Despite the high prevalence of nonalcoholic fatty liver disease (NAFLD), little is known of its pathogenesis based on study of human liver samples. By the use of Affymetrix GeneChips (17,601 genes), we investigated gene expression in the human liver of subjects with extreme steatosis due to NAFLD without histological signs of inflammation (liver fat 66.0 ± 6.8%) and in subjects with low liver fat content (6.4 ± 2.7%). The data were analyzed by using sequence-based reannotation of Affymetrix probes and a robust model-based normalization method. We identified genes involved in hepatic glucose and lipid metabolism, insulin signaling, inflammation, coagulation, and cell adhesion to be significantly associated with liver fat content. In addition, genes involved in ceramide signaling (MAP2K4) and metabolism (UGCG) were found to be positively associated with liver fat content. Genes involved in lipid metabolism (PLIN, ACADM), fatty acid transport (FABP4, CD36), amino acid catabolism (BCAT1), and inflammation (CCL2) were validated by real-time PCR and were found to be upregulated in subjects with high liver fat content. The data show that multiple changes in gene expression characterize simple steatosis.
- liver fat
- nonalcoholic fatty liver disease
- lipid metabolism
nonalcoholic fatty liver disease (NAFLD) covers a spectrum of liver disease ranging from simple steatosis (NAFL) to nonalcoholic steatohepatitis (NASH) (44). Liver fat content is increased in subjects with insulin resistance and the metabolic syndrome compared with subjects without the syndrome independent of age, sex, and body mass index (BMI) (34). An insulin-resistant fatty liver overproduces glucose (38, 46, 51), VLDL (2, 3), coagulation factors (12), and inflammatory markers (35). The pathogenesis of hepatic fat accumulation is poorly understood. This is partly because the technical and ethical limitations associated with study of human liver tissue.
Only a few studies have investigated gene expression in the human liver by high-throughput technologies (4, 10, 32, 43, 45, 61, 62). In all except two (10, 61) of these studies, control subjects were compared with patients with NASH (4, 45, 61, 62) or with NAFLD without subdivision to NAFL and NASH (32, 43). In the two studies comparing subjects with steatosis to those without (10, 61), the groups did not differ with respect to features of insulin resistance (10, 61). In the study of Chiappini et al. (10), the etiology of steatosis was not reported. In addition, the microarray results might not be fully reliable because of problems related to Affymetrix probe design and annotation (13, 48). Since insulin resistance is the most common cause of steatosis (34) and may indeed contribute to the pathogenesis of NASH (1, 16, 21, 39), it would be of interest to compare insulin-resistant subjects with severe steatosis to those with no or less severe insulin resistance and no or mild steatosis. For this purpose, we studied 30 consecutive patients from whom a liver biopsy was taken (58). We then chose subjects with as low as possible liver fat content and patients with as high as possible degree of steatosis due to NAFLD without any histological signs of inflammation or fibrosis to determine gene expression by Affymetrix GeneChips.
Subjects and design.
The subjects represented low (6.4 ± 2.7%) and high (66.0 ± 6.8%) extremes of liver fat content of a group of previously described subjects undergoing laparoscopic gastric bypass surgery or who were referred to the gastroenterologist because of elevated liver function tests, and from whose remaining liver samples sufficient amounts of high-quality RNA could be prepared (58). In brief, these patients fulfilled the following inclusion and exclusion criteria: 1) age 18–60 yr; 2) alcohol consumption less than 2 drinks/day; 3) no histological evidence of inflammation or fibrosis; and 4) no evidence of hepatitis B or C, thyroid dysfunction, autoimmune hepatitis (smooth muscle and antinuclear antibodies), primary biliary cirrhosis (antimitochondrial antibodies), primary sclerosing cholangitis, α1-antitrypsin deficiency, use of hepatotoxic medications or herbal products, or use of medications known to be associated with steatohepatitis. Fat content of the liver biopsy specimens (% of fat-laden hepatocytes) was analyzed by an experienced liver pathologist in a blinded fashion.
The nature and potential risks of the study were explained to all subjects prior to obtaining their written, informed consent. The study was carried out in accordance with the principles of the Declaration of Helsinki. The protocol was approved by the ethics committee of the Helsinki University Central Hospital.
The subjects were studied after an overnight fast. A blood sample was taken for measurements of plasma glucose, serum insulin, C-peptide, free fatty acids (FFA), triglyceride, HDL cholesterol, adiponectin, and serum alanine aminotransferase (S-ALT) concentrations as previously described (34). Serum adiponectin concentrations were measured by using the B-Bridge International ELISA kit (San Jose, CA). Serum FFA concentrations were measured by an enzymatic colorimetric method (Wako Chemicals, Neuss, Germany). Body weight was recorded to the nearest 0.1 kg by use of a calibrated weighting scale (Soehnle, Monilaite-Dayton, Finland) with subjects barefoot and wearing light indoor clothing. Body height was recorded to the nearest 0.5 centimeter by use of a ruler attached to the scale. Waist circumference was measured midway between spina iliaca superior and the lower rib margin and hip circumference at the level of the greater trochanters (40). After these measurements, a liver biopsy was taken as previously described (58). Approximately half of each sample was used for routine histopathological examination, and the rest was immediately frozen and stored in liquid nitrogen.
Microarray sample preparation, hybridization, and scanning.
Frozen liver tissue (1–13 mg) was homogenized in RNeasy lysis buffer, and total RNA was isolated and purified after DNase treatment by using the RNeasy Micro or RNease Mini Kit (Qiagen). RNA concentrations were measured by use of the RiboGreen fluorescent nucleic acid stain (RNA quantification kit; Molecular Probes, Eugene, OR). The quality of RNA was checked by an Agilent Bioanalyzer. Average yields of total RNA were 13 μg/10 mg liver tissue weight. The isolated RNA was stored at −80°C until quantification of the target mRNAs. A total of 1 μg RNA was transcribed into cDNA by use of Moloney murine leukemia virus reverse transcriptase (Life Technologies, Paisley, UK) and oligo(dT)12–18 primers (13). Microarray hybridization and scanning were performed following recommendations of the manufacturer. Briefly, Affymetrix GeneChip HGU133plus2 microarrays (Affymetrix, Santa Clara, CA) were loaded with the fragmented target sample buffer mix. Each sample was individually hybridized for 18 h at 45°C with vigorous mixing. All of the washing and staining steps were performed by using the Affymetrix FS450 fluidics station. The arrays were then scanned via GeneChip Scanner 3000 7G (Affymetrix).
Quantitative PCR validations.
The mRNA concentrations of selected genes were quantified by real-time PCR using the ABI 7000 sequence detection system instrument and software (Applied Biosystems, Stockholm, Sweden). cDNA synthesized from 10 ng of total RNA was mixed with TaqMan Universal PCR master mix (Applied Biosystems) and a gene-specific primer and probe mixture (predeveloped TaqMan gene expression assays, Applied Biosystems) in a final volume of 25 μl. All samples were run in duplicate. Relative expression levels were determined by using a seven-point serially diluted standard curve, generated from cDNA of the human liver. The mRNA expression levels of specific genes were expressed in arbitrary units and normalized to the mean of the mRNA expression levels of RPLP0 and TATA-binding protein to correct for differences in cDNA loading.
Computational analysis of Affymetrix GeneChips.
Reannotation of the probes was done according to the latest release of the Entrez gene database (13). In the original setting, the Affymetrix HGU133 plus2 chip set contains 54,675 probe sets, whereas after reannotation they are grouped in 17,663 probe sets, representing 17,601 unique Entrez gene IDs and 62 quality control probe sets. A meta-annotation package for UNIX R environment was created de novo by using BioConductor facilities (http://bioconductor.org/packages/2.0/bioc/html/AnnBuilder.html) to link the Entrez gene IDs to the major biological databases. The meta-annotation package is available from one of the authors (D. Greco) upon request.
Preprocessing of raw data (CEL files) was performed by using the software R (http://www.r-project.org) and the package BioConductor (http://www.bioconductor.org). The expression values for each gene were calculated by using the robust multiarray average (5, 27) implemented in the package Affy (26). The effect of liver fat content and the possible effect of sampling technique on gene expression were analyzed by linear models and empirical Bayes methods (52). Genes presenting a P value <0.05 for high- vs. low-liver fat groups were considered for further analysis. The DAVID gene annotation system (24) was used to select overrepresented Gene Ontology categories (19) and KEGG pathways (30) in the significant genes compared with all genes represented on the reannotated chip set. Default statistical parameters were employed.
The subjects with high liver fat compared with low liver fat were matched for age and sex but had higher BMI and waist circumference (Table 1). The high-liver fat group also had higher fS-insulin and fS-C-peptide and lower fS-HDL concentrations than the low-liver fat group.
Associations between liver fat content and gene expression in the liver.
A total of 1,060 genes were significantly associated with liver fat content. Of these genes, 419 were positively and 641 negatively correlated with liver fat (Supplementary Table S1). The fold changes, calculated as the ratios of gene expression between the subjects with the most and the least liver fat content, varied between +5.39 (FABP4) and −2.33 (DIP). The genes significantly associated with liver fat covered a wide range of biological functions and could be annotated in 147 overrepresented gene ontology families and metabolic pathways (Supplementary Table S2). They included 41 genes involved in carbohydrate metabolism, 34 genes in lipid metabolism, and 12 genes in amino acid metabolism. In addition, 10 genes were annotated in the insulin signaling pathway, 14 genes in inflammation, and 19 in the MAPK signaling pathway. When the significant genes were clustered according to the cellular component database, 24 genes appeared to be associated with extracellular matrix and 23 with mitochondria.
Quantitative PCR validations were carried out on six genes of interest involved in lipid metabolism (FABP4, CD36, PLIN, and ACADM), amino acid metabolism (BCAT1), and inflammation (CCL2). The PCR-measured expression levels correlated positively with the liver fat content (Fig. 1).
We have previously reported associations between hepatic gene expression, quantitated by real-time PCR, and histologically determined liver fat content in a total of 24 subjects (58). Of these subjects (58), two subgroups with either low or extremely high liver fat content were chosen for the present study to investigate changes in hepatic gene expression by using microarray analysis. The group with low liver fat content includes some patients who have more than what is defined as normal (5–10%; Ref. 44), but this is not unexpected given that all patients had some abnormality such as elevated S-ALT that made the liver biopsy clinically justified. Histologically, the patients had no histologically detectable inflammation or fibrosis. The study is novel as the Affymetrix probes were reannotated and model-based preprocessing methodology was employed. This is important because a large number of Affymetrix probes (30–50%) are misannotated and prone to cross-hybridization (13). This was not taken into account in the previous studies in which patients with NAFLD (10) or NASH (45) were compared with normal subjects. We found genes involved in hepatic glucose and lipid metabolism, insulin signaling, as well as inflammation and cell adhesion to be significantly associated with liver fat content. Increased expression of FABP4, CD36, perilipin, ACADM, BCAT1, and CCL2 in the high liver fat group was verified by PCR.
Perilipin coats lipid droplets and protects triglycerides from the lipolytic action of hormone-sensitive lipase (37). We found the expression of perilipin to increase and that of hormone-sensitive lipase to decrease with increasing liver fat content. The transport of fatty acids into the liver occurs via fatty acid binding proteins (FABPs) (11) and fatty acid transporters such as fatty acid translocase CD36 (8, 55). In obese mice, expression of FABPs has been shown to be increased in the liver and adipose tissue (42), and specific induction of CD36 leads to hepatomegaly and a fatty liver (36). As shown in Table 2 and Fig. 1, expressions of FABP4 and CD36 were positively related to liver fat content in the present study. Decreased lipolysis and increased fatty acid uptake may thus contribute to hepatic fat accumulation in humans.
Transgenic mice with deletion of key enzymes involved in the regulation of fatty acid oxidation develop hepatic steatosis (17, 31). However, studies in humans have suggested hepatic β-oxidation, when measured indirectly by using serum β- hydroxybutyrate concentrations, to be increased rather than decreased in subjects with a fatty liver compared with healthy controls (7, 49). In the present study, expression of the enzyme catalyzing the initial step of β-oxidation of fatty acids [medium chain acetyl-CoA dehydrogenase (ACADM)] (41), was increased when determined by both Affymetrix analyses and PCR in subjects with a fatty liver.
Hepatic lipase controls hepatic lipoprotein metabolism by catalyzing hydrolysis of triglycerides and phospholipids of all major classes of lipoproteins (14, 29). In humans, hepatic lipase activity has been shown to be positively correlated with intra-abdominal fat content (9). In a meta-analysis comprising 25 publications and 24,000 individuals, low activity of hepatic lipase was associated with an increase in serum HDL cholesterol concentrations (28). The positive correlation between liver fat and hepatic lipase may thus provide one molecular link between features of the metabolic syndrome and hepatic fat accumulation.
The human fatty liver is insulin resistant (46, 51). In the present study of nondiabetic subjects, we found several genes involved in insulin signaling to be related to liver fat content. In addition, expression of members of a family of genes involved in ceramide metabolism [UDP-glucose ceramide glucosyltransferase (UGCG)] and signaling [mitogen-activated protein kinase kinase 4 (MAP2K4)] were found to be positively correlated with liver fat content. UGCG catalyzes the first glycosylation step in glycosphingolipid biosynthesis from ceramides (25). Glycosphingolipids have the ability to suppress phosphorylation of the insulin receptor (54). In addition, in vitro studies have shown that ceramides initiate apoptosis by activating MAP2K4 (56), leading to activation of stress-activated protein kinase (SAPK) (47). A recent study in mice showed inhibition of ceramide synthesis to ameliorate both saturated fat and obesity-induced insulin resistance (22). These data raise the possibility that hepatic fat accumulation is associated with an increase in ceramide-mediated cell damage.
Insulin resistance is frequently accompanied by low-grade inflammation as determined by circulatory markers, such as serum C-reactive protein (18) and monocyte chemoattractant protein 1 (MCP-1/CCL2) (50) concentrations. Few data are, however, available regarding the presence of signs of inflammation in the human liver. In the present study, we found several genes involved in inflammation, extracellular matrix formation and remodeling, and coagulation to be altered in subjects with high liver fat content without histologically detectable inflammation. In particular, the chemokines CCL2/MCP-1 and CCL4 were positively correlated with liver fat content. We have previously shown by PCR CCL2/MCP-1 to be increased in the human fatty liver (58). Studies in mice and humans have indicated that the number of macrophages in adipose tissue is increased in obesity and accompanied by increased expression of proinflammatory factors, such as TNF-α and MCP-1 (6, 15, 57, 60). Inflammation is also present in subcutaneous adipose tissue independent of obesity in subjects with high compared with those with normal liver fat content (33). Recently, Haukeland et al. (20) showed that serum CCL2/MCP-1 concentrations correlate with the degree of steatosis and that CCL2/MCP-1 is overexpressed in both simple steatosis and NASH, when determined by immunohistochemistry. These data suggest that inflammation is detectable by gene expression analysis and immunohistochemistry in simple steatotic livers in the absence of histologically detectable inflammatory changes.
In conclusion, we found genes involved in various steps of lipid metabolism, insulin and ceramide signaling, and inflammation to be associated with simple steatosis due to nonalcoholic causes. These data await verification at the level of protein expression. The increased expression of genes involved in ceramide signaling and metabolism is of particular interest, since a high-fat diet can increase liver fat content independent of total caloric intake (59), and high saturated fat intake induces insulin resistance, apoptosis, and inflammatory changes by increasing ceramide synthesis (23, 53).
This study was supported by research grants from the Academy of Finland, the Sigrid Juselius Foundation, the Novo Nordisk Foundation, the Finnish Diabetes Association, Biovitrum, and the Swedish Research Council (project 15352).
This work is part of the project “Hepatic and adipose tissue and functions in the metabolic syndrome” (www.hepadip.org), which is supported by the European Commission as an Integrated Project under the 6th Framework Programme (contract LSHM-CT-2005-018734).
We gratefully acknowledge Mia Urjansson and Katja Tuominen for excellent technical assistance, Dr. Panu Somervuo for valuable advice, and the volunteers for their help.
Present address for O. Puig: Molecular Profiling, Merck & Co., Rahway, NJ.
↵* D. Greco and A. Kotronen contributed equally to this work.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2008 the American Physiological Society