-specific criteria for GWA-genotyping QC and analyses (as described above). We meta-analyzed these summary data by using a fixed-effects model and inverse-variance weighted averages of beta-coefficients to obtain a combined estimate of the overall beta-coefficient and its standard error. Among-study heterogeneity was assessed with the χ² test.

Web Resources

The URLs for data presented herein are as follows:

• Affymetrix genotyping protocol, http://www.affymetrix.com/support/technical/whitepapers/brlmmwhitepaper.pdf

• EIGENSOFT software, http://genepath.med.harvard.edu/∼reich/EIGENSTRAT.htm

• IMPUTE, http://www.stats.ox.ac.uk/∼marchini/software/gwas/gwas.html

• International Hapmap Project, http://www.hapmap.org

• MACH, http://www.sph.umich.edu/csg/abecasis/MACH/

• Online Mendelian Inheritance in Man (OMIM), http://www.ncbi.nlm.nih.gov/Omim

• PLINK software package, http://pngu.mgh.harvard.edu/purcell/plink/

• SNPTEST, http://www.stats.ox.ac.uk/∼marchini/software/gwas/gwas.html

• WGAviewer,⁴¹ http://www.genome.duke.edu/centers/pg2/downloads/wgaviewer.php