Serum transferrin as a biomarker of hepatocyte nuclear factor 4 alpha activity and hepatocyte function in liver diseases

Forty liver samples from patients who underwent a liver biopsy at the Universities of Aachen and Ulm between the years 2006 and 2018 were analyzed. The diagnosis of HCV was confirmed by a positive HCV RNA test, and all patients were treatment-naive. The diagnosis of ALD/NAFLD was based on laboratory parameters, histological findings, personal interview, and an exclusion of other liver disease etiologies including viral infection, hereditary hemochromatosis, and autoimmune liver disease. NAFLD patients had characteristic histological findings without significant alcohol consumption (Table 1). Among them, a matched serum pair was available for 23 samples that were used for correlation analysis. Four liver samples from surrounding, non-affected tissue of liver metastases from patients without known chronic liver disease had been used as controls (Table 1). Staging of fibrosis was done on 4-μm-thick liver sections using Desmet (HCV, controls) and Kleiner (ALD, NAFLD) scores. Publicly available data on transferrin levels in liver disease were obtained from GEO database and in silico analysis was performed on the GSE103580 dataset in 13 patients with alcoholic hepatitis [18]. The study was conducted in compliance with the Declaration of Helsinki (Hong Kong amendment) and Good Clinical Practice (European guidelines) as reflected in an approval by the local ethics committees of the participating centers. All participants, or their legally appointed representatives, provided written informed consent.

RNA-seq was performed on liver tissue biopsies obtained from patients with early alcoholic steatohepatitis (n = 12), non-severe AH (n = 11), severe AH (n = 18), explants from patients transplanted with severe AH (n = 9), and normal background liver specimens from patients resected for benign lesions (n = 10), from the InTEAM Consortium - Alcoholic Hepatitis Liver RNA Sequencing study, sponsored by the National Institute of Alcohol Abuse and Alcoholism (NIAAA, USA). The study details and sequencing data can be found in the Database of Genotypes and Phenotypes (dbGAP, phs001807.v1.p1) of the National Institutes of Health (NIH, USA). The basic clinical and laboratory data of the patients included in this study, the methods used to extract RNA and perform deep RNA sequencing, and the bioinformatic pipelines used to determine transcript counts have been described elsewhere [9]. In order to uncover the transferrin co-expressed transcriptome, the Kendall rank correlation coefficient was calculated. For pathway analysis, directly and reversely correlated genes with p value inferior to 10⁻¹² were selected. The Kendall coefficient ranges for selected genes was + 0.76 to + 0.91 for the directly correlated (n = 309 genes) and − 0.76 to − 0.84 for the reversely correlated (n = 94 genes). Ingenuity Pathway Analysis (IPA, Qiagen Inc., version 2017) was used to uncover main upstream regulators.

Liver DNA methylation studies have been described previously [9]. Briefly, genomic DNA (gDNA) was extracted from flash-frozen liver tissue with PureLink Genomic DNA Mini Kit (Thermo) and quantified using Nanodrop (Thermo). One microgram of isolated gDNA was bisulfite converted, denatured, fragmented, and hybridized to Infinium Methylation Bead Chip, following the manufacturer’s protocol (Infinium MethylationEPIC kit, Illumina). BeadChips were imaged using an Illumina Scan System, and intensity was determined by the iScan Control Software (Illumina). Sample intensities were normalized using functional normalization from the minfi package (v1.24.0) [19]. Probes failing a detection p value threshold (0.01) in at least 50% of samples were removed, as were probes identified as containing a SNP with a MAF > 0.05. Differentially methylated probes were identified by applying limma (v3.34.3) contrasts to M values (absolute change in beta value > 0.1, FDR-corrected p value < 0.05) [20]. Differentially methylated regions were identified using DMRcate (v1.14.0) [21] setting a threshold of absolute change in beta value in > 0.1 and of Stouffer’s value in < 0.05. The package gviz (v 1.28.0) [22] was used to visualize transferrin genomic region, including methylation beta values. The information downloaded from the University of California Santa Cruz (UCSC) server of the Human Genome Sequencing Consortium included HepG2 HNF4A ChIP-seq BigWig data (Sample GSM803460 of series GSE32465) and DNAse I hypersensitive sites Bed data (wgEncodeRegDnaseClusteredV3.bed, UW and Duke ENCODE data).

ChIP-seq was performed in Mayo Epigenomics Development Laboratory (EDL) [23] with the liver tissue from 5 controls and 7 severe AH explants (provided by University of Lille, France). It was carried out for four histone modifications, using antibodies against histone H3 Lysine 27 acetylation (H3K27ac, Cell Signaling #8173), histone H3 Lysine 27 tri-methylation (H3K27me3, Cell Signaling #9733), histone H3 Lysine 4 mono-methylation (H3K4me1, EDL, Mayo Clinic, Lot#1), and histone H3 Lysine 4 tri-methylation (H3K4me3, EDL, Mayo Clinic, Lot#1). For the next-generation sequencing, ChIP-seq libraries were prepared from 10 ng of ChIP and input DNAs with the Ovation Ultralow DR Multiplex System (NuGEN). The ChIP-seq libraries were sequenced to 51 base pairs from both ends using the Illumina HiSeq 2000 in the Mayo Clinic Medical Genomics Core. Data were analyzed by the HiChIP pipeline [24]. Briefly, reads were aligned to the hg19 genome assembly using BWA and visualized using the Integrative Genomics Viewer (IGV). Mapped reads were post-processed to remove duplicates and pairs of reads mapping to multiple locations. The MACS2 and Sicer algorithm was used for peak calling in relation to the input DNA. In this study IGV was then used to visualize the peak changes on transferrin genomic region in this study.

Four control and 6 cirrhotic samples were used for biochemical analysis. Tissue lysates were prepared in homogenization buffer containing 3% sodium dodecyl sulphate (SDS) and were diluted with 4× reducing Laemmli buffer afterwards. Proteins were separated via 10% SDS polyacrylamide gel electrophoresis and transferred to PVDF membranes (GE Healthcare/Amersham Biosciences, Germany) for immunoblotting. After incubation with the primary and horse radish peroxidase (HRP)-coupled secondary antibodies, the resulting HRP signal was detected with an ECL Detection kit (GE Healthcare/Amersham Biosciences, UK). The following antibodies were used: human anti-transferrin (Sigma; HPA005692, Stockholm, Sweden), mouse anti-transferrin (Abcam; Ab82411), anti-HNF4alpha (Sigma; HPA004712, Bromma Sweden), and anti-GAPDH (Novus; NB300-221, Nordenstadt, Germany).

For immunohistochemistry, 3–4-μm-thick liver sections were deparaffinized and rehydrated, and a citrate-based antigen retrieval was performed (Vector H-3300, Vector Laboratories, Petersborough, UK) for 30 min, as recommended by the supplier. The immunodetection was carried out with anti-transferrin as a primary (11,000 dilution; Sigma; HPA005692, Stockholm, Sweden) and a biotin-conjugated anti-rabbit secondary antibody (Vector BA1000). The signal was visualized with peroxidase-labeled streptavidin and the 3,3′-diaminobenzidine as a substrate (Vector SK4100).

Serum transferrin (turbidimetry), iron, and ferritin levels were measured with routine assays using the Cobas 8000 system (Roche Diagnostics, Mannheim, Germany) available at the Clinical Chemistry Department of University Hospital Aachen. Serum apolipoprotein B (APOB; R&D Systems DAPB00, Minneapolis, MN, USA), human tranthyretin (Prealbumin; Avivasysbio, OKIA00081, San Diego, CA, USA), and complement C3 (Abcam, ab108822, Berlin, Germany) concentrations were measured by using commercial sandwich ELISAs.

Primary hepatocytes were isolated from 3- to 4-month-old C57BL/6 mice as described previously [25]. Primary hepatocytes were kept in culture medium (DMEM; PAN Biotech, Aidenbach, Germany) supplemented with 10% FBS, 1% penicillin-streptomycin, and l-glutamine (PAN Biotech). After overnight incubation, the medium was replaced by William’s E medium (PAN Biotech), and cells were treated for 6 h with IL-1β (20 ng/ml; Peprotech, Hamburg, Germany), IL-6 (40 ng/ml; Sigma, St. Louis, MO, USA), TNFα (30 ng/ml; Peprotech), and the HNF4α antagonist BI6015 (50 μM; Millipore, Burlington, MA, USA) or for 24 h with TGFβ1 (10 ng/ml; R&D Systems, Minneapolis, MI, USA). Alternatively, the cells were transfected with the siRNA against HNF4α (60 pmol; Thermo Fisher Scientific ID: 158081 and 158082, Germany) according to the manufacturer’s protocol (Lipofectamine® RNAiMAX™ Transfection Reagent, Thermo Fisher Scientific, Germany) and cultured for 48 h. RNA was isolated via RNeasy tissue mini isolation kit (Qiagen, Hilden, Germany). The RNA samples were translated to cDNA with the M-MLV reverse transcriptase kit (Promega, Madison, WI, USA) and random hexamers (Thermo Scientific, Waltham, MA, USA). The relative expression of genes of interest was determined using specific primers (Additional file 1: Table S1). The mouse ribosomal gene L7 was used as an internal loading control.

Based on the results of a normality test, continuous variables were compared by an unpaired t test. Correlations between clinical variables and serum proteins were tested by Spearman’s rank correlation test. Nominal two-sided p values were reported for all tests and were considered to be statistically significant when p < 0.05 unless mentioned otherwise. Statistical analyses were performed, and graphs were created with GraphPad Prism (GraphPad, La Jolla, CA, USA).

To delineate the factors underlying the predictive usefulness of transferrin in severe liver disease, we compared the hepatic expression of transferrin and transferrin serum levels in 23 patients with advanced liver disease and available matched mRNA-serum sample pairs (Table 1) (cohort i). Immunoblotting revealed decreased transferrin protein levels in cirrhotic livers (Fig. 1a, b). Accordingly, immunohistochemistry manifested a strong, homogeneous hepatocellular staining in control livers, whereas a patchy, diminished signal was seen in cirrhotic patients (Fig. 1c) (cohort i). In individuals with different stages of AH (cohort ii), a progredient decrease in normalized transferrin gene transcripts was seen at more severe stages (Fig. 1d). Spearman’s correlation showed a moderate correlation between the serum transferrin levels and hepatic mRNA expression in individuals with advanced liver fibrosis and available matched serum-mRNA sample pairs (Fig. 1e) (cohort i). Notably, serum transferrin levels did not correlate with serum levels of other established hepatocellular products such as alpha1-antitrypsin, ceruloplasmin, albumin, or apolipoprotein B (not shown). On the other hand, in cohort (i), we observed a moderate negative correlation between serum ferritin and transferrin levels, while no correlation was seen between serum transferrin and serum iron levels (Additional file 1: Fig. S1).

To determine the factors responsible for altered hepatic transferrin expression, we analyzed RNA-seq data from healthy individuals as well as patients with different stages of alcohol-related liver disease (ALD). This cohort comprised the full spectrum of ALD, from early asymptomatic disease in heavy drinkers to patients with alcoholic hepatitis (Figs. 1d and 2a) (cohort ii). The expression of TF in the liver of patients with AH was significantly reduced in patients with severe AH (Fig. 1d). We then performed a gene-gene correlation analysis of TF expression with the whole transcriptome, stratifying by disease severity (Fig. 2a). The most correlated genes were then used to infer transcription factor footprint using Ingenuity Pathway Analysis. The analysis of upstream regulators revealed HNF4α and TGFβ1 as the factors that are most tightly linked to the transferrin-correlated transcriptome (Fig. 2b, c) (cohort ii). This result suggests that transferrin expression could be a bona fide marker of hepatic HNF4α downregulation. The established drivers of acute phase response, i.e., TNFα, IL-1β, and IL-6 also displayed a strong negative correlation with transferrin mRNA in our upstream regulation analysis (Fig. 2b). Since these analyses were made using whole liver RNA, the impact of infiltrating non-parenchymal cells on these transcriptomic footprints cannot be excluded.

To further understand the regulation of transferrin in patients with AH, we analyzed the transferrin genomic region using liver methylomic data as well as chromatin immunoprecipitation coupled to DNA sequencing data (ChIP-Seq) from a previous study in patients with different stages of alcoholic liver disease (Figs. 3 and 4) [9] (cohort ii). We compared these genomic regions with different epigenetic marks and HNF4A ChIP-seq available in ENCODE. The methylome analysis revealed a differentially methylated region in a regulatory region of the transferrin promoter, which was bound by HNF4α. Patients with AH had higher levels of methylation in different regulatory CpG islands in this region, indicating a possible cause of transferrin downregulation (Fig. 3). In a different set of AH patients, ChIP-Seq showed increased levels of H3K27me3 mark and a decrease on H3K4me3 and H3K27ac marks in the same region, indicating that the chromatin conformation promotes repression of transferrin gene expression in these patients (Fig. 4). To test whether these associations imply a direct regulation, we used primary mouse hepatocytes (Fig. 5a). In these, TGFβ1 treatment markedly suppressed transferrin expression (Fig. 5b) whereas IL-1β, TNFα, and IL-6 treatment did not have a significant effect (Fig. 5c–e). Treatment with the HNF4α antagonist BI6015 resulted in an almost complete suppression of transferrin expression (Fig. 5f), while no significant alterations were seen in HNF4γ levels that are known to be affected by high doses of BI6015 (not shown). RT-PCR for HNF4A, which was used to confirm the efficacy of the described treatments, revealed the suppression of HNF4α expression in both the antagonist- and TGFβ1-treated hepatocytes (Fig. 5b, f). Finally, a treatment with HNF4A siRNA resulted in an almost complete suppression of HNF4A mRNA levels and significantly reduced transferrin protein levels (Additional file 1: Fig. S2).

Finally, to further corroborate the correlation between HNF4A and transferrin expression, we analyzed publicly available data on 13 patients with AH (GSE103580) [18]. In these individuals, we observed a strong association between hepatic transferrin and HNF4A levels (r = 0.70 p = 0.007; Additional file 1: Fig. S3A). A comparably strong correlation was seen in 40 patients with advanced liver disease (r = 0.59 p < 0.0001; Additional file 1: Fig. S3B). These data strongly suggest that transferrin decrease reflects an impaired HNF4α axis. In line with that, immunoblotting demonstrated decreased HNF4α protein levels in cirrhotic vs. control livers (Additional file 1: Fig. S3C).

To identify additional markers of impaired HNF4α signaling, we assessed three established HNF4α target genes, i.e., apolipoprotein B (ApoB), transthyretin (TTR), and complement C3 (C3). Spearman’s correlation coefficient revealed a moderate correlation between the mRNA levels of HNF4A and its targets ApoB and C3 but no significant correlation with TTR (Additional file 1: Fig. S4). Moreover, in none of the analyzed species, a significant correlation between serum and hepatic mRNA was detected (Additional file 1: Fig. S4). Notably, similar results were seen when only subjects with ALD/NAFLD were included in the analysis (not shown). Collectively, our data indicate that transferrin may constitute an attractive serological marker reflecting the hepatic HNF4α signaling.