Nonalcoholic fatty liver disease (NAFLD) is a clinicopathological syndrome characterized by excessive fat deposition in hepatocytes except for alcohol and other clear liver damage factors, closely related to insulin resistance and genetic susceptibility. It is also the most common cause of chronic liver disease, with a global prevalence of 25%. (Younossi et al., 2016; Cheung et al., 2019; Harrison et al., 2021). NAFLD can progress to non-alcoholic steatohepatitis (NASH) (Barb et al., 2016; Yin et al., 2023) and even hepatocellular carcinoma (HCC) (Feldstein et al., 2009). The prevalence of NAFLD is increasing yearly, but less than 5% of people with NAFLD are aware of their disease status (Alqahtani et al., 2021). No recognized and reliable drug therapies exist, posing a substantial global public health challenge (Younossi et al., 2019; Paternostro and Trauner, 2022). Early recognition and prevention of NAFLD are, therefore, significant.

Diagnosing NAFLD requires costly imaging and invasive procedures, which impose a significant socioeconomic burden (Dorairaj et al., 2021). Patients with NAFLD often have a combination of type 2 diabetes mellitus (T2DM) (Tilg et al., 2017; Targher et al., 2021), metabolic syndrome (MS) (Tapper and Loomba, 2018), and cardiovascular events (Mantovani et al., 2022; Muzurović et al., 2022), with relatively significant changes in their blood metabolomics. However, as there are no susceptible and specific tests to diagnose NAFLD and differentiate NASH from pure steatosis16, there is a lack of reliable biomarkers to assess the progression of NAFLD (Masoodi et al., 2021). Therefore, further relevant studies (e.g., blood metabolomics) are needed to identify biological markers associated with NAFLD, which can provide a basis for diagnosing and treating NAFLD (Piazzolla and Mangia, 2020; Dorairaj et al., 2021).

NAFLD has a complex and multifaceted biochemical metabolic process. In recent years, more studies have suggested that blood biomarkers of NAFLD that are not regulated by secondary non-causal pathways may be promising candidates for identifying individuals at risk (Gobeil et al., 2022). The National Institutes of Health Medical Library (NIH) shows that 228 of the current 1,230 studies on NAFLD are related to blood metabolism. These studies suggested that certain metabolites are involved in the progression of NAFLD. Some guidelines (Vos et al., 2017; Cusi et al., 2022) demonstrate that blood transaminase levels, gamma-glutamyl transferase, serum triglycerides, and a 4-factor-based fibrosis index (FIB-4) can be applied to aid in the assessment and diagnosis of NAFLD but continue to lack specificity. In a study of NAFLD patients who fasted overnight (Kalhan et al., 2011), significantly elevated levels of glycocholate, taurocholate and glycoglycolate were found in NAFLD participants. Masoodi et al. (2021) found changes in circulating fatty acids, triglycerides, phospholipids, and bile acids in NAFLD patients; Gorden et al. (2015) used linear discriminant analysis of a set of 20 plasma metabolites (including glycerophospholipids, sphingolipids, etc.) that can be used to differentiate NASH from simple steatosis offers the potential to improve the clinical diagnosis of NAFLD and facilitate therapeutic interventions. In addition, Gobeil et al. (2022) aimed to identify novel biomarkers of NAFLD through Mendelian randomization. This analysis suggests that a potential causal relationship was revealed between tyrosine levels and NAFLD in a positive manner, which may represent a novel clinical biomarker for NAFLD. However, systematic studies to assess the causal relationship between blood metabolites and NAFLD are still lacking and translating these metabolic findings into pathophysiological mechanisms and new therapies is an enormous challenge. Therefore, a more comprehensive analysis of the interactions between genetics and blood-circulating metabolites in the pathogenesis of NAFLD is still needed.

Mendelian randomization (MR) is a recent and widely used method for epidemiological investigations in which single nucleotide polymorphisms (SNPs) are pooled to infer the causal effect of exposure factors on outcomes (Emdin et al., 2017; Zheng et al., 2017; Hemani et al., 2018). MR uses a genetic variation to simulate the design of randomized controlled trials (RCTs), and genome-wide association studies (GWASs) can be used for flexible two-sample MR analysis. In the case of high cost, time-consuming, and even low feasibility of RCT trials, MR can be used as an alternative to RCT because it relies on the natural random classification of genetic variation during meiosis to generate a random distribution of genetic variation in the population (Richmond and Davey Smith, 2022). It provides reliable evidence for the causal relationship between phenotypes (Zuccolo and Holmes, 2017). MR can also be used to identify biomarkers of disease-related characteristics by determining whether genetic susceptibility to certain diseases affects other biological characteristics, such as blood metabolomics (Mohammadi-Shemirani et al., 2019; Ritchie et al., 2021).

Some MR studies have been performed to explore the relationship between exposure and NAFLD. However, the main focus was single exposures or common exposure factors, such as interleukin-6 (IL-6) (Li et al., 2022a), seven sleep characteristics (Fan et al., 2022), serum uric acid levels (Li et al., 2022b), iron status (He et al., 2022), and coronary artery disease (Ren et al., 2022). Few studies have focused on blood metabolites and NAFLD. Given that the causal relationship between blood metabolites and NAFLD is not well understood, this study used a two-sample MR approach to assess the causal relationship between 486 human blood metabolites and the risk of NAFLD to provide a deeper understanding of the pathogenesis of NAFLD.

Following the strict instrument selection steps, we performed MR analyses on 486 blood metabolites. Five of the 486 blood metabolites appeared in two forms, totaling 491 MR analyses. F statistics were all greater than the empirical threshold 10, with a minimum of 17, suggesting that all SNPs had sufficient validity. The F statistic for all SNPs was shown in Supplementary Table S2.

As NAFLD is a metabolic stress liver injury closely related to genetic susceptibility, the influence of genetic factors on hepatic steatosis has been reported in recent experimental and observational studies (Younossi et al., 2016; Martin et al., 2021; Oliveira et al., 2021). In this study, we performed an unbiased two-sample MR analysis to causally assess 486 blood metabolites and the risk of NAFLD. We collected the most extensive mGWAS and large NAFLD GWAS summary data from public databases. We performed an initial IVW analysis of 486 metabolites using genetic variants as IVs and ultimately identified a causal relationship between 23 metabolites and NAFLD, 10 known metabolites. We then performed heterogeneity tests and sensitivity analyses on these metabolites. In addition, to further ensure the reliability and stability of the results, we used other databases for validation and performed a meta-analysis and metabolic pathway analysis. The results suggest that higher levels of biliverdin (OR = 1.45; 95% CI 1.20-1.75; p = 0.0001) are causally associated with an increased risk of NAFLD and that higher levels of myristoleate (OR = 0.57; 95% CI 0.39-0.83; p = 0.0030) play a protective role in the development of NAFLD. The graphical summary of this study shows in the Figure 5. To our knowledge, this is the first MR study to assess the causal role of human blood metabolites systematically and comprehensively in NAFLD.

The prevalence of NAFLD and the lack of pharmacological therapies place a heavy burden on the world, making the screening and prevention of NAFLD particularly important. Although guidelines (Cusi et al., 2022) have mentioned several complementary diagnostic indicators for NAFLD (ALT, GGT, etc.), there is still a lack of susceptible and specific diagnostic indicators for NAFLD and biomarkers to assess the progression of NAFLD. Given that the gold standard for diagnosing NAFLD remains liver biopsy, which is costly and invasive, further research is needed to find specific diagnostic indicators for NAFLD. Previous studies (Saeed et al., 2017; Zhao et al., 2020; Notarnicola et al., 2021) have identified various blood metabolite changes in NAFLD patients, such as imbalances in triglyceride metabolism, disturbances in vitamin A metabolism, and elevated plasma N-trimethyl-5-aminovaleric acid (TMAVA) levels in mice with hepatic steatosis. Despite the increasing number of studies on the blood metabolism of NAFLD, there is still a lack of comprehensive and systematic studies to assess the causal relationship between NAFLD and blood metabolites. Inspired by the work of Cai et al. (2022b), Yu et al. (2022), and Wang et al. (2021b) in exploring the causal relationship between metabolites and disease, we designed this MR analysis of NAFLD and blood metabolites to assess the causal relationship between the two systematically.

Our findings suggest that high biliverdin levels increase the risk of NAFLD progression, but there is a paucity of research on the direct correlation between biliverdin and NAFLD. Bilirubin is a by-product of hemoglobin catabolism. Hemoglobin is degraded by heme oxygenase-1 (HO-1) to biliverdin, which is rapidly converted to bilirubin by the action of biliverdin reductase A. Bilirubin and biliverdin reductase A (BVRA) have been found to protect the liver from lipid accumulation as well as disease (Lin et al., 2009; Kwak et al., 2012; Puri et al., 2013). Hepatic BVRA inhibits glycogen synthase kinase-3β by enhancing serine nine phosphorylation, thereby preventing hepatic steatosis (Hinds et al., 2016). After oxidation of bilirubin to biliverdin in the mitochondria, biliverdin must be exported to the cytoplasmic lysate for reduced bilirubin, and studies have found enhanced redox of bilirubin increases insulin resistance and steatosis in obese patients (Shum et al., 2021). From this, we can speculate that the increase in biliverdin levels indirectly reflects an enhanced bilirubin redox process, which affects the progression of NAFLD. However, a related study (Ikeda et al., 2011) found that biliverdin prevented the deterioration of abnormal glucose tolerance in mice with T2DM. NAFLD is commonly associated with metabolic risk factors such as obesity, dyslipidemia, hypertension, and diabetes, and the global rise in the prevalence of obesity and type 2 diabetes has coincided with an increase in the prevalence of NAFLD (Loomba et al., 2021); Biliverdin has antioxidant and anti-inflammatory properties (Shiels et al., 2020), and the progression of NAFLD can be affected by inflammation and oxidative stress damage, although the results of these studies may be limited by methodological flaws such as residual confounding and other factors. Colocalization analysis has been proven a powerful tool in revealing the pleiotropic effects of certain loci on multiple traits (Wang et al., 2021a). Proteins are more likely to be used as drug targets than other molecular traits and MR analysis combined with the use of pQTL as colocalization for IV will be valuable to the wider community of human genetics. The results of pQTL in this study suggest that BLVRAD is a protein with high supporting colocalization evidence (PP.H4 = 0.9579), and this association provides a reference for further exploration of the pathogenesis of NAFLD and revealing novel drug targets and biomarkers. Although the colocalisation results of eQLT are negative, it does not necessarily mean that the study is meaningless.

Up to now, research on the relevance of myristoleate to NAFLD is minimal. One study has shown that myristoleate is expressed at high levels during mid-development in oyster larvae and is one of the metabolites associated with fatty acid metabolism (Liu et al., 2020). A related study found that myristoleic acid produced by E. faecalis reduced obesity through brown fat activation and beige fat formation (Quan et al., 2020). In contrast, a study found that fatty acids (myristoleic acid) during adipogenesis were associated with an increased risk of T2DM (Qureshi et al., 2019). Since obesity and T2DM are risk factors for the development of NAFLD, we hypothesize that myristoleate, which is involved in fatty acid metabolism, may play a role in controlling the progression of NAFLD. And through our MR analysis study, we found that genetic predisposition towards higher levels of myristoleate played a protective role in NAFLD development and can inhibit the progression of NAFLD. However, experimental studies on NAFLD and myristoleate are lacking. Therefore, the protective mechanism of myristoleate needs to be further explored.

In addition, the association of caffeine with the risk of NAFLD remains highly controversial, and although nearly 41% of studies have concluded that caffeine is protective against NAFLD, there are also studies showing that caffeine increases hepatocyte damage in mice and definitive research evidence that prenatal caffeine exposure increases susceptibility to NAFLD in rat offspring (He et al., 2019; Hu et al., 2019; Dungubat et al., 2020). Therefore, our genome-wide association study finding that caffeine may be a risk factor for NAFLD is not contrary to the facts and may be related to the sample size and sample population involved in this study. This suggests that the association between caffeine and NAFLD needs to be examined in a more thorough and comprehensive study.

Our study has certain advantages. Firstly, from the perspective of molecular mechanism, it has a solid theoretical basis and important clinical research value to explore the causal relationship between metabolites and NAFLD by using blood metabolites as exposure factors. Secondly, this study used strict quality control conditions and rational, analytical methods, including various models, to evaluate causal effects, which largely avoided reverse causality and residual confounding. Also, the use of large-scale GWAS data gives it greater statistical validity. In addition, statistical methods such as meta-analysis and Stegall’s test (Hemani et al., 2017) were used to test the validity of the MR results. Therefore, the results of this study are mainly reliable and stable. Thirdly, unlike previous MR analyses of single or conventional exposure factors, analyzing 486 blood metabolites is a more difficult task and presents statistical analysis challenges. The analytical strategy we propose may be informative for similar studies in the future. However, several limitations should be noted in our study. First, all mGWAS and NAFLD GWAS data were from European populations. Although this largely avoids population heterogeneity, the MR results should be further validated in other populations to verify their generalisability in future studies with more GWAS data. Second, more than half of the NAFLD risk predictors obtained through preliminary analyses were unknown metabolites whose functional structures are unknown. Therefore, the results of this study are limited. Third, we revealed that cholestyramine and methyl myristate are nominally causally related to NAFLD using a two-sample MR approach. However, this relationship is theoretical, and we failed to confirm it mechanistically. Therefore, the results of the MR study should be further validated in a robust RCT to demonstrate the existence of a causal relationship.

In conclusion, we have identified a causal relationship between two blood metabolites and NAFLD by MR analysis, providing preliminary evidence of the effect of these two metabolites on the progression of NAFLD. This may help establish individualized explanations or markers for biological differences in disease states and serve as candidate molecules for future mechanistic exploration. However, due to the limited experimental studies on biliverdin, myristoleate, and NAFLD, the mechanisms by which the two metabolites affect NAFLD progression are unclear. Therefore, more studies may be needed to explore whether these two metabolites can be used as clinical circulating biomarkers for the screening and prevention of NAFLD.