The National Health and Nutritional Examination Survey (NHANES) is a cross-sectional nationwide program aimed at monitoring the physical condition and nutritional status of the US population conducted by the National Center for Health Statistics (NCHS) at the Centers for Disease Control and Prevention (CDC) (Available online: https://wwwn.cdc.gov/nchs/nhanes/Default.aspx), as described in our published study (16). The data in our study include socio-demographic issues, dietary schemes, and medical data, which were obtained through interviews, professional physical examinations, and laboratory measurements. The program was approved by the Institutional Review Board (IRB) of the NCHS, and all participants or participants' agents signed informed consent forms.

We collected data from 9,254 participants from the NHANES Public Data 2017–2018. Subjects with missing information on iron metabolism and Vit C (36.8%) were excluded. Partially completed and ineligible liver ultrasound transient elastography (LUTE) samples were also excluded. Other exclusion criteria were as follows: (1) those with HBV antigen or HCV RNA positivity and (2) participants who were under the influence of alcohol. Overall, 3,614 eligible subjects were included in the analysis. The data details of patients were shown in Supplementary Figure 1.

During the 2017–2018 NHANES cycle, liver transient elastography was performed to assess the degree of liver steatosis and sclerosis. All NHANES health technicians and examiners completed a 2-day training program using FibroScan^® and passed the exam. MEC staff have been regularly supervised by investigators and medical epidemiologists specializing in chronic liver disease, thus ensuring the quality of LUTE.

Based on a recent study done by Eddowes et al. median CAP values ≥274, 290, and 302 dB/m were considered to be indicative of S1, S2, and S3 steatosis, respectively (17). Median stiffness values ≥8.2, ≥9.7, and ≥13.6 kPa were considered markers of significant fibrosis (≥F2), advanced fibrosis (F3), and cirrhosis (F4), respectively (17).

To create a real fit model as possible, potential confounders were also included as covariates in our study, which include age, sex (male or female), race (non-Hispanic white, non-Hispanic black, Mexican American, Hispanic, and other races), education level (below high school, high school or equivalent, college, or above), marital status (married and single), and household poverty-to-income ratio (PIR) ( ≤ 1. 30, 1.31–3.5, >3.5, and others), BMI (<25 and ≥25 kg/m²), strenuous recreational activity (yes or no), moderate recreational activity (yes or no), smoking status, history of hypertension, diabetes, and chronic kidney disease. Smoke exposure (active or passive smoking) status was determined by serum cotinine concentration with a cut-off value of 14 ng/ml. The diagnoses of hypertension, diabetes mellitus, and chronic renal failure in the study subjects either met the clinical criteria or were based on the medical history (history of relevant pharmaceutical use) from the questionnaire. Participants' dietary data were sourced from the What We Eat in America (WWEIA) project. The average of two 24-h nutrient intakes was used to estimate the daily nutrient intake of participants in the past month.

Information regarding diabetes, hypertension, and chronic renal failure was based on self-report questionnaires. Hypertension was defined as systolic blood pressure (SBP) >140 mmHg or diastolic blood pressure (DBP) >90 mmHg (18). Diabetes was diagnosed as fasting plasma glucose (FPG) levels ≥7.0 mmol/l or 2-h postload blood glucose (PBG) ≥11.1 mmol/l (19, 20). Dyslipidemia was defined as fasting plasma triglyceride (TG) ≥1.7 mmol/l, fasting high-density lipoprotein cholesterol (HDL-C) <1.04 mmol/l, low-density lipoprotein cholesterol (LDL-C) ≥3.37 mmol/l, or total cholesterol (TCH) ≥5.18 mmol/l (21). Hyperuricemia was defined as serum uric acid (UA) ≥7 mg/dl (≥417 μmol/L) in men and ≥6 mg/dl (≥357 μmol/L) in women (22). Additionally, MetS was defined by three out of four components as follows (23): (1) overweight and(or) obesity [body mass index (BMI) ≥25 kg/m²], (2) FPG ≥6.1 mmol/l and(or) PBG ≥7.8 mmol/l and(or) diagnosed diabetes, (3) hypertension: BP ≥140/90 mmHg or positive history of hypertension or on any hypertensive treatment, (4) dyslipidemia.

HepG2 cell lines were obtained from Shanghai Tenth People's Hospital of Tongji University and cultured at 37°C and 5% CO₂ using DMEM-HG (Gibco, USA) containing 10% fetal bovine serum (Gibco, USA) and 1% v/v penicillin–streptomycin. Oleic acid (OA) (A502071, Sangon Biotech, Shanghai, China) and palmitic acid (PA) (A600497, Sangon Biotech, Shanghai, China) were dissolved in 5% BSA. Cellular steatosis was induced by the addition of 600 μM OA and 300 μM PA. Intracellular lipids were verified with the Oil Red O staining kit (G1262, Solarbio, Beijing, China). Cells were treated with different concentrations of Ascorbic acid (Vitamin C, A92902, Sigma-Aldrich, USA) for 24 h and harvested for subsequent experiments to test the effect of Vit C.

HepG2 cells (5 × 10³/well) were plated into 96-well plates. Cells were then treated with fatty acids and different concentrations of Vit C for 24 h. Cell viability was measured using a CCK8 reagent (No. C0005, Topscience, Shanghai, China) according to the manufacturer's instructions.

Intracellular TG and TCH concentrations were measured according to the instructions of the TG (F001-1-1, Nanjing Jiancheng Bioengineering Institute, Nanjing, China) and TCH assay kits (F002-1-1, Nanjing Jiancheng Bioengineering Institute, Nanjing, China), and the values were corrected using the protein concentration of the samples. The glucose concentration in the medium was measured according to the instructions of the glucose assay kits (A154-1-1, Nanjing Jiancheng Bioengineering Institute, Nanjing, China), and the values were corrected using the cell numbers of the samples. The cell numbers of the samples were tested by an automated cell counting chamber (SD100-002, Nexcelom Bioscience, Shanghai).

Total proteins were extracted from cells using RIPA lysis buffer (PC101, Epizyme, Shanghai, China) with protease inhibitor cocktail (GRF101, Epizyme, Shanghai, China). Equal amounts of proteins were transferred to polyvinylidene fluoride membranes and imaged on an Odyssey scanner (LI-COR Biosciences, USA) after incubation with primary ferritin heavy chain (FTH) (T57085, Abmart), ferritin light chain (FTL) (T56955, Abmart), and β-tubulin (M30109, Abmart) and secondary antibodies (#7054, #7056, Cell Signaling Technology). The primary antibodies were diluted at 1:1,000 and secondary antibodies were diluted at 1:2,000.

The iron-sensitive fluorescent dye Calcein-AM (40719ES, Yeason, Shanghai, China) was used to measure intracellular labile iron pool (LIP) levels according to the previously described method (24, 25). Briefly, cells were stained in 6-well plates using 0.1 μM Calcein-AM at 37°C for 15 min and washed three times with PBS to remove excess dye, then cells were detached using 0.25% trypsin and resuspended in 500 μl 1X Binding Buffer. The intracellular fluorescence intensity was immediately measured by flow cytometry (BD Biosciences, λ_ex 488 nm; λ_em 530 nm). Following this the cells were incubated with 100 μM deferiprone (HY-B0568, MedChemExpress, USA) at 37°C for 40 min (Shake every 5 min to avoid cell clumping) fully bind the free iron in the cells and run the flow assay again. The mean fluorescence intensity (MFI) was calculated by FlowJo software (v10.4) for each population of at least 10,000 cells. The difference in MFI (ΔF) after and before deferiprone treatment was used to reflect the level of LIP.

Continuous variables are presented as the mean ± standard deviation (X ± SD). Categorical variables are presented as numbers (percent). All data analyses were performed using NANS-recommended weighting correction. Normally distributed data were compared with the independent sample t-test, while the non-normally distributed continuous data were compared using the Mann–Whitney U test. For comparison of categorical variables, the chi-square test was used. Pearson's correlation coefficient was used to analyze the correlations between the markers.

Restricted cubic spline (RCS) was performed to fit changes in odds of NAFLD across the Vit C and ferritin range. In addition, mediating effect analysis was used to investigate whether Vit C affects the development of NAFLD or liver fibrosis through serum ferritin mediation. We hypothesized that the dose-response relationship between Vit C (X) and NAFLD (Y) is mediated by serum ferritin (M). The total effect (TE) of Vit C was split into a direct effect (DE) and an indirect effect (IE) on NAFLD. According to Baron and Kenny et al. (26), an intermediate effect of M is considered to exist when the conditions of “statistically significant association between X and M” and “statistically significant association between M and Y” are met.

All statistical analyses were performed using R (version 4.0.2) and SPSS software (version 24.0) and involved the use of R packages such as “mediation,” “rms,” and graphical visualizations were generated by “ggplot2.” Statistical significance was set at P < 0.05.

Three thousand six hundred and fourteen eligible subjects were divided into four steatosis degree groups, S0 (normal, <274 dB/m), S1 (mild, 274–290 dB/m), S2 (moderate, 290–302 dB/m), and S3 (severe, ≥302 dB/m) according to the CAP values. The age (P < 0.001), the percentage of males (P < 0.001), and married populations (P = 0.002) tended to be higher in more severe hepatic steatosis. The obesity indicators: body weight, BMI, WC, and waist-to-hip ratio (WHR), liver injury markers: alanine transaminase (ALT), aspartate aminotransferase (AST), gamma glutamyl transpeptidase (GGT), and alkaline phosphatase (ALP) levels, glucose metabolic markers: FPG, fasting insulin (FINS), homeostasis model assessment of insulin resistance (HOMA-IR), and glycosylated hemoglobin (HbA1c), blood pressure related SBP, DBP, and comorbidities of metabolic disorders such as the rate of hypertension, diabetes, hyperlipidemia, hyperuricemia, and MetS increased with the increasing severity of hepatic steatosis (all P < 0.001). Lipid metabolic markers, TC, LDL-C, and TG increased while HDL-C decreased in the group with more severe hepatic steatosis (all P < 0.05). Additionally, factors related to the steatosis severity of NAFLD is complex. Interestingly, we found ferritin also increased significantly (P < 0.001) while serum Vit C decreased with more serve hepatic steatosis (P < 0.001) (Table 1).

All subjects were also divided into four fibrosis groups according to the degree of fibrosis assessed by the E value. Similar to steatosis, the percentage of men (P = 0.049) and age were higher (P < 0.001) in more severe fibrosis. Systolic blood pressure (SBP, P < 0.001), DBP (P = 0.016), body weight, BMI, WC, and WHR, the rates of hypertension, diabetes, hyperlipidemia, hyperuricemia, and MetS, liver injury makers ALT, AST, and GGT levels, glucose metabolic markers, FPG, FINS, HMOA-IR, and HbA1c also increased with the increasing degree of hepatic fibrosis (all P < 0.001). Triglyceride increased while HDL-C decreased in more severe hepatic fibrosis group (all P < 0.001). Likewise, serum ferritin significantly increased; while Vit C significantly decreased with more severe hepatic fibrosis (all P < 0.001) (Table 2).

Serum ferritin levels were significantly higher in subjects with hypertension, diabetes, dyslipidemia, hyperuricemia, or MetS than in subjects without corresponding metabolic disorders (all P < 0.001). Vitamin C was significantly lower in in subjects with diabetes, dyslipidemia, hyperuricemia, or MetS than subjects without corresponding metabolic disorders (all P < 0.001) (Supplementary Figure 2).

Based on the results above about serum Vit C, we were wondering about the effect of Vit C dietary intake on NAFLD. Therefore, we analyzed the dietary data of the enrolled samples. However, the risk of NAFLD was not reduced in the high-dose Vit C intake group (Q4, Supplementary Table 1). The correlation analysis showed serum Vit C levels were not only correlated with Vit C intake, but also with intake of nutrients including fat, saturated fatty acids, sugar, caffeine, iron, and Vitamin B12 (Figure 1A). Further analysis revealed that although Vit C intake was high in participants, the intake of other nutrients including sugar, iron, and saturated fatty acids, which would lower serum Vit C were also increased (Figure 1B). Additionally, the metabolic disorder in the liver diseases complicates the regulation of Vit C. This explains why the high dose of Vit C intake in the study cohort fails to produce a protective effect against NAFLD and cirrhosis. Therefore, serum Vit C level is more reasonable and reliable for clinical analysis and guidance for NAFLD patients.

To explore the roles of ferritin and Vit C on hepatic steatosis, we adjusted for age, sex, race, education level, marital status, and PIR, increased serum ferritin was a risk factor for hepatic steatosis (1.49 [1.18–1.88], 1.81 [1.43–2.30], 2.15 [1.68–2.74], all P < 0.001), while increased Vit C was a protective factor for NAFLD (0.83 [0.68–1.01], 0.63 [1.43–2.30], 0.47 [0.38–0.58], all P < 0.001). Adjusted for the above parameters plus BMI, smoking status, and vigorous recreational activities, increased serum ferritin was still a risk factor for NAFLD (1.59 [1.24–2.05], 1.71 [1.33–2.20], 2.02 [1.56–2.62], all P < 0.001), while increased Vit C was a protective factor for NAFLD (0.82 [0.66–1.02], 0.65 [0.52–0.81], 0.55 [0.44–0.70], all P < 0.001). Adjusted for hypertension, diabetes, kidney failure, hyperlipemia, hyperuricemia, and MetS, increased serum ferritin was still a risk factor for NAFLD (1.52 [1.17–1.96], 1.72 [1.33–2.23], 1.95 [1.49–2.55], all P < 0.001), while increased Vit C was a protective factor for NAFLD (0.82 [0.66–1.03], 0.70 [0.56–0.88], 0.61 [0.48–0.77], all P < 0.001), as presented in Table 3. To avoid the bias caused by any possible leverage values, we have also used the RCS model to fit the non-linear relationship between Vit C, ferritin, and NAFLD, we found that the odds ratio of NAFLD increased significantly with increasing serum ferritin after adjusting for covariates (P < 0.001, Figure 2A), while serum Vit C showed the opposite trend (P < 0.001, Figure 2B).

Similarly, adjusted for confounding covariates, the highest ferritin group (176.1–243.0 ng/ml) was a risk factor for hepatic fibrosis (1.83 [1.12–2.99], P = 0.016), while the highest serum Vit C group (69.31–359 μmol/L) was a protective factor for hepatic fibrosis (0.39 [0.24–0.61], P < 0.001) (Table 4). By using the RCS model we found that the odds ratio of hepatic fibrosis increased significantly with increasing serum ferritin after adjusting for covariates (P < 0.001, Figure 2C), while elevated serum Vit C significantly decreased the odds ratio of hepatic fibrosis (P < 0.001, Figure 2D).

According to the correlation analysis, serum Vit C was negatively associated with E and CAP values (r = −0.073, P < 0.001; r = −0.140, P < 0.001), while ferritin was positively associated with the E and CAP values (r = 0.148, P < 0.001; r = −0.174, P < 0.001). We further raised the question that whether there was a connection between Vit C and ferritin. Surprisingly, Vit C was found negatively associated with ferritin levels (r = −0.045, P < 0.01) (Figure 2E). Mediation analysis was performed to verify whether ferritin partly mediates the impact of Vit C on NAFLD. The results showed that Vit C may act on NAFLD by acting on ferritin, as shown in Figure 3. Ferritin partly mediated efficacy account for 5.35% of the association of Vit C and steatosis grading (IE = −0.00395, 95% CI: −0.00776 to 0.00; DE = −0.06847, 95% CI: =-0.09964 to −0.04, Figure 3A), while account for 5.67% of the association of Vit C and NAFLD (IE = −0.0109, 95% CI: −0.0203 to 0.00; DE = −0.1798, 95% CI: −0.2488 to −0.10, Figure 3B), account for 3.19 of the association of Vit C and fibrosis grading (IE = −0.00461, 95% CI: −0.00871 to 0.00; DE =-0.14082, 95% CI: −0.19766 to −0.09, Figure 3C), and account for 2.78 of the association of Vit C and cirrhosis (IE = −0.01071, 95% CI: −0.002034 to 0.00; DE = −0.35670, 95% CI: −0.50473 to −0.17, Figure 3D).

To verify the protective effect of Vit C on NAFLD, we induced lipid accumulation with PA/OA in liver cell line hepG2 and treated it with Vitamin C. We found that intracellular lipid deposition gradually decreased with increasing Vit C concentration in the HepG2 cells (Figure 4A) and 200 μg/ml concentration of Vit C significantly suppressed the elevated levels of intracellular cholesterol and triglycerides (Figures 4C,D). However, extremely high concentrations of Vit C levels did not provide further protection against cellular lipid deposition, which may be due to the inhibitory effect on cellular activity, that is adverse effects or toxicity (Figure 4B). This is also consistent with the published animal study (4).

To answer whether Vitamin C treatment has an influence on ferritin, ferritin levels were detected with or without Vitamin C stimulation in HepG2 cells. We observed that FTH levels were elevated by about 23% after the addition of PA/OA and Vit C inhibited ferritin light and heavy chain levels in a dose dependent manner (Figures 5A–C). We also noticed a significant reduction of the LIP within HepG2 cells after PA/OA addition. The reduction of LIP due to palmitic and OA could be restored by supplementation of Vitamin C (Figures 5D,E). Moreover, Vit C reversed the metabolic dysfunction caused by lipotoxic substances and improved the cellular uptake of glucose, while its effect may be mediated by influencing intracellular LIP levels (Figure 5F).