Alterations in gut microbiota and plasma metabolites: a multi-omics study of mild cognitive impairment in Parkinson’s disease

The study participants were recruited from the Affiliated Hospital of Xuzhou Medical University in Xuzhou, China, between May 2022 and June 2023. We initially recruited 100 PD patients and 50 HCs with no history of PD. All patients with PD met the International Parkinson and Movement Disorder Society clinical diagnostic criteria in 2015 (Postuma et al., 2015).

Participants were omitted from the study with: (a) diagnosed psychiatric disorders; (b) atypical or secondary parkinsonism; (c) antibiotic, probiotic, or prebiotic use within the last 1 month; (d) co-infection; (e) severe chronic acute illness or severe cognitive impairment that hindered their ability to complete the study questionnaires; (f) acute and chronic primary gastrointestinal disorders; and (g) DBS surgery. These exclusion criteria were formulated according to the patient's clinical background and pertinent hospitalization records.

Following initial recruitment, participants underwent cognitive assessment. To minimize potential interference from insufficient education or illiteracy on the MoCA scale, we excluded 24 PD patients and 4 HCs with less than 6 years of education. Additionally, to focus on early-stage cognitive changes, we excluded participants with MoCA scores ≤ 20 points (38 PD patients and 6 HCs were excluded). After applying these criteria, the final study cohort consisted of 38 PD patients and 40 HCs. The PD patients were further classified into two groups: 18 with normal cognition (PD-NC; MoCA >25) and 20 with mild cognitive impairment (PD-MCI; MoCA 21–25).

Each patient provided informed consent. The study received ethical approval from the Ethics Committee of the Affiliated Hospital of Xuzhou Medical University (approval number: XYFY2022-KL262-01, approval date: 26 July 2022; approval number: XYFY2023-KL266-01, approval date: 02 August 2023). This investigation adhered to the Declaration of Helsinki principles.

Clinical data evaluation was performed by ZH.L., YDY.L., YN.L. at the Affiliated Hospital of Xuzhou Medical University (Xuzhou, Jiangsu Province, China). Our study recorded demographic and clinical characteristics, comprising age, disease duration, gender, Body Mass Index (BMI), Levodopa Equivalent Daily Dose (LEDD), Hoehn and Yahr stage (H-Y stage), and years of education.

MoCA scale was utilized to assess the patients' cognitive functions during the "on" period, patients scoring between 20 and 25 points were identified as PD-MCI, while those scoring 20 points or below were classified as Parkinson's disease dementia (PDD) (Skorvanek et al., 2018). The Movement Disorder Society-Unified Parkinson's Disease Rating Scale Part III (MDS-UPDRS III) was used to assess motor symptoms (Postuma et al., 2015).

Peripheral venous blood was collected from each participant. The samples were then centrifuged at 3000 × g for 10 min within an hour after collection. Fecal samples were obtained from subjects during the early hours of the day. Both plasma and fecal samples remained preserved at −80 °C pending analysis. Gut microbiota in fecal specimens were sequenced using 16S rRNA sequencing and metabolites contained in plasma specimens were analyzed using LC-MS.

Continuous demographic variables appear as mean (standard deviation) or median (interquartile range), while categorical variables are shown as percentages. For two-group comparisons, Student's t-test was used for normal distributions and Mann–Whitney U-test for non-normal distributions. For three-group comparisons, one-way ANOVA was used for normal distributions and Kruskal-Wallis test for non-normal distributions. Chi-square test was used for categorical data comparisons. Statistical significance was defined as p < 0.05, with analyses conducted using SPSS (version 29.0).

Calculating the alpha-diversity indices with Mothur v1.30.2 (Schloss et al., 2009), the Kruskal-Wallis rank sum test was used to compare the differences between groups. Microbial community similarities between samples were assessed via Principal Coordinate Analysis (PCoA) using both unweighted and weighted UniFrac in the Vegan v2.4.3 software package. And to compare between-group differences using Adonis. Beta-diversity indices differences between groups were analyzed using Partial Least Squares Discriminant Analysis (PLS-DA). To evaluate within-group dispersion, we calculated each sample's Bray–Curtis dissimilarity distance to its group centroid (geometric median) and compared these distance-to-centroid values between groups using the Wilcoxon rank-sum test in R (version 3.3.1) (Abbas et al., 2019). The Gut Microbiota Health Index (GMHI) and the Microbial Dysbiosis Index (MDI) were applied to further evaluate the gap in gut microbiota between the cohorts. It distinguishes between healthy and unhealthy populations more effectively than alpha-diversity indices and ecological measures like Shannon diversity and richness, with consistent stratification performance across populations and it is generally regarded as an indicator of gastrointestinal health and the presence of dysbiosis (Gupta et al., 2020). MDI is an index used to assess the degree of microbial ecological disturbance, with higher scores indicating more severe disturbance (Gunathilake et al., 2020). The Linear Discriminant Analysis (LDA) effect size (LEfSe) (Segata et al., 2011), Microbiome Multivariable Associations with Linear Models 2 (MaAsLin2) (Mallick et al., 2021), and Microbiome Multivariable Associations with Linear Models 3 (MaAsLin3) (Nickols et al., 2024) were utilized to identify significantly enriched bacterial taxa (from phylum to genus) in different groups. LEfSe analysis was performed with LDA scores > 2 and p < 0.05 as significance thresholds. MaAsLin2 and MaAsLin3 analyses were conducted with adjusted p < 0.05 after adjusting for potential confounders including age, gender, BMI, LEDD, UPDRS III score, year of education and disease duration. Microbial taxa identified by at least two algorithms were considered as core differential microbes.

The R package ropls (version 1.6.2) was used for Principal Component Analysis (PCA) and PLS-DA. Notable distinct metabolic compounds were identified through Variable Importance in Projection (VIP) values (VIP > 1) derived from PLS-DA analysis and confirmed via Student's t-test with statistical threshold of p < 0.05. Using the Kyoto Encyclopedia of Genomes (KEGG) database, distinguishing metabolic compounds between the comparison groups were linked to their biological pathway through pathway enrichment evaluation. Metabolite enrichment analysis conducted via the "scipy.stats" package in Python. Assessment of Metabolite Diagnostic Performance via Receiver Operating Characteristic (ROC) Analysis. Correlations between differentially expressed metabolites and differentially abundant gut microbiota (genus level) were evaluated using Spearman correlation coefficients, with statistical significance set at p < 0.05 (Figure 1).

Our study initially enrolled 100 individuals with PD alongside 50 healthy participants of comparable age. Subsequent to the implementation of exclusion criteria (education <6 years and MoCA scores ≤20), the final cohort comprised 38 PD patients (18 PD-NC and 20 PD-MCI) and 40 HCs. No statistically significant differences were observed among groups (PD vs. HC, and PD-MCI vs. PD-NC) in terms of age, gender, years of education, disease duration, LEDD, MDS-UPDRS III, H-Y stage and BMI (Table 1).

To characterize the microbial community richness and diversity among HC, PD-NC and PD-MCI groups, we performed comprehensive alpha-diversity indices analyses utilizing multiple ecological indices, including Sobs, Shannon, Chao, ACE, Simpson, and coverage. Statistical analyses revealed no significant differences in any of the aforementioned alpha-diversity indices between the PD-MCI and PD-NC groups (Table 2). These results indicate that there were no differences in gut microbiota diversity or richness among HC, PD-NC and PD-MCI groups.

In parallel, to examine the microbial community structure among HC, PD-NC and PD-MCI groups, we performed beta-diversity indices analysis using PCoA. While weighted UniFrac-based PCoA showed no clear separation among groups (R² = 0.03971, p = 0.053, Figure 2A), unweighted UniFrac-based PCoA revealed significant compositional differences (R² = 0.04169, p = 0.019, Figure 2B), particularly along PC1 axis (p = 0.005). To further discriminate between PD-NC and PD-MCI groups, we performed PLS-DA at the genus level. The PLS-DA successfully distinguished the two groups, demonstrating clear separation between PD patients with and without cognitive impairment (Figure 2C). The PLS-DA loading plot (Supplementary Figure S1) reveals the key gut microbial taxa contributing to the separation between PD-MCI and PD-NC groups, with taxa exhibiting high loading values representing the most discriminative features. Furthermore, to assess within-group heterogeneity, we calculated each sample's Bray–Curtis dissimilarity to its respective group centroid. Wilcoxon rank-sum test comparing these distance-to-centroid values revealed significant differences between groups (adjusted p = 0.037, Figure 2D), these results suggest differences in both gut microbiota structure and internal variability between PD-MCI and PD-NC groups.

Given these structural differences, we further analyzed both GMHI and MDI to comprehensively evaluate the gut microbiota health status, among HC, PD-NC and PD-MCI groups. GMHI analysis revealed significantly higher scores in the PD-NC group compared to the PD-MCI group (adjusted p < 0.001, Figure 3A). Similarly, in the comparison between HC and both PD-NC and PD-MCI groups, HC exhibited significantly higher scores (adjusted p < 0.001, Supplementary Figures S2A,B). Further analysis revealed significantly elevated MDI values in the PD-MCI group compared to the PD-NC group (adjusted p < 0.001, Figure 3B). We also analyzed the MDI differences between both PD-NC and PD-MCI groups versus HC. We found that both PD-NC and PD-MCI groups exhibited higher MDI values compared to the HC group (adjusted p < 0.001, Supplementary Figures S2C,D). Collectively, these findings suggest that PD-NC group maintain a more favorable gut microbiota profile compared to PD-MCI group.

To better understand the compositional differences, we investigated the taxonomic composition of the gut microbiota between PD-MCI and PD-NC groups. At the phylum level, comparative analysis showed significant differences in microbial abundances between groups. Specifically, p__Bacteroidota and p__Proteobacteria showed higher relative abundances in the PD-MCI group, whereas p__Firmicutes, p__Actinobacteriota, and p__Verrucomicrobiota were reduced compared to the PD-NC group (Figure 3C). To further elucidate these taxonomic differences at a finer resolution, we performed LEfSe, MaAslin2, and MaAslin3 analyses at the genus level. LEfSe analysis identified differential microbiota, with g__Blautia, g__Eggerthella, g__Lachnoclostridium, g__Erysipelatoclostridium, and g__norank_f__norank_o__Oscillospirales being significantly enriched in the PD-NC, while g__Senegalimassilia and g__Eubacterium were significantly more abundant in the PD-MCI (Figures 3D,E). MaAsLin2 identified 11 differentially abundant genera between PD-MCI and PD-NC groups (FDR < 0.05, Figure 3F). Among these, 4 genera were significantly enriched in PD-MCI: g__Sutterella, g__Megamonas, g__Barnesiella, and g__Rothia. Conversely, 7 genera showed significant enrichment in PD-NC group: g__Sellimonas, g__norank_f__Oscillospiraceae, g__Eggerthella, g__Enterococcus, g__Megasphaera, g__Lactococcus, and g__Slackia. We additionally performed MaAsLin3 analysis to further validate our findings. Under the stringent criterion of FDR < 0.05, no differentially abundant genera were identified. However, when applying a less conservative threshold of p < 0.05, g__Senegalimassilia emerged as significantly enriched in the PD-MCI group (Supplementary Table S1). Among these, g__Eggerthella was identified as a differential microorganism by both MaAsLin2 and LEfSe, and was therefore considered a key taxon.

To identify plasma metabolic changes and potential diagnostic markers in PD-MCI, we performed untargeted metabolites analysis on plasma samples from PD-MCI group and PD-NC group. Using PLS-DA and VIP analysis, we identified 123 differential metabolites between PD-MCI group and PD-NC group (VIP >1, p < 0.05). Among these metabolites, 40 were increased and 83 were decreased in the PD-MCI group compared to the PD-NC group (Figure 4). These metabolic alterations suggest significant metabolic disturbances in PD-MCI patients.

Further analysis revealed 30 metabolites with biological significance in the KEGG database. KEGG enrichment analysis showed that these metabolites were significantly enriched in 3 metabolic pathways: phenylalanine metabolism (p < 0.001), choline metabolism in cancer (p = 0.002), and PPAR signaling pathway (p = 0.034) (Figures 5A,B). Among these, the metabolites in the phenylalanine metabolism and PPAR signaling pathway were downregulated in the PD-MCI group, while choline metabolism in cancer was upregulated. KEGG pathway enrichment discovered 7 key metabolites: 8-hydroxy-5Z,9E,11Z,14Z-eicosatetraenoic acid, 2-hydroxycinnamic acid, N-acetyl-L-phenylalanine, phenylacetylglycine, phenylacetylglutamine, PC(18:1(11Z)/18:3(6Z,9Z,12Z)), and LysoPC(18:3(6Z,9Z,12Z)/0:0) (Table 3). Among them, the first 5 metabolites significantly decreased in the PD-MCI group compared to the PD-NC group, while the latter 2 metabolites increased. Among these metabolites, 8-hydroxy-5Z,9E,11Z,14Z-eicosatetraenoic acid belongs to the PPAR signaling pathway; PC(18:1(11Z)/18:3(6Z,9Z,12Z)) and LysoPC(18:3(6Z,9Z,12Z)/0:0) are involved in choline metabolism in cancer; 2-hydroxycinnamic acid, N-acetyl-L-phenylalanine, phenylacetylglycine, and phenylacetylglutamine are associated with phenylalanine metabolism.

ROC analysis was conducted on 7 distinct metabolites. The analysis revealed varying degrees of diagnostic potential: 8-hydroxy-5Z,9E,11Z,14Z-eicosatetraenoic acid (AUC = 0.7139), 2-hydroxycinnamic acid (AUC = 0.7000), N-acetyl-L-phenylalanine (AUC = 0.6917), phenylacetylglycine (AUC = 0.6889), PC(18:1(11Z)/18:3(6Z,9Z,12Z)) (AUC = 0.7222), and LysoPC(18:3(6Z,9Z,12Z)/0:0) (AUC = 0.7333) (Table 4). Notably, phenylacetylglutamine emerged as the most promising single biomarker for PD-MCI identification, demonstrating superior diagnostic capability (AUC = 0.8222) (Table 4; Figure 5C).

Spearman correlation analysis was conducted between KEGG-annotated differential metabolites and differentially abundant bacterial genera (genus level) identified by LEfSe (LDA > 2.0, p < 0.05) and MaAsLin2 (adjusted p < 0.05) between PD-NC and PD-MCI groups. This approach aimed to elucidate specific associations between bacterial genera and metabolites related to cognitive decline in PD, as visualized in Figure 6. In the PD-MCI group, g__Senegalimassilia negatively correlated with 2-hydroxyfelbamate and pseudoephedrine, g__Eubacterium showed a negative correlation with 2-oxo-4-methylthiobutanoic acid and positive correlations with L-Alanine and endomorphin-1. g__Sutterella positively correlated with picolinic acid and (S)-5-Amino-3-oxohexanoate but negatively correlated with terephthalic acid. g__Barnesiella exhibited a positive correlation with 3-oxoadipic acid, and g__Rothia negatively correlated with inosinic acid while positively correlating with endomorphin-1. In the PD-NC group, g__Eggerthella exhibited opposing correlations with uridine 2′,3′-cyclic phosphate, while positive correlations with pseudoephedrine and 2-hydroxyfelbamate. g__Blautia negatively correlated with uridine 2′,3′-cyclic phosphate. g__Lachnoclostridium positively correlated with 2-hydroxyfelbamate, (S)-5-amino-3-oxohexanoate, and phenylacetylglutamine, while showing negative correlations with uridine 2′,3′-cyclic phosphate, endomorphin-1, and PC(18:1(11Z)/18:3(6Z,9Z,12Z)). Additionally, g__Erysipelatoclostridium positively correlated with 2-hydroxyfelbamate, 2-hydroxycinnamic acid, and (2R,3S)-2,3-dimethylmalate. g__norank_f__norank_o__Oscillospirales showed positive correlations with multiple metabolites including 8-hydroxy-5Z,9E,11Z,14Z-eicosatetraenoic acid, (S)-5-amino-3-oxohexanoate, picolinic acid, 2,4-Dihydroxy-2H-1,4-benzoxazin-3(4H)-one and phenylacetylglutamine, while negatively correlating with L-Alanine, LysoPC(18:3(6Z,9Z,12Z)/0:0) and terephthalic acid. g__Sellimonas negatively correlated with uridine 2′,3′-cyclic phosphate and 3-oxoadipic acid, but positively correlated with pseudoephedrine. g__norank_f__Oscillospiraceae negatively correlated with L-alanine, glyceric acid, and uridine 2′,3′-cyclic phosphate, while positively correlating with sinapyl alcohol and 4-methylcatechol. g__Enterococcus positively correlating with (2R,3S)-2,3-dimethylmalate, sinapyl alcohol, N-acetyl-L-phenylalanine, vanillin, and ceftizoxime. g__Megasphaera positively correlated with 2-hydroxycinnamic acid, and g__Slackia negatively correlated with phenylacetylglycine.