Stratified dietary inflammatory potential identifies oral and gut microbiota differences associated with cognitive function in older adults

Abstract

Diet is a crucial factor that shapes the composition of the microbiota throughout the life cycle. Systemic chronic inflammation and microbial imbalance may play a key role in the pathogenesis of cognitive disorders. Inflammatory diets can influence the host microbiome and inflammatory state. This study investigated the impact of the inflammatory potential of the diet on the diversity and composition of the oral-gut microbiome, as well as on cognitive performance, in older adults over 60 years of age. The Energy-adjusted Dietary Inflammatory Index (E-DII) and 16S rRNA sequencing were used to analyze dietary inflammatory properties and oral-gut microorganisms in 54 patients. The results showed that significant differences in the diversity of oral microbiota among different E-DII groups was detected (p < 0.05), whereas gut microbiota diversity didn’t exhibit variations. In the anti-inflammatory diet group, the class Saccharimonadia, the order Corynebacteriaceae, the genera TM7x, Eubacterium_yurii_group, and Centipeda were more abundant in oral microbiomes, and lower abundance of Holdemanella and Haemophilus was observed in the gut microbiomes. Specific oral and gut genera were associated with MMSE, MoCA, AVLT-LR, BNT, and VFT scores (p < 0.05). These results provide insights into anti-inflammatory diets were associated with an increased abundance of beneficial microbes, and a specific oral and gut microbial composition was associated with cognition.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-02292-5.

Introduction

Mild cognitive impairment (MCI) represents a transitional state between normal aging and dementia, including Alzheimer’s disease (AD) and other types of dementia, characterized by mild deficits in the cognitive domains of memory, attention, and executive function, with the individual’s daily living activities remaining unaffected^1,2. Studies have shown that the prevalence of MCI among Chinese older people > 60 years of age is approximately 15.54%, with an annual conversion rate to dementia ranging 6–15% for MCI patients, which is about ten times higher than the conversion rate for cognitively normal older adults^3,4. Given the irreversible nature of dementia and the limitations of treatments, the MCI stage presents a crucial opportunity for risk factor intervention to decelerate cognitive decline.

Diet is an important modifiable factor in cognitive impairment, playing a vital role in the prevention of cognitive decline⁵. Diet can regulate the inflammatory state of the body, with specific food components and nutrients exhibiting anti-inflammatory or pro-inflammatory effects⁶. For instance, the consumption of polyphenols, unsaturated fatty acids, and vitamins inhibits oxidative stress and neuroinflammation, whereas saturated fatty acids may provoke an inflammatory response through the activation of hypothalamic signaling pathways^7,8. Consequently, inflammation might partially mediate the association between diet and cognitive function. The dietary inflammation index (DII) was developed to quantify and assess the overall inflammatory potential of an individual’s diet⁹. Subsequently, to further account for variations in energy intake between individuals, an energy-adjusted dietary inflammation index (E-DII) has been devised¹⁰. Compared with the unadjusted DII, the E-DII has a better general applicability in integrating the measurement of dietary inflammation, thereby effectively enhancing the predictive power for disease outcomes^11–13.

The diet is a key factor that influences the composition of the microbiota throughout life, and its impact on cognitive function through modulation of the microbiota has garnered increasing attention in recent years. Consumed foods affect an individual’s microbiome, particularly diets with inflammatory components^8,14. Diets rich in saturated fats, carbohydrates, and highly processed foods lead to a reduction in microbiota diversity, neuroinflammation, and cognitive impairments¹⁵. However, flavanols in red wine and cocoa positively affect gut microbes and reduce inflammation by increasing beneficial bacteria¹⁶. Animal studies have revealed that oligofructose from fruits and vegetables alters the gut microbiota composition, decreasing pathobionts associated with dementia-related immune responses and inflammation, such as Proteobacteria and Helicobacteriaceae¹⁷. In one study, participants with higher DII scores exhibited lower microbiome diversity and had a positive correlation with C-reactive protein (CRP) levels ≥ 3 mg/L, implying that pro-inflammatory diets reduce microbial diversity while elevating the host’s inflammatory status¹⁸. Thus, the diet can modulate the composition of the gut microbiota and its metabolites through its dietary inflammatory potential, leading to altered gut function and microbial activity^19–21. These changes in the gut microbiota can, in turn, stimulate immune responses and metabolic activities, contributing to chronic and low-grade systemic inflammation, which subsequently triggers neuroinflammatory responses and dysregulation of cognitive functions in the brain^22,23. These findings suggest a potential relationship between an altered gut microbiota regulated by dietary inflammation and cognitive function.

Dietary inflammation not only influences the composition of the gut microbiota, but is also closely linked to the oral microbiota²⁴. Disruptions in the oral microbiota can cause systemic diseases beyond the oral cavity, perpetuating chronic inflammation over extended periods^25,26. Given the proximity of the oral cavity to the brain, the oral microbiome may have a profound connection with the central nervous system²⁷. Epidemiological studies have suggested an association between periodontal disease and cognitive impairment^28,29. Older individuals with chronic periodontitis exhibit an elevated risk of cognitive decline, with periodontitis-related oral microbiota dysbiosis implicated in this cognitive deterioration^30,31. Our previous study showed the significant impact of altered subgingival microbiota diversity on oral health in patients with AD³². Other studies have detected higher levels of oral bacteria in the brains of patients with AD^33,34. Furthermore, oral microecology dysbiosis, coupled with intestinal inflammation, is directly correlated with impaired intestinal barrier function and heightened intestinal permeability, potentially leading to deleterious effects on neurons^27,35,36. Therefore, maintaining oral and gastrointestinal health can significantly benefit the general health of the brain of the host.

Diet-microbiome interactions represent a most promising target for improving brain cognitive function. However, existing studies have not yet fully examined the association between dietary inflammation, oral-gut microbiome, and cognitive function in older adults. Given that dietary inflammation may influence cognitive function through modulatory mechanisms of the microbiome. Therefore, this study aims to investigate the impact of dietary inflammatory potential on the diversity and composition of the oral-gut microbiome, as well as cognitive performance, in older adults aged ≥ 60 years.

Results

Descriptive characteristics of participants

Based on the E-DII score, participants were divided into tertiles from low to high: T1 (tertile 1: − 2.39 to − 0.01) representing the most anti-inflammatory diet group, T2 (tertile 2: 0.01 to 0.97) representing neither anti-inflammatory nor pro-inflammatory diet group, and T3 (tertile 3: 0.98 to 2.34) representing the most pro-inflammatory diet group. The socioeconomic and clinical characteristics of the participants stratified by tertiles of E-DII are illustrated in Table 1. Patients with higher E-DII scores (indicating more pro-inflammatory diets) tended to have lower BMI, MoCA, and VFT scores. However, there was no statistically significant differences in E-DII level in the distribution of other characteristics.

Table 1.

Characteristics of participants based on tertiles of E-DII.

Variables	T1 n = 19 (− 2.39 to − 0.01)	T2 n = 17 (0.01 to 0.97)	T3 n = 18 (0.98 to 2.34)	p-value
Age, years, median (IQR)	67.50 (10.00)	67.50 (7.00)	66.50 (12.00)	0.977
Sex, male, n (%)	6 (33.33)	8 (44.44)	8 (44.44)	0.736
Education level, n (%)				0.994
None	1 (5.56)	1 (5.56)	1(5.56)
Primary school	4 (22.22)	3 (16.67)	5 (27.78)
High school/GED	10 (55.56)	10 (55.56)	9 (50.00)
More than high school	3 (16.67)	4 (22.22)	3 (16.67)
Energy intake (kcal/day), mean (SD)	1435.83 ± 103.14	1421.31 ± 87.93	1440.49 ± 103.45	0.304
BMI (kg/m2), mean (SD)	23.43 ± 0.68	23.95 ± 0.67	21.42 ± 0.61	0.021
MoCA, median (IQR)	23.05 ± 0.98	22.65 ± 0.97	19.78 ± 0.61	0.039
MMSE, median (IQR)	27.00 (3.00)	27.50 (2.00)	26.00 (4.00)	0.173
AVLT-IR, mean (SD)	14.63 ± 1.19	12.12 ± 0.85	12.72 ± 0.72	0.157
AVLT-LR, median (IQR)	4.00 (2.00)	3.50 (3.00)	3 (2.00)	0.398
AVLT-REC, median (IQR)	20.50 (5.00)	21.00 (3.00)	20.50 (3.00)	0.749
VFT, median (IQR)	15.00 (4.00)	12.00 (5.00)	11.50 (4.00)	0.035
BNT, median (IQR)	21.50 (6.00)	22.00 (4.00)	20.00 (7.00)	0.171
STT-A, median (IQR)	62.00 (23.00)	70.00 (42.00)	76.50 (54.00)	0.058
STT-B, median (IQR)	174.00 (83.00)	204.50 (128.00)	236.00 (129.00)	0.139
Income (yuan/m), n (%)				0.786
< 5000	2 (11.11)	5 (27.78)	3 (16.67)
5000–10,000	7 (38.89)	6 (33.33)	8 (44.44)
> 10,000	9 (50.00)	7 (38.89)	7 (38.89)
Smoking status, n (%)				0.051
Current	9 (50.00)	6 (33.33)	3 (16.67)
Former	4 (22.22)	3 (16.67)	1 (5.56)
Never	5 (27.78)	9 (50.00)	14 (77.78)
Drinking status, n (%)				0.120
Current	13 (72.22)	15 (83.33)	12 (66.67)
Former	0 (0.00)	2 (11.11)	4 (22.22)
Never	5 (27.78)	1 (5.56)	2 (11.11)
Hypertension, n (%)	9 (50.00)	12 (66.67)	8 (44.44)	0.483
Diabetes, n (%)	5 (27.78)	8 (44.44)	3 (16.67)	0.224
Hyperlipidemia, n (%)	3 (16.67)	2 (11.76)	1(5.56)	0.861

BMI, body mass index; MMSE, Mini Mental State Examination; MoCA, Montreal Cognitive Assessment; AVLT, auditory verbal learning test; BNT, boston naming test; VFT, verbal fluency test; STT, shape trails test; E-DII, energy-adjusted dietary inflammatory index; SD, standard deviation; IQR, interquartile range; T1, the most anti-inflammatory diet group, T2, the non-anti-inflammatory/pro-inflammatory diet group, T3, the most pro-inflammatory diet group.

Table 2 provides a general description of the variables subsequently analyzed in the study according to cognitive function. Individuals in the MCI group had significantly higher E-DII, STT-A, and STT-B scores but lower MoCA, MMSE, AVLT, and VFT scores than controls with normal cognition. Variables such as BMI, sex, drinking status, and education levels were similar among the groups.

Table 2.

Demographic characteristics of NC and MCI patients.

Variables	NC n = 18	MCI n = 36	p-value
Age, years, median (IQR)	66.50 (8.50)	68.00 (11.00)	0.613
Sex, male, n (%)	8 (44.44)	14 (38.89)	0.695
Education level, n (%)			0.380
None	1 (5.56)	2 (5.56)
Primary school	5 (27.78)	7 (19.44)
High school/GED	7 (38.89)	22 (61.11)
More than high school	5 (27.78)	5 (13.89)
Energy intake (kcal/day), mean (SD)	1404.02 ± 85.22	1447.21 ± 73.33	0.721
E-DII, mean (SD)	− 0.50 ± 0.20	0.70 ± 0.18	< 0.001
BMI (kg/m2), mean (SD)	23.52 ± 0.74	22.63 ± 0.47	0.293
MoCA median, mean (SD)	25.11 ± 0.91	20.19 ± 0.47	< 0.001
MMSE, median (IQR)	29.00 (3.00)	26.00 (3.00)	< 0.001
AVLT-IR, mean (SD)	14.63 ± 1.19	12.12 ± 0.85	0.157
AVLT-LR, median (IQR)	4.00 (2.00)	3.00 (2.00)	0.031
AVLT-REC, median (IQR)	21.00 (4.00)	20.5 (4.75)	0.295
VFT, median (IQR)	15.50 (6.75)	12.00 (4.00)	< 0.001
BNT, median (IQR)	21.00 (5.50)	21.00 (6.00)	0.600
STT-A, median (IQR)	58.50 (22.25)	76.50 (43.50)	0.003
STT-B, median (IQR)	160.00 (69.75)	235.50 (100.75)	0.002
Income (yuan/m), n (%)			0.387
< 5000	3 (16.67)	7 (19.44)
5000–10,000	5 (27.78)	16 (44.44)
> 10,000	10 (55.56)	13 (36.11)
Smoking status, n (%)			< 0.001
Current	13 (72.22)	5 (13.89)
Former	5 (27.78)	3 (8.33)
Never	0 (26.32)	28 (77.78)
Drinking status, n (%)			0.795
Current	14 (77.78)	26 (72.22)
Former	1 (5.56)	5 (13.89)
Never	3 (16.67)	5 (13.89)
Hypertension, n (%)	9 (5.00)	20 (55.56)	0.700
Diabetes, n (%)	6 (33.33)	10 (27.78)	0.673
Hyperlipidemia, n (%)	1 (5.56)	5 (13.89)	0.651

BMI, body mass index; MMSE, Mini Mental State Examination; MoCA, Montreal Cognitive Assessment; AVLT, auditory verbal learning test; BNT, boston naming test; VFT, verbal fluency test; STT, shape trails test; E-DII, energy-adjusted dietary inflammatory index; SD, standard deviation; IQR, interquartile range.

Quality control and basic analysis

After quality pruning and filtering sequencing results of samples from the subgingival bacterial floor, 5,445,683 valid sequences were obtained, with an average length of 422 bp. A total of 4234 OTUs were obtained after OTU clustering at 97% similarity level, and the average number of OTUs was 78. In the gut bacterial floor, we obtained 5,597,349 high-quality sequences, each with an average length of 415 bp. Through OTU clustering at a 97% similarity threshold, a cumulative total of 10,489 distinct OTUs were identified, averaging approximately 194 OTUs per sample. The Good’s coverage of all samples was higher than 99%. It reflected the results of the sequencing and represented the actual composition of the subgingival and gut microbiota.

Altered diversity of subgingival and gut microbiota in participants

We used well-established indexes, including the Shannon and Simpson indexes, to estimate α-diversity among groups. Subgingival microbial richness (as measured by the Shannon index) showed a progressively decreasing trend from the T1 to the T3 group. In contrast, the Simpson index was significantly higher in the T3 group compared with the T2 and T1 groups (Fig. 1A). In contrast, there was no difference in α-diversity of gut microbiota across tertiles of E-DII (Fig. 1B). Furthermore, the PCoA based on the Bray–Curtis distance (Fig. 1C) showed that the subgingival microbiomes of T1 largely overlapped with those of T2, whereas there were statistically significant differences in the microbial distribution between the T2 and T3 groups (PERMANOVA, Bray–Curtis: T2 vs T3, R² = 0.054, p = 0.040). In the gut bacterial flora, no significant microbial structural differences were observed in the PCoA among the T1, T2, and T3 groups (Fig. 1D).

Fig. 1.

The diversity of subgingival and gut microbiota across tertiles of E-DII. (A) Comparison of the Shannon and Simpson indexes of subgingival microbiota and (B) gut microbiota in participants among different E-DII groups. (C) PCoA based on the Bray–Curtis of subgingival microbiota in participants among different E-DII groups. (D) PCoA based on the Bray–Curtis distance analysis of gut microbiota in participants among different E-DII groups. PERMANOVA tests were performed if the centroids, similar to means, of each group were significantly different from each other. R² statistic showing community variations between the compared groups with significant p-values.

Changes of subgingival and gut microbiota composition

To describe the relationship between dietary inflammation and the subgingival and gut microbiomes, we examined changes in the microbiota of participants based on tertiles of E-DII. Stacked bar plots of relative abundance demonstrated evident differences in subgingival and gut microbiota according to different E-DII levels (Fig. 2A and B). At the phylum level, Proteobacteria, Firmicutes, Bacteroidetes, Fusobacteria, and Actinobacteria were predominant phyla in subgingival plaques and feces (relative abundance > 1%). Additionally, 23 and 32 dominant genera, respectively, accounted for more than 80% of subgingival and gut bacterial compositions. We further analyzed the altered composition of microbiota at the genus level.

Fig. 2.

Comparison of subgingival and gut microbiota structures among different E-DII groups. (A) The composition and relative abundance of subgingival microbiota and (B) gut microbiota at the phylum and genus levels. (C) The comparison of relative abundance of oral and (D) gut predominant microbiota at the genus level. Kruskal–Wallis test analysis indicated the significant differences among the three groups, and significant taxa classifications obtained by post hoc test using Bonferroni adjustment. *p < 0.05, **p < 0.01, ***p < 0.001.

The top five subgingival bacterial genera showed significant differences in relative abundance among different E-DII groups, namely Aggregatibacter, Centipeda, Corynebacterium, norank_Saccharimonadaceae, and TM7x (Fig. 2C). Among them, the levels of norank_Saccharimonadaceae, and TM7x gradually decreased from the T1 to T3 groups, suggesting that these two bacteria may be negatively associated with dietary inflammation. Additionally, in the gut microbiota, the relative abundance of five bacterial genera differed significantly in different E-DII groups. Escherichia-Shigella presented a continuous increasing trend with higher E-DII scores. Compared with the T1 group, samples from the T3 group exhibited a slightly higher relative abundance of three bacterial genera: Haemophilus, norank_RE39, and Porphyromonas (Fig. 2D).

Differences in specific subgingival and gut microbiota

To further explore alterations in the subgingival and gut microbiota across different E-DII score levels, we performed LEfSe analysis at the genus level to identify key taxa. In the taxa of subgingival bacteria, the class Saccharimonadia, the orders Corynebacteriales and Saccharimonadales, the families Corynebacteriaceae and Saccharimonadaceae, the genera TM7x, Eubacterium.yurii.group, norank.Saccharimonadaceae, Centipeda and Corynebacterium were more abundant in the T1 group. The most enriched taxa in the T3 group were the genus Lacticaseibacillus (Fig. 3A and B).

Fig. 3.

Crucial taxa that contribute to differences in subgingival and gut microbiota composition in patients with different E-DII levels based on LEfSe analysis. (A) Taxonomic cladogram from LEfSe, depicting taxonomic association between subgingival microbiome communities across different E-DII groups. (B) Histogram of LDA scores from LEfSe analysis showing differentially abundant subgingival bacteria across different E-DII groups. (C) Taxonomic cladogram from LEfSe, depicting taxonomic association between gut microbiome communities among different E-DII groups. (D) Histogram of LDA scores from LEfSe analysis showing differentially abundant gut bacteria across different E-DII groups.

In the taxa of gut microbiota, the abundance of two taxa (the family Porphyromonadaceae and its corresponding genus Porphyromonas) were enriched in the T3 group. However, the abundances of the order RF39, the families Pasteurellaceae and norank.RF39, the genera Haemophilus, Holdemanella and norank.RF39 were depleted (Fig. 3C and D).

Associations between cognitive function and altered microbiomes

The dbRDA test included MMSE, MoCA, AVLT-IR, AVLT-LR, AVLT-REC, VFT, BNT, STT-A and STT-B scores (Fig. 4A and B). To explore the correlation between the relative abundance of altered subgingival and gut microbiota and different cognitive domains, Spearman’s correlation analysis was performed using the cognitive function scores and altered genera according to different E-DII levels. A heatmap was generated to visualize the potential associations. In the subgingival microbiota, the genera Corynebacterium and TM7X (enriched in the T1 group) showed a positive correlation with the MoCA score: the Eubacterium_yurii_group was positively correlated with AVLT-LR and BNT scores (Fig. 4C). In the fecal microbiota, we found that the genus Haemophilus was negatively correlated with MoCA and VFT scores. Conversely, Holdemanella and Porphyromonas showed a positive relationship with MoCA and VFT scores (Fig. 4D).

Fig. 4.

Correlation heatmap analysis between the subgingival and gut microbiota and cognitive function. (A) The dbRDA analysis of subgingival microbial population distribution stratified by cognitive function at the genus level. Arrows represented different cognitive function scores. The long arrow indicates a high correlation with the distribution of the microbiota. The acute angle of the two arrow lines indicated a positive correlation between the clinical characteristics, whereas the obtuse angle indicates a negative correlation. (B) Spearman’s correlation heatmap based on altered subgingival genera and the scores of different cognitive domains. Red represents a positive correlation, and blue represents a negative correlation. (C) The dbRDA analysis of gut microbial population distribution stratified by cognitive function at the genus level. (D) Spearman’s correlation heatmap based on altered gut genera and scores of different cognitive domains.

Altered subgingival and gut microbiota may predict MCI

To further evaluate the potential of the altered subgingival and fecal microbiota based on the E-DII levels as diagnostic biomarkers in differentiating MCI from NC, we constructed random forest models based on the subgingival and fecal microbiota at the genus level. The model built on subgingival microbiota demonstrated moderate accuracy, with an AUC of 0.750 (95% CI 0.527–0.973) (Fig. 5A). In addition, we found that the differential gut microbiota distinguished MCI from NC with high accuracy, yielding an AUC of 0.866 (95% confidence interval 0.669–0.989) (Fig. 5C). Figure 5B and D show the top 3 subgingival or gut genera with highest contributions to the model classification performance in predicting MCI.

Fig. 5.

Subgingival and fecal bacterial biomarkers based on different E-DII levels for predicting MCI. (A) ROC curve analysis evaluated the discriminatory potential of subgingival bacteria in differentiating MCI from NC. (B). The top 3 subgingival bacteria genera most important for discriminating between MCI and NC. (C) ROC curve analysis evaluated the discriminatory potential of fecal bacteria in differentiating MCI from NC. (D) The top 3 fecal bacteria genera most important for discriminating between MCI and NC. Each genus is ranked according to an importance score (mean decrease accuracy). Abbreviations: ROC, receiver operating characteristic; AUC, area under the ROC curve.

Functional changes of subgingival and gut microbiota

As shown in Table 3, KEGG functional orthologs were predicted with PICRUSt among the different E-DII groups. In the subgingival microbiota, compared to the T2 group, the excretory system was significantly altered in the T3 group. In the gut microbiota, the functional orthologs of cardiovascular diseases and the circulatory system were altered in T3 compared to T1 patients.

Table 3.

PICRUSt-based examination of the subgingival and gut microbiota in level 2 KEGG pathways among different E-DII groups.

	T1 mean% (SD%)	T2 mean% (SD%)	T3 mean% (SD%)	p-value	T1 vs. T2	T1 vs. T3	T2 vs. T3
					p-value	p-value	p-value
Oral microbiota
Excretory system	9499.50 (12,980.950)	6726.16 (4593.42)	12,124.38 (10,884.93)	0.011	–	–	0.008
Gut microbiota
Cardiovascular diseases	5.738 (0.275)	5.878 (0.285)	5.880 (0.277)	0.039	–	0.047	–
Circulatory system	2.008 (0.099)	2.046 (0.122)	2.048 (0.099)	0.009	–	0.012	–

PICRUSt, phylogenetic investigation of communities by reconstruction of unobserved states; KEGG, Kyoto encyclopedia of genes and genomes; SD, standard deviation.

Discussion

In the present study, significant differences in the diversity of oral microbiota among different E-DII groups was detected, whereas gut microbiota diversity didn’t exhibit variations. Intriguingly, distinct microbial compositional abundances were observed in both the oral and gut microbiomes of individuals adhering to anti-inflammatory versus pro-inflammatory diets. These differential microbial populations in the oral cavity and gut were implicated in various cognitive domains. The differential microbiota profiles between the oral and gut compartments as defined by E-DII may serve as efficacious indicators for predicting MCI. Furthermore, KEGG pathway analysis showed that the oral and gut functional pathways exhibited significant changes among the three groups.

We found significant differences in oral microbial diversity between groups stratified by E-DII level, although there was no statistically significant difference in terms of gut microbial diversity. Similar to our study, Anderson et al. observed differences in β-diversity of the oral microbiome associated with total carbohydrate, fiber, sucrose, and galactose intake^37,38. Significant changes in diversity are present in patients with gingivitis on a nitrate-rich diet compared with healthy participants³⁹. These findings suggest that the community structure of the oral cavity may be influenced by pro-inflammatory diets, and that variations in the distribution of microbial abundance are significantly impacted by food intake. Little is currently known about the significance of these changes, and more mechanistic studies are needed to explore its role. Zheng et al. found in a study of a healthy population that E-DII was not associated with the overall diversity of the gut flora, and despite initial control for the effect of BMI confounders, no significant association between α and β diversity and E-DII scores was observed⁴⁰. A previous study by our group on dietary inflammation levels and gut microbiota in AD patients have also yielded similar results, with no significant differences in diversity across DII subgroups⁴¹. Studies have shown that stability of the microbiome over a certain period of time may mitigate the effects of the E-DII, suggesting that the E-DII may influence only certain specific types of microorganisms, rather than the overall diversity of the microbiome⁴². Given the limited number of studies exploring the association between the E-DII and oral-gut microbiota, further investigation in larger sample populations is warranted.

This study demonstrated that the composition of the subgingival microbiota was markedly different among participants with varying E-DII levels, and the class Saccharimonadia, the orders Corynebacteriales and Saccharimonadales, the families Corynebacteriaceae and Saccharimonadaceae, the genera TM7x, Eubacterium_yurii_group, norank_Saccharimonadaceae, Centipeda, and Corynebacterium were more abundant in the T1 group. Limited information is available concerning the Saccharibacteria phylum (formerly known as TM7) in mammalian health and disease. Members are generally regarded as symbionts, indicating a mutually beneficial or at least harmless relationship with their host⁴³. Mice with a healthy diet had a higher abundance of TM7 and its corresponding genera^44,45. This phylum acts a modifier that shifts the microenvironment towards a more inflammatory microbiota⁴⁶. However, due to the recalcitrance of culturing TM7, few causal studies investigating its role in inflammatory diseases and calls for additional research to determine their specific roles in different environments. Interestingly, the order Corynebacteriales from the phylum Actinobacteria in our study was more abundant in the most anti-inflammatory diet group. Binda et al. suggested that Actinobacteria play a crucial role in maintaining intestinal barrier homeostasis⁴⁷ and serve as a significant source of antibiotics essential for treating bacterial infections⁴⁸. However, the specific role in dietary inflammation remains unclear, and further detailed studies are required to confirm how it contributes to the anti-inflammatory effects in elderly participants. Moreover, participants with the anti-inflammatory diet in this study had a higher abundance of Eubacterium_yurii_group. Eubacterium has been widely colonized in the gut of most people and plays an important role in nutrient metabolism and the healthy gut microenvironment⁴⁹. It inhibits the growth of potential pathogens, participates in the production of short-chain fatty acids, especially butyrate, which can act as an energy source for colon cells, and exerts anti-inflammatory effects^50,51.

Comparing the gut microbiotas among the three groups revealed lower abundance of Holdemanella and Haemophilus in the anti-inflammatory diet group. Conversely, the abundance of Porphyromonas in the pro-inflammatory diet group was increased. Previous studies have found pro-inflammatory foods (e.g., fermented dairy product and carbohydrate intake) and the production of IL-6, IL-1β, and TNF-α were positively correlated with Holdemanella^52–54. The genus Holdemanella is genetically associated with an increased risk of periodontitis⁵⁵. Similarly, Haemophilus is a sulfur-reducing commensal bacteria of the human oropharynx and has been associated with chronic obstructive pulmonary and periodontal disease^56,57. Furthermore, higher adherence to a plant-based diet and fiber intake was negatively associated with the relative abundance of Haemophilus^58,59. Interestingly, we found that members of the genera Haemophilus and Porphyromonas, which typically colonize the oral cavity, were detected in the gut. Porphyromonas belongs to the respiratory core microbiome⁶⁰. Of these, Porphyromonas gingivalis is the most widely described species and may increase systemic inflammation by inducing intestinal dysbiosis and enhancing the production of IL-17⁶¹. This finding further confirms that it is common for oral microbes to translocate to and then colonize the gut, suggesting that the oral cavity serves as an endogenous reservoir for gut microbiota⁶² and highlighting the bidirectional interconnection between the oral and gut microbiomes.

The diet may alter the oral and gut microbial composition and its metabolites to further drive neuroinflammatory states and thus improve cognitive function^{19,23,32,63,64}. To further support this speculation, we correlated cognitive tests scores with oral microbiota and found that Corynebacterium and TM7X were correlated with higher MoCA scores, the Eubacterium_yurii_group was correlated with higher AVLT-LR and BNT scores, while Lacticaseibacillus and ligilactobacillus were negatively related to MMSE, MoCA, AVLT-LR, and VFT scores. Additionally, different oral bacterial genera associated with varying E-DII levels can serve as effective predictors of MCI. A previous study⁶⁵ reported a significant increase in Lactobacillus in patients with moderate dementia. Notably, Ligilactobacillus, Lacticaseibacillus, and Lactobacillus all belong to the Lactobacillaceae family, which is capable of fermenting carbohydrates into organic acids and metabolizing amino acids into acids and ammonia. These metabolic processes contribute to tooth surface demineralization and dental caries, leading to systemic inflammation that affects cognitive function. Similarly, a previous study found TM7 in the AD group, strongly linking it to periodontitis⁶⁶. Animal experiments have shown that a significant increase in TM7 is associated with inflammation and cognitive decline in older mice⁶⁷. Eubacterium plays a role in regulating inflammation and metabolic functions, and produces butyrate, which can promote intestinal mucosal repair, prevent the formation of inflammatory cytokines, inhibit human neuroinflammation, and delay the decline of cognitive function⁶⁸. In AD mouse models, butyrate intervention can amplify the expression of learning-related genes and restore histone acetylation, effectively improving learning and memory ability⁶⁹.

Furthermore, a lower abundance of Haemophilus in the gut microbiota was negatively correlated with MoCA and VFT scores. Conversely, Holdemanella and Porphyromonas showed a positive relationship with MoCA and VFT scores. These are the top 3 fecal genera with highest contributions to the model classification performance in predicting MCI. Haemophilus has been identified as an opportunistic pathogen and is involved in the inflammatory response⁷⁰. A pro-inflammatory diet leads to harmful bacterial reproduction and increased intestinal permeability, causing intestinal leakage and the transfer of immune cells and bacterial components from the intestinal environment into the circulation, resulting in systemic inflammation that ultimately triggers neuroinflammatory responses and cognitive dysfunction⁷¹. Overall, these data highlight that a pro-inflammatory diet may affect the composition of the gut microbiome and induce inflammatory response of the peripheral circulation and central nervous system through the bidirectional communication pathway “microbe-gut-brain” axis, which may harm brain cognitive function. However, few studies have investigated the relationship between diet and oral-gut microbiota and cognition, and its specific mechanism and pathway deserve further exploration in the future.

Apart from compositional changes in the bacterial taxa, we also predicted functional alterations based on PICRUSt. Gene functions were not altered in the T2 group, whereas T3 patients exhibited varying degrees of genetic changes. A higher level of E-DII was associated with broader modulation of the functional KEGG pathway. The excretory system, cardiovascular diseases, and circulatory system were the key enriched pathways. Thus, the potential of the diet to modulate systemic inflammation may occur through the regulation of the gut microbiota, which uses metabolites produced to exert regulatory effects on both the peripheral and central nervous systems⁷², thereby influencing cognitive functions in the brain. However, current research on the regulation of the microbiota and metabolites by dietary inflammation is limited, and more studies are needed to verify these relationships.

This is the first study to investigate the role of dietary inflammatory potential on the diversity and composition of the oral-gut microbiome, as well as cognitive performance in older adults. However, several limitations should be acknowledged. First, despite the use of a validated FFQ to assess diet, there may have been recall bias among study participants. Second, due to the inherent limitations of cross-sectional studies, causal associations, and potential pathways between the E-DII and oral-gut microbiota and cognitive function must be confirmed in randomized controlled trial studies. Third, although rarefaction curve analysis suggested sufficient sampling for microbial diversity, the relatively small sample size from a single cohort may limit statistical power and generalizability. Additionally, the lack of adjustment for confounders like age and gender leaves the possibility of residual confounding, despite most of the demographic information was not statistically different among the different groups. Fourth, this study utilized 16S rRNA sequencing for microbiome analysis, which has lower taxonomic resolution and limited capacity for functional prediction compared to shotgun sequencing. Moreover, 16S rRNA sequencing has inherent limitations in identifying microbial communities at the species level. Finally, the study focused on individuals of Chinese ethnicity, and generalizing these findings to other ethnic or environmental backgrounds may not accurately reflect the observed associations.

In conclusion, our study indicates that E-DII is significantly associated with oral microbiome diversity, while the diversity of gut microbiota in older adults is not affected. Moreover, the composition of the oral and gut microbiota differs across E-DII levels. Specific bacteria in the anti-inflammatory diet group are associated with better cognitive function, whereas the opposite is observed with a pro-inflammatory diet, suggesting their potential role in MCI prediction. These findings highlight the need for further clinical and mechanistic studies to explore the impact of personalized anti-inflammatory diets on microbiome regulation and cognitive function.

Methods

Study design and population

Fifty-four subjects were recruited from the Memory Clinic of Fujian Provincial Hospital (Fujian, China) between December 2022 and October 2023. Inclusion criteria for participants in this study were: age ≥ 60 years and providing complete dietary information necessary to calculate E-DII scores. The exclusion criteria were the following: (1) a diagnosis of dementia, including Alzheimer’s disease, vascular dementia, Lewy body dementia, and frontotemporal dementia; (2) patients with severe acute diseases, somatic diseases, or tumors; (3) neuropsychiatric disorders (e.g., anxiety, depression, and schizophrenia), or the presence of other neurological disorders associated with cerebral dysfunction and serious medical illnesses; (4) patients who had received periodontal therapy; (5) patients who had been treated with antibiotics or probiotics within the last month, or who had undergone oral or gut surgeries or treatments within the last 2 months; (6) individuals with intestinal diseases, such as irritable bowel syndrome, or severe cardiac, clinically significant liver, kidney, or lung dysfunction; (7) patients with known active infections (e.g., viral, bacterial, or fungal); (8) patients with SARS-CoV-2 infection; (9) patients with severe hearing, visual, or motor impairment that may affect cognitive testing; (10) patients who were taking dietary supplements (such as vitamins, minerals, herbs, and amino acids) or medications known to affect cognitive function. The study protocol was approved by the Ethics Committee of Fujian Provincial Hospital (Ref No. K2022-09-023). All participants gave their informed consent in writing for participating in the study and the research was conducted in accordance with the Declaration of Helsinki. All methods were performed in accordance with the relevant guidelines and regulations for human research, and the study was conducted in compliance with ethical standards.

Cognitive assessment

All tests were conducted according to standard procedures by experienced clinical psychology staff members who had undergone training prior to the study. The test battery, which required approximately 1.0 h to complete, consisted of two screening tests for overall cognitive function, the Montreal Cognitive Assessment (MoCA), and Mini Mental State Examination (MMSE)⁷³. In addition, cognitive domains auditory verbal learning test (AVLT) for memory⁷⁴, Boston Naming Test (BNT)⁷⁵ and Verbal Fluency Test (VFT) for language⁷⁶, and shape trails test A&B (STT-A&B) for attention and executive function⁷⁷ were evaluated.

Diagnosis of MCI

Participants were diagnosed with MCI through a screening and clinical diagnosis⁷⁸. First, patients, insiders, or experienced clinicians reported suspected cognitive impairment. Second, complex functional tasks were intact or minimally impaired, with preserved activities of daily living. In addition, patients with a junior high school or higher education have a MoCA score of < 24, those with an elementary school education have a score of < 19, and those who are illiterate have a score of < 13⁷⁹. Furthermore, after clinical evaluation, participants should be diagnosed with MCI and should not meet the diagnostic criteria for dementia. This study included 36 patients with MCI and 18 in the normal control (NC) group.

Calculation of E-DII scores

Dietary intake was assessed using a validated Chinese annual semi-quantitative food frequency questionnaire (FFQ)⁸⁰. The E-DII was used to assess the inflammatory potential of the diet according to the design of Shivappa et al.⁹ and additional details are available in our previous study¹³. Here, E-DII was calculated using data from 24 of the 45 variables, including pro-inflammatory components (saturated fat, protein, energy, cholesterol, carbohydrates, iron, and fat) and anti-inflammatory components (fiber, ginger, caffeine, green tea, carotenoids, monounsaturated fatty acids, riboflavin, polyunsaturated fatty acids, vitamin A, vitamin E, vitamin C, onions, garlic, zinc, selenium, thiamine, and magnesium). E-DII scores were calculated per 1000 kcal consumed daily, to balance the differences in total energy intake among participants.

Sample collection

Participants were instructed to avoid brushing their teeth the morning of sample collection day and the preceding night, and to fast for ≥ 2 h before sampling. To assess the oral health of all participants, we performed thorough visual inspections of their dentition. Following the comprehensive periodontal assessment, the investigation focused on identifying the deepest pocket within each of the four quadrants for targeted subgingival plaque sampling. Before sampling, any residual gingival debris was meticulously removed using sterile cotton swabs to ensure the purity of the sample. Subsequently, the subgingival plaque was meticulously collected using 40# sterile paper dots (Gapadent, Tianjin, China) by gently inserting the dots into deep periodontal pockets for 20 s, as described in our previous study⁸¹. All samples collected from each participant were mixed in a 1-mL microcentrifuge tube.

Furthermore, each participant was asked to collect a fresh fecal sample in the morning. Because several older participants 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.

Subgingival plaques and feces were transported to laboratory on ice packs within 2 h of collection and stored at − 80 °C. A total of 54 subgingival plaque samples and 54 fecal samples were collected and subjected to 16S rRNA sequencing.

DNA extraction, amplification, and high-throughput sequencing

The details of DNA extraction, amplification, and high-throughput sequencing have been described in our previous study³². Sequencing was performed at the DNA Sequencing and Genomics Laboratory of Sangon BioTech (Shanghai, China) and all operations followed 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 DNA extraction. The 16S rRNA V3–V4 amplicon was amplified KAPA HiFi Hot Start Ready Mix (2 ×) (TaKaRa Bio Inc., Japan) using two universal bacterial 16S rRNA gene amplicon PCR primers (PAGE purified). PCR was performed in a thermal cycler (Applied Biosystems 9700, USA). Sequencing was performed using the Illumina MiSeq system (Illumina MiSeq, USA).

Sequencing and bioinformatics analysis

After sequencing, two short Illumina readings were assembled by PEAR (v0.9.8). The overlap and fastq files were processed to generate individual fasta and qual files, which were analyzed by standard methods. The effective tags were clustered into operational taxonomic units (OTUs) of ≥ 97% similarity using Usearch (v11.0.667). Chimeric sequences and singleton OTUs (with a single read) were removed; the remaining sequences were sorted into each sample based on the OTUs. The sequences were processed using SILVA (https://arb-silva.de/) and Human Oral Microbiome Database (HOMD) (https://homd.org/) as references. The α-diversity was measured using Shannon and Simpson indices; the β-diversity, indicating community structure, was determined using Bray–Curtis. Principal coordinate analysis (PCoA) was conducted using curtis similarity clustering analysis. Their association between microbial population distribution and cognitive function was tested with partial distance-based redundancy analysis (dbRDA). PICRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States) is a tool for predicting bacterial community function⁸², and was employed to predict the probable functions of subgingival and gut microbial communities. The anticipated functional genera were classified based on the level 2 Kyoto Encyclopedia of Genes and Genome (KEGG) orthology (KO).

Statistical analyses

IBM SPSS (v27.0) and R (v4.3.2) software were used for statistical analyses, and p-values < 0.05 (two-tailed) were considered statistically significant. The Shapiro–Wilk test was applied to test the normality of data distribution. Mean with standard deviation (x̅ ± SD) was used for normally distributed measures, whereas median with interquartile range (IQR) was used for non-normally distributed measures. The frequency (percentage) was calculated for count data. Statistical differences between the two groups were determined by the Student’s t-test or Mann–Whitney U test. Multiple groups were compared using one-way analysis of variance (ANOVA) or the Kruskal–Wallis test. Post-hoc comparisons were performed by applying Bonferroni adjustment to multiple comparisons. Pearson’s chi-squared test or Fisher’s exact test was performed to compare categorical variables. Permutational multivariate analysis of variance (PERMANOVA) was employed to identify different microbial communities. Linear discriminant analysis (LDA) effect size (LEfSe) was used to identify the microbial taxa that significantly differed among different E-DII groups, applying a threshold LDA score of > 2.0. Spearman’s correlation coefficients were calculated between the relative abundance of differentially abundant taxa based on different E-DII levels and the scores of different cognitive domains. The random forest model was built through the caret R package. Five-fold cross-validation and area under the receiver operating characteristic (ROC) curve (AUC) were used to evaluate the prediction performance of the model and was implemented using the pROC package in R.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

J. Liu: Conceptualization, writing—original draft preparation. Y. Zhang: Conceptualization, writing—original draft preparation. X. Li: Investigation, Methodology. Z. Hou: Investigation, Methodology. B. Wang: Data curation. L. Chen: Writing—review and editing. M. Chen: Writing—review and editing. All authors reviewed the manuscript.

Funding

This work was supported by the Joint Funds for the innovation of science and Technology, Fujian province [grant number 2023Y9283], Natural Science Foundation of Fujian Province, China [grant number 2024J011029] and the Medical Innovation Project of Fujian Province [grant number 2023CXA004].

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

The study protocol was approved by the institutional review board of Fujian Provincial Hospital (Ref no. K2022-09-023). All participants gave their informed consent in writing for participating in the study and the research was conducted in accordance with the Declaration of Helsinki. All methods were performed in accordance with the relevant guidelines and regulations for human research, and the study was conducted in compliance with ethical standards.