The Guangzhou Medical Ethics Committee of the Chinese Medical Association approved the Guangzhou Biobank Cohort Study (GBCS) and all participants gave written, informed consent before participation (IRB No. 20030210). All methods were performed in accordance with the relevant guidelines and regulations. Details of the GBCS have been reported previously [17,18]. Briefly, the GBCS is a three-way collaboration among Guangzhou Twelfth People’s Hospital and the Universities of Hong Kong and Birmingham. Participants were drawn from the Guangzhou Health and Happiness Association for the Respectable Elders (GHHARE), from September 2003 to January 2008. About 7% of Guangzhou residents aged 50 years or above were included in the GHHARE.

All surviving participants were invited to return for the first (March 2008 to December 2012) and second follow-up examinations (March 2013 to January 2020). Both the baseline and follow-up examinations included a face-to-face, computer-assisted interview conducted by trained nurses to collect information on demographic characteristics, lifestyle factors, and family and personal medical history. Anthropometric and clinical parameters such as fasting plasma glucose (FPG), blood pressure and lipids were measured. The follow-up questionnaire and clinical and laboratory examinations were largely similar to those conducted at baseline. The reliability of the questionnaire was tested by randomly recalling 200 participants for re-interview and the results were satisfactory [17].

MetS components including high-density lipoprotein cholesterol (HDL-C), systolic blood pressure (SBP), waist circumference (WC), FPG, and triglycerides (TG) were measured at baseline and during two follow-up examinations. Levels of SBP, in mm Hg, were recorded as the mean of the last two readings from three measurements, taken using a digital sphygmomanometer (Omron 705CP, Omron, Kyoto, Japan) [19]. Levels of HDL-C, FPG, and TG, in mmol/L, were measured using a Roche COBAS automatic biochemical analyzer (Basel, Switzerland) in the clinical laboratory of the Guangzhou Twelfth People’s Hospital [19]. Levels of WC, in cm, were measured horizontally around the narrowest part of the torso, between the lowest ribs and the iliac crest [19]. To reduce bias from dichotomous variable in the observational study, we constructed a MetS index based on these five components [20,21]. Each component was standardized based on the mean and standard deviation (SD) at baseline. As lower levels of HDL-C indicate poorer health, HDL-C scores were reversed after standardization. The MetS index was derived as the mean of these five standardized variables [20,21]. Furthermore, patterns of changes in the MetS index and its components were all classified according to trajectory analysis.

In the GWAS and MR analysis, MetS was used as a dichotomous variable (case/control). The standard assessment of MetS status has been previously reported in the GBCS studies [22]. Briefly, the MetS was defined by the presence of ≥3 risk factors, which included raised blood pressure (130/85 mm Hg) or treatment of previously diagnosed hypertension; FPG ≥5.6 mmol/L or treatment of previously diagnosed type 2 diabetes mellitus; TG ≥1.7 mmol/L or treatment for the lipid abnormality; HDL-C <1.0 mmol/L for men or HDL-C <1.3 mmol/L for women, or medication use; and central obesity defined as a WC ≥90 cm for men and ≥80 cm for women.

Immediate 10-word recall test (IWRT), delayed 10-word recall test (DWRT), and mini-mental state examination (MMSE) were used to assess the immediate, delayed memory recall and cognitive function, respectively, at both baseline and two follow-up examinations, as reported in our previous GBCS papers [18,23]. Greater scores indicated better function and reduction in scores indicated a decline in function. DWRT or IWRT was a test of verbal learning and memory requiring recall a list of ten words. To better fit Chinese culture, the adapted 10-word list included ‘letter,’ ‘ticket,’ ‘grass,’ ‘arm,’ ‘corner,’ ‘stick,’ ‘book,’ ‘stone,’ ‘chairman,’ and ‘soy sauce.’ During the interview, these 10 words were read out to participants one by one. Then the participants were immediately asked to recall the words. Participants were given one point for each word that they could correctly recall. This process was repeated three times and summed scores for these three recalls were IWRT (0–30). After 5 minutes of answering other questions for distraction, participants were asked to recall as many words as they could remember. The last recall was DWRT (0–10). The total number of words was denoted by IWRT and DWRT scores, respectively. Memory impairment was defined by DWRT scores of less than 4, corresponding to one SD below the mean (mean±SD, 5.5±1.8) [23]. MMSE consisted of 11 items (0–30), as reported in our previous GBCS paper [24]. MMSE can be divided into five dimensions, including orientation (0–10), registration (0–3), attention and calculation (0–5), recall (0–3), and language (0–9) [25]. Poor cognitive function was defined by MMSE scores of less than 25, corresponding to one SD below the mean (mean±SD, 27.5±2.6).

The Guangzhou Biobank contains genetic data from 3,137 participants. DNA was extracted from buffy coat stored at –80°C at the Guangzhou Twelfth People’s Hospital, using a standard magnetic bead extraction procedure. DNA concentrations were measured using Nanodrop (Thermo Scientific, Waltham, MA, USA). In cases where the concentrations were below 15 ng/μL, DNA was re-extracted manually using a silica-based column method (Hipure Blood DNA Mini Kit, Magen Biotechnology, Guangzhou, China). For genotyping, we used the Illumina ASA (BeadChip Array Asian Screening Array-24+v1.0 HTS ASAMD-24v1-0, San Diego, CA, USA) genotyping platform (array). The ASA array includes a broad spectrum of pharmacogenomics markers (n=5,588) obtained from Clinical Pharmacogenetics Implementation Consortium (CPIC) guidelines (www.cpicpgx.org) and the Pharmacogenomics Knowledge Base (PharmGKB) database (www.pharmgkb.org). Additional information about the ASA array is available on the official Illumina website (https://www.illumina.com/products/by-type/microarray-kits). Genotyping assays were performed at Guoke Biotechnology Co., LTD in Beijing, China (www.bioguoke.com). Further details regarding DNA extraction, genotyping methods and quality control measures are provided in the Supplementary Methods.

Baseline sociodemographic, lifestyle and biological factors, self-rated health, and self-reported history of disease and medication were analyzed as potential confounders. Sociodemographic factors included age, sex, education (junior middle or below, senior middle or above), occupation (manual, nonmanual, others), marital status (married, others), and family income (<30,000 Chinese yuan [CNY]/year, ≥30,000 CNY/year; US$1=6.95 CNY). Lifestyle factors included physical activity (inactive, minimally active, active), smoking status (never, ever), and drinking status (never, ever). The biological factor considered was body mass index (BMI). Physical activity was assessed by the International Physical Activity Questionnaire, which was validated in our cohort previously [26]. Additionally, self-reported health status (poor, very poor) and history of disease and medication use including cardiovascular disease, hyperlipidemia, diabetes, hypertension, and drugs for managing hypertension, glucose and lipids, were assessed by experienced nurses.

In Fig. 1, the MetS index was classified into low-stable (18.3%), moderate-stable (49.5%), high-stable (28.0%), and higher-stable (4.2%) trajectory groups. Regarding the components of MetS index, HDL-C, SBP, and WC were classified into four trajectory groups, while FPG and TG were into two trajectory groups. HDL-C included low-stable (54.3%), moderate-stable (34.2%), high-stable (7.7%), and moderate-increase (3.8%) trajectory groups. SBP included low-stable (24.5%), moderate-stable (45.8%), high-stable (25.1%), and higher-stable (4.6%) trajectory groups. WC included low-stable (13.4%), moderate-stable (41.9%), high-stable (36.3%), and higher-stable (8.4%) trajectory groups. FPG included low-stable (94.9%) and high-increase (5.1%) trajectory groups. TG included low-stable (95.2%) and high-stable (4.8%) trajectory groups.

Table 1 shows that at baseline, compared to the low-stable MetS index group, the higher-stable group was older and had a greater proportion of men, ever smokers, drinkers, and poor self-rated health. This group also had higher levels of BMI, and had a higher prevalence of cardiovascular disease, hyperlipidemia, diabetes, hypertension, and medication use. Additionally, the higher-stable group had a lower proportion of physically active individuals, lower levels of education, and lower baseline scores in DWRT and IWRT (all P<0.05). Similar patterns were found in the high-stable MetS index group, except for a lower proportion of poor self-rated health. No associations were found between the MetS index trajectory groups and occupation, marital status, family income or baseline MMSE scores (P=0.050 to 0.712).

In Table 2, model 2, after adjusting for potential confounders, higher MetS index was associated with decreasing annual change rates (decrease) in DWRT, IWRT, and MMSE scores, and the associations with decreases in DWRT and MMSE scores were statistically significant (adjusted β=–0.014 score/year [95% CI, –0.025 to –0.003] and –0.026 score/year [95% CI, –0.042 to –0.010], respectively), while with a decrease in IWRT scores was non-significant (–0.015 score/year [95% CI, –0.039 to 0.008]). MetS index trajectory groups were identified, including low-stable, moderate-stable, high-stable, and higher-stable groups (Fig. 1). Compared to the low-stable MetS index group, the higher-stable group showed significant decreases in DWRT and IWRT scores (–0.048 score/year [95% CI, –0.088 to –0.008] and –0.093 score/year [95% CI, –0.180 to –0.006], respectively). Similar patterns were also found in both moderate and high-stable MetS index groups. No associations were found of the MetS index trajectory groups with changes in MMSE scores (P=0.081 to 0.743). Regarding the five dimensions of MMSE, the MetS index was significantly associated with decreases in attention and calculation, and recall scores (–0.014 score/year [95% CI, –0.022 to –0.005] and –0.021 score/year [95% CI, –0.029 to –0.012], respectively). Conversely, it was associated with a slight increase in registration scores (0.004 score/year [95% CI, 0.002 to 0.006]). Furthermore, compared to the low-stable MetS index group, the high-stable group showed statistically associated decreases in attention and calculation, and recall scores (–0.016 score/year [95% CI, –0.036 to –0.001] and –0.024 score/year [95% CI, –0.041 to –0.007], respectively). However, no associations of the MetS index and its trajectory groups with changes in orientation and language scores were found (P=0.066 to 0.449).

Moreover, the trajectory groups of the MetS index components including HDL-C, SBP, WC, FPG, and TG are shown in Fig. 1, and the associations with annual change rates in DWRT, IWRT, MMSE, and its dimensions were also explored as shown in Fig. 2. Compared to the low-stable status group, the high-stable SBP group and the high-increase FPG group were associated with significant decreasing annual change rates (decrease) in both DWRT and IWRT scores (–0.055 score/year [95% CI, –0.090 to –0.021] and –0.050 score/year [95% CI, –0.082 to –0.019] for DWRT, and –0.077 score/year [95% CI, –0.152 to –0.003] and –0.084 score/year [95% CI, –0.152 to –0.015] for IWRT, respectively). Conversely, the moderate-increase HDL-C group was associated with significant increases in DWRT and IWRT scores (0.087 score/year [95% CI, 0.052 to 0.123] and 0.110 score/year [95% CI, 0.033 to 0.187], respectively). HDL-C was positively associated with an increase in MMSE scores (0.052 score/year [95% CI, 0.035 to 0.070]), while WC showed a negative association (–0.002 score/year [95% CI, –0.003 to –0.001]). Such associations were also found in the moderate-increase HDL-C and higher-stable WC group, relative to their respective stable groups (Fig. 2).

Regarding the five dimensions of MMSE, the positive associations of HDL-C with increasing annual change rates (increase) in attention and calculation, and recall scores were found (0.020 score/year [95% CI, 0.011 to 0.029] and 0.033 score/year [95% CI, 0.024 to 0.042], respectively) (Fig. 2). Conversely, an association of WC with decreases in attention and calculation, and recall scores was found (both –0.002 score/year [95% CI, –0.003 to –0.001]). Compared to the low-stable status group, the moderate-increase HDL-C group was significantly associated with increases in attention and calculation, and recall scores (0.052 score/year [95% CI, 0.028 to 0.077] and 0.050 score/year [95% CI, 0.026 to 0.074], respectively), while the higher-stable WC group was associated with a decrease in recall scores (–0.032 score/year [95% CI, –0.057 to –0.007]). The high-increase FPG group was associated with a decrease in registration scores (–0.007 score/year [95% CI, –0.014 to –0.003]). No associations of SBP and TG with the five dimensions of MMSE were found (Fig. 2).

In the modification analysis, as shown in Supplementary Fig. 1, the associations of the MetS index with annual change rates in DWRT and IWRT scores did not vary by sex, age, education nor family income (P for interactions from 0.17 to 0.73). However, the association with MMSE scores varied by family income (P for interaction=0.03), with a more pronounced decreasing annual change rates in MMSE scores associated with a higher MetS index being observed in the low family income group (–0.050 score/year [95% CI, –0.078 to –0.022]). Additionally, a significant modification effect of family income was observed in the attention and calculation dimension of the MMSE, indicating a more pronounced association in the low family income group (P for interaction=0.02).

A total of 2,705 participants had baseline data on MetS and genotyping. In Fig. 3 and Supplementary Table 2, the Manhattan plot identified 46 SNPs achieving genome-wide significance associated with MetS (P<5×10^–6). For example, rs662799 in chromosome 11, which tags the apolipoprotein A5 (APOA5) allele, exhibited the strongest effect size (P=1.67×10^–9). After excluding SNPs that deviated from the Hardy-Weinberg equilibrium and showed high LD (r² ≥0.1), a weighted genetic score was constructed using two SNPs, with their beta-coefficient derived from the GWAS (Table 3). Of them, rs662799 has been previously reported, while rs1989154, tagging the 5-hydroxytryptamine receptor 4 (HTR4) gene, represented a novel locus for MetS. When using the weighted genetic score as IV in the one-sample MR analysis, the results indicated a non-significant causal association between MetS and decrease in DWRT, IWRT nor MMSE scores (P=0.29 to 0.98) (Supplementary Table 3).

To date, our study is the first to establish the longitudinal association of changes in MetS, expressed as an index, with cognitive decline in 7,775 older Chinese adults over a 12-year follow-up. We found that greater MetS index was significantly associated with declines in memory and cognitive function, especially in delayed memory recall. This association was also evident in the higher-stable MetS index group compared to the low-stable MetS index group. Additionally, our results showed that in the components of MetS index, HDL-C was positively associated with annual change rates in memory function, while SBP and FPG were negatively associated. Moreover, HDL-C showed a positive association, whereas WC was negatively associated with changes in cognitive function, including the dimensions of attention, calculation and recall. The association of the MetS index on cognitive decline was predominantly observed in participants with low family income (as an indicator of social deprivation), with no such association noted in those with high family income. The GWAS of MetS (case-control) identified some significant SNPs in the Chinese population. The one-sample MR results showed the non-significant causal association between MetS and memory and cognitive decline. Our results suggest the importance of managing MetS and its components in older adults, potentially aiding in delaying or preventing cognitive impairment, especially in delayed memory recall.

Our results of a significant association of MetS with memory and cognitive decline were in line with some [31-33] but not all [34,35] previous studies, which all used dichotomous classification of MetS. For example, a longitudinal study of 4,150 British participants with an average age of 60 years, measured MetS at three different time points and found that participants with persistent MetS showed poorer cognitive performance than their healthy participants over a 10-year follow-up [33], though cognitive function was only assessed at the end of follow-up. Another cohort study on 4,106,590 Korean participants aged over 40 years and with an average follow-up of 4.9 years, indicated that those with persistent or developing MetS had a higher risk of dementia compared to healthy participants [31]. A similar association was reported in another longitudinal study involving 1,492,776 Korean participants with an average follow-up of 5.2 years [32]. Notably, these studies excluded participants previously diagnosed with dementia, but those with pre-clinical symptoms (i.e., memory loss) might not be identified and subsequently lead to over-estimated results. Conversely, one study based on 5,693 Taiwan participants with a mean age of 63 years and measuring MetS at two different time points, found a non-significant association between developing MetS and dementia risk over a 10-year follow-up [34]. Similarly, another cohort study based on 3,458 Taiwan participants aged over 40 years, also measuring MetS at two different time points, reported a non-significant association between persistent MetS and cognitive decline [35]. The lack of significance in these studies may be attributed to the use of dichotomous classification of MetS, which could reduce the statistical power. Additionally, the heterogeneity in the reference groups could explain the inconsistent results across studies. Therefore, the current dichotomous classification of MetS may not optimally explore the risk of cognitive decline. Previous studies showed that the MetS index was an accurate predictor of the 10-year incidence of cardiovascular disease, suggesting that this index, when used as a standardized continuous variable, was more sensitive in detecting the association with cognitive function [10,36]. Hence, our study provides complementary evidence suggesting that MetS should be considered as a continuum rather than dichotomy.

Moreover, most previous studies describing the association of MetS and its components with cognitive function did not specifically examine the dimensions of cognitive function [5,34]. For example, a prospective study of 5,693 participants in Taiwan with a 10-year follow-up indicated that, compared to healthy participants, those with lower HDL-C levels and higher blood pressure and WC levels at baseline had a higher risk of dementia [34]. Another prospective study of 1,519 participants in Singapore with a 6-year follow-up also showed that higher WC and lower HDL-C levels at baseline showed a positive association with the risk of mild cognitive impairment [5]. Given that both abnormal WC and FPG levels are indicative of insulin resistance, our results are in corroborate with these previous studies. Regarding the dimension of cognitive function, our results suggest that MetS and its components mainly affect attention, calculation, and recall abilities, aligning with previous studies. A systematic review of 19 cohort studies indicated that MetS-related decline in attention, calculation and recall abilities was evident prior to the onset of dementia [37]. However, the Taiwan study reported non-significant associations of baseline MetS and its components with these cognitive dimensions when compared to healthy participants [34]. Another cohort study of 599 Dutch participants over a 5-year follow-up indicated a positive association of baseline MetS status with increasing annual change rates in attention and memory function [38]. Notably, as the average age of participants in this Dutch study was 85 years, survival bias may be a concern.

In our GWAS, we identified significant genetic variants associated with MetS located in the genes of APOA5, zinc finger protein 259 (ZNF259), BUD13 homolog (BUD13) and HTR4. For example, rs662799 in the APOA5 gene, rs6589566 in the ZNF259 gene, and rs10790162 in the BUD13 gene are strongly associated with MetS in our study, aligning with findings from previous studies [16,39,40]. The genetic variants in the APOA5 gene on chromosome 11 are known to influence lipid metabolism [41]. Both the ZNF259 and BUD13 genes, situated in the APOA5 gene cluster on chromosome 11q23.3, have functions similar to the APOA5 gene, affecting lipid metabolism [42]. The role of the APOA5 gene cluster involves either intracellular inhibition of the very low-density lipoprotein (VLDL) assembly or activation of lipoprotein lipase, enhancing lipolysis and VLDL clearance [41]. Therefore, variants in the APOA5 gene could contribute to or exacerbate the dyslipidemia components of MetS, leading to decreased HDL-C levels and increased TG levels.

Moreover, our study is the first to report that the HTR4 gene variant rs1989154 is significantly associated with MetS in Chinese participants. The HTR4 gene at chromosome 5 has been associated with obesity [43]. Previous studies showed that 5-hydroxytryptamine, regulated by the HTR4 gene, was a monoamine neurotransmitter acting as a satiety-generating signal in the brain tissue which regulated food intake in both experimental models and humans [43,44]. Variants in the HTR4 gene might increase the propensity for eating, thereby contributing to the obesity components of MetS. However, due to resource constraints, replication of this genetic variant could not be tested in another sample. Future studies are needed to replicate this locus in other populations.

Our results showed that the association of an elevated MetS index with cognitive decline was more pronounced in participants with low family income, indicating that individuals who were more socially deprived might suffer more from the metabolic abnormality. A cross-sectional study of 5,200 United States participants with high socioeconomic position, as expressed by high educational attainment (median, 16.0 years; interquartile range, 16.0 to 18.0 years of education attained), showed a non-significant association between MetS and cognitive impairment [45]. The authors further examined differences in the MetS-cognitive impairment association by levels of sociodemographic (age, sex, education) and clinical factors but did not observe significant variation [45]. A possible explanation is that individuals who are more socially deprived (i.e., low family income or low socioeconomic position) might increase vulnerability to disease through direct physiological process and unhealthy behaviors [46]. Additionally, limited resources also play a role. Participants with low family income are less likely to have adequate healthcare utilization or access to other essential services. These factors may collectively exacerbate the impact of MetS and the related burdens.

The mechanism underlying the association between MetS and cognitive decline may be attributed to the multifactorial pathogenesis of its components. For example, dyslipidemia can lead to dysfunction of the cellular lipid membranes. This dysfunction may augment the enzymatic activity of beta-site amyloid precursor protein cleavage enzyme-1 (BACE-1) and gamma-secretase, thereby accelerating the cleavage of amyloid precursor protein (APP). Consequently, this process could result in increased production of β-amyloid plaques [47]. Hypertension may cause alterations in cerebral vessels, leading to endothelial dysfunction and an elevated risk of atherosclerosis. Such vascular changes can directly damage brain tissue (i.e., white matter), leading to cognitive impairment. They may also lead to ischemic stroke, subsequently causing post-stroke dementia [48]. Additionally, both hyperglycemia and obesity are associated with insulin resistance and inflammation, which can cause overexpression of pro-inflammatory cytokines, and lead to neurodegeneration and neurotoxicity in brain tissue [49].

The strengths of the present study include a population-based longitudinal study design with an adequate follow-up period and sample size, repeated measurement of MetS and its components, memory function and cognitive function, and conducting a GWAS of MetS in a homogenous Chinese population. Additionally, different methodologies were used to verify the association of MetS with memory and cognitive decline (prospective cohort and MR studies), complemented by comprehensive adjustment for multiple potential confounders. However, several limitations should be acknowledged. Firstly, the sample sizes for the GWAS and MR analyses were relatively small compared to the overall cohort, which might have led to insufficient statistical power in detecting significant causal links between MetS and cognitive decline. Future studies with larger sample sizes are needed to validate our findings and potentially reveal stronger genetic associations. Meanwhile, our MR analysis might be constrained by the use of a limited number of SNPs to construct the weighted genetic score, potentially limiting the comprehensiveness of the genetic profile for MetS. Further studies could explore the use of polygenic risk scores that incorporate a larger number of SNPs to provide a more holistic genetic assessment. Secondly, the direction of the association of MetS with memory recall and registration appeared to be opposite, which could be due to the different assessment methods. Note that memory recall relies heavily on the ability to consolidate information, while registration involves simply repeating words. Additionally, a recent cohort study on 1,037 Australian adults examining the validity of the MMSE dimensions showed that orientation, attention and recall scores in the dementia group were significantly lower than those in the healthy group. In contrast, scores for registration and language did not differ significantly between the two groups [50]. This result suggests that MMSE dimensions may not be robust indicators of specific cognitive domains, such as language and registration, indicating the need for more specific neuropsychological tests to assess these aspects [50]. Therefore, memory recall is considered a more reliable indicator of memory function than registration. Regarding the language dimension, our unexpected results of a negative association between HDL-C and language underscores the necessity for future studies to use more specific tests for language and examine the association between MetS and language function. Thirdly, the history of diabetes was self-reported, and the lack of HbA1c measurement could lead to an underdiagnosis of preexisting diabetes, potentially confounding the observed associations between MetS components and cognitive decline. Finally, all GBCS participants were recruited in Guangzhou, which may not be representative of the general Chinese population. However, given that the prevalence of chronic diseases such as diabetes in our cohort was quite similar to the nationally representative samples of urban Chinese [17], the generalizability of our findings to a broader population might not be a concern.

In conclusion, our study showed a significant association of MetS and its components with declines in memory and cognitive function, especially in delayed memory recall. Given that Asia has a high prevalence of metabolic disease, our findings underscore the importance of effectively managing MetS and its components in older adults to delay or prevent cognitive impairment, with a specific emphasis on mitigating the impact on delayed memory recall.