Demographics and clinical data
Fifty-seven isolated dystonia patients and 27 HCs were recruited and an integrative analysis of the gut microbiome and serum metabolome was performed. Clinical details of the study cohort are shown in Table 1.
Totally, 4,397,348 high-quality 16S rDNA V4 sequences were obtained, with 52,349 ± 8462 (mean ± SD) per sample after de-multiplexing and quality control (Supplementary Table S1). A total of 997 OTUs (> 97% similarity, excluding singletons and low-abundance taxa with frequency <3 for samples, range = 204–461, median = 306 taxa per sample) were identified, representing 22 taxonomic phyla. The dominant phyla were Firmicutes (66·4%), Bacteroidetes (17·5%), Proteobacteria (9·4%) and Actinobacteria (5·3%). At the genus level, Bacteroides (13%), Faecalibacterium (10%), unidentified Lachnospiraceae (7%), Agathobacter (6%) and Blautia (5%) were the most abundant, of which all but Bacteroides belong to Clostridiales in Firmicutes (Supplementary Figure S1).
Gut microbiota revealed by 16s rDNA amplicon sequencing
Although gut microbial α-diversity was not significantly different as calculated by Shannon and Simpson indices (Figure 1A), NMDS analysis demonstrated a significant difference between isolated dystonia and HCs (Figure 1B, PERMANOVA, R2 = 0·157, p = 0·002). We also observed higher β-diversity in the microbiota of dystonia patients based on ANOSIM analysis, indicating a more heterogeneous community structure among dystonia patients compared with HCs (Figure 1C). LEfSe analysis showed that Clostridiales and Bacteroidetes were the key taxa distinguishing isolated dystonia from HCs. Dystonia patients were significantly enriched with Blautia, Bifidobacterium and unidentified Lachnospiraceae, while depleted with Bacteroidetes and unidentified Clostridiales as compared with HCs (Figure 1D). The relative abundance of the most abundant taxa (≥1%) at different taxonomic levels was further compared (Supplementary Table S2 and Figure S2). At the species level, Eubacterium hallii, Blautia obeum and Dorea longicatena were enriched in dystonia subjects, while Bacteroides vulgatus, Bacteroides_plebeius and Fusobacterium varium were enriched in HCs (Figure 1E).
Stratification of bacterial community in dystonia patients
Given more heterogeneity, we applied DMM to our cohort to investigate possible gut microbial subgroups in dystonia, and identified two significantly distinct MCSs (n = 8 and n = 49; R2 = 0·138; p <0·001) using a Laplace approximation (Figure 2A). Specific co-occurring bacterial families were characteristically enriched in these groups; the MCS1 (n = 8) microbiota was characteristically enriched by Bacteroidaceae, Ruminococcaceae and Lachnospiraceae. The second larger group, which we designated MCS2A (n = 20) and MCS2B (n = 29), respectively，exhibited a sharp decreased abundance of Bacteroidaceae, especially in MCS2B (Figure 2B). NMDS analysis confirmed a strong and significant relationship between MCS class and bacterial β-diversity (PERMANOVA, R2 = 0·29, p <0·001), corroborating the existence of compositionally distinct microbial states (Figure 2C). These distinct microbial states exhibited significant differences in diversity (Shannon, Kruskal-Wallis tests, p = 0·002; Figure 2D), with MCS2B exhibiting the lowest index. We also investigated whether specific factors (illness duration, age at onset, BMI and severity evaluation scores) that was differentially associated with microbial community states. No significant correlation was detected.
Functional analysis of the isolated dystonia by metagenomic sequencing
As shown above, 29 of 57 (51%) the isolated dystonia underwent a gut microbial dysbiosis. To further explore microbial functional feature of dystonia, we performed metagenomic sequencing of a subgroup consisting of 13 patients from MCS2B and 13 gender- and age-matched HCs. An average of 10·7 ± 2·1 Gb clean reads were generated per sample，except for one HC sample removed due to low depth of sequencing (Supplementary Table S3).
Consistent with the 16S rDNA analysis, B. obeum, D. longicatena and E. hallii were significantly increased in dystonia, while B. vulgatus and B. plebeius were decreased (Figure 1F and Supplementary Figure S3). We estimated the abundance of metabolic pathways using metagenomic reads mapped to functional orthologs from the KEGG databases to explore differences in the metabolic potential of gut microbiomes between dystonia and HCs. Dystonia communities were functionally different from healthy communities, and less closely clustered together among individuals, suggesting that inter-individual functional variation was higher in dystonia than in HCs (Figure 3A).
A total of 22 metabolic pathways (level 3) were found to be significantly different (p <0·05, q <0·02) in abundance (>0·01%) between dystonia patients and HCs, including those involved in nucleotide, amino acid, carbohydrate and lipid metabolism (Supplementary Table S4). Interestingly, we identified several different pathways that appeared to be more active in the microbiome of dystonia patients, including purine metabolism (ko00230), peptidoglycan biosynthesis (M00550), glycerolipid/glycerophospholipid metabolism (ko00561, ko00564), phenylalanine, tyrosine and tryptophan biosynthesis (ko00350, ko00400) and sulphur metabolism (ko00920) (Figure 3B). While genes related to TCA cycle (ko00020, ko00630, ko00720), glycan biosynthesis and metabolism pathways (ko00511, ko00531, ko00540, ko00600, ko00603) and vitamin B6 metabolism (ko00750) were less abundant in dystonia (Figure 3C). Furthermore, we traced the contributing genes and determined their likely taxonomic origin to determine which bacteria are involved in these pathways (Figure 3B, C).
Associations between gut microbial species and serum metabolites
To investigate the extent to which the altered microbiome in the dystonia patients was associated with serum metabolites in the host, we performed non-targeted metabolomics profiling of serum from the 13 dystonia patients and the 12 HCs in metagenomic analysis, and yielded 1543 features (Supplementary Table S5). A supervised PLS-DA using two components (R2X = 0.525, R2Ycum = 0·99, Qcum2 = 0·975, p <0·001) was performed, resulting in some separation tendencies between dystonia cases and HCs (Figure 4A). Based on the PLS-DA models of metabolite profiling data, 242 metabolites were found to be significantly different in abundance between dystonia cases and HCs, with 87 metabolites having a higher concentration and 169 metabolites having a lower concentration in dystonia compared with HCs (p <0·05, VIP >1 and FC >2 or <0·5, Figure 4B, Supplementary Table S5).
Subsequently, spearman’s correlation coefficients were computed for relationships between the relative abundance of the identified dystonia-associated species and the different 242 normalised individual metabolomic features. Metabolites were grouped into two clusters depending on the correlations, and correlation coefficients with significant p-values (<0·01) are shown in Figure 4C. Clostridiales species enriched in dystonia patients correlated positively with the first metabolite cluster, including L-glutamic acid, phenylalanylphenylalanine and taurine. While Bacteroides species with decreased abundance in dystonia patients correlated positively with the second cluster, including D-tyrosine, D-(-)-aspartic acid and N, N-dimethylacetamide (DMAc).