Population-based human study (NHANES)Study population

The NHAENES database provides a large, sophisticated, stratified, ongoing analysis of nutrition and health data for the entire U.S. population [29]. NHANES data (2017–2018) were used to investigate the relationship between iron metabolism and NAFLD. Those participants who lacked serum iron (SI), serum ferritin (SF), transferrin saturation (TSAT), soluble transferrin (sTfR), and NAFLD data were excluded from this study (n = 918). There was a total of 5483 individuals considered. The National Center for Health Statistics Research Ethics Review Board approved all protocols and informed consent was obtained from all subjects. More information on the survey design, methodology, and data can be found on the NHANES website (https://www.cdc.gov/nchs/nhanes/).

NAFLD outcomes

NAFLD was defined using the U.S. fatty liver index (FLI), a well-validated diagnostic index [30], which was employed utilizing NHANES III data and calculated as an equation according to a previous study [31, 32] that included information on body mass index (BMI), gamma glutamyl transferase (GGT), triglycerides (TG), and waist circumference. All the information was collected concurrently with the status of iron metabolism. NAFLD was defined as an FLI score of ≥ 60. The FLI formula is expressed as follows:

$$} = (}^}) + 0.139 * } + 0.718 * \ln (}) + 0.053 * }^ }} )/(1 + }^}) + 0.139 * } + 0.718 * \ln (}) + 0.053 * }^ }} ) * 100.$$

SI, SF, sTfR, and TSAT measurements

The serum iron concentration was measured using the DCX-800 system, which is a timed endpoint method [33]. Acetic acid releases iron from transferrin during the process, and hydroxylamine and thioglycolate reduce it to the ferrous state. FerroZine Iron Reagent quickly complexes with ferrous irons. At a fixed-time interval, the system tracks changes in absorbance at 560 nm. The concentration of iron in the sample is exactly proportional to this change in absorbance. Using the Roche/Hitachi 912 Clinical Analyzer, immunoturbidimetry was the technique of choice for measuring ferritin [34]. Antigen/antibody complexes were created when latex-bound ferritin antibodies interacted with the antigen in the sample. This was turbidimetrically measured after agglutination. The complexes produced were measured at 700 nm and were proportional to the ferritin content (primary wavelength). The method principle for the measurement of soluble transferrin receptor (sTfR) was a particle-enhanced immunoturbidimetric assay that used Roche kits for the Cobas® c501 clinical analyzer [35]. The antigen in the sample reacted with latex particles coated with anti-sTfR antibodies to generate an antigen/antibody combination. The precipitate was identified photometrically after agglutination. Total iron binding capacity (TIBC) was calculated indirectly using the unsaturated iron binding capacity (UIBC) method [36], and the transferrin saturation value was calculated as (iron/TIBC) × 100%. The biomarkers of iron metabolism (SI, SF, TSAT, and sTfR) were divided into quartiles for the assessment of the possible association between iron metabolism and incident NAFLD. The instrument measures SI, SF, TSAT, and sTfR in a range of 11–481 µg/dL, 2–1090 µg/L, 1.4–95% and 1–34.2 mg/L, respectively. Those lower than the detection limits value were missing values and were deleted in this study. The NHANES website (https://wwwn.cdc.gov/nchs/data/nhanes/2017-2018/manuals/2017_MEC_Laboratory_Procedures_Manual.pdf) describes a more thorough processing process for SI, SF, TSAT, and sTfR.

Covariates

This research included a number of covariates: demographic data (age, sex, race/ethnicity, family poverty income ratio (PIR), marital status, education level), dietary data (mean energy intake, protein intake, folic acid intake, vitamin B12 intake, vitamin C intake, iron intake), questionnaire data (hypertension, diabetes mellitus (DM), coronary heart disease (CHD), congestive heart failure (CHF), angina pectoris, heart attack, stroke, smoking, alcohol use, and physical activity (PA, which was collected from the Physical Activity questionnaire (PAQ) in NHANES and categorized into four groups according the intensity of PA: No, Moderate, Both (participants who had a combination of moderate-intensity and vigorous-intensity physical activity,), and Vigorous)), examination data (BMI and waist circumference), laboratory data (total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL cholesterol, HDL-C) and haemoglobin (HB), urinary albumin creatinine ratio (uACR), alanine aminotransferase (ALT), aspartate aminotransferase (AST), GGT, fasting glucose, fasting insulin, glycosylated haemoglobin (HbA1c), high-sensitivity C-reactive protein (hsCRP), estimated glomerular filtration rate (eGFR), uric acid (UA), blood urea nitrogen (BUN), and serum creatinine (Scr)). Detailed covariate information is available publicly from the NHANES database (http://www.cdc.gov/nchs/nhanes/).

Statistical analysis

The NHANES estimations were all based on sample weights [37]. All analyses were performed in version 3.6.4 of R (R Foundation for Statistical Computing, Vienna, Austria) and version 22.0 of SPSS (SPSS Inc., Chicago, IL, USA). Continuous variables are presented as the means ± standard deviations, and categorical variables are expressed as numbers (n) and percentages (%). To investigate the relationship between iron metabolism and NAFLD, multivariable logistic regression was used. First, Model 1 was adjusted for age and sex. Second, based on Model 1, race/ethnicity, level of education, marital status, family PIR, hypertension, DM, smoking status, and drinking status were further adjusted for (Model 2). Finally, Model 3 was updated as our main model and included Model 2 variables plus BMI, waist circumference, PA, mean energy intake, protein intake, folic acid intake, vitamin B12 intake, vitamin C intake, iron intake, the complication of CHD, CHF, angina pectoris, heart attack, stroke, and TC, TG, HDL-C, HB, HbA1c, hsCRP, ALT, AST, GGT, eGFR, uACR, UA, BUN, and Scr. Subgroup analyses were used to evaluate the relationship between iron status and NAFLD based on age, sex, race/ethnicity, hypertension, DM, and BMI. We used regularization technique (Least absolute shrinkage and selection operator (LASSO) regression) to solve the potential overfitting problem. In addition, multivariate stepwise regression analysis was performed to deal with multiple testing problem. P-value < 0.05 was considered statistically significant.

Laboratory-based controlled animal study

Five-week-old male C57BL/6J mice were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd., treated according to the guidelines of the China Pharmaceutical University Institutional Animal Care and Use Committee (approval number: 2019-03-001), and all mouse studies were reported according to the ARRIVE guidelines [38]. Mice were housed in a specific pathogen-free (SPF) facility and adapted to the new conditions for 2 days before the experiment. Mice were maintained in a temperature-controlled (22–23 °C) room with a 12:12-h light/dark cycle.

Animals and experimental protocols

To induce NAFLD, mice were fed a high-fat diet (D12492, 60% kcal from fat; research diet) and fructose (Y0002132, 2.31 g/100 ml; sigma) water (high-fat-fructose diet) for 30 weeks (n = 5). Mice fed normal chow served as controls (n = 5). When the mice were sacrificed, blood was collected from the inferior vena cava, and the entire middle lobe of the liver was fixed with formalin. The largest lobe in the middle of the liver was divided into 8 small pieces after cutting the edge, and one of those 8 pieces was directly added to TRIzol (9109, TaKaRa, Dalian, Liaoning, China) for RNA extraction.

Assays for biomarkers of iron metabolism

Transferrin saturation (TSAT) was defined as the ratio of serum iron (SI) and total iron-binding capacity (TIBC). Unsaturated iron-binding capacity (UIBC) is calculated by subtracting SI from TIBC [39]. SI and TIBC were measured using commercial kits (all from Nanjing Jiancheng Bioengineering Institute) according to standard procedures [11, 40]. Serum ferritin, sTfR, and transferrin were determined using ELISA kits (Jiangsu Meibiao Biotechnology Co., Ltd) according to the manufacturer’s instruction [41]. Serum ferrous irons (Fe2+) serum was measured using the phenanthroline colorimetric method [40] by colorimetric assay kit purchased from Elabscience following the manufacturer’s instructions.

Measurements of hepatic TG, TC and MDA

Three pieces of liver were selected for the quantification of triglyceride (TG), total cholesterol (TC), and malonic dialdehyde (MDA) respectively, which were determined by commercial kits (Nanjing Jiancheng Bioengineering Inst ferrous iron itute) according to the manufacturer’s protocols as previously described [5, 42]. Thiobarbituric acid (TBA) method was used for the determination of MDA, which was based on that MDA can react with TBA at high temperature and acidity to produce the red-brown product with the maximum absorption peak at 532 nm [42].

Histological analysis

The fixed liver lobes were embedded in paraffin, then sectioned at 5 µm and stained with hematoxylin–eosin (H&E) according to the standard protocol described [43,44,45]. Scores for steatosis, inflammation, and ballooning were assessed by a four-member research team according to the NAFLD Activity Score (NAS) system [46].

RNA extraction and quantitative real-time PCR (qRT‒PCR) analysis

Total RNA extraction and cDNA reversion were conducted according to methods described previously [5, 7, 47]. qRT-PCR was performed using SYBR Premix (Vazyme, Q331-02, China) according to the manufacturer’s instructions for the ABI StepOnePlus real-time PCR machine (Applied Biosystems). Custom primers were designed for mouse hepcidin (forward, 5′-TTGCGATACCAATGCAGAAGA-3′; reverse, 5′-GATGTGGCTCTAGGCTATGTTTTG-3′), ferroportin (forward, 5′-TTGCAGGAGTCATTGCTGCTA-3′; reverse: 5′-TGGAGTTCTGCACACCATT-3′), and 18S (forward, 5′-AGTCCCTGCCCTTTGTACACA-3′; reverse, 5′-CGATCCCAGGGCCTCACTA-3′). All real-time PCR primers were synthesized by Sangon Biotech (Shanghai, China). Gene expression was normalized to 18S and calculated as previously described [5, 7, 47].

Statistical analysis

The results are presented as the means ± SEMs and were statistically analysed by the unpaired Student's t test or Pearson product-moment correlation coefficients. Differences were considered significant when the P-value was < 0.05.

Microarray and RNA-seq data

Hepatic gene expression was compared among 20 patients with simple steatosis, 19 with non-alcoholic steatohepatitis (NASH), and 24 healthy controls in the GSE89632 dataset [48] (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE89632). In the GSE126848 dataset, RNA sequencing was performed on liver biopsies obtained from healthy normal weight (n = 14) and healthy obese (n = 12) individuals, and simple steatosis (n = 15) and NASH (n = 16) patients [49] (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE126848). In the GSE185051 dataset, liver biopsies from 51 paediatric NASH patients and 5 normal subjects were analysed by RNA sequencing [50] (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE185051). 12 biopsy diagnosed NASH patients and 5 different subjects for healthy control groups were included in the GSE24807 dataset [51] (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE24807).

To obtain these data, we used the package 'GEOquery' installed in R, especially the function getGEO(),  importing the dataset GSE number, and obtained the list object of the corresponding dataset. Then, the pData() and expr() functions were used to view the sample information and expression matrix, respectively. The chip platform used in the dataset was confirmed by viewing the web page and checking the list of objects. The gene annotation information of the platform was obtained by using the GEOquery software package and matched with the array information. The gene symbol was used to replace the probe name. Genes with |log2-fold changes (FCs)|> 0.5 and P-value < 0.05 were considered differentially expressed genes (DEGs). Then, the hepcidin and ferroportin genes were selected, and the expression information of these genes in each sample and the grouping information of each sample were extracted and stored in an Excel table for subsequent analysis. The differences between different groups were compared by the Mann–Whitney test and considered statistically significant at the P-value < 0.05.