A total of 60 subjects with AD aged between 60 and 80 years and community-dwelling Han residents were recruited from the Memory Clinic of Fujian Provincial Hospital (Fujian, China) between February 2020 and December 2021. The participants with AD in this study were diagnosed according to the 2018 National Institute of Aging and Alzheimer's Association (NIA-AA) guidelines and the Diagnostic and Statistical Manual of Mental Disorders (fifth edition) by experienced neurologists (45). And the patients with AD were divided into mild, moderate, and severe AD groups according to the mini-mental state examination (MMSE) scores (mild: 21–26, moderate: 11–20, and severe: 0–10) (46). The exclusion criteria included: (1) other causes of dementia or other types of dementia; (2) a family history of dementia; (3) any kind of neurodegenerative disease, such as Parkinson's disease; (4) confirmed mental illness, such as schizophrenia; (5) severe cardiac, pulmonary, hepatic, or renal disease, or any tumor; (6) intestinal diseases, such as irritable bowel syndrome; (7) taking antibiotics, glucocorticoids, or probiotics within 1 month; (8) known active infections, such as viral, bacterial, or fungal infections, or other autoimmune diseases; (9) infected with a SARS-CoV-2; (10) with severe auditory, visual, or motor deficits that may interfere with cognitive testing were also excluded. In addition, the oral health status of the participants with AD was measured to exclude subjects with the following conditions in the 2 months before the study: oral or dental surgery, inflammation of the oral or perioral tissues, or other chronic diseases of the oral cavity.

The study was registered at the Chinese Clinical Trial Registry (ChiCTR2100041749) and approved by the Ethics Committee of Fujian Provincial Hospital (ref no. K2020-09-025). All participants provided written informed consent before participation.

A validated Chinese annual semi-quantitative food frequency questionnaire (FFQ), which has good reliability and validity and contains 17 food groups with a total of 149 food items, was repeated six times over one-year to collect 24-h recalls for 3 days from each subject every 2 months to elucidate daily variation and seasonality of food changes, as well as medium-term changes in dietary habits during the study period. In our study, FFQ was used to assess subjects' dietary intake data for the past year (47). Subjects quantified their food intake by referring to the Retrospective Dietary Survey Supplementary Reference Physical Atlas, which provided photographs of food items in different portion sizes. For example, consumption of edible oil and condiments by those in the household was measured monthly. Participants with >20% of missing FFQ items and abnormal total energy intake were excluded from this study to avoid outlier effects, i.e., an overall intake for males <800 or >4,000 kcal/day and for females <500 or >3,500 kcal/day (48). It is worth mentioning that each AD patient had a relative or guardian present at the time of the survey to confirm the patient's dietary habits. And when patients were unable to fill out the questionnaire on their own, we completed it by consulting their family members, with priority given to partners and co-residents, to confirm the authenticity of the questionnaire content.

The DII is a literature-based tool that computes the inflammatory properties of a diet, based on the association of certain food and dietary constituents with defined inflammatory hallmarks: CRP, TNF-α, IL-1β, IL-4, IL-6, and IL-10 (28). In this study, based on the actual food parameters obtained from the FFQ, 24 food parameters were ultimately included to calculate the DII. Pro-inflammatory ingredients include carbohydrates, energy, fat, protein, saturated fat, cholesterol, and iron, and anti-inflammatory ingredients include carotenoids, caffeine, fiber, monounsaturated fatty acids, polyunsaturated fatty acids, riboflavin, green tea, onions, garlic, ginger, vitamin A, vitamin C, vitamin E, magnesium, zinc, selenium, thiamin (Table 1) (28). Components such as eugenol, saffron, isoflavones, pepper, thyme/oregano, and rosemary were excluded because of a lack of related information in the FFQ recordings.

To avoid randomness caused by the use of individual initial intake values, the actual intake of each food parameter was normalized to a z-score based on the global average intake and standard deviation for 11 countries. Individual z-scores were then converted to centered percentiles. Each centered percentile was multiplied by the standardized overall inflammatory effect score. Finally, the DII scores for all food parameters were summed to obtain the individual DII scores. The final score is a continuous measure, interpreted as strongly anti-inflammatory (the lowest score) to strongly pro-inflammatory (the highest score).

Physical activity (PA) in the most recent week was assessed using the short form of the International Physical Activity Questionnaire (49). The questionnaire asked whether subjects had performed any activities from the following categories during the previous week: walking; moderate activity (household activity or child care); and vigorous activity (running, swimming, or other sports activities). Metabolic equivalent (MET) hours per week were calculated using corresponding MET coefficients (3.3, 4.0 and 8.0, respectively) according to the following formula: MET coefficient of activity × duration (hours) × frequency (days). Total PA levels were assessed by combining the scores for different activities.

Participants were instructed not to brush their teeth on the morning of the sampling day and the night before. After a trained dentist scratched the supragingival debris with sterilized cotton balls, a subgingival plaque was collected with 40# sterilized paper points (Gapadent, Tianjin, China) that were gently inserted into the deep periodontal pocket for 20 s. Once removed from the periodontal pocket, the paper point was placed into a 1-mL sterile cryopreservation tube (50). No sample was collected if a patient had no teeth or dental implants. All subgingival plaque samples were stored at −80°C before processing.

Each participant was asked to collect a fresh fecal sample in the morning. Because several community-dwelling older subjects could not send their samples to the hospital immediately, they were given fecal collection containers (SARSTEDT, Germany) with approximately 5 mL of special cytoprotective agents to preserve the DNA in the stool at an approximate temperature for 10–14 days. The fecal samples were then transferred to the laboratory and stored at −80°C before processing.

Blood samples were collected from the participants after an overnight fast by trained laboratory staff. Immediately after the sample of venous blood was collected, it was centrifuged and stored at −80°C before processing. The serum concentrations of IL-6, complement 3 (C3), high-sensitivity (hs-CRP), IL-1β, IL-4, IL-10, IL-12, as well as TNF-α, were determined with an enzyme-linked immunosorbent assay method according to the manufacturer's directions (Elabscience, Wuhan, China). Oral and blood samples were collected on the same day, except for fecal sample.

DNA extraction and 16S rRNA gene amplicon sequencing DNA extraction, PCR amplification, and the sequencing of the V3–V4 hypervariable regions of the bacterial 16S rRNA gene based on all oral and gut DNA samples were undertaken at the DNA Sequencing and Genomics Laboratory of Sangon BioTech (Shanghai). Following the manufacturer's instructions, total community genomic DNA extraction was performed using an E.Z.N.A. Soil DNA Kit (Omega, USA). PCR was started immediately after the DNA was extracted. The 16S rRNA V3–V4 amplicon was amplified using KAPA HiFi Hot Start Ready Mix (2×) (TaKaRa Bio Inc., Japan). Two polyacrylamide gel electrophoresis-purified universal bacterial 16S rRNA gene amplicon PCR primers were used: the amplicon PCR forward primer (5′-CCTACGGGNGGCWGCAG-3′) and the amplicon PCR reverse primer (5′-GACTACHVGGGTATCTAATCC-3′). PCR was performed using a thermal cycler (Applied Biosystems 9700, USA) using the following program: one cycle of denaturing at 95°C for 3 min; five cycles of denaturing at 95°C for 30 s, annealing at 45°C for 30 s, and elongation at 72°C for 30 s; 20 cycles of denaturing at 95°C for 30 s, annealing at 55°C for 30 s, and elongation at 72°C for 30s; and a final extension at 72°C for 5 min. The PCR products were checked through a separation with electrophoresis in 1% (w/v) agarose gels in Tris, boric acid, and EDTA (TBE) buffer, staining with ethidium bromide, and visualizing under ultraviolet light.

Sequencing was then performed using the Illumina MiSeq system (Illumina MiSeq, California, USA). The raw sequencing reads were detected using FastQC software to remove the primer region and low-quality sequences. The chimera sequences arising from the PCR amplification were detected and excluded using Mothur (http://www.mothur.org) based on the GreenGenes database. The high-quality reads that reached a 97% nucleotide similarity were clustered into operational taxonomic units (OTUs) according to the Ribosomal Database Project database. Summaries of the taxonomic distributions of OTUs were constructed using these taxonomies and were used to calculate the relative abundance of microbiota at the phylum and genus taxonomic levels.

The α- (Shannon, Simpson, Chao1, and ACE index) and β-diversity analyses (Bray–Curtis dissimilarity and principal coordinate analysis [PCoA]) were conducted using QIIME and R software to compare the similarity among samples in terms of the diversity of species. Analysis of variance (ANOVA) or the Kruskal–Wallis test was performed to evaluate α-diversity among the different groups. Permutational multivariate analysis of variance (PERMANOVA) was employed to identify the different microbial communities among groups. The relative abundance diagram of flora species was mainly used to visualize the results of species annotation. STAMP analysis was used to identify species that differed in abundance between two or more groups. The key taxa responsible for the differences in the oral and gut microbiota between AD groups were identified using the linear discriminant analysis effect size (LEfSe) for biomarker discovery (51). Spearman correlation coefficients were used to detect relationships between the taxa and inflammatory markers. A significant α of 0.05 and an effect size threshold of 2 were used for all biomarkers discussed in this study.

Baseline characteristics of participants are presented as means ± SD, medians (interquartile ranges [IQRs]), and numbers (percentages). Comparisons of continuous variables between groups were made using ANOVA or rank sum tests, and group differences were compared using a chi-squared test or Fisher's exact test for categorical variables. Kruskal-Wallis test was used to analyze the skewness distribution data. Furthermore, when the distribution of variables was skewed (as for IL-6, C3, hs-CRP, IL-1β, IL-4, IL-10, IL-12, and TNF-α), the values were converted to their natural logarithm.

DII scores were ranked and split into approximately equal tertiles, with tertile 1 (T1) representing the most anti-inflammatory diet group and tertile 3 (T3) representing the most pro-inflammatory group. We performed multiple linear regression analysis to examine associations between each tertile of DII scores and inflammatory indicators, and tertile 1 (T1) was considered the reference. In Model 1, we did not adjust for covariates. In Model 2, we adjusted for age and gender. In Model 3, we adjusted for the variables in Model 2 with the addition of education, body mass index (BMI), smoking, alcohol consumption, physical activity, hypertension, diabetes, hyperlipidemia, coronary heart disease, and cerebrovascular disease. Tests for trends were performed by assigning the median value for each tertile and modeling this as a continuous variable. Energy adjustment was done using the residual approach that was explained previously by Willett et al. (52). For calculating energy-adjusted dietary intakes, each of the dietary components is regressed on their total energy intake and residual values were added to their actual mean intake to estimate energy-adjusted values. Statistical analysis was performed using SAS version 9.4 software (SAS Institute Inc., Cary, USA) and STAMP v2.1.3 (53). The R package and GraphPad Prism v6.0 were used to prepare the graphs. All tests of significance were two-sided and p < 0.05 was considered statistically significant.

The DII range was −0.021 (the most anti-inflammatory diet) to 1.38 (the most pro-inflammatory diet). According to the DII score, the tertiles were divided from low to high, T1 (tertile 1) means the most anti-inflammatory diet group, T2 (tertile 2) means no anti-inflammatory/pro-inflammatory diet group, and T3 (tertile 3) means the most pro-inflammatory diet group. The baseline characteristics of 60 patients with AD including T1 (n = 20), T2 (n = 20), and T3 (n = 20) are shown in Table 2. Among all the variables examined, significant differences were observed only for the DII score, mean total daily energy intake, and education level (p < 0.05). Patients with higher DII scores (i.e., more pro-inflammatory diets) tended to have lower total energy intakes and education levels. However, there was no statistically significant difference across DII levels in the distribution of other baseline characteristics and inflammation markers (p > 0.05). To further analyze whether there is an association between each tertile of DII scores and inflammatory indicators, we used multiple linear regression with model adjustment for con-founders. The results still showed that none of the inflammatory indicators was statistically significant among the groups (see Supplementary material for details).

The energy-adjusted dietary intakes for participants in the different inflammatory diet groups are presented in Table 3. Compared to the most anti-inflammatory diet group, individuals with the most pro-inflammatory diet group consumed significantly lower amounts of energy, protein, fat, carbohydrate, fiber, vitamin A, carotenoids, thiamin, riboflavin, vitamin C, vitamin E, magnesium, iron, zinc, selenium, and PUFA (all p < 0.05).

Considering the progression of AD, we further analyzed the association between baseline characteristics and inflammatory factors with different severity of AD (Table 4), significant differences were observed only for the education level, MMSE scores, and IL-4 (p < 0.05). After pairwise comparison (Figure 1), the IL-4 levels in patients with moderate and severe AD were significantly higher than those in mild AD (p < 0.05).

As shown in Figure 2, when the number of samples reached 60, the number of species observed was nearly parallel, indicating that the sample size of our experiment was sufficient. There were no significant differences among the three groups in the oral microbiomes in the Shannon, Simpson, Chao, or ACE indices (Figure 3, p > 0.05). Furthermore, there were no differences in the α-diversity of gut microbiomes across the different DII tertiles in patients with AD (Figure 4, p > 0.05). Interestingly, the PCoA based on weighted UniFrac distances showed (Figure 5) that the oral microbiomes of PT1 largely overlapped with the microbial distribution of PT2, whereas there were statistical differences in the microbial distribution of PT2 and PT3 (PERMANOVA, Bray–Curtis: PT2 vs PT3, R² = 0.061, p = 0.013). As for the gut microbiomes, no differences were found among the FT1, FT2, and FT3 groups.

The relative abundance at the phylum and genus levels revealed similar overall microbiome composition among the three groups in the oral microbiomes. The most abundant phyla were Bacteroidetes, Proteobacteria, Firmicutes, Fusobacteria, and Actinobacteria; these five phyla accounted for more than 95% of the total abundance of all the species (Figure 6A). At the genus level, Neisseria, Prevotella, Streptococcus, Fusobacterium, and Leptotrichia were the top five most abundant bacterial taxa, as shown in Figure 6B. We also found that in the gut microbiomes, Firmicutes, Proteobacteria, Bacteroidetes, Actinobacteria, and Verrucomicrobia were identified as the most abundant sequences at the phylum level (Figure 6C). At the genus level, Escherichia, Shigella, Bacteroides, Klebsiella, Faecalibacterium, and Prevotella were the most abundant bacterial taxa, as shown in Figure 6D.

The STAMP results showed that there were no significant differences in the percent relative abundance of any phylum among the groups. At the genus level, the abundance of Prevotella and Olsenella in the oral microbiomes was much higher in PT1 (the most anti-inflammatory diet group) than in PT3 (the most pro-inflammatory diet group). The differences in the relative abundance of Abiotrophia, Neisseria, and Parvimonas between the PT2 (no anti-inflammatory/pro-inflammatory diet group) and PT3 groups were statistically significant (Figure 7A). In addition, the abundance of genera Alistipes, Ruminococcus, Odoribacter, and unclassified Firmicutes in the gut microbiomes were lower in the FT3 group than in the FT1 group. The differences in the relative abundance of Sphingobium, Microbacterium, Centipeda, and Gp6 were statistically significant between the FT2 and FT3 groups, and the other branching genera Pseudoxanthomonas, Firmicutes, Bacillariophyta, Oxalobacter, Alistipes, and Rhodocyclaceae were also found to be significantly different between the FT1 and FT2 groups (Figure 7B).

To further explore the differences in oral and gut microbiomes between the anti-inflammatory diet and pro-inflammatory diet groups, we used LEfSe (linear discriminant analysis score cut off >2.0) analysis to identify the key taxa responsible for the differences in the compositions of the oral and gut microbiota. However, there were no significant differences among the three groups of patients with AD (Supplementary material for details).

Spearman correlation analysis showed significant correlations between inflammatory markers and specific oral and gut microbiota (Figure 8). In the analysis of oral microorganisms, IL-4 was positively correlated with Fusobacterium (p = 0.008, r = 0.34) and Selenomonas (p = 0.012, r = 0.32); IL-1B was positively correlated with Peptostreptococcaceae_incertae_sedis (p = 0.018, r = 0.31); IL-12 was positively correlated with Haemophilus (p = 0.013, r = 0.32) and Rothia (p = 0.005, r = 0.36) and negatively correlated with Selenomonas (p = 0.001, r = −0.4); TNF-α was positively correlated with Alloprevotella (p = 0.013, r = 0.32) and Selenomonas (p = 0.014, 0.32) and negatively correlated with Haemophilus (p = 0.018, r = −0.31); CRP was positively correlated with Gemella (p = 0.006, r = 0.35); and IL-6 was negatively correlated with Granulicatella (p = 0.022, r = −0.3). Furthermore, in the analysis of intestinal microorganisms, IL-4 was positively correlated with Megamonas (p = 0.016, r = 0.31) and Anaerostipes (p = 0.014, r = 0.31); IL-12 was negatively correlated with Weissella (p = 0.013, r = −032); TNF-α was positively correlated with Lactobacillus (p = 0.019, r = 0.3); and IL-6 was positively correlated with Clostridium_XlVb (p = 0.008, r = 0.34), Morganella (p = 0.005, r = 0.36), and Providencia (p = 0.002, r = 0.39).