Sugary beverage intake and genetic risk in relation to brain structure and incident dementia: evidence from the UK Biobank

Multivariable-adjusted through


Introduction
Dementia, a major type of neurodegenerative disease, poses great burdens on the well-being and economic cost of caring for older people worldwide as lifespan continues to increase [1][2][3]. Considering its limited therapy, it is thus crucial to identify the risk factors, especially modi able ones, of dementia for prevention measures. Recent population-based studies [4][5][6] supported that dietary habits are associated with dementia risk in late life. Among multiple dietary factors identi ed, sugary beverages (SBs) intake is widely recognized worldwide as a driver of poor cardiometabolic health [7][8][9], with uncertain relation to neurological health.
Prior ndings [10,11], although scarce in well-designed cohort settings, generated a plethora of discourse on whether high consumption of SB is a risk factors for dementia. A prospective study [10] of 1,484 adults found that arti cially-sweetened beverage (ASB) intake (>0 serving/week vs. none), but not sugar-sweetened beverage (SSB) intake, was associated with a 98%-159% increased risk of dementia after adjustment for major potential confounders including other dietary confounders. Another cross-sectional study [11] suggested that SBs, including fruit juice, are associated with lower brain volume and poorer cognitive performance.
However, evidence is con icting, with other studies nding that fruit intake and fruit juice intake was inversely associated with poor cognitive function [12,13] and dementia [14,15]. In addition, due to the relatively limited sample size, estimates in these studies often came with wide con dence intervals. Hence, we investigated the long-term association of SBs, including SSB, ASB, and naturally sweet juices (NSJ), with incident dementia and brain structural markers in UK Biobank, a large population-based prospective study in the UK.

Study population
This study was based on the UK Biobank (UKB), a population-based cohort study in the UK with deep genetic and phenotypic data collected [16]. Commenced in 2006, UKB recruited over 500,000 UK residents aged over 40 at 22 assessment centers, as described in more detail elsewhere on the UK Biobank website (http://www.ukbiobank.ac.uk/resources/). Ethical approval was granted by the North West-Haydock Research Ethics Committee (REC reference: 16/NW/0274).
We included 187,994 participants in the UKB ( Figure S1) who: 1) nished the Oxford WebQ, a web-based 24-h dietary assessment tool, at least once; and 2) did not exit the program until March 2021. To minimize measurement error, we excluded self-reported non-typical day of diet recalls and those with extreme energy intake (>20MJ/d) [17]. To reduce reverse causation, we excluded 1067 participants who developed dementia or died within the two years subsequent to baseline dietary assessment.

Sugary beverage intake
Repeated web-based 24h diet recall questionnaires (Oxford WebQ) were introduced to the assessment, with invitations being emailed to participants every 3-4 month (mean repetition = 1.93). Participants recalled how much SBs they consumed in questionnaires that have been validated using biomarkers [18] and by interviewer-administered 24 h recalls [19] and showed good reproducibility [20]. In this study, SBs consisted of sugar-sweetened beverage (SSB, i.e., zzy drinks and squash), arti cially-sweetened beverage (ASB, such as diet zzy), and naturally sweet juices (NSJ, consisting of fruit and vegetable juices). The intake level of them were comparable with national data in UK [21]. We categorized daily intake of these beverages into 4 groups: 0, 0~1 (i.e. >0 and <=1) unit/d, 1~2 (i.e. >1 and <=2) units/d, or >2.0 units/d, in which a unit refers to one glass/can/carton or 250 mL of the beverages, and their correlation was presented in Table S2.

Dementia and its subtypes
Dementia cases in the UK Biobank were linked to hospital admissions and death registries. We identi ed Alzheimer's Dementia (AD), and vascular dementia (VaD), and other types or unde ned dementia from International Classi cation of Diseases (ICD) codes of hospital inpatient admission data (Table S1). In this study, health conditions were updated in the linkage to the healthcare system to March 2021.

Brain structure
Brain structure data in the UK Biobank sample were obtained from magnetic resonance imaging (MRI) since 2014 [22]. Our study used imaging-derived phenotypes generated by an image-processing pipeline developed and ran on behalf of UKB. In this study, the volumes (in mm 3 ) of the whole brain, white matter, grey matter were derived from T1 structural brain MRI, and the volume of white matter hyperintensities (WMHs) was derived from T2-weighted brain MRI. The external surface of the skull was estimated from the T1 and used to normalize brain tissue volumes for head size. Head-size adjusted brain volumes were added correspondingly (left and right sides) and then z-standardized. For WMH, the volume was transformed by taking the logarithm before z-standardized because of its skewed distribution.

Polygenic Risk Score
A polygenic risk score (PRS) capturing each participant's load of common genetic variants related to the risk of Alzheimer's disease and dementia was constructed in the UK Biobank sample [23], with details being described in a previous study [4]. Brie y, the analysis was constrained to participants with white backgrounds.
Single-nucleotide polymorphisms (SNPs) were selected using "clumped" results so that the remaining SNPs were the most signi cant variant per linkage disequilibrium block, common, and available in the UK Biobank.
The threshold of inclusion for a P-value was <0.5. The number of associated alleles at each SNP was weighted according to the regression coe cient with AD in the discovery stage of GWAS results, summed, zstandardized, and then divided into tertiles.

Covariates
In this study, we included multiple covariates for confounding adjustment, which were selected based on previous literature and prior knowledge on the potential causal pathways [4,10,17]. Demographic characteristics, including age, sex de ned by self-reported identity, race, education level, and Townsend deprivation index (indicating the social deprivation status), and lifestyle factors, including smoking status, body mass index (BMI), alcohol drinking status, and physical activity levels, were all collected at baseline.

Statistical analyses
Participants' baseline characteristics were presented by their SB intake. Continuous variables were displayed as means (standard deviations, SDs), and categorical variables were shown as numbers (percentages). Cox proportional hazard models were used to estimate the hazard ratios (HRs) and con dence intervals (CIs) for SBs intake in categories and incident dementia, with person-years being calculated from date of the rst 24-h diet recall report to the diagnosis of dementia, the ascertainment of death, or loss to follow-up, whichever came rst. Missing values were imputed to median or class with the most participants. Proportional hazard assumption was tested by entering an exposure-time interaction term in the model. The HR were adjusted for age, age-square and sex in model 1. Model 2 was additionally adjusted for Townsend deprivation index (low, medium, or high deprivation), education (college or above, or high school or below), physical activity (low, medium, high), smoking (ever smoked or not), alcohol intake (currently drinking or not), total energy intake, and alternative healthy eating index (AHEI) excluding SB component. Bodyweight status categories (underweight, normal weight, or overweight) were further adjusted for in model 3. Penalized splines were used to explore the potential non-linearity by treating SB intake as a continuous variable, with maximal Akaike information criterion (AIC) being used to choose an optimal degree of freedom [24]. We tested the mediation effect of incident diabetes using "mediation" package [25] and reported the quasi-Bayesian estimates.
In the secondary analysis, we investigated the association of SBs with AD and VaD using model 3 mentioned above. To explore the joint association of polygenic risk and SBs with dementia, we categorized participants into 12 groups by crossing 3 PRS quantiles and four intake levels for each type. Specifying the non-intakers with low PRS as the reference group, we estimated the HRs using model 3 mentioned above and further adjusted the association for 20 principal components of population structure, number of risk bases, and kinship. To explore potential effect modi cations by major covariates, we conducted subgroup analyses strati ed by age, sex, Townsend deprivation index levels, smoking status, alcohol drinking status, body weight status, educational levels, and PRS. P for interactions between these covariates and SBs were calculated by entering a multiplication term in model 3.
In assessing the relation of SBs to the brain structure, since total and regional brain volume was only available for participants undergoing image assessment, we used data of a subgroup of participants (N=12,566) for these analyses. Linear regression was used to estimate the beta coe cients of SBs with brain volume measurements, with the differences in z-score being presented.
To assess the robustness of our ndings, we conduct several sensitivity analyses in several steps: 1) adding baseline comorbidities (cancer, cardiovascular diseases, hypertension, and diabetes), which may lie within the causal pathway of SBs and dementia, in the models; 2) excluding participants with baseline cancer, cardiovascular diseases, or diabetes, because may also have changed their dietary intake because of their disease status; 3) mutually adjusting the models for three types of SBs; 4) further adjusting the relation for total sugar intake; 5) adjusting the models for BMI and BMI square instead of BMI categories; and 6) assessing the association of SBs with brain structure restricted to participants aged over 60 years at baseline. Statistical analyses were performed using R 3.6.0, and two-sided P-values below 0.05 were considered statistically signi cant.

Participant Characteristics
Of the 187,994 dementia-free participants, the mean (SD) age at baseline was 56.2 (7.9). Among them, 44.9 % were female, and 96.4 % were White/Caucasian (Table 1 & Table S3). During the follow-up period (mean = 9.5 y), a total of 1,351 dementia cases were reported. Participants who consumed more SSB were more likely to be female, younger, and had higher income.

Sugary beverages and incident dementia
All three types of SBs were associated with incident dementia (Table 2). Participants reporting to be consuming >2 units/d of SSB were at higher dementia risk (HR=1.47; 95% CI, 1.13~1.92) compared to those who did not drink any, partially mediated by type 2 diabetes (proportion = 1.81%, P-mediation=0.02). Higher ASB intake was also associated with a higher dementia risk, with its intake of 0~1 unit/d being associated with 1.21-fold hazard (95%CI, 1.03-1.43), intake of 1~2 units/d with 1.50-fold hazard (95%CI, 1.19~1.90), and the HR for intake over 2 units/d being 1.41 (95%CI, 1.00~1.99). On the contrary, participants with moderate NSJ intake (0~1 unit/d) were at a decreased risk (HR=0.80; 95%CI, 0.71~0.90) compared with non-NSJdrinkers. We did not observe a signi cantly higher or lower risk for participants who consumed NSJ more than 1 unit per day as compared to none. No mediation effect of type 2 diabetes was detected for ASB or NSJ.
When merging all three types of beverages into one (Table S4), we found that higher total SB intake was associated with an elevated dementia risk (HR=1.25 per unit/d; 95%CI, 1.06~1.46).
We used penalized splines to estimate the potential non-linear associations (Figure 1), we found that SSB and ASB were linearly associated with higher incident dementia risk (P-nonlinearity=0.09 for SSB, and 0.17 for ASB). Also, daily sugar intake over 100 g was associated with a higher dementia risk. We found a non-linear Jshaped curve (P-nonlinearity<0.001) for NSJ, with the trough of HR being observed at approximately 1 unit/d.
The corresponding associations were signi cantly modi ed by genetic risk of dementia (P-interaction=0.0016 for SSB and PRS, 0.0013 for ASB and PRS, and 0.0010 for NSJ and PRS), with stronger associations of all three types of SBs being observed among individuals with medium and higher genetic risk ( Figure S2). Viewed differently, the genetic risks were signi cantly magni ed by higher intake of SSB (HR=1.70; 95%CI: 1.05~2.75 for >2 unit/d and high PRS) and ASB (HR=2.16; 95%CI: 1.24~3.77 for >2 unit/d and high PRS), and was instead attenuated by moderate intake of NSJ (Figure 2).

Subgroup and sensitivity analyses
Generally, the primary ndings were consistently observed across major subgroups of participants strati ed by age, sex, Townsend deprivation index, education level, smoking status, alcohol drinking, and BMI categories (Table S7). ASB consumption was less associated with dementia in participants with low deprivation level (P-interaction<0.001). The association of SSB and NSJ with dementia was consistent among all subgroups (P-interaction>0.05 in all tests). In the sensitivity analyses (Table S8) the association of SBs with dementia remained similar.

Discussion
In this prospective study of adults in the UK, higher SSB (>2 unit/day) and ASB (>0 unit/day) intake were associated with higher dementia risk, while consuming moderate NSJ (0~1 unit/day) was associated with a lower risk and lower level of suboptimal brain structural markers. These associations were similar across major subgroups but was signi cantly altered by genetic risk of dementia. In aggregate, the ndings of this study underscored the detrimental role of SSB and ASB and the potential bene cial role of moderate intake of NSJ in the prevention of dementia.
To our knowledge, this study is one of the few to explore the relation of type-speci c SBs with dementia.
Looking at prior ndings, among 2,888 participants aged over 60 years [10], researchers discovered that higher consumption of ASB was associated with a higher risk of Alzheimer's disease during 9.5 years of follow-up. The estimated HR was 2.89 (95% CI, 1.18~7.07) for Alzheimer's disease, while SSB was not signi cantly related. Another study conducted among 1,865 participants in Framingham Heart Study [26] added that consuming sugar from beverages over 7 servings/week was associated with a substantially higher risk of allcause dementia (HR=2.80, 95%CI, 2.24~3.50). Our study, using data of a well-administered European cohort, provided further and strong evidence on brain structure to support the hypothesis that ASB intake as well as SSB intake is associated with a higher risk of dementia, although both HRs were not as high as in previous ndings, potentially due to the younger population or different approach to assess the beverage intake.
Additionally, we observed that SSB was associated with higher risk of AD but not VaD, possibly because SSB consumers with an extensively high vascular risk died earlier [10]. Meanwhile, we only observed the protective association of NSJ with AD, which may be accounted for by a limited number of VaD cases.
In our research on NSJ, which have been consumed as an substitute for SSB [27], moderate intake of it was associated with decreased risk of dementia, which was in concordance with several previous studies. For example, in a prospective study of 1,836 Japanese Americans, drinking juices >=3 times/week was related to a 76% reduced hazard of Alzheimer's disease (HR= 0.24, 95% CI 0.09~0.61) [28]. Other short-term interventional studies also presented similar results that juice intake was associated with slowed cognitive decline and lowered risk of cognitive impairment [29,30]. In another prior study among men in US, fruit juice intake was related with subjective cognitive decline in a dose-response manner from 0~1 serving/d [12]. Our study extended that while moderate intake of 0~1 unit/d was associated with lower dementia risk, participants taking NSJ > 2 units/d was approximately had a similar dementia risk as those who did not drink any. Given that excessive fruit juice intake could be associated with higher diabetes risk [31] or other comorbidity, there awaits further investigation to de ne the optimal level.
Although the biological pathway of the relation between SBs and dementia is not fully understood, several possible mechanisms could shed light on the results. For SSB, excessive sugar intake might induce a rapid rise in blood glucose and insulin [32], thus causing brain dysfunction [33]. For ASB, aspartame could be linked to energy production disruption and increased oxidative stress [34], and thus contributed to a higher risk of dementia [35] Also, the phenylalanine in aspartame could directly affect the synthesis of inhibitory monoamine neurotransmitters and induce neural degeneration [36]. Therefore, its neurological harm may outweigh bene ts from reduced caloric intake. Conversely, rich contents of vitamins [37][38][39], minerals [40], carotenoids [41][42][43], and avonoids [44] in NSJ may have advantaged the effects of excessive sugar and thus protected brain health. Furthermore, oxidative damage caused by the β-amyloid peptide in the pathogenesis of dementia may be hydrogen peroxide mediated [45].
Findings in the present study have timely social and public health implications. For SSB which has long been considered as an excessive energy source, our ndings provided evidence of it being a risk factor of dementia. While food administration departments in some western countries have advocated sugar reduction [46], the risk of excessive added sugar intake remains inadequately noticed by more developing countries. In the meantime, the widely used arti cial sweetener aspartame as a substitution for sugar is quite controversial.
Although aspartame has been suggested to be related to neural dysfunction 39 through abnormal blood glucose level and direct neurological effect, epidemiological evidence remained very scarce in the past. Our ndings suggested that the use of aspartame as a 'healthy' and 'zero-calorie' sweetener may need to be reassessed. For NSJ recommended as a potential healthy beverage alternative, our results also suggest that excessive intake may not play a protective role. Therefore, it is necessary to emphasize a moderate quantity when recommending fruit and vegetable juice intake.
The present study has several strengths. First, the large sample size and relatively long-term follow-up enabled us to explore a comprehensive relation between SBs and dementia incidence. To the best of our knowledge, our research is one of the largest of its kind, and the population-based design with high representativity ensured the generalizability of the results. The availability of genetic data and brain images in the database also allowed in-depth analyses. Secondly, linkage to registered healthcare records, low rate of loss to followup, less affected by selection bias. Insu cient reliability of results due to underestimation of dementia cases in other studies is theoretically avoided in this study. Third, the availability of multiple covariates and careful control in regression models minimized potential confounding effects.
Several limitations should be noted in interpretation of our ndings. First, 24h diet recall might be poor at representing a long-term dietary habit. Although using repeated daily records over 2 years, the measurement error was attenuated [18,19], and the intake levels in this study were comparable with the national data of UK [21], there warrants further investigation using dietary assessments that could better represent a long-term status, such as food frequency questionnaire. Secondly, a relatively young age at baseline meant that a large proportion of participants had not yet reached an age of being at risk of dementia, but subgroup analyses among participants with a baseline age over 60 did demonstrate similar results as that of the whole population. And because the participants undergoing MRI may be healthier and more tolerant to sugar or aspartame than the general population, our results on brain structure need to be further veri ed. Third, milder dementia cases were likely to be underreported when patients did not not seek medical care in this study, considering only registration data were used to de ne dementia cases in this study. Moreover, our results may still be subjective to reverse causality, although we have excluded the dementia incidences within the two years after dietary assessment. Due to the long-term development of dementia, early cognitive decline may precede and induce dietary changes, and there awaits further research to con rm these associations and help explain the underlying mechanisms.

Conclusions
The present study demonstrated that higher SSB and ASB intake were associated with higher dementia risk, while moderate intake of NSJ was associated with a lower risk. These associations were similar across major subgroups de ned by sociodemographic features but were modi ed by genetic predisposition. Our observational ndings provide evidence for revised thinking in the balance of reducing the consumption of sugar and arti cially-sweetener. <0.001 SD, standard deviation; AHEI, Alternative Heathy Eating Index (with sugary beverages being excluded) a Body weight status was defined by BMI (<=20 kg/m2 to be underweight, >20 and <=25 to be normal weight, >25 to be overweight) b Beverages intake was categorized into 4 groups: 0, 0~1 (i.e. >0 and <=1) unit/d, 1~2 (i.e. >1 and <=2) units/d, or >2.0 units/d, in which a unit refers to one glass/can/carton or 250 mL.  Figure 1 Curvesa for the association of sugary beverages intake and total sugar intake with risk of incident dementia (N=187,994). a Penalized splines of hazard ratio adjusted for Townsend deprivation index, education level, physical activity, smoking, alcohol intake, total energy intake, alternative healthy eating index (AHEI) excluding sugary beverages, and body weight status.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.