Background

Dementia is characterized by an inexorably progressive impairment of cognition and the capacity to carry out activities of daily life. It is a heterogeneous syndrome posing a substantial burden on patients, their proxies, and national health-care systems [1]. In the UK, over 850,000 individuals suffer with dementia [2]. Globally, roughly 50 million individuals have dementia, with this figure expected to rise to 152 million by 2050 [1]. In the absence of effective pharmacological treatments for dementia, the identification and detailed investigation of potentially modifiable protective factors have gained considerable attention in recent years.

Glucosamine is a widely used non-vitamin, non-mineral supplement for relieving both osteoarthritis and joint discomfort [3]. It is an approved osteoarthritis prescription medication in most European nations and is widely used as a nutritional supplement in countries like the USA and Australia, where roughly 20% of adults use it daily [4, 5]. Despite the controversy regarding the efficacy of glucosamine supplements on osteoarthritis and joint discomfort [6, 7], glucosamine has been proved to have anti-inflammatory properties [8] and may prevent a wide range of diseases [9, 10]. In this instance, a variety of epidemiological studies have revealed that glucosamine consumption may protect against colorectal cancer [11, 12], lung cancer, [13], cardiovascular disease [14, 15], diabetes [16], and all-cause death [17]. Importantly, a cross-sectional research recorded the association between glucosamine consumption and better cognitive function [18]. However, research regarding the association between glucosamine use and dementia risk remains scant.

The importance of glucosamine in brain function has been highly supported by previous studies [19, 20]. Glucosamine mimicked the effects of a low-carbohydrate diet in a prior animal research, resulting in increased lifespan [21], and studies consistently showed that a low-carbohydrate diet protects against dementia [22, 23]. An animal study suggested that glucosamine may promote cognitive function by impacting energy metabolism [20]; other animal models have indicated the neuroprotective and anti-neuroinflammatory effects of glucosamine [24]. In addition, glucosamine participates in the O-linked N-acetylglucosaminylation of various proteins, which was verified to be related to many neurological or neurodegenerative diseases [25, 26]. Therefore, we hypothesize that regular use of glucosamine may have a causal influence on incident dementia.

Based on the UK Biobank study of nearly 500,000 British people, we investigated the relationship between regular use of glucosamine and the risk of all-cause dementia, Alzheimer’s disease (AD) and vascular dementia. We also explored potential modifying effects by several established risk factors (including APOE ε4 genotypes) for dementia.

Traditional observational studies include drawbacks such as residual confounding and/or reverse causation, inadequate adjustment (e.g., healthy lifestyle or other factors), and a focus on correlation rather than causation. By employing genetic variants as proxy for glucosamine use, Mendelian randomization (MR) avoids some of these limitations and provides genetic support for causal associations [27]. Thus, in addition to observational analysis, we performed MR to give additional insights for the assessment of potential causal relationships.

Methods

Study design

This study analyzed data from UK Biobank (application 55,794), a large prospective cohort study enrolling over 500,000 participants between the ages of 40 and 70 from 22 research centers in the UK (England, Wales, and Scotland) between 2006 and 2010 [28]. Through detailed electronic questionnaires, face-to-face interviews, and physical assessments, participants provided personal data on health-related variables. Participants who dropped out of the study (n = 1298), had dementia (n = 224), or lacked data on glucosamine use (n = 6171) were excluded from the analysis. We also excluded 15,339 participants from further analysis owing to a lack of quality-controlled genotyping data (Fig. 1).

Fig. 1
figure 1

Flow diagram of the participant selection process

Full size image

Exposure assessment

At one of 22 assessment sites across the UK, the participants filled out a touch-screen questionnaire. In answer to the question “Do you usually take any of the following?”, a list of supplements, which included glucosamine, was given to the participants to choose from. Based on these data, we established a binary classification for regular glucosamine use: 1 = yes, 0 = no. This evaluation method was used in previous studies [16, 29, 30].

Ascertainment of incident dementia

We used participants’ baseline survey information, hospital admission diagnosis records, and death registration records to define outcomes, which included all-cause dementia, AD, and vascular dementia. Diagnoses were recorded using the International Classification of Diseases (ICD) coding system (Additional file 1: Table S1) [31]. The incident disease in this study was determined by the primary or secondary diagnoses from hospital admission data or primary or secondary causes of inducing death after baseline data collection. A subsample of the population was also retrieved from primary care data using Read Codes (version 2 or 3) in the sensitivity analysis [32]. Participants were followed up from the time of the baseline to the first diagnosis, death, or February 25, 2018, in Wales and England and February 28, 2017, in Scotland, whichever came first. Detailed information on the APOE genotyping is presented in the Additional file 1: Supplemental Methods [33,34,35].

Covariates

Various potential confounders were assessed using a baseline touch-screen questionnaire. Age, gender, ethnicity, the Townsend Deprivation Index (TDI), level of education, and annual household income were included as the sociodemographic factors. Lifestyle behavior included smoking status, alcohol intake, being physically active, body mass index (BMI), vegetable intake and fruit intake. Health-related variables included hypertension, cardiovascular disease, cancer, digestive disease, depression, diabetes, emphysema or chronic bronchitis, high cholesterol, chronic kidney disease, chronic liver disease, and Elixhauser Comorbidity Index. Medication utilization included antihypertensive drugs, insulin treatment, statin, opioids, aspirin, and other non-steroidal anti-inflammatory drugs (NSAIDs). We also included chondroitin, dietary supplements for minerals, vitamins, and other nutrients (fish oil, calcium, iron, zinc, and selenium), memory, and reaction time in the analysis. The details on calculating the Elixhauser Comorbidity Index are shown in the Additional file 1: Supplemental Methods [36, 37]. We calculated participants’ BMI by dividing their weight by the square of their height in meters. The TDI is a comprehensive poverty index, which is calculated by the following factors: ownership of a home, ownership of a car, being unemployed or not, and whether there are too many people living together [38]. It shows the socioeconomic status of a participant. According to WHO guidelines on physical activity for health [39], we classified individuals as < 150 or ≥ 150 min/week based on total minutes of moderate physical activity per week (collected by touchscreen question, one vigorous physical activity minute equals two moderate physical activity minutes). This assessment method was widely used in prior studies [40, 41]. Patients with any of the following situations are classified as having hypertension: using hypertensive drugs, systolic blood pressure higher than 140 mmHg, diastolic blood pressure higher than 90 mmHg, or self-reported hypertension. Health status was determined through self-reporting combined with ICD-10 codes from hospital records. Memory and reaction time assessments were conducted through touch screen [42,43,44]. A pair matching test was used to measure memory, in which participants had to recall six pairs of shapes and their positions in 5 s. The number of mistakes made during matching was used to evaluate performance. The test method of reaction time is shown below: a series of figures will be displayed on the screen; the participant was asked to press the button as rapidly as possible when two identical figures appear. The mean response time (ms) across eight rounds for properly selected matching groups was used to measure performance. The UK Biobank website has further information on these variables (www.ukbiobank.ac.uk).

Statistical analysis

Observational analysis

For continuous variables, the mean (SD) is used, and for categorical variables, the number (%) is used. We performed multiple imputation with chained equations to cope with missing variables to reduce the possibility of inferential bias [45, 46]. There were five datasets imputed. The imputation model contained all variables used in the analysis. Additional file 1: Table S2 provides detailed data on missing variables.

Cox proportional hazard models were used to calculate the hazard ratios (HR) and 95% confidence intervals (CIs) for the relationships between regular glucosamine use and all-cause dementia, AD, and vascular dementia. We tested the proportional hazards assumption by Schoenfeld residual tests [47], and no violations of this assumption were identified. Two models were used. We only included sex and age in Model 1. Additional factors, such as ethnicity, education, TDI, annual household income, BMI, fruit intake, vegetable intake, smoking status, alcohol intake, being physically active, medical conditions, drug use, other supplement use, memory, and reaction time were adjusted in Model 2.

In order to evaluate potential effect modifiers, subgroup analyses based on sex (female or male), age (< 60 or ≥ 60 years), obesity (BMI ≥ 30 kg/m2, no or yes), current smoking status (no or yes), diabetes (no or yes), hypertension (no or yes), aspirin use (no or yes), use of non-aspirin NSAIDs (no or yes), use of vitamin supplementation (no or yes), use of other non-vitamin supplementation (no or yes) and APOE ε4 carrier (no or yes) were performed. To investigate the differential effects of glucosamine on the likelihood of dementia in subgroups, we calculated the p-value for interaction by including the cross-product term of the stratifying variables with glucosamine use in the fully adjusted model.

We evaluated the robustness of our findings by a sequence of sensitivity analyses (Additional file 1: Table S3). Firstly, we conducted an analysis of competing risks which considered all-cause mortality as a competing event for dementia. Secondly, since individuals who took glucosamine were more likely to take chondroitin than those who did not, we conducted sensitivity analyses by removing chondroitin users. Thirdly, to minimize the possibility of reverse causality, we excluded people who died within the first 2 years after baseline assessment. Fourthly, we excluded individuals who had missing covariate values. Fifthly, we calculated a propensity score for each participant and further adjusted for the score in the fully adjusted model. The multivariate logistic regression model was applied to estimate propensity scores taking all covariates into account. Sixthly, we added a subsample of the population retrieved from primary care data using Read Codes (version 2 or 3) in the analysis. We performed all analyses using R version 4.0.3, and p less than 0.05 (two-sided) was deemed significant.

Mendelian randomization

We performed a two-sample MR design using summary-level data. Single-nucleotide polymorphisms (SNPs) served as risk factor instruments. The analysis relied on public summary-level data. All original studies received ethical approval. A detailed description of the data sources, the selection of genetic instrumental variables, and the test on instrument strength and statistical power are shown in the Additional file 1: Supplemental Methods [48,49,50,51,52,53,54,55,56]. Data sources and instruments are listed in Additional file 1: Table S4-12. The GWAS from Ben Neale Lab round 2 used a linear regression model in Hail for large-scale phenotypes in the UK Biobank, even for binary variables. As a workaround, we used BOLT-LMM, a software package widely used to deal UK Biobank data, to calibrate the effect. The detailed information on BOLT-LMM is shown in the Additional file 1: Supplemental Methods [57].

The “MendelianRandomization” and “TwoSampleMR” R packages were used for all statistical analyses. Inverse variance-weighted (IVW) MR was the major analysis we applied for single-variable MR analysis. In order to deal with the issue of the robustness of the IVW result, we used the MR-Egger and weighted median-based regression, both of which assume distinct instrumental variables assumptions [54, 58]. When all genetic variants are invalid instrumental variables, the MR-Egger regression produces consistent results; the weighted median needs valid IVs to contribute 50% of the weight. The accuracy of weighted median estimates and IVW estimates are almost the same, which are much higher than that of MR-Egger estimates because the accuracy of MR-Egger estimates is particularly imprecise when all the IVs are about the same strength [59]. To assess potential IV violations, we carried out the MR-Egger intercept test [60], MR pleiotropy residual sum and outlier (MR-PRESSO) test [61], and Cochran Q heterogeneity test [62]. To identify high-influence points, we used a leave-one-out validation [63]. The MR Steiger test was also performed to assess the potential reverse causal effect of glucosamine on dementia (Additional file 1: Table S13) [64].

The likelihood of using additional supplements is higher among glucosamine users than in non-glucosamine users. Taking these associations into account, we conducted multivariable MR to assess the direct effect of regular use of glucosamine on dementia under the condition of controlling vitamin, chondroitin supplements intake, and osteoarthritis. We aggregated the genetic instruments used in the related GWASs—glucosamine, vitamin, chondroitin supplements, and osteoarthritis. SNPs were clumped by linkage disequilibrium within a window of 10,000 kb (R2 < 0.001) to confirm their independence. Then we derived SNP effects and standard errors from the GWAS summary statistics and harmonized them with GWAS data on dementia. Measured and unmeasured pleiotropy were taken into account by using multivariable MR extension of the IVW MR approach [65] and the MR-Egger method [66].

Results

The mean age of the 494,814 participants was 56.5 years (SD 8.1) and the proportion of female was 54.4%. At baseline, 94,259 (19.0%) of participants reported using glucosamine. There was a higher percentage of older, female, non-smoking, and physically active glucosamine users than nonusers. In addition, glucosamine users had a lower TDI, higher prevalence of comorbidities such as cancer, hypertension, arthritis, and depression, but less cardiovascular disease, emphysema or chronic bronchitis, diabetes, high cholesterol, chronic kidney disease, and chronic liver disease. The percentage of using statin, opioids, non-aspirin NSAIDs, chondroitin, vitamins, minerals, and other dietary supplements was higher in glucosamine users than non-users (Table 1).

Table 1 Baseline characteristics of study participants by glucosamine use
Full size table

Associations of glucosamine use with incident dementia

During the 8.9-year (IQR 8.3–9.7 years) median follow-up, we recorded 2458 cases of all-cause dementia, 924 cases of AD, and 491 cases of vascular dementia. Table 2 shows the associations of regular use of glucosamine with the outcomes. A statistically significant inverse relationship was found between glucosamine use and risk for all-cause dementia (HR 0.81; 95% CI 0.73–0.90), AD (HR 0.78; 95% CI 0.65–0.92), and vascular dementia (HR 0.68; 95% CI 0.54–0.87). The hazard ratios of glucosamine users in multivariable-adjusted models were 0.84 (95% CI 0.75 to 0.93) for all-cause dementia; 0.83 (95% CI 0.71 to 0.98) for AD; and 0.74 (95% CI 0.58 to 0.95) for vascular dementia (Table 2).

Table 2 Associations of regular glucosamine use with incident dementia
Full size table

Subgroup and sensitivity analyses

To investigate potential subgroup effects, we conducted several specified subgroup analyses (Fig. 2). We found that the protective effect of glucosamine on AD was stronger among participants aged below 60 years, compared with those above 60 years (p = 0.04 for interaction). Other stratifying variables have not modified the association of glucosamine use with incident dementia (p for interaction > 0.05).

Fig. 2
figure 2

Relationship between glucosamine use and risk of dementia stratified by potential risk factors. Findings were adjusted for age, sex, ethnicity, education, Townsend Deprivation Index, household income, body mass index, fruit consumption, vegetable consumption, smoking status, alcohol consumption, physical activity, health condition, antihypertensive drugs, insulin treatment, statin use, opioids use, chondroitin use, aspirin use, non-aspirin NSAID use, vitamin supplementation, mineral and other dietary supplementation, memory, and reaction time

Full size image

When we excluded participants who had outcomes within 2 years of follow-up, participants who used chondroitin, and participants with missing values for variables, the relationships of glucosamine use with all-cause dementia, AD, and vascular dementia persisted. After adding cases that were retrieved from the primary care data using Read Codes, the results did not alter. During the follow-up in participants without dementia, 19,082, 19,654, and 19,763 deaths were documented as competing events for all-cause dementia, AD, and vascular dementia, respectively. The competing risks analysis produced results which were consistent with the Cox proportional hazards model (Additional file 1: Table S2).

Mendelian randomization

According to Univariable MR analysis, genetically determined regular glucosamine use was associated with a decreased risk for all-cause dementia (IVW odds ratio, 0.85; 95% CI 0.76 to 0.95), AD (IVW odds ratio, 0.85; 95% CI 0.78 to 0.93) and vascular dementia (IVW odds ratio, 0.64; 95% CI 0.42 to 0.96) (Table 3). Weighted median and the MR-Egger provided similar estimates to those of IVW. The accuracy of the MR-Egger estimations was much lower. We use forest plots to display the MR results for the impacts of SNPs related to glucosamine use on dementia risk (Additional file 1: Fig. S1). No pleiotropy across instruments has been found by the Cochran’s Q statistic. We did not find directional pleiotropy by MR-Egger intercept analysis. No potential outliers were found by MR-PRESSO. No high leverage, high impact points were found using conventional IVW leave-one-out analysis (Additional file 1: Fig. S2). Absence of weak instrument bias is shown by F statistics for genetic instruments (Additional file 1: Table S5). Using the MR Steiger test, we detected no evidence of reverse causality (Additional file 1: Table S13).

Table 3 MR results for the relationship between regular glucosamine use and incident dementia
Full size table

In MVMR, the genetic liabilities for regular glucosamine, vitamin, chondroitin use, and osteoarthritis were evaluated. Use of glucosamine continued to have a significant effect on all-cause dementia (IVW odds ratio, 0.88; 95% CI, 0.81–0.95; P < 0.001), AD (IVW odds ratio, 0.78; 95% CI, 0.72–0.85; P < 0.001) and vascular dementia (IVW odds ratio, 0.73; 95% CI, 0.57–0.94; P < 0.001). These results align with those derived from the MVMR-Egger sensitivity analyses. Again, no horizontal pleiotropy was found in the MR-Egger intercept analysis. Details on the instruments used in MVMR can be found in Additional file 1: Table S14-15.

Discussion

We observed that regular glucosamine use was related to a 15% decreased risk of all-cause dementia, 17% for AD, and 26% for vascular dementia in this large population-based study including 494,814 participants. These associations remained after adjusting for variables including sociodemographic factors, lifestyle behavior, comorbid conditions, medication, and other dietary conditions. Moreover, the beneficial effect of glucosamine use on AD seemed to be larger in participants aged below 60 years than in those aged above 60 years. The APOE genotype did not modify this association. In the MR analysis, we again observed protective causal effects of regular glucosamine use on dementia risk. Our findings were mostly consistent among various MR methods that made various assumptions regarding horizontal pleiotropy, demonstrating that horizontal pleiotropy is not probable to be a sufficient explanation for our findings.

We found that 19.0% of participants used glucosamine; this number is close to the 22.0% of the Australians over 45 who also take glucosamine [5]. Our findings are in line with a prior cross-sectional investigation that found glucosamine intake to be related to better cognitive function [18]. Glucosamine users had a higher reasoning score and faster reaction speed than non-users [18]. Furthermore, in a mouse model, glucosamine exerted a cognition-enhancing function [20], which implicated the beneficial impact of glucosamine use on dementia prevention.

Because glucosamine and chondroitin supplements are typically used simultaneously once daily [6], our observed relationships might be attributed to either of these supplements. To address this concern, a sensitivity analysis was conducted to test whether glucosamine alone (without chondroitin) could prevent dementia. No substantial change occurred in the sensitivity analyses. Thus, we speculate that glucosamine use might have a preventive role in the development of dementia, independent of chondroitin co-administration.

In our study, a stronger effect was found between glucosamine use and AD among participants aged below 60 years compared with those above 60 years. The weaker effect of glucosamine use in older participants may be related to the gradual atrophy of the hippocampus and the reduction of cortical density as the age increases, resulting in the reduction of brain cell membrane receptors and the decreased sensitivity to drugs [67]. This result underscores the age-modified connection between glucosamine use and dementia and emphasizes the importance of early prevention of dementia.

The protective association between glucosamine use and dementia may be explained by a few different processes. As a popular supplement that can pass through the blood–brain barrier, glucosamine may get to the hippocampus, striatum, and cortex [68, 69]. Meanwhile, several glucosamine transporters were identified in the brain [70]. For instance, glucose transporter 2 (GLUT2) was found in neurons and exhibited the greatest affinity for glucosamine [71, 72]. Intriguing evidence indicates that specific neuronal populations rely on GLUT2 to regulate glucose levels, thereby affecting their vulnerability to pathogenic mechanisms underlying AD [73, 74]. These studies highly support the important role of glucosamine on dementia. C-reactive protein, an indicator of systemic inflammation, was significantly lower in those who regularly took glucosamine, according to data from the National Health and Nutrition Examination Survey (NHANES) [8]. Animal studies also showed that glucosamine might suppress neuroinflammation [75], which is proved to increase the risk of dementia [76]. Furthermore, a prior research discovered that glucosamine might simulate a low-carbohydrate diet in mice through lowering glycolysis and enhancing amino acid catabolism [77]: consequently, glucosamine has been considered a mimicking agent for energy restriction [21]. Recent works demonstrated that a low-carbohydrate diet protects against the development of dementia [78, 79]. In addition, glucosamine could reverse the imbalanced gut microbiota [80]. Through the gut–brain axis, the gut microbiota modulates the brain functioning of the host and plays a significant role in dementia pathogenesis [81, 82]. Thus, glucosamine might have a beneficial effect on dementia pathology by regulating the gut microbiota. Other pathways may possibly be relevant and warrants further studies to explore the functional roles of glucosamine in dementia.

Our research had a number of advantages, such as a large number of participants and abundant data on dietary, health-related behaviors, and various factors that enabled us to examine the robustness of the findings and explore the effects of exposure in several subgroups. Furthermore, the MR analysis offered a superior method of obtaining somewhat less confounded estimates of causal associations that were not impacted by reverse causation or confounding. We admit that our research has limitations. Firstly, the “regular glucosamine use” was defined as self-reported at the baseline only, which might have changed in the follow-up period. Details on glucosamine use, such as dose and use duration, were not collected in the UK Biobank, which may weaken the study findings. Hence, further research that incorporates the glucosamine intake pattern and cross-validates the data on glucosamine for accuracy is required to delve into these connections. Secondly, UK Biobank did not record the adverse side effects participants suffered after using glucosamine. Nonetheless, glucosamine has been proved to be a safe supplementation for individuals with osteoarthritis due to its low risk of side effects including rare allergic reactions and gastrointestinal reactions [3]. Although people at high risk of diabetes showed reduced glucose tolerance after taking glucosamine [83, 84], studies have proved that in healthy people and diabetic patients, any oral dose of glucosamine will not affect the glucose metabolism and lipid status [85, 86]. Thirdly, in general, 20–100 imputed datasets are recommended, while in this study 5 datasets were imputed. Due to rather low proportions of missing data, we consider five imputed datasets to operate well. Fourthly, despite the SNPs we used were significantly correlated with the exposure, the genetic variants reflected only a modest portion of the overall variance in glucosamine intake, limiting them from being precise proxies of exposure. Given that we do not yet know how the genetic instruments work biologically, we cannot totally eliminate out breaches of the independence and exclusion restriction assumptions, especially with regard to pleiotropy [63]. Nevertheless, to infer reliable causal estimates, we used a variety of techniques, including Cochran's Q statistic, MR-PRESSO, weighted median, and MR-Egger. Fifthly, the interpretation of genetic liability of supplement use should be cautious as genetic predictors of glucosamine may capture participants with worse joint health [87]. We further adjusted osteoarthritis in the multivariable MR analysis to reduce bias. Sixthly, MR is a useful option for validating results; nevertheless, genetic variants reflect lifetime exposures rather than brief treatment modalities, which may create a bigger impact than a time-limited intervention [88]. Therefore, our findings should be taken cautiously, since they are hypothesis generating and warrant more clinical data to further investigate the connection between glucosamine intake and dementia. Seventhly, although the current definition for dementia was widely used in previous studies and the true positive rate for all-cause dementia collected in the UK Biobank was as high as 82.5% [89]; the true positive rates of Alzheimer’s disease and vascular dementia were lower than 75%. Thus, the results on the subtypes of dementia should be taken cautiously.

Conclusions

Regular glucosamine use was associated with a lower risk of all-cause dementia, AD, and vascular dementia, based on data from the UK Biobank cohort and a mendelian randomization study. The potential implications of our findings for dementia prevention need additional confirmation in well-powered randomized controlled trials. We also recommend additional basic scientific research to investigate the underlying mechanisms.