A comprehensive literature search of related studies was conducted using the PubMed, Embase database, and Web of Science (published on or before April 15, 2023), with the following keywords: “liver inflammatory disease OR autoimmune hepatitis OR autoimmune liver disease” AND “Genetic OR SNP OR polymorphism OR genotype OR variant OR allele OR mutation OR insertion OR deletion OR copy number OR genome-wide association study OR GWAS”. The title, abstract, or full text of the studies were reviewed as necessary to identify all relevant articles. In addition, reference lists of all the included studies, reviews, and meta-analyses were manually screened for any additional potential studies.

Inclusion criteria: 1) original articles published in English; 2) observational studies; 3) investigating associations between genetic variants and risk of AIH; 4) providing risk estimates [including odds ratio (OR) and relative risk (RR)] and 95% confidence intervals (CIs) or data to calculate them. Exclusion criteria: 1) participants complicated with other liver diseases; 2) reviews, abstracts, case reports, and letters. Two investigators (DG and YW) independently assessed the eligibility of each publication, and disagreement was discussed with the principal author (MZ).

Two authors (DG and MZ) independently extracted the data using a predesigned collection sheet including PMID, first author, publishing year, study design, sample size of cases and controls, source of population, ethnicity, variants, gene, major and minor alleles, genotype and allele counts, risk estimates and corresponding 95% CIs, or P value (for studies using multiple adjusted models, the most fully adjusted estimates were extracted).

Meta-analyses were performed to calculate the pooled OR and 95% CIs under an additive genetic model. The statistical heterogeneity among the studies was assessed using the Cochran Q statistic (P < 0.10 was considered statistically significant) and I² statistic (I² ≤ 25% represented mild heterogeneity, 25% - 50% represented moderate heterogeneity, and ≥ 50% represented large heterogeneity) [11]. A random-effects model was used when I² ≥ 50%, while a fixed-effects model was used when I² < 50%. For variants associated with AIH, sensitivity analyses were conducted by excluding the initial published or initial positive study. Furthermore, we assessed potential publication bias using Begg’s test [12] and small-study bias using Egger’s test [13].

Associations with P < 0.05 in the primary meta-analyses were evaluated using the Venice criteria to assess the epidemiological credibility, and the detailed methods were described in our previous research [14]. Finally, epidemiological credibility was categorized as strong, moderate, or weak, based on the grade level of A, B, or C in three criteria: the amount of evidence, replication, and protection from bias. Furthermore, FPRP was calculated for these associations [15]. Specifically, FPRP values < 0.05, 0.20 - 0.05, and >0.20 were considered as strong, moderate, and weak evidence of a true association, respectively. We would upgrade the cumulative evidence if the FPRP result was strong, whereas downgrade the cumulative evidence if the FPRP result was weak.

To provide biological insights into the significant variants identified by meta-analysis, we mapped these SNPs to genes and conducted functional annotation with the Encyclopedia of DNA Elements (ENCODE) tool HaploReg v4.1 [16]. To identify tissues most relevant to the significant genes, GTEx tissue enrichment analysis was conducted based on 54 tissue types available from GTEx (version 8) through functional mapping and annotation of genome-wide association studies (FUMA) GENE2FUNC process [17]. In addition, we assessed the enrichment of the significant mapped genes in Gene Ontology (GO) biological processes using the WebGestalt tool [18]. We adopted the Benjamin-Hochberg procedure to correct for multiple testing and considered a false discovery rate (FDR) corrected P < 0.05 as a statistical difference.

Statistical analysis was conducted using Stata version 15 (StataCorp, College Station, TX), and a two-tailed P < 0.05 was considered statistically different unless otherwise specified.

In total, we screened 1213 publications after duplicates excluded in the literature search, and the selection process is presented in Figure 1. After screening, a total of 62 publications involving 11,068 cases and 45,482 controls were included for quantitative analysis. Among these articles, 57 were candidate-gene association studies (including 56 case-control studies and 1 cohort study), and 5 were GWASs. Thirty-three articles investigated the association between human leukocyte antigen (HLA) genes and the risk of AIH, while 34 articles examined the relationship between non-HLA genes and the risk of AIH.

In total, we performed meta-analyses to investigate the association between 75 variants in 11 genes or loci and the risk of AIH. Among these variants, 17 in the HLA gene and 7 in non-HLA genes were associated with AIH (Table 1 and Table 2). However, no association was found between 51 variants and AIH.

Seventeen variants in the HLA gene were found to be associated with the risk of AIH. Among them, 7 associations were graded as strong, 4 were graded as moderate, and 6 were graded as weak (Table 1). In particular, strong evidence suggested that rs2187668 (OR: 2.695, 95% CI: 2.418 - 3.004, P = 1.10 × 10⁻⁷¹), DRB1*0405 (OR: 3.223, 95% CI: 2.602 - 3.993, P = 8.20 × 10⁻²⁷), DRB1*0301 (OR: 3.023, 95% CI: 2.443 - 1.678, P = 2.81 × 10⁻²⁴), DRB3*0101 (OR: 3.667, 95% CI: 2.649 - 5.075, P = 4.69 × 10⁻¹⁵), and DQB1*0201 (OR: 2.554, 95% CI: 1.863 - 3.503, P = 5.88 × 10⁻⁹) were associated with AIH at genome-wide significance level (P < 5.0 × 10⁻⁸). Importantly, DRB1*0301 and DRB3*0101 were the newly identified genome-wide significant risk loci (Table 1).