Causal Relationships between Iron Status and Nonalcoholic Fatty Liver Disease: Two-Sample, Multivariable, and Two-Step Mendelian Randomization

Abstract

Background:

Nonalcoholic fatty liver disease (NAFLD) was clinically documented to be accompanied by iron homeostasis imbalances, however, the causal relationship between them remains unclear. Therefore, this study aimed to examine the relationship between iron homeostasis indicators (serum iron, ferritin, transferrin, total iron binding capacity (TIBC), and transferrin saturation (TSAT)) and NAFLD risk.

Methods:

We applied two-sample Mendelian randomization (MR) to assess the effects of genetic liability to iron homeostasis indicators (N = 43,220–246,139) on NAFLD risk (N = 377,988) in individuals of European ancestry. Reverse direction MR, multivariable MR, and two-step MR were performed to estimate reverse association, causal effects independent of smoking or drinking, and the mediating effect of lipid metabolism, respectively. Smoking and drinking as confounders were considered confounders.

Results:

Genetically predicted serum iron, ferritin, and TSAT were significantly associated with a higher risk of NAFLD (odds ratio (OR): 1.286, 95% confidence interval (CI): 1.075–1.539; p = 0.0059; OR: 1.260, 95% CI: 1.050–1.500, p = 0.0195; and OR: 1.223, 95% CI: 1.067–1.402; p = 0.0039, respectively). Reverse direction MR analysis suggested that genetic liability to NAFLD had no significant causal effect on iron homeostasis. Sex-specific MR exhibited a stronger effect size for the association of elevated ferritin with NAFLD risk in males (OR: 1.723, 95% CI: 1.338–2.219; p = 2.48 × 10^-5). Two-step MR revealed that elevated triglycerides (TGs) mediated approximately 3%–5% of the observed effect of serum iron and TSAT on NAFLD risk, while decreased low-density lipoprotein cholesterol (LDL-C) mediated 9%–10%.

Conclusion:

Genetic liability to iron status imbalance may causally affect NAFLD. This evidence may support the clinical treatment of NAFLD in the target population.

Graphical abstract

Keywords

iron status / lipid metabolism / nonalcoholic fatty liver disease / Mendelian randomization / mediation

1. Introduction

Nonalcoholic fatty liver disease (NAFLD) is characterized by hepatic steatosis in the absence of significant alcohol consumption and other secondary causes of hepatic fat accumulation [1]. It has become the most prevalent form of chronic liver disease, ranging from nonalcoholic fatty liver (NAFL) to advanced histological stages. A systematic review estimated that NAFLD afflicted 29.8% of the global population in 2019 [2]. Despite its growing prevalence, effective noninvasive diagnostic strategies and treatments for NAFLD are still lacking, underscoring the need to identify modifiable factors associated with disease progression. Among these, iron status may be a promising target for NAFLD prevention and treatment.

Iron plays key roles in various physiological processes, while cellular iron overload can induce oxidative stress and lipid peroxidation via the Fenton reaction [3]. Biochemical measures of iron status, such as serum iron, ferritin, transferrin, total iron binding capacity (TIBC), and transferrin saturation (TSAT), serve as essential clinical indicators for iron homeostasis. Several observational studies have consistently found that elevated serum ferritin levels are independently associated with elevated severity and mortality among patients with NAFLD [4, 5, 6, 7]. However, the association between other iron status indicators and NAFLD severity is limited and controversial [8, 9, 10, 11]. Although iron homeostasis imbalances are clinically documented to accompany NAFLD, the causal relationship between them remains unclear.

Smoking and alcohol consumption are the most common lifestyle factors often accounted for as confounders in association studies. Cigarette smoking has been identified as a risk factor for the onset and severity of NAFLD [12, 13], and smoking status could also affect iron status [14, 15]. Modest alcohol consumption (defined as 210 g per week for men and 140 g per week for women) is associated with lower risks for steatohepatitis and advanced fibrosis among healthy people [16]. Meanwhile, alcohol consumption is also associated with a reduced risk of iron deficiency and an elevated risk of iron overload [17]. Therefore, smoking and alcohol consumption may act as confounders in the association between iron status and NAFLD.

Observational epidemiological studies are limited by their susceptibility to confounding and reverse causation. An alternative approach that can address these challenges is the Mendelian randomization (MR) study in which genetic instruments are used as the proxy of iron status or NAFLD. MR is based on the hypothesis that genetic instruments are determined at conception by random chromosome distribution and are unrelated to potential confounders. In this study, we provide evidence on the causal effect of iron status, including serum iron, ferritin, transferrin, TIBC, and TSAT, on NAFLD risk by conducting two-sample MR and meta-analysis based on multiple sets of genome-wide association study (GWAS) summary statistics. Considering sex differences in the prevalence, risk factors, severity, and clinical outcomes of NAFLD [18], we further investigated sex-specific causal relationships. Multivariate MR (MVMR) was used to adjust for smoking status and alcohol consumption to confirm the results. Dyslipidemia plays a key role in the development of NAFLD [19], and two-stage MR was performed to explore whether lipid metabolism (low-density lipoprotein cholesterol [LDL-C] and triglyceride [TG]) mediated the association of iron homeostasis imbalance with NAFLD.

2. Materials and Methods

2.1 Study Design

Fig. 1 shows the study design. Independent instrumental variables for iron homeostasis were retrieved separately from the GWAS performed by Bell et al. (2021) [20] and Benyamin et al. (2014) [21], whereas the summary statistics of SNP-NAFLD associations were retrieved separately from the GWAS performed by de Fairfield et al. (2022) [22], FinnGen (2021) (https://gwas.mrcieu.ac.uk/datasets/finn-b-NAFLD/) and Anstee et al. (2020) [23] to represent the discovery cohort, replication-1 cohort, and replication-2 cohort, respectively. We first examined the associations of iron homeostasis indicators (2021) with NAFLD using two-sample MR in the large discovery dataset and then performed two replication analyses in two independent datasets. To increase the power of the analysis, we combined estimates from three data sources. We performed the same procedure for iron homeostasis indicators (2014). Reverse direction MR, multivariable MR, and two-step MR were performed to estimate reverse causation, causal effects independent of cigarettes or drinking, and the mediating effect of lipid metabolism, respectively. The analytic process followed the STROBE-MR guidelines [24]. MR analyses depend on three key assumptions: assumption one, instrumental variants (IVs) are strongly related to the risk factor (i.e., the relevance assumption); assumption two, genetic variants are not associated with any confounding factors in the exposure-outcome association (i.e., the independence assumption); assumption three, IVs affect outcomes only through exposure (i.e., the exclusion restriction assumption).

2.2 Genetic Instrument Selection and Data Sources

Detailed information on GWASs of studied exposures, mediators and outcomes is presented in Table 1. There is no sample overlap between exposures and outcomes in this study. For each phenotype, we directly extracted independent genome-wide significant (p 5 \times 10^-8) single-nucleotide polymorphisms (SNPs) reported in the corresponding source literature as IVs. The complete list of IVs is summarized in Supplementary Tables 1–3 (B) . In our analysis, when IVs were not present in the outcome GWAS, we used LDproxy to identify linkage disequilibrium (LD) proxies (r² > 0.9). Finally, in univariable MR analysis, we used the TwoSampleMR R package to undertake a harmonization procedure for integrating IV information between exposure and outcome [25]. We also removed IVs that were palindromic with intermediate allele frequencies. Then, we applied Steiger filtering to ensure that each IV explained more phenotypic variance in the exposure than in the outcome and removed genetic variants that did not satisfy this criterion [25]. The IVs with a “False” Steiger direction were excluded. Supplementary Tables 4–12 (B) contain further information on the harmonized datasets utilized in the current MR analysis.

Summary-level data (i.e., beta coefficient and corresponding standard error) for the associations of exposure-associated IVs with NAFLD were extracted from GWAS including 4761 NAFLD cases and 373,227 noncases (discovery stage), GWAS including 894 cases and 217,898 noncases (replication-1 stage), and another GWAS including 1483 NAFLD cases and 17,781 noncases (replication-2 stage). The case definition and exclusion criteria in the included NAFLD GWASs are shown in Supplementary Table 13 (B).

2.3 Statistical Analysis

2.3.1 Univariable Mendelian Randomization

Univariable two-sample MR was used to estimate the total effect of genetic liability to iron homeostasis indicators on NAFLD. We used the multiplicative random effects inverse-variance weighted (IVW) method to generate effect estimates as the main outcome [26]. If the essential assumptions of the MR are satisfied, this technique delivers the best statistical power. To examine and validate these assumptions, we ran a series of sensitivity analyses using the simple median [27], weighted median [27], weighted mode [28], MR-RAPS [29], MR-DIVW [30], and Mendelian randomization Pleiotropy Residual Sum and Outlier (MR-PRESSO) [31] techniques, as each approach makes different assumptions on instrument validity. Univariable two-sample MR analysis was performed using the TwoSampleMR (version 0.5.6), MendelianRandomization (version 0.6.0), Mr.raps (version 0.3.1), MRPRESSO (version 1.0) and MR-CAUSE (version 1.2.0) packages in the R environment (version 3.6.1, R Foundation for Statistical Computing, Vienna, Austria). Combined estimates from three data sources were achieved by the meta package in R. All statistical tests were two-sided. To account for multiple comparisons, we applied Bonferroni correction to adjust the p values in our analysis.

2.3.2 Multivariable Mendelian Randomization

We also performed multivariable Mendelian randomization (MVMR) to assess the causal effect of genetic liability to iron homeostasis indicators on NAFLD independent of adjustment variables. We selected cigarettes per day or drinking per week as adjustment variables. For each MVMR analysis, 54 and 91 IVs for cigarettes per day and drinking per week were added to the models, respectively. We performed an IVW-MVMR regression, incorporating one exposure and one adjustment variable into the regression model. We used conditional F statistics to assess the validity and strength of the instrumental variables. MVMR analyses were performed using the MVMR package (https://github.com/WSpiller/MVMR, version 0.3.0) in the R environment.

2.3.3 Two-Step Mendelian Randomization

To further assess the mediation effects of lipid metabolism on the causal relationship between iron homeostasis indicators and NAFLD, we performed a two-step MR mediation analysis. We selected low-density lipoprotein cholesterol (LDL-C) and triglyceride (TG) as potential mediators. Two-step MR is based on the coefficient product method to calculate indirect (or mediator) effects. This process calculates two MR estimates, one for the causal effect of exposure on the mediator and the other for the causal effect of the mediator on the outcome. These two estimates are then multiplied together to estimate the indirect effect [32].

2.3.4 Assessing Instrument Strength

We assessed the strength of the instruments by calculating their F statistic. F statistics can be calculated using the formula F=\frac{n-k-1}{k}\mathrm{·}\frac{r^2}{1-r^2} (n is the sample size, k is the number of IVs, and r^2 refers to how much phenotypic variation can be explained by the set of genetic instruments) [33]. Given that r^2 is not generally provided in GWAS summary data, we used the formula r^2=\sum 2\mathrm{·}{\beta }^2\mathrm{·}f\mathrm{·}\left(1-f\right) (f is the minor allele frequency, and \beta is the effect estimate for each SNP) [34] to obtain r^2 estimates.

2.3.5 Assessing Horizontal Pleiotropy and Effect Heterogeneity

To further assess the robustness of the findings, we conducted further tests for horizontal pleiotropy. Estimates of intercept in MR-Egger regression were used to evaluate the directional pleiotropy of IVs [35]. The MR-PRESSO method can detect and correct for horizontal pleiotropy by removing outliers [31]. To quantify heterogeneity across individual causal effects, Cochran’s Q was calculated, with a p value \le 0.05 indicating the presence of heterogeneity; as a result, a random-effects IVW model should be used [36]. The I^2 statistic was also used to assess whether causal estimates from different genetic variants were comparable (i.e., heterogeneous) [37]. Typically, a value of I^2 between 0 and 0.25 is considered low heterogeneity; between 0.25 and 0.5 is considered moderate heterogeneity, and above 0.5 is considered high heterogeneity.

2.3.6 Graphical Visualization

We also created informative graphs to show the results. Scatter plots contrast SNP-outcome relationships with SNP-exposure associations in a graph that can provide a quick image of the causal effect estimates of different MR approaches. Forest plots are mostly used to determine the occurrence of heterogeneity in IVs. Funnel plots provide a visual assessment of the degree of pleiotropy balance of the instruments utilized, with symmetry providing evidence against directional pleiotropy. The leave-one-out plots were created to evaluate influential outliers.

3. Results

3.1 Statistical Power, Heterogeneity and Pleiotropy Assessment

The mean F statistic and power calculation are shown in Supplementary Table 1 (supplementary material A). The mean F statistic for each set of instruments exceeds 10, indicating the absence of weak instrument bias. Across combinations of the iron homeostasis indicators and NAFLD outcome, we had 80% power to detect ORs as small as 1.12–1.46. High heterogeneity was observed in the analyses for ferritin (2021) and TAST (2021) in the discovery analyses, as well as for TIBC (2021) in the sex-specific analysis (Supplementary Table 2, A). The results of graphical diagnostics also revealed no significant heterogeneity in the analysis of the four indicators (Supplementary Figs. 1–4, A). Thus, we should note whether these associations remained consistent after removal of outlier variants in MR-PRESSO analysis. No pleiotropy (p value for Egger intercept >0.05) was detected in the associations in the discovery, replication, and sex-specific analysis.

3.2 Mendelian Randomization between Iron Homeostasis and NAFLD

The primary (IVW) and sensitivity analyses in the discovery dataset are shown in Fig. 2. The IVW method indicated that a genetically predicted one-SD increase in serum iron (7.76 µmol/L) was significantly associated with an increased risk of NAFLD (OR [95% CI], 1.286 [1.075–1.539], p =0.0059). The OR of NAFLD was 1.223 [1.067, 1.402] (p = 0.0039) for genetically predicted one-SD increase of TSAT (13.25%). IVW indicated that no significant associations were found between ferritin or TIBC and the risk of NAFLD. High heterogeneity was observed in the analysis for ferritin (p value for Q = 1.27 \times 10^-5, I square = 0.581, Supplementary Table 2, A). However, the results from MR-PRESSO after outlier correction showed that the risk of NAFLD was slightly increased (OR [95% CI], 1.260 [1.050, 1.500], p = 0.0195) for a one-SD increase in ferritin (1.08 µg/L). The effect estimates of sensitivity analyses were directionally consistent with IVW.

The combined results were similar to the results in the discovery datasets, as exhibited in Supplementary Fig. 5 (A) and Supplementary Table 14b–f (B). In the reverse MR analysis, genetic liability to NAFLD (2022) was not significantly associated with four iron homeostasis indicators (2021) (Supplementary Table 14a, B), indicating that there was no causal effect of genetic liability to NAFLD on iron homeostasis. As shown in Supplementary Table 3 (A), when cigarettes per day or drinking per week were included in the MVMR model, there was still evidence that a genetically predicted serum iron or TAST had a significant effect on the risk of NAFLD.

3.3 Sex-Specific Association between Iron Homeostasis and NAFLD

Using the IVW method, sex-specific association found that the OR of NAFLD dramatically increased to 1.723 [1.338, 2.219] (p = 2.48 \times 10^-5) for genetically predicted ferritin increase in males, while the result in females was similar to that in the whole cohort (OR [95% CI], 1.383 [1.068, 1.790], p = 0.0139) (Fig. 3 and Supplementary Table 14g,h, B). Genetically predicted increase in serum iron led to 1.327 [1.047, 1.682] (p = 0.0194) times the risk of NAFLD among males and 1.324 [1.022, 1.716] (p = 0.0336) times the risk among females. Additionally, genetically predicted increases in TSAT related to 1.236 [1.042, 1.465] (p = 0.0149) times among males and 1.253 [1.038, 1.512] (p = 0.0186) times among females. The Cochran Q test showed no evidence of heterogeneity in the sex-specific effect of ferritin, serum iron or TAST on NAFLD.

3.4 Mediation Effect of Dyslipidemia

Two-step MR analysis was performed to explain the mediation effect of dyslipidemia (LDL-C and TG) on the causal relationship between iron homeostasis and NAFLD (Table 2a,2b and Supplementary Table 15, B). Genetically predicted one-SD increase in serum iron was associated with 0.108 mg/dL lower LDL-C, and genetically predicted one-SD increase in LDL-C (44.1 mg/dL) was associated with 0.802 times the risk of NAFLD. In contrast, a genetically predicted one-SD increase in serum iron was associated with a 0.029 mg/dL increase in TG, and a genetically predicted one-SD increase in TG (100.2 mg/dL) was associated with a 1.420-fold increased risk of NAFLD. The proportion of the effect of serum iron on NAFLD mediated by LDL-C or TG levels was 9.52% and 3.97%, respectively. The effect of TAST on mediators (LDL-C and TG) and mediators on NAFLD was similar to that of serum iron, and the mediation proportion was 9.95% and 4.98% for LDL-C and TG, respectively.

4. Discussion

To the best of our knowledge, no previous study has carried out a MR analysis to explore the effect of iron status on NAFLD. This MR study found that genetic predisposition to higher levels of serum ferritin, serum iron or TSAT was associated with an increased risk of NAFLD, independent of genetically predicted smoking and drinking, and mediated by LDL-C and TG. There were no inverse associations of genetically predicted NAFLD risk with iron homeostasis indicators. Notably, the causal relationships between ferritin and NAFLD showed sex-specific differences and were particularly statistically significant in males.

The epidemiological investigations of iron homeostasis indicators and NAFLD outcomes are summarized in Supplementary Table 4 (A). Most research has focused on the association between serum ferritin and NAFLD severity in patients, with limited attention to its relationship with NAFLD risk in the general population [4, 5, 6, 7]. Only one large population-based study uncovered a positive association between serum ferritin and NAFLD risk, as well as a negative association between serum iron and NAFLD [11]. Serum iron, on the other hand, has been linked to a higher risk of hepatocellular carcinoma in NAFLD patients [8] and abnormally elevated ALT in the general population [9]. According to the current research background, the causal relationship between iron status indicators and NAFLD is ambiguous. This study established the causal direction of increased serum ferritin, serum iron, and TSAT to an increased risk of NAFLD.

As shown in Fig. 4, iron status involves multiple-hit progression of NAFLD. The initial hit is insulin resistance, which leads to an increased uptake and synthesis of free fatty acids stored as TG in hepatocytes, resulting in simple steatosis. Increased hepatic iron levels aggravated insulin resistance, and increased lipogenesis and gluconeogenesis in a genetically obese and iron overload mouse model [38]. The second hit is the production of oxidative stress and activation of inflammatory pathways, which accelerate the progression from steatosis to nonalcoholic steatohepatitis (NASH). Handa et al. [39] found that dietary iron induced hepatic oxidative stress, inflammation, and immune cell activation in a genetically obese mouse model. Iron accumulation promotes inflammation by altering macrophage polarization status [40]. The third hit is fibrosis, which progresses from NASH to cirrhosis. Persistent activation of hepatic stellate cells (HSCs) is an early and key event in fibrosis. Several studies have reported that iron promotes fibrosis by activating HSCs [41, 42]. Additionally, the experimental study in Caenorhabditis elegans revealed that iron overload activates sgk-1, encoding serum- and tissue-specific glucocorticoid-inducible kinase, which upregulates acs-20 and vit-2/3—homologous to mammalian fatty acid transport proteins (FATP1/4) and yolk lipoprotein genes [43]. This regulation enhances lipid uptake and facilitates lipid storage in lipid droplets, establishing a direct link between iron metabolism and lipid accumulation. Complementarily, clinical investigations in dialysis patients demonstrate a correlation between hepatic iron overload and increased hepatic proton density fat fraction (PDFF), as measured by MRI [44]. These findings indicate that iron deposition aggravates hepatic lipid accumulation, potentially through oxidative stress and dysregulation of lipid metabolism pathways. Together, these studies suggest that iron overload not only acts as a “first hit” by initiating lipid accumulation but also serves as a “second hit” by amplifying hepatic injury and accelerating NAFLD progression.

The causal relationship from genetically predicted ferritin to NAFLD exhibited sex differences, implying that males were more vulnerable to increased ferritin, which could be explained by the protective role of endogenous estrogens in female. Among younger patients, the prevalence of both NAFLD and NASH was 2–3 times higher in men than in women, while the prevalence of NASH was higher in women after 60 years of age [45]. An epidemiologic study consistently found that menopause, as a natural state of estrogen deficiency, was associated with an increased risk of hepatic steatosis [46].

The study discovered that elevated TG mediated approximately 3%–5% of the observed effect of serum iron and TSAT on NAFLD risk, while decreased LDL-C mediated 9%–10%. A previous MR study also found that dyslipidemia characterized by low levels of LDL-C and high levels of TG was associated with an increased risk of NAFLD [47]. Moreover, sufficient evidence found that iron overload caused lipid metabolism disorders, especially elevated circulating TG [48, 49, 50, 51]. The plasma nonesterified fatty acid (NEFA) pool is the primary source of fatty acids for TG synthesis in the liver. Adipose tissue via TG lipolysis is the primary contributor to the plasma NEFA pool. A retrospective study, including 740 patients with NAFLD, found that steatosis severity was higher in those with low LDL-C [52], which impaired lipid export from hepatocytes. Elevated TG and decreased LDL-C mediate iron overload-induced imbalances in fatty acid uptake, synthesis, and oxidative and secretory pathways in the liver, eventually leading to NAFLD.

From a clinical perspective, this study provides valuable insights into the clinical implications of iron homeostasis in the pathogenesis of NAFLD, offering evidence of its significant role in disease progression and potential therapeutic avenues. The identification of TG and LDL-C as partial mediators’ points to the potential utility of targeted lipid-lowering interventions to mitigate the risk of NAFLD in individuals with elevated iron parameters. Furthermore, the observed sex differences in the causal association between ferritin and NAFLD risk emphasize the necessity of sex-specific considerations in disease prevention and management strategies.

The most notable strength of the study is that extensive sensitivity analyses and replication studies were performed. In addition, we used sex-specific MR, MVMR, and mediation analysis to dig deeper into the correlation mechanism between iron homeostasis and NAFLD. However, limitations should be kept in mind when interpreting our results. The major issue for any MR study is horizontal pleiotropy, which means that certain genetic instrument variables influence the risk of outcome not through exposure but through other alternative pathways. However, various measures were employed to minimize the effects of pleiotropy, although residual confounding may still exist. In addition, the nonlinear association and gene‒environmental interaction could not be estimated in summary-level genetic statistics. At last, although we performed a stratified analysis of sex by making full use of the data, a stratified analysis of other factors could not be achieved, such as age and BMI. Future research should aim to explore gene-environment interactions. Additionally, clinical interventions targeting iron metabolism, such as iron chelation therapies or modulation of key iron-regulatory genes, may offer potential therapeutic strategies for NAFLD, particularly in individuals with iron overload.

5. Conclusion

In conclusion, this study presents MR evidence supporting the causal roles of elevated serum iron, ferritin and TAST in the development of NAFLD, as well as the mediation effects of increased TG and decreased LDL-C. The causal relationship between serum ferritin and NAFLD risk was stronger in males. These findings underscore the importance of managing iron homeostasis and lipid homeostasis as an effective strategy of NAFLD treatment and concerns for sensitive populations.

Availability of Data and Materials

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

References

[1] Chalasani N, Younossi Z, Lavine JE, Diehl AM, Brunt EM, Cusi K, et al. The diagnosis and management of non-alcoholic fatty liver disease: practice Guideline by the American Association for the Study of Liver Diseases, American College of Gastroenterology, and the American Gastroenterological Association. Hepatology. 2012; 55: 2005–2023. https://doi.org/10.1002/hep.25762.

[2] Le MH, Yeo YH, Li X, Li J, Zou B, Wu Y, et al. 2019 Global NAFLD Prevalence: A Systematic Review and Meta-analysis. Clinical Gastroenterology and Hepatology. 2022; 20: 2809–2817.e28. https://doi.org/10.1016/j.cgh.2021.12.002.

[3] Muñoz M, Villar I, García-Erce JA. An update on iron physiology. World Journal of Gastroenterology. 2009; 15: 4617–4626. https://doi.org/10.3748/wjg.15.4617.

[4] Kowdley KV, Belt P, Wilson LA, Yeh MM, Neuschwander-Tetri BA, Chalasani N, et al. Serum ferritin is an independent predictor of histologic severity and advanced fibrosis in patients with nonalcoholic fatty liver disease. Hepatology. 2012; 55: 77–85. https://doi.org/10.1002/hep.24706.

[5] Hagström H, Nasr P, Bottai M, Ekstedt M, Kechagias S, Hultcrantz R, et al. Elevated serum ferritin is associated with increased mortality in non-alcoholic fatty liver disease after 16 years of follow-up. Liver International. 2016; 36: 1688–1695. https://doi.org/10.1111/liv.13144.

[6] Angulo P, George J, Day CP, Vanni E, Russell L, De la Cruz AC, et al. Serum ferritin levels lack diagnostic accuracy for liver fibrosis in patients with nonalcoholic fatty liver disease. Clinical Gastroenterology and Hepatology. 2014; 12: 1163–1169.e1. https://doi.org/10.1016/j.cgh.2013.11.035.

[7] Bugianesi E, Manzini P, D’Antico S, Vanni E, Longo F, Leone N, et al. Relative contribution of iron burden, HFE mutations, and insulin resistance to fibrosis in nonalcoholic fatty liver. Hepatology. 2004; 39: 179–187. https://doi.org/10.1002/hep.20023.

[8] Yu YC, Luu HN, Wang R, Thomas CE, Glynn NW, Youk AO, et al. Serum Biomarkers of Iron Status and Risk of Hepatocellular Carcinoma Development in Patients with Nonalcoholic Fatty Liver Disease. Cancer Epidemiology, Biomarkers & Prevention. 2022; 31: 230–235. https://doi.org/10.1158/1055-9965.EPI-21-0754.

[9] Ruhl CE, Everhart JE. Relation of elevated serum alanine aminotransferase activity with iron and antioxidant levels in the United States. Gastroenterology. 2003; 124: 1821–1829. https://doi.org/10.1016/s0016-5085(03)00395-0.

[10] Bertol FS, Araujo B, Jorge BB, Rinaldi N, De Carli LA, Tovo CV. Role of micronutrients in staging of nonalcoholic fatty liver disease: A retrospective cross-sectional study. World Journal of Gastrointestinal Surgery. 2020; 12: 269–276. https://doi.org/10.4240/wjgs.v12.i6.269.

[11] Yang HH, Chen GC, Li DM, Lan L, Chen LH, Xu JY, et al. Serum iron and risk of nonalcoholic fatty liver disease and advanced hepatic fibrosis in US adults. Scientific Reports. 2021; 11: 10387. https://doi.org/10.1038/s41598-021-89991-x.

[12] Hamabe A, Uto H, Imamura Y, Kusano K, Mawatari S, Kumagai K, et al. Impact of cigarette smoking on onset of nonalcoholic fatty liver disease over a 10-year period. Journal of Gastroenterology. 2011; 46: 769–778. https://doi.org/10.1007/s00535-011-0376-z.

[13] Zein CO, Unalp A, Colvin R, Liu YC, McCullough AJ. Smoking and severity of hepatic fibrosis in nonalcoholic fatty liver disease. Journal of Hepatology. 2011; 54: 753–759. https://doi.org/10.1016/j.jhep.2010.07.040.

[14] Lee CH, Goag EK, Lee SH, Chung KS, Jung JY, Park MS, et al. Association of serum ferritin levels with smoking and lung function in the Korean adult population: analysis of the fourth and fifth Korean National Health and Nutrition Examination Survey. International Journal of Chronic Obstructive Pulmonary Disease. 2016; 11: 3001–3006. https://doi.org/10.2147/COPD.S116982.

[15] Chełchowska M, Ambroszkiewicz J, Gajewska J, Jabłońska-Głąb E, Maciejewski TM, Ołtarzewski M. Hepcidin and Iron Metabolism in Pregnancy: Correlation with Smoking and Birth Weight and Length. Biological Trace Element Research. 2016; 173: 14–20. https://doi.org/10.1007/s12011-016-0621-7.

[16] Wongtrakul W, Niltwat S, Charatcharoenwitthaya P. The Effects of Modest Alcohol Consumption on Non-alcoholic Fatty Liver Disease: A Systematic Review and Meta-Analysis. Frontiers in Medicine. 2021; 8: 744713. https://doi.org/10.3389/fmed.2021.744713.

[17] Ioannou GN, Dominitz JA, Weiss NS, Heagerty PJ, Kowdley KV. The effect of alcohol consumption on the prevalence of iron overload, iron deficiency, and iron deficiency anemia. Gastroenterology. 2004; 126: 1293–1301. https://doi.org/10.1053/j.gastro.2004.01.020.

[18] Lonardo A, Nascimbeni F, Ballestri S, Fairweather D, Win S, Than TA, et al. Sex Differences in Nonalcoholic Fatty Liver Disease: State of the Art and Identification of Research Gaps. Hepatology. 2019; 70: 1457–1469. https://doi.org/10.1002/hep.30626.

[19] Bessone F, Razori MV, Roma MG. Molecular pathways of nonalcoholic fatty liver disease development and progression. Cellular and Molecular Life Sciences. 2019; 76: 99–128. https://doi.org/10.1007/s00018-018-2947-0.

[20] Bell S, Rigas AS, Magnusson MK, Ferkingstad E, Allara E, Bjornsdottir G, et al. A genome-wide meta-analysis yields 46 new loci associating with biomarkers of iron homeostasis. Communications Biology. 2021; 4: 156. https://doi.org/10.1038/s42003-020-01575-z.

[21] Benyamin B, Esko T, Ried JS, Radhakrishnan A, Vermeulen SH, Traglia M, et al. Novel loci affecting iron homeostasis and their effects in individuals at risk for hemochromatosis. Nature communications. 2014; 5: 4926. https://doi: 10.1038/ncomms5926.

[22] Fairfield CJ, Drake TM, Pius R, Bretherick AD, Campbell A, Clark DW, et al. Genome-Wide Association Study of NAFLD Using Electronic Health Records. Hepatology Communications. 2022; 6: 297–308. https://doi.org/10.1002/hep4.1805.

[23] Anstee QM, Darlay R, Cockell S, Meroni M, Govaere O, Tiniakos D, et al. Genome-wide association study of non-alcoholic fatty liver and steatohepatitis in a histologically characterised cohort. Journal of Hepatology. 2020; 73: 505–515. https://doi.org/10.1016/j.jhep.2020.04.003.

[24] Skrivankova VW, Richmond RC, Woolf BAR, Yarmolinsky J, Davies NM, Swanson SA, et al. Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization: The STROBE-MR Statement. JAMA. 2021; 326: 1614–1621. https://doi.org/10.1001/jama.2021.18236.

[25] Hemani G, Tilling K, Davey Smith G. Orienting the causal relationship between imprecisely measured traits using GWAS summary data. PLoS Genetics. 2017; 13: e1007081. https://doi.org/10.1371/journal.pgen.1007081.

[26] Bowden J, Spiller W, Del Greco M F, Sheehan N, Thompson J, Minelli C, et al. Improving the visualization, interpretation and analysis of two-sample summary data Mendelian randomization via the Radial plot and Radial regression. International Journal of Epidemiology. 2018; 47: 1264–1278. https://doi.org/10.1093/ije/dyy101.

[27] Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genetic Epidemiology. 2016; 40: 304–314. https://doi.org/10.1002/gepi.21965.

[28] Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. International Journal of Epidemiology. 2017; 46: 1985–1998. https://doi.org/10.1093/ije/dyx102.

[29] Zhao Q, Wang J, Hemani G, Bowden J, Small DS. Statistical inference in two-sample summary-data Mendelian randomization using robust adjusted profile score. Annals of Statistics. 2020; 48: 1742–1769. https://doi.org/10.1214/19-AOS1866.

[30] Ye T, Shao J, Kang H. Debiased inverse-variance weighted estimator in two-sample summary-data Mendelian randomization. The Annals of Statistics. 2021; 49: 2079–2100. https://doi.org/10.1214/20-AOS2027.

[31] Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nature Genetics. 2018; 50: 693–698. https://doi.org/10.1038/s41588-018-0099-7.

[32] Carter AR, Sanderson E, Hammerton G, Richmond RC, Davey Smith G, Heron J, et al. Mendelian randomisation for mediation analysis: current methods and challenges for implementation. European Journal of Epidemiology. 2021; 36: 465–478. https://doi.org/10.1007/s10654-021-00757-1.

[33] Burgess S, Thompson SG, CRP CHD Genetics Collaboration. Avoiding bias from weak instruments in Mendelian randomization studies. International Journal of Epidemiology. 2011; 40: 755–764. https://doi.org/10.1093/ije/dyr036.

[34] Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genetic Epidemiology. 2016; 40: 597–608. https://doi.org/10.1002/gepi.21998.

[35] Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. International Journal of Epidemiology. 2015; 44: 512–525. https://doi.org/10.1093/ije/dyv080.

[36] Bowden J, Del Greco M F, Minelli C, Davey Smith G, Sheehan N, Thompson J. A framework for the investigation of pleiotropy in two-sample summary data Mendelian randomization. Statistics in Medicine. 2017; 36: 1783–1802. https://doi.org/10.1002/sim.7221.

[37] Ioannidis JPA, Patsopoulos NA, Evangelou E. Heterogeneity in meta-analyses of genome-wide association investigations. PLoS ONE. 2007; 2: e841. https://doi.org/10.1371/journal.pone.0000841.

[38] Altamura S, Müdder K, Schlotterer A, Fleming T, Heidenreich E, Qiu R, et al. Iron aggravates hepatic insulin resistance in the absence of inflammation in a novel db/db mouse model with iron overload. Molecular Metabolism. 2021; 51: 101235. https://doi.org/10.1016/j.molmet.2021.101235.

[39] Handa P, Morgan-Stevenson V, Maliken BD, Nelson JE, Washington S, Westerman M, et al. Iron overload results in hepatic oxidative stress, immune cell activation, and hepatocellular ballooning injury, leading to nonalcoholic steatohepatitis in genetically obese mice. American Journal of Physiology. Gastrointestinal and Liver Physiology. 2016; 310: G117–27. https://doi.org/10.1152/ajpgi.00246.2015.

[40] Handa P, Thomas S, Morgan-Stevenson V, Maliken BD, Gochanour E, Boukhar S, et al. Iron alters macrophage polarization status and leads to steatohepatitis and fibrogenesis. Journal of Leukocyte Biology. 2019; 105: 1015–1026. https://doi.org/10.1002/JLB.3A0318-108R.

[41] Houglum K, Bedossa P, Chojkier M. TGF-beta and collagen-alpha 1 (I) gene expression are increased in hepatic acinar zone 1 of rats with iron overload. The American Journal of Physiology. 1994; 267: G908–G913. https://doi.org/10.1152/ajpgi.1994.267.5.G908.

[42] Carthew P, Edwards RE, Smith AG, Dorman B, Francis JE. Rapid induction of hepatic fibrosis in the gerbil after the parenteral administration of iron-dextran complex. Hepatology. 1991; 13: 534–539.

[43] Wang H, Jiang X, Wu J, Zhang L, Huang J, Zhang Y, et al. Iron Overload Coordinately Promotes Ferritin Expression and Fat Accumulation in Caenorhabditis elegans. Genetics. 2016; 203: 241–253. https://doi.org/10.1534/genetics.116.186742.

[44] Rostoker G, Loridon C, Griuncelli M, Rabaté C, Lepeytre F, Ureña-Torres P, et al. Liver Iron Load Influences Hepatic Fat Fraction in End-Stage Renal Disease Patients on Dialysis: A Proof of Concept Study. eBioMedicine. 2019; 39: 461–471. https://doi.org/10.1016/j.ebiom.2018.11.020.

[45] Hashimoto E, Tokushige K. Prevalence, gender, ethnic variations, and prognosis of NASH. Journal of Gastroenterology. 2011; 46: 63–69. https://doi.org/10.1007/s00535-010-0311-8.

[46] Völzke H, Schwarz S, Baumeister SE, Wallaschofski H, Schwahn C, Grabe HJ, et al. Menopausal status and hepatic steatosis in a general female population. Gut. 2007; 56: 594–595. https://doi.org/10.1136/gut.2006.115345.

[47] Yuan S, Chen J, Li X, Fan R, Arsenault B, Gill D, et al. Lifestyle and metabolic factors for nonalcoholic fatty liver disease: Mendelian randomization study. European Journal of Epidemiology. 2022; 37: 723–733. https://doi.org/10.1007/s10654-022-00868-3.

[48] Kim J, Jia X, Buckett PD, Liu S, Lee CH, Wessling-Resnick M. Iron loading impairs lipoprotein lipase activity and promotes hypertriglyceridemia. FASEB Journal. 2013; 27: 1657–1663. https://doi.org/10.1096/fj.12-224386.

[49] Brunet S, Thibault L, Delvin E, Yotov W, Bendayan M, Levy E. Dietary iron overload and induced lipid peroxidation are associated with impaired plasma lipid transport and hepatic sterol metabolism in rats. Hepatology. 1999; 29: 1809–1817. https://doi.org/10.1002/hep.510290612.

[50] Tang Y, Wang D, Zhang H, Zhang Y, Wang J, Qi R, et al. Rapid responses of adipocytes to iron overload increase serum TG level by decreasing adiponectin. Journal of Cellular Physiology. 2021; 236: 7544–7553. https://doi.org/10.1002/jcp.30391.

[51] Setoodeh S, Khorsand M, Takhshid MA. The effects of iron overload, insulin resistance and oxidative stress on metabolic disorders in patients with β- thalassemia major. Journal of Diabetes and Metabolic Disorders. 2020; 19: 767–774. https://doi.org/10.1007/s40200-020-00560-x.

[52] Mouzaki M, Shah A, Arce-Clachar AC, Hardy J, Bramlage K, Xanthakos SA. Extremely low levels of low-density lipoprotein potentially suggestive of familial hypobetalipoproteinemia: A separate phenotype of NAFLD? Journal of Clinical Lipidology. 2019; 13: 425–431. https://doi.org/10.1016/j.jacl.2019.02.002.

Funding

Sanming Project of Medicine in Shenzhen(SZSM202111001)