Relationship between the Japanese-style diet, gut microbiota, and dementia: a cross-sectional study

Previous studies have shown associations between the gut microbiota, microbial metabolites, and cognitive decline. However, the effect of the dietary composition on such associations has not been fully investigated. We performed a cross-sectional sub-analysis of data from our prospective hospital-based cohort study (the Gimlet study) to evaluate the relationships between dietary composition, cognitive decline, and the gut microbiota. All the participants of the Gimlet study had been provided with information regarding this sub-study in 2018. Patients were excluded if they were unable to provide sucient data in the questionnaire regarding their dietary composition. We assessed their demographics, dietary composition, risk factors, cognitive function, results of brain imaging, gut microbiome, and microbial metabolites. On the basis of previous studies, a nine-component traditional Japanese diet index (JDI 9 ), a 12-component modern JDI (JDI 12 ), and a 12-component revised JDI (rJDI 12 ), were dened. Higher JDI scores indicated greater conformity to the traditional Japanese diet. We then evaluated the relationships between the JDI scores, cognitive function, and the gut microbiome and microbial metabolites using multivariable logistic regression analyses. perivascular rate. Model multivariable logistic regression analyses, which were adjusted for the model 2 factors, years of education, and risk factors (hypertension, dyslipidemia, diabetes mellitus, ischemic heart disease, chronic kidney disease, smoking, alcohol consumption, and ApoE). Model 4: backward stepwise multivariable logistic regression analyses adjusted for the factors in model 3, plus enterotype I, silent lacunar infarct, white matter hyperintensity, cerebral microbleeds, enlarged perivascular space, score of the voxel-based specic regional analysis system for Alzheimer’s disease, single-photon emission-computed tomography ndings (presence or absence of low blood ow in the area of the posterior cingulate gyrus and/or precuneus), ankle brachial index, and pulse wave velocity. Continuous, ordinal, and categorical variables are expressed as the mean ± standard deviation, median and interquartile range, or frequency or proportion (percentage), respectively. The Wilcoxon rank-sum test and the χ 2 test were used. blood flow in the area of the posterior cingulate gyrus and/or the precuneus.

magnetic resonance imaging (MRI) abnormalities [20]. Furthermore, metabolites of the gut microbiota may play important roles in these associations [15]. Therefore, we hypothesized that the consumption of a Japanese-style diet would be inversely associated with cognitive decline and that there would be an association between this diet and the gut microbiome and/or microbial metabolites.
In the present study, we aimed to evaluate the relationships between adherence to a Japanese-style diet, the gut microbiota, and cognitive decline by means of a sub-analysis of data from the ongoing clinical study. Furthermore, we aimed to evaluate the three forms of the JDI (the conventional (JDI 9 ), updated (JDI 12 ), and a newly-modi ed JDI) to determine which would show the closest relationships with cognition and the gut microbiota.

Study design
We performed a cross-sectional sub-analysis of data from a hospital-based prospective cohort study, the Gerontological Investigation of Microbiome: a Longitudinal Estimation Study (the Gimlet study), which has been conducted at the National Center for Geriatrics and Gerontology (NCGG) in Japan. Detailed information regarding the Gimlet study is provided in our previous reports [12,13,15]. Brie y, we enrolled patients visiting the Memory clinic at the NCGG who agreed to undergo both a medical assessment of their cognitive function and a fecal examination. The activities of daily living and cognitive function of the participants were assessed annually after their enrollment.

Participants
Between March 2016 and March 2017, we enrolled consecutive patients who visited the Memory clinic at the NCGG and agreed to undergo both a medical assessment of their cognitive function and a fecal examination. Participants in the Gimlet study were eligible for this sub-study if they met the following criteria: (1) they were able to undergo brain MRI and complete a questionnaire regarding their dietary composition; and (2) they provided their informed consent in writing. Patients were excluded from this sub-study if they: (1) were unable to undergo MRI and/or complete the questionnaire; or (2) they were unable to provide su cient data in the questionnaire regarding their diet. Patients who had potential confounders and effect modi ers for the variables of interest (for example, the recent use of antibiotics) had been excluded at the time of enrollment in the Gimlet study. All the patients who had enrolled in the Gimlet study and their families had been provided with information regarding this sub-study in 2018, after their enrollment in the Gimlet study.

Baseline assessments
All the participants underwent a comprehensive geriatric assessment [21] that was based on the following: (1) demographic characteristics; (2) risk factors; (3) basic and instrumental activities of daily living (ADL) scales; (4) global cognitive function, assessed using the Mini-Mental State Examination (MMSE) [22] and Clinical Dementia Rating (CDR) [23] scales; (5) neuropsychological testing; (6) behavioral and psychological symptoms; (7) assessment of the burden for caregivers; (8) depression status; (9) laboratory parameters; (10) arterial stiffness, as an indicator of arteriosclerosis [24], and the 'impact' of pulse [25]; and (11) the results of brain imaging, such as MRI and single-photon emission computed tomography (SPECT). All the clinical samples and data were provided by the NCGG Biobank, which collects clinical data for research.

Dietary assessments
The questionnaire regarding dietary composition consisted of 12 items. All the components of a typical Japanese diet were grouped on the basis of the de nitions used in the Japanese National Health and Nutrition Survey of 2011 [26]. As in previous studies [8,9], we identi ed the following 12 components of the diet: rice, miso, sh and shell sh, green and yellow vegetables, seaweed, pickles, fruit, soybeans and soybean-derived foods, mushrooms, beef and pork, chicken, green tea, and coffee. The participants and their families answered questions regarding their consumption of these items using the following options: (1) always (on 6-7 days per week), (2) usually (on 3-5 days per week), (3) sometimes (on 1-2 days per week), and (4) rarely (on <1 day per week).

Japanese dietary indices
We evaluated three Japanese dietary indices. The rst was the conventional JDI (JDI 9 ) [8]: (1) for each of the seven bene cial components (rice, miso, sh and shell sh, green and yellow vegetables, seaweed, pickles, and green tea), the participants were assigned one point if their daily intake of the item was equal to or greater than the sex-speci c median dietary intake; and (2) for each of the two less bene cial components (beef and pork, and coffee), the participants were assigned one point if their daily intake was below the sex-speci c median intake because there are sex differences regarding the dietary intake of these items [27]. Thus, the JDI 9 score ranged from 0 to 9, with higher scores indicating greater conformity to the traditional Japanese diet.
The second index was de ned recently [9,28] and comprises 12 components (the JDI 12 ) because three further bene cial components (soybeans and soybean-derived foods, fruit, and mushrooms) had been added to the JDI 9 . Therefore, the JDI 12 score ranged from 0 to 12, and was indicative of an expanded traditional Japanese diet. However, according to recent reports [29,30], higher intake of coffee is associated with better health status and contributes to a lower risk of dementia. Therefore, we also de ned a 12-component revised JDI (rJDI 12 ), in which one less bene cial component (coffee) in the JDI 12 was changed to be a bene cial component. The score of the rJDI 12 , which is more representative of a modern Japanese diet, also ranged from 0 to 12 points.
Classi cation of cognitive function Dementia was de ned as an MMSE < 20 and/or a CDR ≥ 1. The participants who did not have dementia were further categorized as having mild cognitive impairment (MCI) or normal cognition (NC). MCI was de ned as an MMSE ≥ 20 and a CDR = 0.5, which implies possible, very mild dementia, and suggests that the patient has a higher risk of developing dementia in the future [31]. In contrast, NC was de ned as an MMSE ≥ 20 and a CDR = 0.

Brain imaging
The participants underwent a 1.5-T MRI of their brains (Philips Ingenia, Eindhoven, the Netherlands). MRI scans were obtained, including diffusion-weighted images, uid-attenuated inversion recovery images, T2-weighted images, T2 * -weighted gradient-echo images, three-dimensional T1-weighted sagittal and axial coronal views, and 3D time-of-ight magnetic resonance angiography scans. The presence and components of cerebral small vessel disease (SVD), such as silent lacunar infarct (SLI), white matter hyperintensity (WMH), cerebral microbleeds (CMB), and enlarged periventricular space (EPVS), were categorized using previously published standards for reporting vascular changes on neuroimaging [32].
The voxel-based speci c regional analysis system for Alzheimer's Disease (VSRAD) software (Eisai Co., Ltd., Tokyo, Japan) was used to quantify cortical and hippocampal atrophy, using standardized z-scores [33]. A high VSRAD score suggests the presence of Alzheimer's disease (AD) because this score re ects hippocampal atrophy, which is one of the characteristics of the brain of a patient with AD. The participants also underwent N-isopropyl-p-[ 123 I]-iodoamphetamine-SPECT, in which the presence of low blood ow in the area of the posterior cingulate gyrus and/or the precuneus was regarded as a surrogate marker of AD [34].

Total SVD score
As in a previous study [35], we rated the MRI burden of SVD on an ordinal scale from 0 to 4 by recording the presence of each of four features of cerebral SVD. This score consisted of the following:

Gut microbiome
Fecal samples were collected at the participants' homes, and the samples were then frozen and stored at −81°C at the NCGG Biobank. After all the samples had been collected, the gut microbiome of each participant was analyzed by the TechnoSuruga Laboratory (Shizuoka, Japan) using terminal restriction fragment-length polymorphism (T-RFLP) analysis [36]. The T-RFLP analysis was used to classify gut microbes into the following 10 groups: Prevotella, Bacteroides, Lactobacillales, Bi dobacterium, Clostridium cluster IV, Clostridium subcluster XIVa, Clostridium cluster IX, Clostridium cluster XI, Clostridium cluster XVIII, and 'others'. By referencing the Human Fecal Microbiome T-RFLP pro le [37,38], each gut microbiome was categorized as representing one of three enterotypes: enterotype I, which included Bacteroides at > 30%; enterotype II, which included Prevotella at > 15%; and enterotype III, which comprised other combinations of microorganisms. The Firmicutes/Bacteroidetes (F/B) ratio was also calculated because a high ratio is indicative of dysbiosis [39].

Analysis of microbial metabolites in feces
In previous studies, the concentrations of gut microbial metabolites, including short-chain fatty acids (SCFAs) [6,17] and lipopolysaccharide [17], have been quanti ed. In the present study, we measured the fecal concentrations of organic acids, SCFAs, ammonium ions, indoles, phenol, skatole, and p-cresol, as previously described [15]. The concentrations of organic acids and SCFAs (acetic acid, propionic acid, butyric acid, iso-butyric acid, succinic acid, lactic acid, formic acid, valeric acid, and iso-valeric acid) were measured using high-performance liquid chromatography. Ammonium ion concentrations were quanti ed using ion chromatography; and the fecal concentrations of indoles, phenol, skatole, and pcresol were quanti ed using gas chromatography/mass spectrometry. Detailed information regarding the analysis of the fecal metabolites is provided in our previous report [15] and the Supplementary le.

Statistical analysis
Continuous, ordinal, and categorical variables are expressed as the mean ± standard deviation, median and interquartile range, or frequency or proportion (percentage), respectively. Data were compared using the unpaired Student's t-test, Wilcoxon rank-sum test, and χ 2 test, respectively. First, the participants were allocated to groups according to the presence or absence of dementia, the presence or absence of enterotype I, and those with MCI or NC, among the participants who did not have dementia, and their clinical characteristics were compared using the Wilcoxon rank-sum test. In addition, their gut microbiomes and gut microbial metabolites were compared using the χ 2 test or the Wilcoxon rank-sum test. Second, we evaluated the relationships between the gut microbiome, microbial metabolites, and dietary composition. Third, the participants were allocated to two groups according to their JDI 12 : a high JDI 12 group (above the median value) and a low JDI 12 group (below the median value), and we compared their clinical characteristics. We also compared the differences between participants with high rJDI 12 scores and those with low rJDI 12 scores. Finally, multivariable logistic regression models were used to identify independent associations between the presence of dementia, JDI score (JDI 12 or rJDI 12 ), and the gut microbiome. Odds ratios (ORs) and 95% con dence intervals (CIs) are presented. All the comparisons were two-tailed, and P < 0.05 was considered to represent statistical signi cance. Data were analyzed using the JMP 12.0 software package and SAS v9.4 (SAS Institute Inc., Cary, NC, USA).

Participant characteristics
We previously enrolled 128 participants in the Gimlet study, but of these, 43 were excluded from this subanalysis because of incomplete data and/or refusal to participate. We therefore analyzed data from the remaining 85 eligible participants (61.1% women; mean age: 74.6 ± 7.4 years; mean MMSE score: 24 ± 5).

Enterotype I vs. other enterotypes
Compared with participants with the other enterotypes, those with enterotype I tended to consume less seaweed, fruit, and milk and dairy products, although these differences were not signi cant (Table S1).

MCI vs. NC patients
Among the participants who did not have dementia, the participants with NC tended to consume more sh, shell sh, and coffee than those with MCI, although these differences were not statistically signi cant (Table S2). In addition, participants with enterotype I tended to consume less miso, sh and shell sh, seaweed, fruit, chicken, and milk and dairy products, but again the differences were not statistically signi cant (Table S3).

Microbial metabolites
Participants who consumed more mushrooms (eg. higher vs. lower intake; median concentration of isobutyric acid: 0.03 vs. 0.11 mg/g, P = 0.017; Table S4), soybeans and soybean-derived foods (Table S5), or coffee (Table S6) had lower fecal concentrations of several gut microbial metabolites than those who consumed smaller amounts of these items.
High vs. Low JDI 12 score Participants with high JDI 12 scores were less likely to be female (High vs. Low JDI 12 : 46.8% vs. 79.0%, P = 0.004), less likely to have dementia (14.9% vs. 42.1%, P = 0.007) and WMH (17.0% vs. 36.8%, P = 0.048), had lower GDS, ZBI, and VSRAD scores, and were more likely to have good neuropsychological test results ( Table 2). The concentrations of the majority of the gut microbial metabolites in participants with high JDI 12 scores were lower than in those with low JDI 12 scores (Table 3). voxel-based speci c regional analysis system for Alzheimer's disease; SPECT, single-photon emission computed tomography; CRP, C-reactive protein; eGFR, estimated glomerular ltration rate. High vs. Low rJDI 12 score Participants with high rJDI 12 scores were less likely to be female, less likely to have dementia (High vs. Low rJDI 12 : 15.0% vs. 37.8%, P = 0.027) and WMH (12.5% vs. 37.8%, P = 0.012), had lower ZBI (5 vs. 14 points, P = 0.010) and VSRAD (median score; 0.74 vs. 1.15, P = 0.031) scores, and were more likely to have good neuropsychological test results (Table S7). In addition, the fecal concentrations of the majority of the gut microbial metabolites in participants with high rJDI 12 scores were lower than in those with low rJDI 12 scores (Table S8).
In multivariable analyses, participants with high JDI 12 scores were inversely associated with the presence of dementia and were signi cantly associated with lower ORs ( Table 4). The increment in both JDI 12 and rJDI 12 scores (per one point increment) were also inversely associated with the presence of dementia and were associated with lower ORs; however, there were no signi cant differences in the stepwise multivariable logistic regression analyses (Table S11). The area under the receiver operating curve (AUC) of rJDI 12 for the presence of dementia was the highest among those of the three JDI indices (AUC for JDI 9 : 0.637; JDI 12 : 0.668; and rJDI 12 : 0.693). Abbreviations: OR, odds ratio; CI, con dence interval; JDI, Japanese dietary index. The dependent variable was the presence of dementia. JDI 9 : conventional JDI score, which comprised seven bene cial components ("rice", "miso", " sh and shell sh", "green and yellow vegetables", "seaweed", "pickles", and "green tea"), and two less bene cial components ("beef and pork" and "coffee"); 0-9 points. JDI 12 : modi ed JDI score, comprising the JDI 9 components and three additional bene cial components ("soybeans and soybean-derived foods", "fruit", and "mushrooms"); 0-12 points. r-JDI 12 : revised JDI 12 score, in which one less bene cial component ("coffee") in the JDI 12 was changed to a bene cial component, such that these became "rice", "miso", " sh and shell sh", "green and yellow vegetables", "seaweed", "pickles", "green tea" and "coffee"; and there was just one less bene cial component ("beef and pork"), making it more consistent with the modern Japanese diet; 0-12 points.
OR 95% CI P Model 1: univariate analyses. Model 2: adjusted for age, sex. Model 3: backward stepwise multivariable logistic regression analyses, which were adjusted for the model 2 factors, years of education, and risk factors (hypertension, dyslipidemia, diabetes mellitus, ischemic heart disease, chronic kidney disease, smoking, alcohol consumption, and ApoE). Model 4: backward stepwise multivariable logistic regression analyses adjusted for the factors in model 3, plus enterotype I, silent lacunar infarct, white matter hyperintensity, cerebral microbleeds, enlarged perivascular space, score of the voxel-based speci c regional analysis system for Alzheimer's disease, single-photon emissioncomputed tomography ndings (presence or absence of low blood ow in the area of the posterior cingulate gyrus and/or precuneus), ankle brachial index, and pulse wave velocity.

Discussion
The main nding of the present study is that adherence to a traditional Japanese diet was inversely associated with cognitive decline and tended to be associated with low concentrations of several gut microbial metabolites. Associations between adherence to a traditional Japanese diet and cognitive decline have previously been reported in community-dwelling older people [4,5,40] and in younger adults [41]. However, this relationship had not previously been evaluated in older adults in a hospital-based cohort. Therefore, our ndings are novel in this respect. Such associations among diet, clinical data, the gut microbiome, and microbial metabolites had not been previously reported.
Traditional Japanese diets include large amounts of sh and shell sh, miso, seaweed, vegetables, soy products, fruit, and green tea [40,42], and some of these ( sh, vegetables, soy products, and fruit) are associated with bene cial effects on cognitive function [40,[42][43][44]. In the present study, we found that greater intake of these foods was inversely associated with cognitive decline in univariate analyses and tended to be associated with lower fecal concentrations of speci c gut microbial metabolites. These ndings are consistent with those of previous studies.
The assessment of dietary patterns using the JDI 12 and rJDI 12 yielded similar results in the present study.
However, the rJDI 12 showed more clearly that the consumption of larger amounts of four foods (soybeans and soybean-derived foods, fruit, mushrooms, and coffee) was associated with a lower OR for the presence of dementia. Therefore, these foods likely make the largest contribution to the association between diet and dementia.
The Japanese dietary pattern shares characteristics with the Mediterranean diet; for instance, the high intake of vegetables, legumes, and sh, and the low consumption of meat [42]. Therefore, the mechanism underlying the association between adherence to a Japanese-style diet and the lower risk of dementia may be similar to that reported following studies of the Mediterranean diet. Speci cally, the Mediterranean diet affects the composition of the gut microbiota and the concentrations of the derived metabolites, and is associated with improvements in biomarkers of AD [6].
In particular, we found that mushroom and soybean consumption was associated with lower ORs for the presence of dementia. Mushrooms and soybeans contain many useful nutrients, such as dietary ber, minerals, B vitamins, vitamin D, and vitamin E [44,45]. Furthermore, mushroom consumption alters lipid metabolism, and has anti-obesity, anti-atherosclerotic, and anti-diabetic effects [45]. In addition, soybeans contain phytoestrogens (iso avones), which have bene cial effects on cognitive function [44]. This is consistent with both JDI 12 and rJDI 12 being inversely associated with the presence of dementia.
Milk and dairy products are considered to be bene cial for health. Previous studies have shown that the consumption of milk and dairy products is associated with better mental health [46] and a lower risk of dementia [5,47] in older adults, but there was insu cient evidence regarding its relationship with cognitive function to draw a conclusion in a previous meta-analysis [48]. In the present study, a signi cant relationship between dementia and the consumption of milk and dairy products was also not identi ed.
The relationships between the consumption of green tea and coffee and cognitive function are also unclear. One previous study showed that the consumption of green tea prevents oxidative stress, but may not signi cantly affect cognitive function in older adults [49], and another showed that green tea, but not coffee, reduces the risk of cognitive decline in older adults [50]. It has also been reported that caffeine consumption, and especially the moderate quantities consumed in coffee or green tea, reduces the risk of dementia [10], probably because caffeine affects neural and vascular activity, including vasoconstriction and cerebral blood ow [10]. In the present study, there was no signi cant association between green tea consumption and dementia, but greater intake of coffee tended to be associated with lower concentrations of microbial metabolites that are associated with dementia. This trend will be investigated further in the future.
Polyunsaturated fatty acid [51] and probiotic [52] consumption have been shown to ameliorate or preserve cognitive function. Previous studies have suggested that intake of the probiotics Bi dobacterium [52] or Lactobacillus [53] inhibits cognitive decline. Dietary diversity is also protective against cognitive decline [54], and nutritional education [55] is essential for patients and their family members. In the present study, the effects of soup stock on the gut microbiota and cognitive function were unclear, and these relationships will be investigated in more detail in the future.
The present study had several strengths. First, we have provided evidence for novel relationships between the Japanese dietary indices, the gut microbiota, and cognitive decline, which are consistent with the existence of a diet-microbiome-gut-brain axis. Speci cally, adherence to a traditional Japanese diet is associated with low fecal concentrations of speci c gut microbial metabolites. Moreover, increments in the JDI scores are associated with decreases in the prevalence of dementia. Second, we systematically assessed the cognitive function of patients using a comprehensive geriatric assessment and a range of neuro-psychological tests. Last, our ndings may lead to greater attention being paid to the relationships between the composition of the diet and cognitive function, through assessment of the gut microbiome.
Th present study also had several limitations. A causal relationship between the gut microbiome and dietary pattern could not be identi ed because of the cross-sectional nature of the study. We are currently conducting a longitudinal observational study that will help us identify causality in these relationships in the near future. In addition, other research groups are conducting randomized, placebo controlled, doubleblind clinical studies regarding the links between the gut microbiota and cognition using antioxidant or probiotic supplements [56,57]. The small number of participants and large number of potential variables in the present study mean that it may have been statistically underpowered; a larger-scale study might have yielded signi cant ndings where we only identi ed trends. In addition, selection bias may also exist because we studied a single, hospital-based cohort. We could not quantitatively evaluate dietary intake, and frequency evaluation is unable to determine the dose-dependency of the relationships. The speci c effects of each microbial metabolite have not been determined. However, there have been recent studies of the associations between metabolites and cognitive function [16,17,58], and those ndings will be complemented by others in the future that will clarify the nature of such relationships. Measurement of the concentration of amyloid-β precursor protein may be also useful because it is associated with the risk of cognitive impairment [59].
Although the present sub-study was of a small number of patients, which renders the analysis preliminary, our ndings provide evidence for relationships between the Japanese dietary pattern, the gut microbiota, and cognitive function. Longitudinal assessments of these relationships should be made in future studies to determine the underlying mechanisms involved.

Conclusions
Adherence to a traditional Japanese diet was inversely associated with cognitive decline and tended to be associated with lower concentrations of several microbial metabolites in an older Japanese population.     The Wilcoxon rank-sum test and the χ 2 test were used.  Comparison of the Japanese diet indices (JDI9, JDI12, and rJDI12) in participants with and without

Declarations
dementia The x-axis shows the presence or absence of dementia and the y-axis the JDI score. * P < 0.05 and ** P < 0.01, according to the Wilcoxon rank-sum test. JDI9: conventional JDI score; JDI12: modi ed JDI score, including three additional components (soybeans and soybean-derived foods, fruit, and mushrooms); rJDI12: revised JDI12 score, in which one less bene cial component (coffee) in the JDI12 was rede ned as a bene cial component. The rJDI12 score more clearly shows the difference between participants with and without dementia than the JDI9 and JDI12 scores.