We present longitudinal fecal microbiome data for WT and HD mice at 4, 6, 8, 10 and 12 weeks of age to determine the onset of gut dysbiosis. A total of 20928 species were detected, and those above a relative abundance of 0.001% were included for further analysis. The remaining 2024 species were not affected by cage or sequencing run effects.
As expected, Bacteroidetes and Firmicutes were the two most abundant bacteria phylum in the mouse fecal microbiome followed by Proteobacteria, Actinobacteria, and Verrucomicrobia. We examined the phylum composition at 5 different time points from the shotgun metagenomics sequencing data (Fig. 1A-D). Two-way repeated ANOVA revealed significant effects of age on the relative abundances of Bacteroidetes (Age:p=0.0019, Geno:p=0.5865), Firmicutes (Age:p=0.0017, Geno:p=0.9378), and Proteobacteria (Age:p=0.0001, Geno:p=0.7592). Overall, there were no significant differences in the relative abundances of Bacteroidetes and Firmicutes between WT mice and HD mice at any of the time points. Similarly, we observed no significant differences in the relative abundances of different bacterial families when comparing WT and HD mice at any of the time points (data not shown).
Subtle differences in bacterial species composition at 12 weeks of age
To further characterize how the phylogenetic and functional differences observed above change with age, we performed a principal coordinates analysis (PCoA) of the obtained taxonomic and gene profiles respectively (Fig. 2). We did not observe any strong differences between samples according to their genotype at 4, 6, 8, and 10 weeks of age. However, at 12 weeks of age, which is prior to overt motor symptom onset, PERMANOVA testing revealed significant effects of genotype in the bacterial species composition based on Bray-Curtis distance (p=0.029).
For alpha diversity measures, we observed no significant effects of age or genotype on the number of bacterial species observed (Two-way repeated-measures ANOVA, Fig. 1E). Age had a significant effect (p=0.006) but not genotype based on the Inverse Simpson index (Fig. 1F).
In terms of beta diversity analysis, Bray-Curtis index indicated significant differences in HD mouse gut microbial structure when compared to their WT littermates at 12 weeks of age (Fig. 2). Thus, we performed sPLS-DA to determine the specific bacterial species that discriminate between the two groups. In total, 50 bacterial species were selected as a signature (classification error rate of 0.3). The differentially abundant bacterial species were found to be Clostridium mt 5, Treponema phagedenis, Clostridium leptum CAG:27, Desulfatirhabdium butyrativorans, Plasmodium chabaudi, Defulfuribacillus alkaliarsenatis, Plasmodium yoelii and Chlamydia abortus.
The known commensal butyrate producers, Faecalibacterium praunitizii and Eubacterium hallii, were more abundant in the gut of HD mice than the WT mice at 12 weeks of age. Akkermansia muciniphila, another well-known species of commensal bacteria was less abundant in the HD gut only at 12 weeks of age. There were no differences in the abundances of other butyrate producers including Roseburia intestinalis, Clostridium symbiosum and Eubacterium rectale.
The remainder of the bacterial species we identified were not well annotated, and thus we were unable to identify their potential effects on gut dysbiosis.
The gut microbiome of HD mice is functionally distinct from their WT littermates at 12 weeks of age
One advantage of whole genome shotgun sequencing is that it allows the profiling of microbial genes to interrogate the function of the gut microbiome, thus enabling us to determine the specific functional effects of the gut microbial composition alteration in the HD mouse. We identified 333 genes above the cut-off relative abundance of 0.1%, corresponding to 245 KEGG orthologs (KOs). We found no effects of cage and sequencing run on the bacterial genes and KOs, so we proceeded to the subsequent analysis without further filtering.
For the KO analysis, the most abundant were “aminoacyl-tRNA biosynthesis”, followed by “alanine, aspartate and glutamate metabolism” and “ABC transporters and purine metabolism”. Focusing on the 12-week time point, sPLS-DA identified a KO signature that characterized the two genotype groups. Five pathways were identified which included galactose metabolism, sulfur metabolism, lysine degradation, glutathione metabolism and butanoate metabolism (classification error rate of 0.2, Fig. 3.
For the gene analysis, sPLS-DA identified 20 genes which form the signature to discriminate the WT and the HD mouse gut (classification error rate of 0.27, Fig. 4). The most discriminant genes were pyre, pepD, oadB, gmd which were also identified in the pathways from the sPLS-DA KO analysis (Fig. 4).
Modelling of bacterial species and KOtrajectories revealed significant volatility in the gut of HD mice
As HD is a progressive neurodegenerative disease, we sought to investigate whether there were any potential effects in the stability of gut microbial structure as well as the function from early adolescence (4 weeks) to the adult stage of life (12 weeks).
LMMS models categorized each species or KO into one of the four different models denoted as 0, indicating a relatively flat profile, 1 indicating either increasing or decreasing over time, 2 and 3 indicating complex curves across time (Fig. 5). After filtering of the noisy profiles, models for 1708 species, 124 KOs and 1228 genes were able to be fitted for both HD and WT mice.
This analysis revealed that in the gut of WT mice, the temporal profile of the bacterial species, genes and KOs remains largely stable throughout adolescence and adult stage: 98%, 97.6% and 100% respectively were categorized into Model 0 (Fig. 5). In the gut of HD mice, the majority of the temporal profiles of bacterial species, genes and KOs were unstable: only 23.5%, 41% and 39.5% respectively were categorized into Model 0 whilst the rest were largely categorized into Model 1 (74.9%, 57.3%, and 58.1% respectively) and the remainder were categorized into Model 2 (Fig. 3). The median proportionality distance showed that most of the members within each model were strongly associated (distance close to 0, Supplementary Table 1).
Targeted metabolomic profiling of plasma revealed distinct differences between HD and WT mice at 12 weeks of age
We further investigated the potential functional differences of the gut microbiome composition in
the circulatory system which could be one of the pathways of bidirectional communication between the brain and the gut microbiome. Due to the nature of the amount of blood required, and the progressive nature of the gut dysbiosis, only the plasma metabolites at 12 weeks of age were examined, following terminal cardiac blood collection. In total, the platform identified 221 metabolites, which belong to the following broad categories: amino acids, carbohydrates, lipids, nucleotides, peptides and xenobiotics. sPLS-DA selected 15 metabolites as a signature which could distinguish the samples based on their genotype (Fig. 6A). The top metabolites selected included adenosine triphosphate (ATP), 3-Methylhistidine, urocanic acid, carnosine, threonic acid, homocitrulline, orotic acid, ADP, p-Aminobenzoic acid, 2-methylbutyrlglycine, trigonelline, alpha-hydroxyisobutyric acid, propionic acid, butyric acid, pipecolic acid, 2-hydroxybutyric acid, isobutyrylglycine, 2-hydroxy-3-methylbutyric acid and ribitol (Fig. 6B).
Significant reduction of propionate and butyrate in the plasma of HD mice compared to their WT littermates
The gut function analysis revealed that butanoate metabolism was affected in the HD gut, hence, we sought to determine a possible change of SCFAs in the blood of the HD mice. Even though SCFA data was detected by the targeted metabolomics, the majority of the SCFAs in the circulatory system would have been metabolized by the liver and a minimal amount would have passed through to the bloodstream. Therefore, the targeted metabolomics approach may not accurately determine the concentration of these SCFAs. Hence, we conducted an independent SCFA assay to validate the SCFA levels in the plasma. Decreases in the plasma concentration of propionate (p= 0.09) and butyrate (p = 0.09, one-way ANOVA) were observed in the HD mice when compared to their WT littermates, but no differences in the concentration of acetate were observed (Fig. 6C).
Integration of plasma metabolomics and shotgun metagenomic sequencing data
Our integrative analysis with DIABLO identified a signature of 30 bacterial species, 20 genes and 8 metabolites that were highly associated (correlation > 0.7 after variable selection, Fig. 7A). Further visualization in Cytoscape revealed that many microbes were found to be highly correlated with butyrate, homocitrulline, ATP, L-asparagine, 3-methylhistidine, orotic acid, and isobutyrylglycine.
We further narrowed our analysis to the metabolites with known functions in the nervous system, including ATP, butyrate and pipecolic acid, and the associated bacterial species which have been classified. Closer inspection of each network revealed many overlaps in the bacterial species between the three networks. Several Bacteroides species including B.pyogenes, B.ihuae, B.oleiciplenus and B.timonensis were positively correlated with butyrate but negatively correlated with ATP and pipecolic acid (Fig. 7B). Similarly, we identified Prevotella ruminicola and Odoribacter laneus to be positively correlated with butyrate levels and negatively correlated with ATP and pipecolic acid.
Blautia producta was uniquely identified to be negatively correlated with butyrate. Prevotella scopos was identified to be negatively correlated with ATP levels and no associations were found with butyrate and pipecolic acid.