Association between Gut Microbiota with Mild Cognitive Impairment and Alzheimer’s Disease in a Thai Population

Background: Mild cognitive impairment (MCI) and Alzheimer’s disease (AD) are common in older adults. Much recent work has implicated the connection between the gut and the brain via bidirectional communication of the gastrointestinal tract and the central nervous system through biochemical signaling. Altered gut microbiota composition has shown controversial results based on geographic location, age, diet, physical activity, psychological status, underlying diseases, medication, and drug use. Objectives: This study aimed to investigate the relationships of gut microbiota with MCI and AD. Methods: 16S metagenome profiles from stool collection of participant groups (normal; n = 20, MCI; n = 12, AD; n = 20) were analyzed. The diagnosis of cognitive conditions was made by standard criteria consisting of clinical interviews, physical examinations, cognitive assessments, laboratory examinations, and neuroimaging by both structural neuroimaging and amyloid positron emission tomography scans. Correlations between medical factors with food frequency and the fecal microbiome were elucidated. Results: A significant difference at the operational taxonomic unit level was observed. The significantly higher abundance of bacteria in nondementia patients belonged to the Clostridiales order, including Clostridium sensu stricto 1 (p < 0.0001), Fusicatenibacter (p = 0.0007), Lachnospiraceae (p = 0.001), Agathobacter (p = 0.021), and Fecalibacterium (p < 0.0001). In contrast, Escherichia-Shigella (p = 0.0002), Bacteroides (p = 0.0014), Holdemanella (p < 0.0001), Romboutsia (p = 0.001), and Megamonas (p = 0.047) were the dominant genera in the AD group. Left and right hippocampus and right amygdala volumes were significantly decreased in the AD group (p < 0.001) and significantly correlated with the groups of bacteria that were significantly different between groups. Conclusion: There was a relationship between the composition of the gut microbiome and neurodegenerative disorders, including MCI and AD. Reduction of Clostridiaceae and increases in Enterobacteriaceae and Bacteroides were associated with persons with MCI and AD, consistent with previous studies. The altered gut microbiome could be potentially targeted for the early diagnosis of dementia and the reduction of AD risk.


Introduction
Currently, older people worldwide are a growing demographic group in society. Statistics from World Population Prospects 2019 showed that there are more than 1 in 11 people aged 65 years old and older [1], leading to common problems in older people, including dementia. The World Health Organization reported that approximately 55 million people have dementia [2], and there are nearly 10 million new cases every year. Moreover, dementia is also the seventh leading cause of death in older people [2].
Alzheimer's disease (AD) is the most common form of dementia and could contribute to 60-70% of cases [2]. However, the exact cause of AD remains unclear. An accumulation of amyloid-beta and hyperphosphorylated tau has been found in the form of extracellular senile plaques and intracellular neurofibrillary tangles in AD. Inflammatory substances and the complement system have been found in patients' brains [3]. Therefore, the major cause of inflammation, which is a potential risk factor for cognitive impairment leading to AD, has been studied.
The current diagnosis of AD relies on clinical features, histories from patients and their care providers, physical examinations, cognitive assessments, laboratory examinations, and appropriate diagnostic neuroimaging. The structural neuroimaging used are cranial computerized tomography and magnetic resonance imaging for hippocampal gyrus volumes and other common pathologies [4]. In people with atypical dementia, functional or molecular imaging could be used to determine the blood supply to different parts of the brain and the metabolism at different brain locations. Molecular imaging tests include tissue-specific imaging markers to view amyloid plaque, which is specific to AD, using a radioisotope-specific positron emission tomography (PET) scan, such as using Pittsburgh compound B and florbetapir F 18 (18F-AV-45) [5], which can be detected in the early stage of AD and appear in the 2018 National Institute of National Institute on Aging and Alzheimer's Association (NIA-AA) recommendations as biomarkers of amyloid-beta plaques [6]. In addition, PET is most useful for excluding AD pathologies if undetected [7].
The human body contains more than 10 14 microorganisms known as the microbiota, consisting of approximately 1,000 species, most of which are in the gastrointestinal tract [8]. Studies have shown that the gut microbiome (GMB) has a significant effect on health in terms of disease progression and prevention, development of immunity, formation of vitamins, and nutrient absorption [9]. Current studies have found a link between the GMB and AD [10]. It is believed that the GMB affects brain function through the brain-gut-microbiome axis [11]. An altered GMB induces increased gut permeability and triggers an inflammatory response, which impairs the blood-brain barrier and promotes neuroinflammation and amyloidosis [12][13][14]. However, the results of current studies of GMB alterations in AD have been controversial and might have been affected by ethnicity, diet composition, and the criteria used to diagnose dementia. In addition to AD, mild cognitive impairment (MCI), which is a transitional stage between normal cognition and dementia, was also reported to be related to a reduction in GMB diversity compared to normal people [15].
The GMB has never been studied in the Thai population concerning the relationship with AD. The detection of amyloid by PET scan would increase the accuracy of AD diagnosis compared to structural neuroimaging. This research, therefore, aimed to study the association of the GMB in Thai people with AD with amyloid PET positivity compared to those with MCI and cognitively normal older people (normal control: NC) with negative amyloid PET scans. We compared the diversity and relative abundance of the GMB in normal older people, MCI patients, and people with positive amyloid AD, along with data from structural neuroimaging amyloid PET and dietary profiles. The association possibly found in the study might increase knowledge of the etiology and factors affecting the progression of AD, which could lead to interventions for the prevention, disease modification, or treatment of AD.

Materials and Methods
This study with a cross-sectional design was conducted in the geriatric clinic in Siriraj Hospital, Bangkok, Thailand.

Selection of Research Participants: Cases and Controls
We selected the participants from research on the use of F-18 florbetapir (F-18-AV-45) PET to assess brain amyloid deposition in AD, MCI, and normal aging [16]. The participants were divided into three groups: the AD group (n = 20) with β-amyloid deposition in the brain demonstrated by PET scan; the MCI group (n = 12) without cerebral amyloid deposition; and the cognitively normal group without amyloid accumulation in the brain (normal control: NC) (n = 20). The AD group included those diagnosed based on the NINCDS-ADRDA criteria for probable AD. The Thai Mental Status Exam (TMSE) results were lower than 26, and the Clinical Dementia Rating (CDR) scale score was 0.5 or higher. The amyloid PET scans in this group were all positive for amyloid accumulation in the brain. The MCI group had cognitive impair-Neurodegener Dis 2022;22:43-54 DOI: 10.1159/000526947 ment according to its history, and the CDR scale was 0.5. Normal controls were those who did not have clinical information about cognitive symptoms, no history of drug use or disease conditions affecting the cognitive symptoms, normal physical examinations, TMSE results of 24-30, CDR scores of 0, no functional limitations, and normal neuroimaging by both structural neuroimaging and amyloid PET. The participants could make decisions; otherwise, legal representatives signed a consent form for the patient's participation in the research project before the patient joined the study [16].

Exclusion Criteria
Participants were excluded if they: (1) had a history of receiving systemic antibiotics within 3 months before sample collection; (2) had a history of gastrointestinal cancer; (3) had brain lesions that indicated stroke or other abnormalities affecting memory function; or (4) had actively uncontrolled digestive disorders, including inflammatory bowel disease (IBD), malabsorption syndrome, recent Clostridioides difficile infection, colitis, persistent or chronic diarrhea of unknown etiology, or untreated Helicobacter pylori infection.

Withdrawal or Termination Criteria
Participants who expressed their intention not to participate in the research after signing the consent form or who requested to cancel participation during the trial were withdrawn from the study.

History Taking, Physical Examination, and Neuroimaging
Demographic data were collected, including age, sex, educational level, body mass index, comorbidities, supplements or medications used, and alcohol consumption. Global cognitive function was measured by the CDR and TMSE. Oral health was assessed by the Oral Health Assessment Tool (OHAT: Thai version) [17]. All of the participants underwent the full neuropsychological tests and 3-T magnetic resonance imaging scans (Ingenia, Philips Medical System, Best, The Netherlands) with a 32-channel head coil and an amyloid PET study with administration of our proprietary [18F] florbetapir biomarker. The brain volume was interpreted by a neuroradiologist for subfield segmentation. The results of amyloid PET scans were determined through a consensus of two nuclear medicine physicians who were blinded to the patients' clinical information. The details of the neuroimaging protocol were previously reported [16].

Collection of Nutrition Information
Before we collected a stool sample, the participants or their primary caregivers provided dietary information over 1 month using a food frequency questionnaire (FFQ) according to the study by La-Ongkham et al. [18].

Collection of Stool Samples
The research assistant explained the stool collection process to the participants or caregivers. Stool samples were collected by the volunteers at their own homes. The Stool Collection and Preservation System (Norgen Biotek Corp., Thorold, ON, Canada) was used as a microbiome DNA-stabilizing device according to the manufacturer's protocol. After the stool sample was collected, the participant returned the stool to the lab within 72 h. The samples were stored at −20°C for further DNA extraction.

DNA Extraction and 16S Metagenomic Sequencing
Total DNA was extracted from the sample using the QIAamp Stool Mini kit (Qiagen, USA) according to the method specified in the manual. The extracted DNA samples were amplified for the 16S rDNA gene using a specific primer at the V3-V4 region. 16S amplicon sequencing and library preparation were performed according to the Illumina MiSeq instructions.
DNA Sequence Processing DNA sequences were trimmed and cleaned up using Quantitative Insights into Microbial Ecology 2 (QIIME2) [19]. Amplicon quality-filtered, paired-end duplex reads were combined to contig by the DADA2 algorithm. Chimeric sequences were removed. The contig sequences were aligned to the Ribosomal Database Project Naive Bayes classifier [20]. Sequences with 97% similarity were clustered into the same operational taxonomic units (OTUs). OTUs with <0.05% of total sequence reads were filtered out from the dataset to account for sequencing errors. Individual OTUs were aligned to the SILVA 16S rRNA gene database for classification at the phylum, class, order family, and genus levels.

Statistical Analysis of Microbial Diversity and Abundance
Richness and alpha diversity (observed OTUs, Faith's phylogenetic diversity, evenness, and Shannon index) between participant groups were calculated using normalized OTUs (52,000 reads) of random subsampling. Beta diversities were computed using normalized OTUs by Jaccard in the QIIME2 program. Kruskal-Wallis (pairwise) test measures at the 0.05 significance level were applied for differences in richness and alpha diversity testing between groups using the QIIME2 program. To detect significant differences in beta diversity metrics between groups, permutational multivariate analysis of variance (PERMANOVA) in the QIIME2 program was used. The differential abundance of taxa among the MCI, AD, and NC groups was determined at the OTU level using the DESeq2 package in R software. Variables independently associated with dementia were statistically analyzed by Spearman's correlation (R package). p < 0.05 was considered to represent statistical significance.

Participants' Demographics, Clinical Characteristics, and Dietary Consumption Frequency
Participant demographic and clinical characteristics are summarized in Table 1. The mean age, BMI, underlying diseases, number of medications used, and OHAT score were similar among the groups. Structural neuroimaging, including the left hippocampus (p < 0.0001), right amygdala (p < 0.0001), and right hippocampus (p < 0.0001), in the MCI group was significantly different from that in the AD group. However, the NC and MCI groups were not significantly different in these brain regions. Dietary consumption frequency was also analyzed, as shown in Table 2. The participants with MCI consumed significantly more fish than the NC group (p < DOI: 10.1159/000526947 0.008), but this amount was not different from the AD group. The participants with MCI consumed significantly more vegetables than both the NC and AD groups (p < 0.01).

Gut Microbiome Composition
The GMB composition of the NC, MCI, and AD groups was measured. The alpha diversity, which represented the abundance of OTUs in each participant, was assessed using the observed OTU and Shannon indices. There were no significant differences among the NC, MCI, and AD groups (Fig. 1a). Beta diversity, which represents the similarity or differences in GMB composition, was computed by Jaccard and showed no significant differences among the three participant groups (NC vs. MCI; p = 0.059, NC vs. AD; p = 0.164, MCI vs. AD; p = 0.063) (Fig. 1b). The comparative bacterial phyla abundance analyzed by the Kruskal and Dunn test was not significantly different among the three groups (Fig. 2a). Across all participants, the dominant phyla were Firmicutes and Bacteroides. Firmicutes were more abundant in the NC group (NC 63.4%, MCI 58.4%, and AD 54.8%). Bacteroides (NC 25.1%, MCI 31.4%, and AD 30.8%) and Actinobacteria (NC 4.8%, MCI 7.3%, AD 6.2%) were less abundant in the NC group (Fig. 2a).
Analysis of differential abundances at the OTU level was also performed. The NC group exhibited a signifi-cantly higher abundance of bacteria belonging to the Clostridiales order, including Clostridium sensu stricto 1, Fusicatenibacter, Lachnospiraceae, Agathobacter, and Fecalibacterium (Fig. 2b). The AD group showed a higher abundance of Escherichia-Shigella, Bacteroides, Holdemanella, Romboutsia, and Megamonas than the NC and MCI groups (Fig. 2b).

Correlation Analysis of Clinical Factors, Food Intake, and Gut Microbiomes
The associations among clinical factors, food intake frequency, and different genera of gut bacteria among the three groups are shown as a heatmap in Figure 3. A correlation between individual MBBs was observed. Clostridium sensu stricto 1 had a significantly positive correlation with Agathobacter and Faecalibacterium. Holdemanella had a significantly positive correlation with Romboutsia and Megamonas.
General factors, including age, BMI, and underlying diseases, showed a statistically significant correlation with some microorganisms. Age was positively correlated with Romboutsia abundance but negatively correlated with Clostridium sensu stricto 1 abundance. BMI was positively correlated with Fecalibacterium and Escherichia-Shigella. Underlying diseases were negatively correlated with Holdemanella.
CDR and TMSE, which represent, respectively, cognitive impairment status and brain regions, including the right amygdala and the left and right hippocampus, showed statistically significant correlations with the microbiota. CDR was positively correlated with Holdemanella but negatively correlated with Clostridium sensu stricto 1 and Lachnospiraceae. TMSE was positively correlated with Agathobacter but negatively correlated with Holdemanella and Romboutsia. The size of the right amygdala positively correlated with Lachnospiraceae, Agathobacter, and Faecalibacterium but negatively correlated with Holdemanella, Romboutsia, and Megamonas. The size of the left and right hippocampus positively correlated with Clostridium sensu stricto 1, Agathobacter, and Faecalibacterium. The size of the right hippocampus was negatively correlated with Holdemanella.
Additionally, a statistically significant correlation between food intake frequency and GMB was also observed. Noodles were positively correlated with Bacteroides and Holdemanella. Whole grains positively correlated with both Fusicatenibacter and Holdemanella. A high vegetable diet positively correlated with Fusicatenibacter and Escherichia-Shigella. Chicken consumption was negatively correlated with Fusicatenibacter. Both egg and soymilk diets were negatively correlated with Bacteroides.

Discussion
This study is the first that elucidated the associations of GMB composition among Thai participants with cognitive impairment (both MCI and AD) compared with those with cognitively normal people (NC). Although no statistically significant differences were found using alpha and beta diversity, particularly at the phylum level, among the groups, there was a difference in the NC group from MCI and AD groups at the genera level of the GMB. Most of the significantly higher abundance of bacteria in the NC group belonged to the Clostridiales order, including Clostridium sensu stricto 1, Fusicatenibacter, Lachnospiraceae, Agathobacter, and Fecalibacterium. In contrast, Escherichia-Shigella, Bacteroides, Holdemanella, Romboutsia, and Megamonas were the significantly dominant genera in the AD group compared to the NC and MCI groups.
Reports of altered GMB composition from different geographic locations have been controversial. Microbial community richness or abundance in terms of alpha diversity was found to be lowered in US people with dementia, whereas it was increased in Japanese patients and did not change in European studies [21][22][23][24]. However, all three studies showed significant microbial composition    differences, as shown by beta diversity. In the present study, there was no significant difference in either alpha or beta diversity in any of the three participant groups. Microbial diversity depends on many factors, especially in older people who are often exposed to factors affecting the alteration of the GMB, such as infections, antibiotic use, polypharmacy, physical activity, and hospital stays [23]. Therefore, apparently normal older people in a control group might not be healthy in a strict sense and could account for the different results in different studies [23]. We observed that cognitively normal participants (NC) showed a higher abundance of the phylum Firmicutes and a lower abundance of Bacteroides and Actinobacteria in the opposite direction from the MCI and AD participants. The reduction in Firmicutes in AD is consistent with a finding in Chinese individuals with AD. Depletion of Firmicutes, including Clostridiaceae, Lachnospiraceae, and Ruminococcus, was found in Chinese patients with AD [15]. All of the genera were previously characterized as short-chain fatty acid (SCFA)-producing bacteria. The correlations of the three family members with obesity, insulin resistance, and diabetes have been reported [25]. Insulin resistance was associated with a reduction in cerebral glucose metabolism; therefore, it might increase amyloid deposition and AD risk in adults [26].
Clostridium, Agathobacter, and Faecalibacterium, which have less abundance in AD, are also positively correlated with the brain volumes of the left hippocampus, right hippocampus, and right amygdala. These results reflect a previous study that indicated that reduction of hippocampal and right amygdala volumes is the first sign of cognitive decline in the healthy elderly, with left hippocampal volume perhaps also showing prognostic shortterm changes in overall cognition [24]. Reduction in the abundance of Clostridium sensu stricto and Faecalibacterium was previously observed in Chinese AD patients [27]. Three genera have been reported as SCFAs, especially butyrate-producing bacteria. SCFAs were speculated to play a prominent role in neuroinflammation by influencing glial cell morphology, modulating neurotrophic factors, enhancing neurogenesis, involvement to serotonin biosynthesis, and maintaining neuronal homeostasis and brain function [28]. Clostridium sensu stricto 1 was previously showed a higher abundance in participants with better cognitive function [29]. Not only SCFAs, but Clostridiaceae were also indole propionic acid producers, and the indole propionic acid was reported as antioxidative agent protecting primary neurons and neuroblastoma cells against amyloid β peptide [15].
Additionally, family Lachnospiraceae including Fusicatenibacter and Agathobacter was reported to have higher abundance in NC than in AD. Members of Lachnospiraceae are SCFAs producing bacteria. However, they produce diverse types of SCFAs. For example, Fusicatenibacter and Agathobacter are butyrate and acetate producers; on the other hand, other members are propionate producers [30]. Among the SCFAs, butyrate has beneficial effect on neuronal function. It acts as an anti-inflammatory agent by repressing the nuclear factor kappa-light-chain enhancer of activated B cells pathway in inflammation [31].
Moreover, it plays a role as a histone deacetylase inhibitor that activates dendritic sprouting and improves a number of synapses, leading to long-term memory enhancement [32]. In contrast to butyrate that increases intestinal motility, propionate decreases motility [33]. These findings strongly support the functional role of butyrate-producing bacteria in AD pathogenesis.
An increased abundance of the proinflammatory bacteria Escherichia-Shigella and Bacteroides was found in AD participants, as previously reported [34][35][36]. In vitro and in vivo studies have suggested that the bacterial amyloid peptide called curli, produced by Escherichia coli, has a structure similar to Aβ42. The curli peptide or human Aβ42 can interact with the human TLR2 receptor, leading to the activation of bone marrow macrophages and the production of proinflammatory cytokines, such as IL-6 and IL-1β. Moreover, bacterial amyloid can activate Tlymphocytes and induce production of the proinflammatory interleukins IL-17A and IL-22. These cytokines can penetrate the blood-brain barrier and cause reactive oxygen species production and activate TLR2/1-mediated nuclear factor kappa-light-chain enhancer of activated B cells neuroinflammation pathway in microglia and astrocytes which associated to neurodegeneration diseases [37]. Curli-producing E. coli-treating rats were also exhibited development of microgliosis and astrogliosis by evaluating TLR2, IL-6, and TNF expression [38]. Curli is also involved in decreased expression levels of epithelial cell tight junction proteins, such as zonula occludens-1 (ZO-1), claudin-1, and occluding [39]. Higher abundance of Bacteroides was previously found in AD. Lipopolysaccharides of Bacteroides can transport from intestine to brain and cumulate in the parenchyma and vessels and colocalize with Aβ plaques around blood vessels as found in AD brains higher than normal control [37]. Additionally, lipopolysaccharides can induce the secretion of proinflammatory cytokines IL-1, IL-6, and TNF-α causing TLR4-mediated inflammation [28].
We found a significantly positive correlation among Holdemanella, Romboutsia, and Megamonas in AD. Not only did the bacteria correlate by themselves, but they also significantly, negatively correlated with right-amygdala brain volume. In addition to right-amygdala brain volume, Holdemanella has displayed a negative association with several factors, including cognitive score, righthippocampus volume, and underlying diseases. with cognitive impairment [40]. In the mice model, fatand cholesterol-enriched diet could enhance Holdemanella and Romboutsia abundance. Holdemanella was positively associated with cecum fatty acids, saturated fatty acids (stearic acid and palmitic acid), and monostearin, which is the by-product of fatty acids [41]. Moreover, mice fed with a palmitate saturated fatty acid diet were found to be correlated with modulation of the levels of α-synuclein, tyrosine hydroxylase, and dopamine, which are neurochemicals involved in the pathogenesis of neurodegenerative diseases [42]. High levels of palmitic acid in the frontal cortex and parietal cortex are related to Parkinson's disease and AD, respectively [43,44].
A study in Chinese people with AD comparing to cognitively normal healthy control showed that Romboutsia decrease in AD-associated fecal microbiota [27]. However, in our study, people with AD had higher level Romboutsia and Megamonas than NC. There are only few studies reveal a role of Romboutsia on cognitive function. A study in a rat model with type 1 diabetes with cognitive decline showed that Romboutsia negatively correlated with lactate and myo-inositol level in hippocampus but positively correlated with serum glutamine and hippocampal GABA levels. This finding suggested that Romboutsia may affect cognitive function by modulating glutamate-glutamine (GABA) cycle [45]. A higher abundance of Romboutsia was found in Chinese epilepsy patients [46]. Megamonas also showed more abundance in Chinese healthy controls than in patients with AD [47] and multiple system atrophy [48]. At the same time, Megamonas showed elevated abundance in patients with primary sclerosing cholangitis with or without IBD, as well as major depressive disorder, coronary artery disease, and obesity, but it was negatively correlated with physical activity level [49][50][51]. Further research conducted on the roles of Holdemanella, Romboutsia, and Megamonas on neurological function will clarify the possible roles of these bacterial genus-mediated brain functions.
Food intake is one factor that affects GMB composition and function. Food frequency profiles were correlated with the GMB. Our finding was that whole grain and vegetable consumption was positively correlated with high bacterial abundance in both the NC and AD groups. Noodle intake positively correlated with Bacteroides and Holdemanella, which had higher abundance in AD. Thai noodles are derived from many ingredients, such as rice flour, wheat flour, eggs, and vegetables, but the major nutritional component is carbohydrates. Soy foods have been reported to have health-promoting effects. Soy food consumption increases Bifidobacteria and Lactobacilli levels and alters the proportions of Firmicutes and Bacteroidetes in the human GMB [52]. In contrast, we found that soymilk consumption was related to Bacteroides, which are dominant in AD.
One strength of this study is that it included well-characterized dementia participants. The data and samples were prospectively collected in a standard manner. The enrollment criteria for the participants included clinical interviews, physical examinations, cognitive assessments, laboratory examinations, and neuroimaging by both structural neuroimaging and amyloid PET.
The current study also has some limitations. The main limitation is the small sample size. The microbiome profile tends to have great variability in the population. Therefore, a larger sample size would be more precise for identifying any significant differences. A small sample size can cause higher variance, as we observed in the MCI group. We used FFQ profiles to determine the effect of foods on the GMB associations. However, the FFQ has limited profiles of food nutrients and sizes per serving, and these data might not be able to assess the effects of dietary metabolites on the GMB. To clarify the association between food intake and the GMB, well-designed human intervention trials are needed to establish causal relationships. However, a cohort study of those with and without specific GMB components that follows them until the disease condition (MCI and AD) occurs is unlikely due to long duration of such a study and the large sample size that would be needed. In addition, the 16S metagenomic profile has limitations since it does not represent the function or activity of the microbiota. Determining the functional role of the microbiota in temporal and causal relationships between dementia and the GMB might be confirmed by metagenomics analysis, along with metabolomic profiles of individual samples.

Conclusion
This study demonstrated that microbial communities are associated with dementia. The reduction in Clostridiaceae and Lachnospiraceae, together with a high abundance of Escherichia-Shigella, Bacteroides, Holdemanella, Romboutsia, and Megamonas, might further predict the severity of AD symptoms or facilitate early diagnosis of AD, allowing for therapeutic interventions that target modification of GMB alterations in older people at high risk for AD.