Systematic review with meta-analysis: the I148M variant of patatin-like phospholipase domain-containing 3 gene (PNPLA3) is significantly associated with alcoholic liver cirrhosis

Authors

  • A.-J. Chamorro,

    1. Alcoholism Unit, Department of Internal Medicine, University Hospital of Salamanca, Institute of Biomedical Research of Salamanca (IBSAL), Salamanca, Spain
    Search for more papers by this author
  • J.-L. Torres,

    1. Alcoholism Unit, Department of Internal Medicine, University Hospital of Salamanca, Institute of Biomedical Research of Salamanca (IBSAL), Salamanca, Spain
    Search for more papers by this author
  • J.-A. Mirón-Canelo,

    1. Department of Epidemiology, University of Salamanca-IBSAL, Salamanca, Spain
    Search for more papers by this author
  • R. González-Sarmiento,

    1. Molecular Medicine Unit-IBSAL, University of Salamanca-SACYL-CSIC, Salamanca, Spain
    Search for more papers by this author
  • F.-J. Laso,

    1. Alcoholism Unit, Department of Internal Medicine, University Hospital of Salamanca, Institute of Biomedical Research of Salamanca (IBSAL), Salamanca, Spain
    Search for more papers by this author
  • M. Marcos

    Corresponding author
    1. Alcoholism Unit, Department of Internal Medicine, University Hospital of Salamanca, Institute of Biomedical Research of Salamanca (IBSAL), Salamanca, Spain
    • Correspondence to:

      Dr M. Marcos, Servicio de Medicina Interna, Hospital Universitario de Salamanca, P° San Vicente, 58-156, Salamanca 37007, Spain.

      E-mail: mmarcos@usal.es

    Search for more papers by this author

  • As part of AP&T's peer-review process, a technical check of this meta-analysis was performed by Mr M. Siddiqui. This article was accepted for publication after full peer-review.

Summary

Background

Several studies have reported an association between alcoholic liver cirrhosis (ALC) or other forms of alcoholic liver disease (ALD) and the genetic variant rs738409 (C>G) in adiponutrin/patatin-like phospholipase domain-containing 3 gene (PNPLA3).

Aim

To evaluate the influence of this variant on ALC and other forms of ALD.

Methods

We performed a systematic review of previous studies on the relationship between rs738409 of PNPLA3 and ALD and meta-analysis was conducted in a random-effects model. Calculations of the odds ratios (ORs) and their confidence intervals (CIs), tests for heterogeneity and sensitivity analyses were performed.

Results

Database search identified 11 previous studies available for inclusion with a total of 3495 patients with ALD (2087 with ALC) and 5038 controls (4007 healthy subjects and 1031 alcoholics without ALD). Patients with ALC compared to controls had a significantly higher prevalence of the G allele when comparing GG vs. CC (OR 4.30, 95% CI 3.25–5.69; P < 0.00001) or GC vs. CC genotypes (GC vs. CC: OR 1.91, 95% CI 1.67–2.17) or under a recessive or dominant model. Similar results were found when comparing separately patients with ALC vs. alcoholics without ALD or healthy subjects. An association of the G allele with ALD emerged when comparing ALD patients vs. alcoholics without ALD and/or healthy subjects although moderate to large heterogeneity was observed. Our data suggested an additive genetic model for this variant in ALD.

Conclusion

Our meta-analysis shows that the rs738409 variant of PNPLA3 is clearly associated with alcoholic liver cirrhosis.

Introduction

Alcoholic liver disease (ALD) refers to a wide spectrum of liver abnormalities, ranging from fatty liver to acute alcoholic hepatitis, and alcoholic liver cirrhosis (ALC). The most severe of these, ALC, causes an estimated 373 000 deaths per year.[1]

Genetic and environmental factors influence the risk of developing ALD in heavy drinkers.[2, 3] However, identifying specific genetic variants associated with ALD has been difficult,[4] despite the large number of association studies and meta-analysis performed.[5-7] One variant has garnered much research attention recently. The single nucleotide polymorphism (SNP) rs738409 C>G, within the adiponutrin/patatin-like phospholipase domain-containing 3 gene (PNPLA3), causes a substitution of methionine for isoleucine at position 148 (I148M). This SNP was initially found to be associated with nonalcoholic fatty liver disease (NAFLD) in a genome-wide association scan (GWAS)[8] and subsequent studies have identified an association of this SNP with ALC and ALD.[9-11] Further, the G allele confers a higher risk of steatosis and/or fibrogenesis in liver diseases, including chronic hepatitis B and C infection, and hereditary hemochromatosis. Therefore, this SNP seems to be a risk factor in the development of advanced liver disease and hepatocellular carcinoma (HCC).[12]

Meta-analyses of genetic association studies are useful to confirm and evaluate the strength of an association when large numbers of genetic studies have been published.[13] To date, two meta-analyses have clearly established the relationship of this SNP with NAFLD[14] and HCC.[15] It is of clinical interest to establish the relationship of this polymorphism with ALD, since the presence of this allelic variant could help to identify patients with a higher risk of developing advanced forms of this disease. Therefore, the purpose of this study was to perform a systematic review and a meta-analysis to analyse the influence of this polymorphism on different forms of ALD, and to estimate the strength of this association.

Material and methods

Inclusion criteria

Case–control, cross-sectional and cohort studies that analysed the relationship between ALD and the polymorphism rs738409 (G>C) of PNPLA3 were included. We required that studies include a group of patients with ALC or a group of patients with other forms of ALD (e.g. alcoholic hepatitis or alcoholic fatty liver) as cases, and a group of alcoholics without ALD or a group of healthy subjects as controls. Cases had to be diagnosed either by liver biopsy or by clinical, ultrasonography and/or gastroscopic findings according to defined criteria. Studies that included cases or controls with different phenotypes had to provide separate genotype information for every group.

Bibliographical search and data extraction

Reports published up to April 2014 and containing information on the relationship between the rs738409 SNP and the presence of ALD were identified through a computerised search in the following electronic databases: MedLine (source PubMed), Web of Science, Scopus and Embase (which includes conference abstracts). Databases were searched, without language restriction, by using the following terms in combination as MeSH terms and text words: ‘rs738409’, ‘PNPLA3’, ‘adiponutrin’, ‘patatin-like phospholipase domain-containing 3’, ‘alcohol’, ‘alcoholism’, ‘alcoholic liver disease’ and ‘liver cirrhosis’. The search was supplemented by reviewing references of original research reports and review articles on the topic. We also retrieved additional studies using the MedLine option ‘Related Articles’. The bibliographical search and data extraction steps were performed by two investigators (AJC, JLT). Disagreements were resolved by consulting with a third author (MM).

The criteria used for the diagnosis of ALC or other forms of ALD were recorded and genotyped individuals were first categorised into the following groups: alcoholics with ALD, alcoholics without liver disease and healthy individuals. Alcoholics with ALD were further classified according to the type of ALD. Genotype frequencies were extracted or calculated from raw data. Corresponding authors were contacted via email for additional or missing data regarding genotype counts when necessary. Additional data extracted included authors' name(s), year of publication, country, ethnicity, demographical information, presence of blind genotyping, report of Hardy–Weinberg equilibrium (HWE) testing and adjustment or matching for confounders.

Statistical methods and comparisons

The main objective of our meta-analysis was to compare the distribution of rs738409 G>C SNP among alcoholics with ALC as cases vs. controls. For this purpose, independent meta-analyses were carried out to combine: (i) studies that had compared alcoholics with ALC vs. alcoholics without ALD, (ii) alcoholics with ALC vs. healthy subjects and (iii) studies that had compared alcoholics with ALC vs. a group of controls (either alcoholics without liver disease or healthy subjects). Our secondary aim was to analyse the distribution of rs738409 among patients with ALD vs. controls. Additional meta-analyses were also performed, if three or more studies were available to do so, to compare: (i) patients with ALC as cases vs. patients with ALD excluding those with ALC or alcoholic hepatitis, (ii) patients with ALD excluding those with ALC or alcoholic hepatitis as cases vs. controls and (iii) patients with alcoholic hepatitis vs. controls. Alcoholic patients lacking information about concomitant liver disease were not included in any comparison. Sub-analysis by ethnicity was also performed.

The strength of the association was estimated by odds ratio (OR) and its 95% confidence interval (CI) for each study. Pooled results are reported as the OR and 95% CI with P-values, obtained with a random-effects model (DerSimonian and Laird method).[16] A P-value <0.05 was considered statistically significant. Cochran's Q-statistic was used to assess heterogeneity. A significant Q-statistic (< 0.10) indicated heterogeneity across studies. The I2 statistic was used to estimate inconsistency in meta-analyses, representing the percentage of the observed between-study variability due to heterogeneity rather than chance. The following suggested cut-off points were used: I2 = 0–25%, no heterogeneity; I2 = 25–50%, moderate heterogeneity; I2 = 50–75%, large heterogeneity; I2 = 75–100%, extreme heterogeneity.[17]

We did not choose any a priori genetic model and analysed potential associations comparing GG vs. CC and GC vs. CC genotypes. We also explored the different comparisons under recessive (GG vs. GC + CC) and dominant (GG + GC vs. CC) models. We further plotted the heterozygote (GC vs. CC) log OR (log ORGg) against the homozygote (GG vs. CC) log OR (log ORGG). The slope (λ) of the regression line drawn in the scatter plot provides an estimate of the genetic model of inheritance, with λ values equal to 0, 0.5 and 1 suggesting a recessive, co-dominant (additive) and dominant genetic model, respectively.[18]

Publication bias was first examined by visual inspection using a funnel plot and by assessing the significance of Egger's test and Begg–Mazumdar's rank correlation test.[19, 20] If publication bias was suspected, Duval and Tweedie's trim-and-fill analysis was used to estimate the number of studies omitted due to publication bias and to perform adjustments by symmetrical imputation of the omitted studies.[21] HWE was tested by means of a χ2 test. We performed a sensitivity analysis to examine the effect of excluding individual studies, as well as the effect of excluding studies in which genotype distribution among controls deviated significantly (P < 0.05) from HWE.[20] The meta-analysis was performed by using the computer software packages RevMan 5.0[22] and Comprehensive Meta-analysis (Biostat, Englewood, NJ, USA).

Results

Study identification and selection

The initial search identified 339 references for possible inclusion in our meta-analysis. Figure 1 presents a flow chart of retrieved and excluded studies. After removing duplicates, 156 studies remained, and 140 of these were excluded after abstract examination. This step left 16 full-text references that were assessed for eligibility. Among these, four studies were excluded because they lacked a case and/or control group, according to our predefined criteria.[23-26] Two additional studies[27, 28] were excluded because we were unable to retrieve enough information regarding genotype count. The authors of two studies[10, 29] kindly provided additional data regarding genotype distribution. Finally, the article by Stickel et al.[9] was considered two separate studies, as the authors reported information from two different populations. Thus, a total of 11 studies from 10 references were included in this meta-analysis. These studies comprised a total of 3495 patients with ALD (2087 with ALC) and 5038 controls (4007 healthy subjects and 1031 alcoholics without ALD).

Figure 1.

Flow chart of the selection of studies for inclusion in the meta-analysis.

Study characteristics

Criteria for defining groups according to the presence or absence of ALD as well as genotype counts for each group are shown in Table 1. Only two studies reported liver biopsy results for all patients with suspected ALD.[11, 30] Regarding ethnicity, nine studies included Caucasian patients,[9, 11, 29-34] one study included North Indian patients,[35] and the study by Tian et al.[10] included Mestizo (mixed European and Native American ancestry) individuals from Mexico. Other demographical variables, such as country, female/male ratio and age, are shown in Supplementary Table 1.

Table 1. Selection criteria and genotype distribution of rs738409 polymorphism of PNPLA3 among studies included in a meta-analysis of rs738409 polymorphism and alcoholic liver disease
First author/yearCriteria for selection of cases and controls N rs738409 genotype distribution
CCCGGG
  1. ALD, alcoholic liver disease; ALC, alcoholic liver cirrhosis; DSM-IV, Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition; ALT, alanine transaminase; AST, aspartate transaminase.

  2. Stickel et al., 11a and Stickel et al., 11b stand for multicenter sample study and population-based sample study, respectively.

Tian et al., 2010[10]1. ALD group: patients with alcohol dependence (DSM-IV) with at least 5 years and 200 g/week of alcohol consumption with abnormal biochemistry indicative of liver disease913102364447
 1.1. ALC group: diagnosed by unequivocal clinical, biochemical, ultrasound and/or endoscopic findings (no histological confirmation) and exclusion of other causes47939179261
 1.2. ALD without ALC group: abnormal biochemistry and some clinical indications of chronic liver disease (i.e. jaundice, hepatomegaly) without biochemical, clinical or imaging findings indicative of cirrhosis43463185186
 2. Alcoholics without ALD group: patients with alcohol dependence (DSM-IV) with at least 5 years and 200 g/week of alcohol consumption and no biochemical or clinical features of chronic liver disease3035915589
Seth et al., 2010[34]1. ALC group36617317023
 2. Alcoholics without ALD group: heavy drinkers with normal liver blood tests182119603
Falleti et al., 2011[29]1. ALC group: confirmed by liver biopsy or diagnosed according to clinical signs, imaging features and laboratory findings.166437548
 2. Healthy controls: blood donors without clinical and/or laboratory evidence of liver disease42821817535
Nguyen-Khac et al., 2011[31]1. ALD group: patients with acute alcoholic hepatitis or ALC105464316
 1.1. Acute alcoholic hepatitis group: diagnosed according to alcohol intake >50 g/day for the previous 3 months, Maddrey score ≥32, and liver biopsy65272612
 1.2. ALC group: diagnosed either by liver histology or a set of clinical, biochemical, echographic and endoscopic features, in absence of acute alcoholic hepatitis or other causes of cirrhosis4019174
 2. Healthy controls: free of liver disease and without alcohol intake10572294
Nischalke et al., 2011[32]1. ALC group: alcoholics with average alcohol consumption >300 g/week with cirrhosis diagnosed either by liver biopsy, transient elastography (stiffness >15 kPa), or signs of portal hypertension160497833
 2. Healthy controls190112699
Stickel et al., 2011a[9]1. ALD group: alcoholics with present heavy alcohol consumption (>60 g/day for women and >80 g/day for men) for more than 10 years diagnosed according to liver biopsy or clinical, laboratory, sonographic and/or radiological findings60431923154
 1.1. ALC group: diagnosed by liver biopsy (fibrosis stage 4) or unequivocal clinical and laboratory evidence210909327
 1.2. ALD without ALC group39422913827
 1.2.1. Alcoholic liver damage group: liver steatosis on ultrasound and elevation of ALT2191138224
 1.2.2. Alcoholic fatty liver group: liver steatosis on ultrasound and normal liver enzyme levels175116563
 2. Alcoholics without ALD group: alcoholic patients with normal appearance of the liver on ultrasound and normal liver enzyme levels43926415322
Stickel et al., 2011b[9]1. ALD group: at-risk drinkers (alcohol consumption >20 g/day) with nonsevere liver disease2691539719
 1.1. Alcoholic liver damage group: sonographic steatosis and increased serum ALT levels2051078117
 1.2. Alcoholic fatty liver group: sonographic steatosis but normal serum ALT levels6446162
 2. Alcoholics without ALD group: at-risk drinkers with normal liver echogenicity and normal serum ALT levels10767382
 3. Healthy controls: abstainers or with alcohol consumption <10 g/day162104553
Trepo et al., 2011[11]1. ALD group: excessive alcohol intake (>30 g/day for males and >20 g/day for females), and abnormal ALT or AST or suspected ALC with ALD-suggestive histology and exclusion of other causes33014014842
 1.1. ALC group: diagnosed based on liver biopsy or both unequivocal clinical-biochemical data and compatible imaging findings26510013035
 1.2. ALD without ALC group6540187
 2. Healthy controls32818111829
Rosendahl et al., 2012[33]1. ALC group: patients consuming >80 g alcohol/day (males) or >60 g/day (females) for ≥10 years, diagnosed by liver biopsy (fibrosis stage 4) or unequivocal clinical and laboratory findings and exclusion of other causes135515925
 2. Healthy controls: blood donors1950113571897
Dutta et al., 2013[35]1. ALC group: patients consuming ≥80 g alcohol/day for ≥10 years, who were diagnosed by clinical, radiological and biochemical parameters, and exclusion of other causes60242115
 2. Healthy controls: blood donors10062299
Way et al., 2013[30]1. ALD group: patients with alcohol misuse for ≥25 years and biopsy-proven ALD of varying severity38720515329
 1.1. ALC group206968822
 1.2. ALD without ALC group181109657
 1.2.1. Alcoholic liver damage group: intermediate biopsy changes11471376
 1.2.2. Alcoholic fatty liver group: minimal steatosis6738281
 2. Healthy controls74444925738

All but two studies[30, 34] reported to have excluded patients with liver diseases from other causes. In three studies, cases and controls were matched by age and sex.[9, 31, 32] Three studies adjusted for confounders in statistical analysis.[10, 11, 32] Blind genotyping was only reported in the article by Stickel et al.,[9] and five studies reported no deviation from HWE.[10, 11, 29, 32, 35] After our analysis, we found no significant deviation from HWE in any of the control groups of the studies used for meta-analysis. Regarding genotyping methodology, five studies used Taqman assays,[9, 11, 31, 33] two studies used KASPar assays,[30, 34] two studies used restriction fragment length polymorphism assays,[29, 35] one study used microarray genotyping,[10] and one used LightSNiP.[32]

Association between rs738409 and ALC

After meta-analysis, the G allele of rs738409 occurred more frequently in individuals with ALC (Table 2). This result was found when comparing GG vs. CC genotypes (OR 4.30, 95% CI 3.25–5.69; P < 0.00001, Figure 2a), GC vs. CC genotypes (OR 1.91, 95% CI 1.67–2.17; P < 0.00001, Figure 2b), and for a dominant (OR 2.25, 95% CI 1.99–2.55; P < 0.00001) or a recessive model (OR 3.10, 95% CI 2.43–3.94; P < 0.0001). These results and the slope of the regression line (λ = 0.44) after plotting logORGg against logORGG suggested an additive genetic model for the G allele (Figure S1A).

Table 2. Summary of meta-analysis results for the association of rs738409 of PNPLA3 with alcoholic liver cirrhosis
 OR95% CIP (random-effects model) I 2 P (heterogeneity)
  1. OR, odds ratio; CI, confidence interval; ALC, alcoholic liver cirrhosis; ALD, alcoholic liver disease; N, number of studies included in each meta-analysis; n, number of individuals in each group.

Patients with ALC (n = 2087) vs. controls (n = 4769) (N = 10)
GG vs. CC4.303.25–5.69<0.0000142%0.08
GC vs. CC1.911.67–2.17<0.000010%0.93
Dominant (GG + GC vs. CC)2.251.99–2.55<0.000010%0.44
Recessive (GG vs. CC +CG)3.102.43–3.94<0.0000139%0.10
Patients with ALC (n = 1055) vs. alcoholics without ALD (n = 924) (N = 3)
GG vs. CC4.192.93–5.99<0.000010%0.81
GC vs. CC1.831.46–2.29<0.000010%0.92
Dominant (GG + GC vs. CC)2.201.78–2.73<0.000010%0.53
Recessive (GG vs. CC +CG)2.912.23–3.79<0.000010%0.87
Patients with ALC (n = 1032) vs. healthy subjects (n = 3845) (N = 7)
GG vs. CC4.372.89–6.62<0.0000160%0.02
GC vs. CC1.951.66–2.29<0.000010%0.77
Dominant (GG + GC vs. CC)2.311.94–2.75<0.0000121%0.27
Recessive (GG vs. CC +CG)3.182.19–4.63<0.0000158%0.03
Figure 2.

Meta-analysis of the association of the rs738409 PNPLA3 polymorphism with ALC. Patients with ALC as cases are compared with controls (healthy subjects or alcoholics without ALD). (a) GG vs. CC genotypes. Test for overall effect: Z = 10.19 (P < 0.00001). Test for heterogeneity: χ2 = 15.61 (P = 0.08), I2 = 42%. (b) GC vs. CC genotypes. Test for overall effect: Z = 9.62 (P < 0.00001). Test for heterogeneity: χ2 = 3.63 (P = 0.93), I2 = 0%. Each study is shown by an OR estimate with the corresponding 95% CI. ALC, alcoholic liver disease.

The comparison of GG vs. CC homozygotes and the recessive model showed moderate heterogeneity (I2 values of 39% and 42%, respectively) whereas the comparison of GC vs. CC genotypes and the dominant model produced no heterogeneity (I2 = 0). Sensitivity analysis excluding individual studies yielded the same pattern of results. As shown in Table 2, these results were consistent when comparing patients with ALC vs. alcoholics without liver disease (Figure 3) and patients with ALC vs. healthy controls. Of note, the meta-analysis comparing patients with ALC vs. alcoholics without liver disease showed no heterogeneity (I2 = 0) for all comparisons.

Figure 3.

Meta-analysis of the association of the rs738409 PNPLA3 polymorphism with ALC. Patients with ALC as cases are compared with alcoholics without liver disease. (a) GG vs. CC genotypes. Test for overall effect: Z = 7.87 (P < 0.00001). Test for heterogeneity: χ2 = 0.43 (P = 0.81), I2 = 0%. (b) GC vs. CC genotypes. Test for overall effect: Z = 5.30 (P < 0.00001). Test for heterogeneity: χ2 = 0.17 (P = 0.92), I2 = 0%. Each study is shown by an OR estimate with the corresponding 95% CI. ALC, alcoholic liver disease.

The inclusion of data from Caucasians (excluding studies by Tian et al.,[10] and Dutta et al.[35]) showed significant results for all comparisons and did not reduce heterogeneity (data not shown). No evidence of publication bias was detected after funnel plot inspection or by Egger's or Begg–Mazumdar's test (Figure S2).

Other results

Regarding the association between this SNP and ALD, the results of these meta-analyses are shown in Table 3. The GG genotype of rs738409 occurred more frequently than the CC genotype among individuals with ALD (OR 3.68, 95% CI 2.56–5.29; P < 0.00001) and the GC genotype was also more frequent than the CC genotype among individuals with ALD (OR 1.62, 95% CI 1.38–1.89; < 0.00001). The G allele (GG or GC genotype) also occurred more frequently in individuals with ALD under a dominant or recessive model. However, a moderate to large level of heterogeneity was found, with I2 values ranging from 47% to 71%. The exclusion of individual studies or restricting the analysis to Caucasians did not change these significant results, nor did it reduce heterogeneity (data not shown).

Table 3. Summary of meta-analysis results for the association of rs738409 of PNPLA3 with alcoholic liver disease
 OR95% CIP (random-effects model) I 2 P (heterogeneity)
  1. OR, odds ratio; CI, confidence interval; ALD, alcoholic liver disease; N, number of studies included in each meta-analysis; n, number of individuals in each group.

Patients with ALD (n = 3495) vs. controls (n = 5038) (N = 11)
GG vs. CC3.682.56–5.29<0.0000171%0.0002
GC vs. CC1.621.38–1.89<0.0000146%0.05
Dominant (GG + GC vs. CC)1.951.60–2.37<0.0000169%0.0004
Recessive (GG vs. CC +CG)2.842.11–3.81<0.0000163%0.003
Patients with ALD (n = 2152) vs. alcoholics without ALD (n = 1031) (N = 4)
GG vs. CC2.722.02–3.66<0.000010%0.43
GC vs. CC1.391.11–1.730.00434%0.21
Dominant (GG + GC vs. CC)1.621.27–2.070.0000150%0.11
Recessive (GG vs. CC +CG)2.281.80–2.89<0.000010%0.57
Patients with ALD (n = 1612) vs. healthy subjects (n = 4007) (N = 8)
GG vs. CC4.142.52–6.83<0.0000176%0.0002
GC vs. CC1.721.42–2.09<0.0000143%0.09
Dominant (GG + GC vs. CC)2.081.61–2.70<0.0000172%0.0009
Recessive (GG vs. CC + CG)3.142.06–4.78P < 0.0000169%0.002

As shown in Table 1, the G allele of rs738409 occurred significantly more frequently among individuals with ALC compared to patients with a milder form of ALD (excluding patients with alcoholic hepatitis). Specifically, the GG genotype was more frequent in individuals with ALC compared to the CC genotype (OR 2.43; 95% CI 1.79–2.31; P < 0.0001), and the CC genotype occurred more frequently in individuals with ALD without a diagnosis of ALC or alcoholic hepatitis. The frequency of the G allele in ALD individuals without ALC or alcoholic hepatitis was greater than that in healthy controls, but was not statistically significant in all comparisons. Only one study[31] analysed the association of rs738409 variant with alcoholic hepatitis and therefore we were unable to perform a meta-analysis.

Discussion

The results of our meta-analysis showed that the G allele of the PNPLA3 rs738409 G>C SNP is associated with ALC. Individuals with the GG genotype had approximately a four-fold risk of having ALC when compared to individuals with the CC genotype. The GC heterozygous genotype was also associated with a smaller, but still significant, risk. This result was consistent when we performed separate comparisons for studies including either alcoholics without liver disease or healthy subjects as controls, which reinforces the strength of the association of the G allele with ALC. Regarding other forms of ALD, although a significant association with the G allele was also found for many comparisons, a significant amount of heterogeneity was observed that precluded a reliable interpretation. It is also of interest that our data fit an additive genetic model for the G allele, which have not been previously reported in this disease. Therefore, the rs738409 SNP seems to be associated with ALC, at least in Caucasians. Although we have not analysed the relationship of this SNP with HCC associated to ALD, a recent meta-analysis has also found a significant association between HCC in patients with ALD and this polymorphism.[14]

We must acknowledge, however, that a potential limitation of our study is the large heterogeneity found in several comparisons. Heterogeneity in meta-analysis may result from phenotype misclassification, differences in genotyping methods or in baseline characteristics of participants (e.g. age, sex), interaction with confounding factors, ethnic diversity or population stratification.[36] In our meta-analysis, much of the heterogeneity found was likely due to the use of different criteria for cases and controls across studies. This is highlighted by the fact that meta-analyses using ALD as cases had more heterogeneity than those using ALC as cases. Furthermore, comparisons using alcoholics without ALD as controls had low or no heterogeneity, while comparisons including healthy subjects as controls had a higher degree of heterogeneity. Indeed, the inclusion of alcoholics without ALD as controls is preferable because it permits the study, in isolation, of the direct association between genetic variants and the risk of liver disease on a background of alcoholism.[37] This approach has been used in previous studies in this field[5, 6] but has the risk of phenotype misclassification if liver biopsy is not performed. We also explored ethnicity as a source of heterogeneity, but restricting analysis to Caucasians did not reduce I2 values. Finally, we were not able to explore potential interactions with variables such as body mass index, age or sex, due to the lack of information available.

Publication bias is also a matter of concern in meta-analysis. We searched in multiple databases for reports, including abstracts presented at conferences, in an attempt to minimise this risk. Further, we have assessed publication bias according to several methods. The findings indicated that this bias was unlikely to have a relevant effect on our results.

Although it is true that conclusions from our study may not be definite due to heterogeneity of cases and controls, our results are similar across different subgroups and comparisons, particularly for ALC. Furthermore, our data are consistent with previous meta-analyses of associations between this SNP and NAFLD, HCC or liver fibrosis.[14, 15, 38] Together, these results strongly suggest that this variant is a common risk factor for advanced liver disease, including ALC, and highlight the potent effect of this SNP on a broad spectrum of liver diseases. However, the effect of this polymorphism on protein function is not yet understood. PNPLA3, also referred to as adiponutrin, is a 481-amino acid protein of the calcium-independent phospholipase A2 family, mainly expressed in human liver. It was initially described as having a role in fat storage. Indeed, the I148M substitution is a loss-of-function mutation and increases cellular triglyceride content in the liver by limiting triglyceride hydrolysis. However, there are also data from basic and clinical studies suggesting that this mutation has an independent effect on inflammation and fibrogenesis.[4, 39, 40] Unravelling the pathogenic role of this variant will give insight into disease mechanisms of several liver disorders, including ALD and could reveal potential therapeutic targets.

In conclusion, the main result of this meta-analysis is that the G allele of the PNPLA3 rs738409 G>C polymorphism is associated with ALC among Caucasians. Despite the potential limitations of genetic studies in this field, this is the allelic variant most clearly associated with ALC susceptibility. Future studies addressing its potential role as a marker to guide clinical decisions are warranted.

Authorship

Guarantor of the article: M. Marcos.

Author contributions: AJC and MM designed the study protocol. AJC, MM and JLT managed the literature searches and performed the statistical analysis. All authors contributed to interpretation of data. AJC, JLT and MM drafted the manuscript. RGS, FJL and JAM provided critical revision of the draft for important intellectual content. All authors critically reviewed content and approved final version for publication.

Acknowledgements

Declaration of personal interests: None.

The authors are grateful to Drs. David Hinds, Pierluigi Toniutto and Edmondo Falleti for providing additional data regarding their studies.

Declaration of funding interests: This study was funded in part by the Spanish Ministry of Science and Innovation, Instituto de Salud Carlos III and the European Union FEDER funds, Una manera de hacer Europa (grant numbers PI10/01692 and I3SNS-INT12/049 to M.M.) the Red de Trastornos Adictivos-RTA (grant number RD12/0028/0008 to F.-J.L.) and the Junta de Castilla y León, Spain (grant number GRS 531/A/10 to M.M.).

Ancillary