Plasma liver enzyme tests are broadly applied to predict biochemical liver functions and to diagnose liver diseases. Alanine transaminase (ALT) and aspartate transaminase (AST) are closely associated with hepatocyte damage and steatosis, including non-alcoholic fatty liver disease (NAFLD) [1]. Elevated liver enzyme levels, as prospective risk factors, have been reported to be involved in type 2 diabetes and coronary heart disease [2, 3]. However, alkaline phosphatase and gamma-glutamyltranspeptidase are related with biliary, cholestatic, and alcoholic diseases [4].

Overweightedness in childhood or adolescence is persistent in adulthood obesity and is closely associated with adverse health effects, including metabolic diseases [5]. A recent cohort study in Korean adolescents has shown strong associations of liver enzyme levels with a cluster of metabolic syndrome components [6]. Thus, finding genetic markers of liver enzyme levels in childhood would be important for early prediction of liver function and its related disease in adulthood.

Liver enzyme levels are affected by environmental and genetic factors. The heritabilities of liver enzyme levels range from approximately 35% to 61% [7, 8]. To date, two genome-wide association studies (GWASs) in European ancestry populations have identified genetic variants located in multiple genes associated with levels of ALT (HSD17B13-MAPK10, TRIB1, CHUK, and SAMM50), AST (NBPF3-ALPL), and both of them (CPN1 and PNPLA3) [9, 10]. In addition, a meta-analysis identified pleiotropic effects between NAFLD and metabolic traits, such as low-density lipoprotein, high-density lipoprotein, triglyceride, fasting glucose, and insulin resistance [11].

Despite the identification of novel genes or loci associated with liver enzymes in cohort populations, the common variants in childhood are still largely unknown. Here, we present findings from a GWAS in childhood subjects to identify new susceptibility loci for liver enzyme levels.

A total of 747,076 SNP genotype assays were tested on the DNA samples of 484 Korean childhood subjects. ALT and AST levels of the subjects according to age and sex are presented in Supplementary Table 1. Linear regression analyses were performed to estimate the p-values for the association between genotype distributions and levels of ALT and AST. Manhattan plots of association analyses of all SNPs with respect to ALT and AST are presented with the chromosomal positions (X-axis) and negative logarithm of p-values (Y-axis) of each SNP (Fig. 1). Quantile-quantile plots comparing the association p-values with those expected for a null distribution are also presented (Fig. 2). The highest-ranked SNPs showing the lowest p-values (p < 1.0 × 10^-5) of association with ALT and AST levels are listed in Table 1. In addition, genotype clusters of the SNPs are presented in Supplementary Fig. 1.

Although there was no SNP satisfying the criteria of genome-wide significance (p < 1.0 × 10^-8), we identified three loci (rs4949718, rs80311637, and rs596406 in ST6GALNAC3, ADAMTS9, and CELF2, respectively) that were suggestively associated with levels of ALT and AST. Of these SNPs, rs4949718 was found to be more strongly associated with ALT levels (p = 1.87 × 10^-7) but less in AST levels (p = 1.49 × 10^-6). The missense rs80311637 (V1597M) and intronic rs596406 SNPs were more strongly associated with an increased risk of AST than ALT.

The levels of ALT and AST, known as transaminases, are important indicators for determining hepatocyte injury or hepatocellular diseases. However, ALT is known to be found in the highest level in the liver; thus, ALT levels are more liver-specific indicators [4]. Moreover, it has been reported that ALT levels are more significantly associated with a cluster of metabolic syndrome than AST levels in a Korean adolescent cohort study [6].

In this study, we identified three novel genes (ST6GALNAC3, ADAMTS9, and CELF2) that were multiply associated with elevated levels of ALT and AST. Of the three loci, rs4949718 in the intron region of ST6GALNAC3 yielded the lowest p-value for the association with ALT, compared to AST levels (Table 1). In addition, rs80311637 (ADAMTS9) and rs596406 (CELF2) were shown to be more strongly associated with AST levels, compared to ALT. These results indicate that the liver-specific feature of ALT was not driven as the secondary association for AST. In recent, a previous GWAS in an Amish population identified two genetic variants (rs17209512 and deletion) in glutamic-oxaloacetic transaminase 1 (GOT1) that were associated with serum AST. However, the rs17209512 SNP, which is in incomplete LD with rs76850691, was not replicated in our study (Table 1).

For the three multiply associated loci, we performed conditional analyses to test whether the association with the other trait was simply secondary to the primary association (Supplementary Table 2). In these analyses, the association signals on rs4949718 for AST and rs80311637 and rs596406 for ALT were abolished by adjustment of the ALT and AST, respectively. It is assumed that the multiple effects of these loci on ALT and AST resulted from the secondary effects by physiologically correlated features between ALT and AST [12]. In addition, when comparing with BMI-unadjusted data, the association signals in ALT were relatively weaker than those of the BMI-adjusted data. However, there was no significant difference in p-values before and after adjusting for BMI in AST (data not shown).

ST6GALNAC3 encodes a type of sialyltransferase that plays a role in the transfer of sialic acids from cytidine 5' monophospho-N-acetylneuraminic acid (CMP-NeuAc) to the carbohydrate termini of glycoproteins and glycolipids [13]. Previous GWASs have revealed the associations of ST6GALNAC3 SNPs with human electroencephalogram and nicotine dependence [14, 15]; however, there has been no report on an association with liver enzyme levels. Although the association with liver enzymes is still unknown, ST6GALNAC3 has been shown to be associated with subscapular skin-fold thickness, a type of obesity-related trait, in a Korean adult population [16]. Obesity affects elevated liver enzyme levels by inducing inflammatory responses in the liver, accompanied by increases in C-reactive protein [17]. The involvements between obesity and liver enzyme levels have been found in cohort studies [18-20]. Therefore, the association of ST6GALNAC3 with obesity may influence liver enzyme levels indirectly. In addition, ST6GALNAC3 protein interacts with α2-Heremans-Schmid glycoprotein (AHSG), which is also known as fetuin-A, by enzyme-substrate specificity [13]. AHSG is abundantly distributed in serum protein and is produced mainly in the liver, and elevated levels of this protein are associated with the risk of NAFLD [21]. Considering that liver enzyme levels are closely involved indicators in NAFLD [22, 23], the interaction with AHSG might induce an association between ST6GALNAC3 and liver enzyme levels.

ADAMTS9 is a metalloproteinase that plays a role in cleavage of proteoglycans and is expressed in skeletal muscle and pancreas [24]. ADAMTS9 has been reported as a type 2 diabetes susceptibility region in a GWAS [24]. This gene has also shown an association with insulin resistance and impaired beta-cell function in the pancreas [25]. Taken together, the association of ADAMTS9 with liver enzymes might be due to its susceptibility to diabetes and insulin resistance.

CELF2 (also known as CUGBP2) induces the skipping of exon 11 of insulin receptor (IR). The mRNA expression level of exon-skipped IR is significantly lower in liver than in other organs in mouse models [26]. It has been reported that a decrease in the number of IRs on target cells can cause insulin resistance [27]. Therefore, it is assumed that CELF2 affects liver enzyme levels by lowering expression of IR in liver that induces insulin resistance.

In summary, our data demonstrate new multiple associations with ALT and AST levels in a Korean childhood cohort. Further replication and functional studies remain to be done to clarify the genetic factors of liver dysfunction.