Introduction
Non-alcoholic fatty liver disease (NAFLD) is the most common cause of chronic liver disease in the world [
1]. NAFLD, which has been recently redefined as part of steatotic liver disease (SLD) under the term MASLD (metabolic dysfunction-associated fatty liver disease), is a rapidly growing contributor to liver mortality and morbidity globally and affects approximately 25% of the adult population [
2‐
4]. This disease is characterised by the accumulation of fat in the liver [
5] and can progress from simple steatosis (≥5.% of liver fat content) to steatohepatitis (≥ 5% of liver fat content and inflammation), and lead to liver fibrosis and cirrhosis [
6‐
8]. These advanced stages are mostly responsible for the substantial economic burden of MASLD [
6,
9], and it has been estimated that MASLD will be the first cause of liver transplant by 2030 [
3,
10,
11]. To date, the clinical management of MASLD is constrained to lifestyle interventions such as maintaining a healthy weight and balanced diet, as excess energy intake and low energy expenditure are key modifiable risk factors, and at present no pharmacological treatment has been approved [
12]. However, it is not yet fully understood how different dietary macronutrients relate to MASLD, independently of energy intake [
13,
14].
It has been proposed that dietary carbohydrates increase liver fat accumulation because they promote de novo lipogenesis (DNL), and when this physiological mechanism is stimulated in excess, it would contribute to MASLD [
15,
16]. In addition, inflammation has also been pointed as a potential mechanism in which carbohydrates are associated with liver fat accumulation [
16]. However, recent studies have suggested that different types and sources of carbohydrates could influence liver fat accumulation differently [
13,
17]. Population-based studies looking at total dietary carbohydrate intake and MASLD have observed positive [
18‐
22], negative [
23] and non-significant associations [
21,
24‐
29]. A recent meta-analysis of 34 observational studies concluded that there were no significant associations between carbohydrates and MASLD [
30]. These inconsistent results may be due to small sample sizes or differences in dietary assessment methods, inclusion criteria, adjustment for confounders or MASLD diagnosis tools. In particular, there are few observational studies that are prospective and have adjusted for total energy intake or assessed different types and sources of carbohydrate intake simultaneously. Therefore, this study sought to study the associations between different types and sources of dietary carbohydrates in the largest prospective study to date with liver fat measured using the most accurate and precise non-invasive method for liver fat quantification, magnetic resonance imaging (MRI) [
31].
Discussion
In the largest observational investigation of macronutrient intake and liver fat to date, associations varied across different types and sources of carbohydrates with liver fat accumulation. Overall, the results from both the cross-sectional and prospective analyses suggested strong inverse and independent associations between the intake of non-free sugars, fibre and starch from whole grains with liver fat. Conversely, free sugars were positively associated with liver fat in the prospective analyses but not the cross-sectional analyses, whereas starch from refined grains was not associated with liver fat in prospective analyses but displayed positive associations in the cross-sectional analyses.
Total carbohydrates were inversely associated with liver fat in both cross-sectional and prospective analyses, albeit weakly in the latter. However, other observational studies from Japan, Iran and Korea reported positive associations between high carbohydrate intake and measurements of liver fat, possibly due to differences in the proportions of subtypes of carbohydrates typically consumed in different populations compared to the UK [
18‐
21]. Since other studies from European populations have likewise found inverse associations with carbohydrate intake akin to the current study, it overall suggests that mixed results may be due to variations in the types and sources of carbohydrate, which were not measured in these previous studies [
23].
When looking in more detail at the types and sources of carbohydrates, the current study found strong, inverse associations between fibre and starch from wholegrains with high liver fat in both the cross-sectional and prospective analyses. Cross-sectional and case-control studies in America and Europe also previously reported inverse associations with fibre, although the current study is the first to show large-scale prospective evidence with MRI-based phenotyping of liver fat [
23,
24]. Fibre, which wholegrains contain high amounts of, may reduce low-grade inflammation, improve lipid profiles, increase satiety and supress ghrelin, a hormone with orexigenic effects; this could explain why it is negatively associated with liver fat [
14,
53]. In addition, it may affect the gut microbiome, by influencing the gut barrier, gastrointestinal immune and endocrine responses, thereby playing a role in whole-body and liver metabolism [
54].
In contrast, the associations with starch from refined grains were less clear, with a weakly positive relationship suggested from the cross-sectional analyses but a generally flat association in the prospective results. A previous systematic review and meta-analysis of observational studies concluded there was not a significant relationship between refined grains and MASLD, although none of these studies were prospective [
55]. Meanwhile, a recent RCT of 50 overweight adults with a 12-week feeding intervention of refined grains reported a 49% increase in liver fat [
56]. While the RCT was small and susceptible to chance findings, there is mechanistic evidence suggesting that refined starchy foods may cause the accumulation of fat in the liver by promoting inflammation [
16]. Our population-based prospective study suggests that the association of liver fat with starch may vary by the source of starch, although while the benefit of fibre was able to be detected, more research is needed to verify whether consuming more refined grains is harmful. It could be that the healthy volunteer bias of UK Biobank may have weakened any risks associated with the consumption of refined grains, and more large-scale prospective studies will need to assess different sources of starch and liver fat accumulation [
50]. Alternatively, the prospective analyses included an adjustment for BMI that was not done in cross-sectional analyses as this variable was contained in the HSI index; this may have contributed to the differing results across time points, if BMI is the main pathway through which starch from refined grains or free sugars is associated with liver fat.
The inverse linear association demonstrated here with non-free sugars is novel, and to the best of our knowledge, no previous study has looked at the relationship between this exposure and liver fat. Sources of non-free sugars may be high in fibre, such as vegetables and fruits - but non-free sugars also come from dairy, which is low in fibre content. This suggests that their role in liver fat accumulation could be independent from fibre. Recent research has also shown an inverse association between dairy products and type 2 diabetes, another important metabolic condition that is also associated with MASLD [
57,
58]. Further research could focus on sources of non-free sugars to understand whether they have different associations with liver fat.
On the other hand, the associations between free sugars and high liver fat were non-significant in the cross-sectional analyses, but positive in the prospective analyses. While the cross-sectional analyses had more power due to sample size, the prospective analysis had a more reliable outcome ascertainment. Previous research is likewise mixed, with a review of observational studies suggesting a positive association, whereas both positive and null results have been reported from RCTs of dietary intervention trials [
59‐
61]. Some of the differences may be due to variation in food groups comprising the term ‘free sugars’, with research on free sugars from sugar-sweetened beverages (SSBs) generally more consistently associated with an increase in liver fat than free sugars from other sources [
59‐
61]. A recent meta-analysis of controlled trials concluded that excess energy from SSBs is associated with large increases in liver fat [
62]. Therefore, looking at the association of free sugar intake with liver fat as part of an isocaloric or hypercaloric diet is also an important source of variation in the previous research, and more research needs to assess if sources of free sugars besides those from beverages are associated with liver fat accumulation, independent of overall energy consumption.
This study had several strengths, such as studying the exposure of dietary carbohydrates as a whole group and as types and sources simultaneously. This was possible due to the availability of detailed dietary data from the Oxford WebQ. Previous research has shown that the dietary assessment methods in the UK Biobank estimate intake with acceptable reproducibility and validity, with the advantage of being feasible to administer in a large population without too much participant burden [
38,
63]. The exclusion of participants with underlying health conditions helped attenuate the influence of reverse causality, although we cannot fully rule out reverse causality whereby subclinical disease may have led the participants to change their diet prior to measurement in this study. Many potential confounders were also adjusted for in the analysis, including energy intake, although the calculation of an E-value indicated that unmeasured confounders with associations of 2.31 with both the exposure and outcome could explain away the inverse relationship of fibre with the odds of liver fat in the cross-sectional analyses [
64,
65].
A key limitation in this study is the potential selection bias arising from the low response rate in UK Biobank (5%), which may have introduced a healthy volunteer bias particularly for those who agreed to answer two or more dietary questionnaires (or four, as in sensitivity analyses) [
32,
63]. However, research has shown that even with such a low response rate, estimated risk factor associations with disease in the UK Biobank appeared reliable [
66‐
69]. Carbohydrate intakes were calculated from self-reported questionnaires, and key confounders like physical activity were estimated from self-reported questionnaires which may have introduced measurement error and information bias [
38]. For example, the 24-h dietary assessment used here does not collect information about food items that were not on the list, which could lead to an underestimation of dietary intake and lead to residual confounding. Using at least two 24-h dietary assessments and removing implausible intakes attempted to minimise this information bias. Importantly, some subtypes of carbohydrates have more within-person variability than others: previous research indicated that within-person variability may be larger for starch than for fibre in UK Biobank 24-h assessments [
70]. Within-person variability in exposures will introduce random error that leads to regression dilution bias and attenuates associations with disease towards the null, and this bias will be greater in the types of carbohydrates that had more variability [
71]. Lastly, the outcome of HSI in cross-sectional analyses was an indirect proxy of liver fat that has not been validated in a UK population and is driven mostly by BMI, which was already high in this population. Thus, while an overestimation of hepatic steatosis using the HSI in this population may be a limitation of the cross-sectional analysis, using an index for hepatic steatosis in the baseline sample of UK Biobank allowed for the large-scale investigation of carbohydrate quality and MASLD, with measurements on approximately 23,000 participants.
It is important to note that in this paper we use the term MASLD when referring to previous data that originally used the term NAFLD. This was done in order to adopt the new nomenclature that has been introduced this year [
4]. While the new definition is slightly different, and includes the presence of one metabolic factor, a recent study showed that it is possible to consider NAFLD cases data as MASLD cases [
72].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.