Relationships of gut microbiota, short-chain fatty acids, inflammation, and the gut barrier in Parkinson’s disease

The study subjects were originally recruited for a pilot study of gut microbiota and PD [31]. Samples for the present study were collected at a follow-up timepoint together with extensive clinical data, including questionnaires on diet, non-motor symptoms, and PD severity; full details have been published previously [32]. The majority of the patients in this study were classified as being in the early to mid-stages of PD (2–3 on Hoehn and Yahr scale). Subjects collected stool samples at home into tubes with DNA stabilizer for microbiota analyses and tubes without preservatives for SCFA and immunological analyses. They kept the samples refrigerated until transport to the clinic (up to 3 days) for storage at -80 °C. Venous blood samples were collected into EDTA tubes, centrifuged (10 min, 3000 rpm), and plasma was divided into aliquots and stored at -80 °C. Samples were distributed to the laboratories by overnight shipping on dry ice. After excluding subjects with missing values for microbiota, acetic, butyric, or propionic acid, or most immune markers, the total number of subjects remaining for analysis was 111 (55 patients, 56 control subjects).

The microbiota data analyzed in this study have been published previously with a detailed description of the workflow [32]. Briefly, stool samples were collected into PSP Spin Stool DNA Plus Kit tubes (STRATEC Molecular). DNA was extracted with the corresponding kit from the same manufacturer. The V3-V4 regions of the 16S rRNA gene were PCR-amplified and sequenced with Illumina MiSeq. Sequence quality control, Operational Taxonomic Unit (OTU) clustering, and taxonomical classification were performed with mothur [33, 34]. Enterotypes were determined with an online tool [35, 36].

Detection and quantification of SCFAs in the stool samples were performed at the Norwegian University of Life Sciences using a Trace 1310 gas chromatograph (Thermo Fisher Scientific).

Factors of interest in stool were measured using the Zonulin Stool ELISA (ALPCO 30-ZONHU-E01), LEGEND MAX™ Human MRP8/14 (Calprotectin) ELISA (Biolegend 439,707), LEGEND MAX™ Human NGAL (Lipocalin-2) ELISA (Biolegend 443,407), and the V-PLEX Proinflammatory Panel 1 Human (Meso Scale Discovery [MSD], Rockville, MD, K15049D) kits. The V-PLEX kit was also used to measure inflammatory mediators in plasma, as was the Human LBP Kit (MSD K151IYC). Stool and plasma samples were prepared, assays run, and results analyzed according to manufacturers’ protocols by the Emory Multiplexed Immunoassay Core.

All statistical analyses were performed in R [37] with packages including phyloseq [38], vegan [39], and DESeq2 [40] for microbial data comparisons. For statistical comparisons between SCFAs/markers, clinical variables, enterotypes, and alpha diversity, we used the Kruskal-Wallis test, Wilcoxon rank sum test, or Pearson correlations depending on the types of variables. When multiple comparison corrections were included, we used the Benjamini & Hochberg false discovery rate (FDR), correcting SCFAs, stool markers, and plasma markers separately. Intercorrelated inflammatory markers were merged based on Principal Component Analysis (PCA). The complete analysis is included as Additional File 2.

Patient and control groups were similar with regard to basic demographics such as age, sex, and body mass index (BMI); however, as we knew from previous analyses of the same subjects [31, 32], the groups differed regarding medications, medical history, and various symptom scores. A higher percentage of controls reported a history of stroke and use of medications for high blood pressure and cholesterol, and PD patients scored higher on scales of non-motor symptoms, gastrointestinal problems, and constipation (Table 1). Contrasting inflammatory markers and SCFAs between PD patients and control subjects, patients had lower levels of butyric and propionic acid and higher levels of calprotectin in their stool and lower levels of CXCL8 in plasma (Additional File 3A). When the data were stratified by sex, the differences were particularly prominent for butyric acid in males and for plasma CXCL8 and stool calprotectin in females (Fig. 1, Additional File 3B). For calprotectin, this sex effect may be explained in part by relative baseline levels of the protein. Among control subjects, calprotectin levels trended lower in females than in males (p=0.051) (Additional File 3B).

The inflammatory markers measured from the same sample material (plasma or stool) were largely intercorrelated, but we found no correlations between measurements of the same markers in stool and plasma (Fig. 2). PCAs for combining the most correlated markers resulted in one principal component (PC) for stool (summarizing IL-10, IL-12p70, IL-13, TNF, IL-6, and IFNγ), and two plasma PCs (PC1: IL-10, IL-4, TNF; PC2: IL-1β, IL-2, IL-12p70, IL-13) (Additional File 2). In stool, the inflammatory markers neutrophil gelatinase-associated lipocalin (NGAL) and calprotectin and the gut permeability marker zonulin were significantly correlated (Fig. 2). In general, calprotectin and zonulin showed fewer correlations with other stool markers in the PD group as compared to the control group. Except for correlations with stool CXCL8 and IL-1β in the control group, none of the three most abundant SCFAs was significantly correlated with any of the markers in stool or plasma (Fig. 2).

Contrasting clinical variables against SCFAs and inflammatory markers suggested inverse correlations between stool SCFAs and several PD-related clinical variables such as the Non-Motor Symptoms Scale (NMSS) total score (which assesses factors including orthostatic hypotension, anosmia, depression, anxiety, cognitive impairment, constipation, and abnormalities in sleep, perception, and urinary patterns), the Rome III constipation/defecation subscore, stool consistency associated with constipation on the Victoria Bowel Performance Scale (BPS), and the Geriatric Depression Scale-15 (GDS15) (Fig. 3a, Additional File 4A). A similar pattern was seen for the Rome III constipation/defecation subscore, Rome III irritable bowel syndrome (IBS) criteria, and NMSS score and plasma CXCL8 (Fig. 3b, Additional File 4A). These correlations were significant for the entire cohort, and some were significant in controls, but not when evaluated only within PD patients. Butyric acid levels were correlated, however, with the age of onset for motor and non-motor symptoms of PD, while CXCL8 and IL-1β levels in stool were inversely correlated with the age of motor symptom onset (Fig. 3a, c, Additional File 4B). Acetic acid, while not significantly associated with constipation-related stool consistency in PD patients, was positively correlated with defecation frequency (Fig. 3a). Higher levels of CXCL8 in stool were also associated with constipation-related stool consistency in PD patients and with lower defecation frequency in controls (Fig. 3c). In PD patients, calprotectin levels were inversely linked to IBS symptoms, and stool zonulin was inversely correlated with variables related to PD severity, including the Unified Parkinson’s Disease Rating Scale (UPDRS) score, Hoehn and Yahr, GDS15, and NMSS score (Fig. 3c, Additional File 4B). Symptom onset- and severity-related correlations remained significant after correcting for disease duration (Additional File 2).

In PD patients, several measures of symptom severity were positively correlated with microbial alpha diversity indices, which indicate richness and evenness of bacterial taxa. These included UPDRS scores, Hoehn and Yahr score, and subscales assessing tremor, rigidity, and postural instability and gait difficulty (PIGD) (Additional File 5A, 5B). Microbial beta diversity describes the degree to which different communities of bacteria differ from one another in composition. We found that beta diversity was also significantly associated with UPDRS scores, rigidity, and PIGD. Additionally, the usage of PD medications was significantly related to beta diversity (Additional File 5C).

Microbial alpha diversity indices were inversely correlated with total SCFAs, acetic, butyric, and propionic acid, and NGAL in stool in control subjects but not in PD patients (Fig. 4). In patients only, a significant inverse correlation between the Shannon diversity index and stool CXCL8 was found (Fig. 4). Inverse correlations between alpha diversity indices and stool calprotectin, zonulin, and IL-1β were significant in the whole cohort, but when the cohort was split by PD status, only the correlations between the Shannon diversity index and calprotectin in PD patients and between the Shannon diversity index and zonulin in control subjects remained statistically significant.

Linear regression modeling corrected for sex and PD/control status suggested that the relationship between alpha diversity and stool inflammatory markers was independent of sex except for calprotectin, for which an inverse relationship was observed in females but not in males (Fig. 6a, Additional File 6A-C). There was a significant interaction between PD/control status and alpha diversity for propionic acid, showing an inverse correlation between microbial alpha diversity and propionic acid concentration in controls but not in PD patients (Fig. 5b). The total SCFAs variable showed a similar trend (Additional File 6B). The interaction in the propionic acid model remained significant after confounder correction (Additional File 6D).

We found significant associations between beta diversity and total SCFAs, acetic, butyric, and propionic acid as well as stool NGAL, zonulin, IL2, and the stool marker PC in full data, with or without correction for PD/control status, sex, constipation, and BMI (Additional File 7A). The effect of calprotectin, while not statistically significant, was worth noting (p = 0.058 for full data, p = 0.079 when corrected for confounders). Control-only comparisons showed significant effects for SCFAs, NGAL, and zonulin; only butyric acid and zonulin were significant in PD-only comparisons (Additional File 7A, 8). These two variables also had a significant interaction with PD/control status when tested with the variable split into two categories by median (Additional File 7B).

One broad characterization method for microbiota composition is defining enterotypes based on certain indicator species in bacterial communities [36]. In our subjects, the Firmicutes enterotype was more common and the Prevotella enterotype less common among PD patients compared to control subjects (Additional File 9A). Stool zonulin, NGAL, propionic acid, and butyric acid had differences in concentrations between enterotypes (Additional File 9B). The difference in propionic acid was significant only in all subjects, not in controls or PD patients separately. Butyric acid levels were higher in PD patients with the Prevotella enterotype than the other two enterotypes, and control subjects with this enterotype had the lowest concentrations of NGAL and zonulin (Fig. 6). The findings regarding NGAL (full data and controls) and zonulin (controls only) remained significant after FDR correction (Additional File 9B).

Guided by the associations with beta diversity, we explored the associations of specific bacterial taxa with the three most abundant SCFAs and with stool NGAL, calprotectin, and zonulin. The abundances of Butyricicoccus, Clostridium sensu stricto, and Roseburia were positively correlated with levels of SCFAs while the abundances of Akkermansia, Escherichia/Shigella, Flavonifractor, Intestinimonas, Phascolarctobacterium, and Sporobacter decreased with increasing SCFA concentrations (Fig. 7, Additional File 10, 11A). Similar patterns could be seen for the parent families of these taxa, such as a positive association between butyric acid and Lachnospiraceae and a negative one for Enterobacteriaceae and Verrucomicrobiaceae (Additional File 11A). A positive association between SCFAs and Bacteroides was observed only in controls, while a negative association with Bifidobacterium was observed only in PD patients (Fig. 7, Additional File 10, 11A).

Microbiota associations with the stool markers differed between PD patients and control subjects. Stool NGAL levels were positively associated with Bifidobacterium in PD but with Bacteroides and Roseburia in controls (Fig. 7, Additional File 11B, 12A). Relative abundance of Prevotella was negatively associated with both NGAL and zonulin in controls only (Fig. 7, Additional File 11B, 12). The only microbiota-immune association found to be significant in both PD patients and controls was the negative relationship between stool zonulin and the genus Coprococcus (Fig. 7, Additional File 11B, 12B), a taxon known to have a strong influence on gut permeability [41] and stool zonulin levels [42]. Other associations with low-abundance taxa were statistically significant but were driven by a small number of data points, precluding confident interpretation (Additional File 12).

Key findings from this study are summarized in Additional File 13.