The data originated from the UK Biobank, an extensive, prospective cohort study encompassing more than 500,000 participants within the age range of 37–63 years. Affirming their voluntary participation, the individuals provided written consent via electronic questionnaires for the collection of their data. The study incorporated comprehensive magnetic resonance imaging (MRI) derived from a multimodal imaging initiative, yielding a rich dataset of imaging information. Adhering to stringent ethical research standards, the UK Biobank study has received clearance from the Northwest Multicenter Research Ethics Committee. Authorization for the utilization of this dataset has also been granted by the Human Ethics Committee of West China Hospital, Sichuan University.

A total of 210,948 participants completed at least one 24-h dietary recall. Participants were excluded if they had existing liver disease (Supplementary Table S1, n = 1,148), implausible energy intake levels (women: <500 or >3,500 kcal/day; men: <800 or >4,000 kcal/day, n = 2,896), or lacked MRI data (n = 181,055). Following these exclusions, the study included 25,885 eligible individuals (Figure 1).

Participants were asked to complete online 24-h dietary assessments, providing detailed recalls of their consumption of 206 common foods and 32 beverages over the preceding 24 h. Daily beverage intake was assessed by following question: “How many glasses, cans, or cartons containing 250 mL of sugar-sweetened beverages, artificially sweetened beverages, or natural juices did you drink yesterday?” Participants were categorized into 3 groups based on the distribution of these products: 0, 0 to 1 and ≥1 serving per day. From 2009 to 2012, participants were asked online five times, and those who completed at least one dietary assessment were included in the study by calculating their mean beverage intake. The beverages intake levels were comparable with the national data in the UK (19).

The Proton Density Fat Fraction (PDFF), ascertained through Magnetic Resonance Imaging (MRI), represents the proportion of protons associated with fat relative to the overall proton count within the liver. This method allows for the direct measurement of LFC, eliminating the need for invasive biopsy, which was considered as the most accurate noninvasive method (20). Previous study showed MRI was an excellent assessment compared with liver biopsy (Spearman correlation coefficient = 0.85) (21) and can classify grades of hepatic steatosis with areas under the summary receiver operating characteristic curves ≥90% (22). A cardiac-gated Shortened Modified Look-Locker Inversion sequence was utilized to measure liver T1 values, and it can be adjusted to account for the influence of iron, resulting in an cT1 score, which is expressed in milliseconds (ms) and serves as an indirect indicator of hepatic fibro-inflammatory activity. This metric, derived from MRI, has been confirmed for accuracy by comparison with liver histology and has shown its practical use in clinical settings going forward (23–26).

Our models were adjusted for several covariates: sex (male and female), age, Townsend deprivation index, education statues, alcohol intake (g/d), smoking status (never, current, former), BMI (<25.0, 25.0 to <30, ≥30 kg/m²), abdominal obesity (yes and no), physical activity (metabolic equivalents [MET] hours per day for all physical activity), hypertension (yes and no), glucose (mmol/L), triglyceride (mmol/L), cholesterol (mmol/L), C-reactive protein (mg/L), and platelet distribution width (%), total energy (kcal/d), total sugar (g/d), and healthy diet score (0–7). Healthy diet score were based on following criterion (27): total vegetables, ≥4 servings per day; total fruit, ≥4 servings per day; total fish, ≥2 servings per week; processed meat, ≤1 servings per week; red meat, ≤1.5 servings per week; whole grains, ≥3 servings per day; refined grains, ≤1.5 servings per day and achieving one of the above criteria is scored as one point (ranged from 0 to 7). Covariates were collected by professionals through questionnaires, physical examinations, and biological samples (28). The detailed descriptions of covariates were performed in Supplementary material according to previous research (27).

Continuous variables were presented as means with standard deviations and categorical variables were depicted as counts and percentages. According to previous study (29), we performed univariate and multivariate linear regression models and calculated the arithmetic mean difference (AMD) and 95% confidence intervals (CIs) to explore the correlation PDFF, cT1, and SBs consumption. In addition, we defined hepatic steatosis as PDFF ≥ 5% (30) and hepatitis as cT1 ≥ 800 ms (31), and investigated their associations with beverage intake. Restricted cubic splines with three knots (knots placed at the quartile of each beverage intake) were used in unadjusted and fully adjusted models to explore the dose–response relationships between SBs intake, PDFF, and cT1, respectively. Additionally, subgroup analyses were conducted to explore the relationship of AMD of PDFF and cT1 with SBs intake across different subgroups, categorized by age (<55 vs. ≥55), sex (male vs. ≥female), alcohol intake (<10 g/d vs. ≥10 g/d), BMI (<25.0 vs. 25.0 to <30 vs. ≥30 kg/m²) physical activity (whether met WHO physical activity guidelines: 150 min of moderate activity per week or 75 min of vigorous activity) (32). Quantile G-computation (QGC) (33) model was employed to comprehensively evaluate the relative importance and joint association of various beverage intakes on Proton Density Fat Fraction (PDFF) and contrast-enhanced T1-weighted imaging (cT1) outcomes. The dietary questionnaire inquires about the servings of beverages consumed within 24 h (250 mL = 1 serving). Therefore, we utilized substitution models to assess the effect of replacing one serving of a beverage with another (34). Moderation analysis was conducted to investigate which covariates moderated the effect of ASB and SSB intake on PDFF and cT1. We also performed a mediation analysis to quantify the proportion of PDFF and cT1 explained by indirect factors as well as the direct association of SSB intake and ASB intake. The detailed descriptions of statistical analysis were provided in Supplementary material.

To robust our findings, several sensitivity analyses were conducted. First, we divided health diet score into the original seven parts for adjustment. Second, we additionally adjusted for carbohydrates intake, use of medication (aspirin, cholesterol lowering, hypoglycemic, and antihypertensive medication) due to potential confounding. Third, the percentage change of PDFF also expressed by linear regression models. Fourth, the linear regression models was repeated in participants who completed at least 2 dietary assessments.

Statistical significance was set at a two-tailed p-value of <0.05. All statistical analyses were performed using SPSS (version 27.0; IBM Corp., Armonk, NY, United States) and R software 3.5.0 (Vienna, Austria).

A total of 25,885 participants (age: 55.3 ± 7.5 years, male: 12,228 [47.2%]) were included in our study with a median follow-up of 10.3 (4.5–16.3) years. Among these population, 5,511 (21.3%) individuals consumed ASB, 8859 (34.2%) individuals consumed SSB, and 14,567 (66.3%) individuals consumed NJs. Compared to those without SBs intake, participants with ≥1 serving/d more likely to be younger, current smoker and had higher BMI and energy intake and less physical activity. For biochemical examination, they had higher glucose, triglyceride, cholesterol, liver enzyme (Table 1 and Supplementary Table S2).

In the fully adjusted linear regression models (Table 2 and Supplementary Tables S3, S4), compared to those without ASB and SSB intake, the arithmetic mean difference (AMD) of PDFF was 0.15 (95% CI, 0.06 to 0.24, p < 0.001) and 0.21 (95% CI: 0.12 to 0.29, p < 0.001), and the AMD of cT1 was 3.86 (95% CI, 1.26 to 6.79, p < 0.001) and 2.43 (95% CI: 1.31 to 3.57, p < 0.001) in individuals with ≥1 serving/d, respectively. Additionally, there were both higher PDFF and cT1 for each additional serving of ASB and SSB intake. Compared to non-NJs intake population, those with 0–1 serving/d had decreased PDFF (AMD: −0.10, 95%CI: −0.19 to −0.01). However, this association was not significant between NJs and cT1. In addition, using PDFF ≥ 5% as the threshold for hepatic steatosis and cT1 ≥ 800 ms as the threshold for hepatitis, we found that compared to those without ASB and SSB intake, the odds ratio of hepatic steatosis was 1.08 (95% CI, 1.03 to 1.15, p = 0.05) and 1.14 (95% CI: 1.02 to 1.23, p = 0.008) and the OR of hepatitis was 1.33 (95% CI, 1.10 to 1.65, p < 0.001) and 1.29 (95% CI: 1.12 to 1.48, p < 0.001) in individuals with ≥1 serving/d, respectively (Supplementary Table S5).

In fully adjusted restricted cubic splines (Figure 2), ASB intake showed linear dose–response association with PDFF and cT1, and there was a non-linear dose–response association of SSB with PDFF (P for nonlinear = 0.048), with PDFF plateauing when SSB intake reaches 2 servings/day. However, significant dose–response association between NJs intake, PDFF, and cT1 were not observed (P for overall >0.05).

In fully adjusted QGC models (Figure 3), for each extra serving of SBs intake, the AMD of PDFF was 0.20 (95%CI: 0.12 to 0.28), and SSB contributed the most (54.7%); While, the AMD of cT1 was 3.18 (95%CI: 1.48 to 4.87), and ASB contributed the most (53.1%).

In substitution analysis (Table 3), the PDFF was decreased when substituting SSB with NJs and water, and substituting ASB with water. Moreover, a decrease in cT1 (AMD: −1.78, 95%CI: −2.76 to −0.80) was observed after substituting ASB with water.

In subgroup analyses (Supplementary Tables S6–S10), the associations of PDFF, cT1, and SBs intake were broadly similar in different subgroups of sex, age, BMI, and physical activity (P for interaction >0.05). However, there was a stronger association among those who consumed more alcohol (P for interaction = 0.002). In moderation analysis (Supplementary Tables S11, S12), we found alcohol intake (ASB: P for interaction = 0.002; SSB: P for interaction <0.001) and sugar intake (ASB: P for interaction = 0.009; SSB: P for interaction = 0.004) had a positive moderating effect (β > 0) on the association of ASB and SSB consumption with PDFF. Moreover, abdominal obesity was also a significant moderate factor for cT1.

In mediation analyses (Figure 4 and Supplementary Tables S13, S14), body fat, healthy diet, and inflammation showed an partial mediation effect on the association of ASB and SSB consumption with PDFF and cT1. On the other hand, sugar intake and triglycerides also served as an adverse mediated factors for SSB. Considering that there was no dose–response association between NJs intake, PDFF, and cT1, the mediation analysis did not conducted for NJs.

In sensitivity analyses (Supplementary Tables S15–S18), the associations remained robust after adjusting additional covariates, including original seven parts of healthy diet score, carbohydrate intake, aspirin, and lowering cholesterol, hypoglycemic, and antihypertensive medication; including population completed at least 2 dietary questionnaires; and representing difference as percentage change of PDFF.