Introduction

With the escalation of the aging population, age-associated cognitive decline has been emerged as a widespread and crucial public health concern worldwide1. The decline in cognitive abilities not only impacts the individual’s quality of life and functional independence, but also gives rise to significant societal and economic implications2,3,4. Prevention was of utmost importance as no causal therapy existed for cognitive decline, therefore, identifications of potential modifiable risk factors for cognitive decline remains a pressing issue.

Metabolic dysfunction-associated fatty liver disease (MAFLD), which incorporates the pathogenic role of metabolic dysfunction in the development and progression of nonalcoholic fatty liver disease, is one the most common chronic liver diseases globally and affects approximately 25% of the adult population5,6,7. The occurrence of MAFLD is often accompanied by obesity, abnormal metabolism of blood lipids and glucose, and microecological imbalance. Additionally, MAFLD has been shown to be associated with a series of metabolic diseases8,9,10. These metabolic disorders associated with MAFLD are also known to be related to extrahepatic manifestations involving the central nervous system, such as cognitive impairment. A connection between MAFLD and cognitive decline has been proposed due to frequent reports of attention, forgetfulness, and memory issues in patients with MAFLD11,12,13,14. However, previous studies generated inconsistent conclusions regarding the association between MAFLD and cognitive function, with some failed to demonstrate a significant result15,16. One of the possible reasons for the inconsistencies may be the single measurement of cognitive function, which neglected the nature change of cognitive function over time. As acknowledged, characterizing longitudinal patterns of cognitive function may provide a more accurate and robust assessment for the association.

Therefore, this prospective cohort study aimed to identity the different trajectories of cognitive function over time, and investigate a possible link of MAFLD with distinct trajectories of cognitive function. When the association was constructed, we further aimed to assess whether and to what extent the association is mediated by MAFLD-related metabolites among Chinese adults.

Results

Baseline characteristics

A total of 845 participants were enrolled, a comparison of baseline characteristics between excluded and included participants was presented in Supplementary Table 1. Among the enrolled participants, 277 (32.78%) participants had MAFLD, the median age was 49.10 (interquartile range, 40.56–59.62) years and 300 (35.50%) were men. Compared with non-MAFLD participants, those with MAFLD were more likely to be older, men, less educated, have a higher proportion of current smokers, current drinkers, a higher prevalence of hypertension, diabetes, dyslipidemia, more likely to take relative medications, have more cardiometabolic risk factors (Table 1).

Table 1 Baseline characteristics
Full size table

Trajectories of MMSE

A total of four times of MMSE measurements during a median follow of 6.21 (interquartile range: 6.08–6.31) years. The number of participants with 1, 2, 3, and 4 times of MMSE measurements was 121, 206, 330, and 180. Two-class model was selected as the preferred model according to the criteria mentioned in the method, with Bayesian Information Criterion of −5193.89, Akaike Information Criterion of −5165.24, average posterior probability of assignment of 78.66% and 21.34%, respectively (Supplementary Table 2). The number of participants in class 1 (normal decrease pattern) was 714 (84.50%) participants, which was 131 (15.50%) in class 2 (large decrease pattern) (Fig. 1).

Fig. 1: The trajectories of MMSE over time with latent class mixture models.
figure 1

The solid lines showed class-specific mean predicted MMSE, and shading around the lines represent 95% confidence interval for the calculated trajectories. MMSE Mini-Mental State Examination.

Full size image

Association of MALFD with trajectories of MMSE

A total of 72 (25.99%) participants in the MAFLD developed large decrease in MMSE score, which was significantly higher than that in non-MAFLD participants (59, 10.39%). After adjusted for potential covariates, MAFLD had an 81% higher risk of developing large-decrease in MMSE (OR, 1.81; 95% CI, 1.16–2.84; P = 0.0090) (Table 2). Subgroup analyses showed that the associations between MAFLD and the risk of developing large decrease in MMSE were consistent across different subgroups, all the P values for interaction were >0.05 (Supplementary Table 3).

Table 2 Association between MAFLD with the trajectories of MMSE (large decrease pattern as the outcome of interest)
Full size table

Differentially expressed metabolites

The comparison of positive ions, negative ions, and NMR metabolites between non-MAFLD and MAFLD group was presented in Supplementary Tables 4–6. The results showed that among 90 kinds of positive ions, the concentration of lysophosphatidylcholine (lysoPC) (20:3(5z,8z,11z)) in MAFLD group was significantly lower than that in non-MAFLD group (FC = 0.31; FDR = 0.02; Fig. 2A). Among 23 kinds of negative ions, the concentration of lysophosphatidylethanolamine (lysoPE) (18:1(9z)/0:0) in MAFLD group was significantly higher than that in non-MAFLD group (FC = 4.34; FDR < 0.001; Fig. 2B). Among 274 kinds of NMR-measured metabolites, the concentration of valine in MAFLD group was significantly lower than that in non-MAFLD group (FC = 0.47; FDR < 0.001; Fig. 2C).

Fig. 2: Volcano plot showed differentially expressed metabolites between non-MAFLD and MAFLD group according to fold change and the significance.
figure 2

The outcome was each metabolite identified. The horizontal dotted line represented the significance threshold (FDR-adjusted P = 0.05). The scatter denoted the up-regulated (red) or down-regulated (blue) metabolites for correlations with MAFLD. Each plot represents a metabolite identified. FDR false discovery rate, MAFLD metabolic dysfunction-associated fatty liver disease.

Full size image

Mediation analysis

The results of mediation analysis were presented in Fig. 3, which showed that the association of MAFLD with large decrease in MMSE pattern was partially mediated by LysoPE(18:1(9z)/0:0) (mediated proportion = 9.93%) and valine (mediated proportion = 11.04%). However, lysoPC(20:3(5z,8z,11z)) did not play a mediated role in the association.

Fig. 3: Mediation analysis of MAFLD-related metabolites for the association between MAFLD with cognitive function trajectories.
figure 3

The outcome in the figure indicates a large decrease pattern of MMSE. A Contribution of LysoPC; B Contribution of LysoPE; C Contribution of Valine. Adjusted for age, sex, body mass index, education, physical activity, smoking, and drinking, hypertension, dyslipidemia, estimated glomerular filtration rate, and alanine transaminase. MAFLD metabolic dysfunction-associated fatty liver disease *P < 0.05; **P < 0.01; ***P < 0.001

Full size image

Discussion

In this prospective cohort study, we identified two longitudinal patterns of cognitive function change over time. The status of MAFLD was associated with a higher risk of developing a large decrease in cognitive function over time. Metabolites of lysoPC(20:3(5z,8z,11z)), lysoPE(18:1(9z)/0:0), and valine were differently expressed in patients with MAFLD. Additionally, the association between MAFLD and the pattern of a large decrease in cognitive function was partial mediated by lysoPE(18:1(9z)/0:0) and valine. These findings indicated that MAFLD might increase cognitive impairment risk by disturbing lipid and amino acid metabolism.

Accumulative evidence with inconsistent results has investigated the associations of MAFLD with cognitive function. Cross-sectional analysis using data from the National Health and Nutrition Examination Survey showed that MAFLD was significantly associated with increased risk of cognitive impairment measured by serial digit learning test17. Similarly, the Rotterdam Study reported that MAFLD was associated with structural and hemodynamic brain markers in a population-based cross-sectional setting18. On the contrary, anther cross-sectional study conducted among patients with severe obesity showed that the cognitive impairment was not associated with the presence of MAFLD19. Post hoc analyses using data from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) and Systolic Blood Pressure Intervention Trial (SPRINT) also demonstrated that markers of chronic liver disease were not associated with cognitive impairment or related brain imaging markers among individuals with diabetes and hypertension15. These conflicting results were limited to the cross-sectional design. One prospective study investigated the role of NAFLD on the risk of incident cognitive impairment, and the result showed that patients with NAFLD had higher 4-year incidence of cognitive impairment than non-NAFLD patients did20. However, whether the longitudinal association persisted for MAFLD was not well-demonstrated, as the cognitive function declined with age.

In our study, we used latent class mixture models to identify distinct trajectories of cognitive function with aging. Two class of cognitive function trajectories were identified with approximately 15.50% participants had the large decline in cognitive function. Thes rate was a little smaller than the reported among older Mexican Americans (20.0%)21, which may be attributable to the younger adults enrolled in our study. Further analysis added longitudinal evidence on the association of MAFLD and cognitive function by demonstrating that participants with MAFLD had a higher risk of developing a large decline trajectory of cognitive function over time. Subgroup analysis showed that this positive association was consistent across different age, sex, and metabolic syndrome groups. Taken together, these findings suggested that strategies on management of metabolic risk factors and liver disease may brought additional benefit in prevention cognitive decline or even the development of dementia.

Another finding of our study was that we identified three metabolites differentially expressed in participants with MAFLD: lysoPC(20:3(5z,8z,11z)), lysoPE (18:1(9z)/0:0), and valine. The results were supported by previous investigations in vivo and in vitro. One review showed that accumulated altered lysoPCs have been implicated in the tissue impairment and dysfunction underlying metabolic disorders, including MAFLD22. One population-based study on MAFLD adults showed that lysoPC levels as a potential noninvasive biomarker for MAFLD23. In the context of lysoPE, Yamamoto et al found that NAFLD patients have significantly elevated levels of LysoPE in their bodies24. Valine was a kind of essential amino acid, and fattening animal experiments showed that supplementing valine could change the fatty acid composition and reduce lipid accumulation in liver25. Additionally, our study further performed the mediation analysis to explore the role of the three metabolites in the association between MAFLD and cognitive function trajectories. The result showed that approximately 10% of the pathway from MAFLD to a large decrease in cognitive function was mediated through lysoPE and valine, which may provide a possible mechanism underlying the association between MAFLD and cognitive impairment.

Although the pathological mechanisms by which MAFLD affects cognitive function have not yet been fully elucidated, several key components contributing to the association have been proposed, including insulin resistance, systemic inflammation, lipotoxicity, vascular dysfunction, and dysbiosis26,27. First, the liver was considered as the central organ responsible for metabolizing sugars, and the occurrence of MAFLD often accompanied by insulin resistance28. The disruption of insulin signaling in the brain can lead to cognitive impairment by downregulating blood–brain barrier insulin receptors and reducing the transport of insulin into the brain. Second, MAFLD exhibited chronic low-grade inflammation, which extended systemically. The proinflammatory cytokines could increase blood–brain barrier permeability and initiating a complex immune response the central nervus system, resulting in a neurotoxic effect and cognitive impairment29. Third, lipotoxicity emerged as accumulation of lipids in non-adipose tissues, particularly manifests in the liver. Dysregulation caused by lipotoxicity in the brain involves loss of orexin signaling, which was associated with memory impairment, learning deficits, and neuroinflammation30. Additionally, one study also showed that MAFLD accelerated the signs of Alzheimer’s disease in central nervous system by inducing neuronal apoptosis and decreased the expression of lipid metabolites which were responsible for beta amyloid plaque clearance in the disease31.

The longitudinal design with repeated measurements of MMSE allowed us to characterize the longitudinal trajectories of cognitive function over time. However, several limitations also needed to be addressed. First, we used abdominal ultrasound to identify fatty liver, which could not detect hepatic steatosis when fat content is <20%. Therefore, misclassification of MAFLD with mild steatosis would underestimate the effect of MAFLD on cognitive function trajectories. Second, cognitive function was measured by MMSE in the present study, which may underestimate the rate of cognitive decline for older adults with low cognitive function. However, MMSE was treated as continuous variable in the analysis, the aforementioned limitation made a relatively small influence on the association. Third, the selection bias maybe existed due to our inclusion and exclusion criteria, further studies are eagerly needed. Fourth, the relatively small sample size may lead to a lower statistical power for the subgroup analysis, hence, the findings of subgroup analysis were interpreted only as explanatory results. Fifth, a limitation of our mediation analysis was that both the exposure and mediator were measured at baseline. However, we thought MAFLD may represent a history health problem with hepatic steatosis, while metabolites represent the state around the time of sampling or at least a narrower time period than MAFLD. We did not expect a substantial alteration in the causal ordering of the exposure and mediation. The findings needed to be validated in future studies with a time-lagged between exposure and mediators. Sixth, due to the observational design, our study could not establish the causality between MAFLD and cognition, thus further research is warranted. Finally, findings of this study may not be generalizable to other populations, as results were based on Chinese population.

In conclusion, two distinct cognitive function trajectories over time were identified, and MAFLD was associated with a higher risk of developing a large decline of cognitive function. Additionally, the association was partially mediated by lysoPEs and valine. The results provided a novel insight for the association between MAFLD and cognitive decline, although large prospective studies are warranted to confirm our findings in different populations and to investigate the biological mechanisms for the observed associations.

Methods

Study population

Data were obtained from the China Suboptimal Health Cohort Study (COACS), which was a prospective community-based study conducted in Jidong Oilfield Staff Hospital, China. The detailed information on the study has been descried previously32. In brief, a total of 1011 participants aged 18–65 years with data on metabolites were enrolled in the baseline survey from September 2013 to June 2014. All the participants completed the questionnaire interview, clinical and laboratory examines, and were followed up visits in year 2015, 2016, 2018, 2019, 2020, and 2021 to updated the afore-mentioned information. In our current study, we analyzed the association of baseline MAFLD, metabolites, with the risk of trajectories of cognitive function based on MMSE score were developed using data in 2015, 2018, 2019, and 2021. We excluded participants with cognitive impairment at baseline (n = 48), with missing data on the assessment of MAFLD (n = 118), leaving 845 participants in the current analysis (Fig. 4). The study was performed according to the guidelines of the Helsinki Declaration and was approved by Ethical Committee of the Jidong Oil-field Hospital of China National Petroleum Corporation. All participants were agreed to take part in the study and provided informed written consent.

Fig. 4: The flowchart of the study.
figure 4

MAFLD metabolic dysfunction-associated fatty liver disease, MMSE Mini-Mental State Examination.

Full size image

Definition of MAFLD

MAFLD was defined as the presence of hepatic steatosis with one of three metabolic dysfunctions: overweight or obese (body mass index [BMI] ≥ 23 kg/m2), type 2 diabetes, or other metabolic abnormalities ≥233,34. Other metabolic abnormalities included (1) waist circumference ≥90 cm in men and ≥80 cm in women; (2) blood pressure (BP) ≥ 130/85 mm Hg or use of antihypertensives; (3) triglyceride ≥150 mg/dL or use of lipid-lowering agents; (4) high‐density lipoprotein cholesterol (HDL‐C) <40 mg/dL in men and <50 mg/dL in women or use of lipid-lowering agents; (5) fasting blood glucose (FBG) 5.6–6.9 mmol/L; (6) homeostasis model assessment of insulin resistance score ≥2.5. For lack of information on homeostasis model assessment of insulin resistance, we used triglyceride glucose index over the 75th percentile as an alternative to the 6th diagnostic criteria35.

Measurement of cognitive function

Cognitive function was assessed using the Chinese version of the Mini-Mental State Examination (MMSE) questionnaire, which comprises 30 items. The MMSE was administered by well-trained personnel. The total possible score ranges from 0 to 30, with a lower score indicates a poor cognitive ability. Following previous studies, participants with a MMSE score of <27 were considered cognitive impairment36.

Metabolomics analysis

Plasma samples were thawed at 4 °C. Once, thawed, plasma metabolomics analyses were performed with untargeted UHPLC-Q-TOF-MS (Shimadzu, Kyoto, Japan) and 1H-NMR (Varian VNMRS 600 MHz spectrometer, Agilent Technologies, USA), respectively. The plasma samples were analyzed by the UHPLC-Q-TOF-MS method in both positive and negative iron modes. Analytes with detection rates <80% or inter-or intra-assay coefficients of variation >20% were excluded. More details on the sample collection and measurements were described previously37.

Assessment of covariates

Information on demographics characteristics, lifestyle, personal medical history was collected using structured questionnaires. Anthropometric measurements were performed by trained staff. After at least a 12-h fating, blood samples were collected using venipuncture in the morning. The plasma samples were separated in the laboratory after centrifugation at 4 °C, for 10 min at 3000 × g. Then, the samples were stored at −80 °C immediately, and freeze–thaw cycles were strictly avoided until metabolomic analysis. The biochemical tests were analyzed using an auto-analyzer (Hitachi 747, Hitachi, Tokyo, Japan). Estimated glomerular filtration rate was calculated using the Chronic Kidney Disease Epidemiology Collaboration creatinine equation.

Statistical analysis

The trajectories of MMSE were determined using latent class mixture models. The models were fitted with MMSE as the dependent variable, age as the time-scale mixed with the ‘lcmm’ R package. The analysis models were checked for between 1 and 4 trajectory classes based on a quadratic adherence trajectory function. The optimal number of classes was determined by Bayesian Information Criteria, Akaike Information criterion, sample size in each class, and clinical interpretation.

Baseline characteristics are presented as mean ± standard deviation, median with interquartile range, or frequency with percentages, as appropriate. Differences in baseline characteristics across non- and MAFLD groups were compared using the t test or Wilcoxon test for continuous variables and chi-square test for categorical variables. The association of MAFLD with different trajectories of MMSE was assessed by logistic regressions. Four models were constructed progressively. Model 1 was unadjusted; model 2 was adjusted for age and sex; model 3 was further adjusted for body mass index, education, physical activity, smoking, and drinking; model 4 was further adjusted for hypertension, dyslipidemia, estimated glomerular filtration rate, and alanine transaminase. To test the consistency of the findings, subgroup analysis stratified by age (≤60 vs >60 years), sex, body mass index (≤25 vs >25 kg/m2), depression status, and metabolic syndrome was performed. A multiplicative term between subgroups and MAFLD status was added into the model and the interaction was tested by a likelihood ratio test.

Metabolites differentially expressed between non-MAFLD and MAFLD participants were defined as meeting the following criterion: |Log2 Fold Change|≥ 1 and false discovery rate (FDR) < 0.05, using univariate analysis of t-test and Wilcoxon rank sum test, as appropriate. Additionally, mediation analyses were performed to evaluated whether and to what extent MAFLD-related metabolites may explain the association between MAFLD and the trajectories with MMSE by using the ‘mediation’ package in R. The mediated proportion was computed as indirect effect/total effect on the log scale × 100%. The models were also adjusted for covariates in model 4 above.

All the analyses were performed using R version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria), and SAS version 9.4 (SAS Institute, Cary, NC, USA). All the statistical tests were 2-sided, and P < 0.05 was considered statistical significance.