signature=20fd33fb253e841a02225ae1c641f184,Microbiome-metabolome signatures in mice genetically pron...

General characterization

We characterized n = 45 mice, grouped in normal diet-fed (ND, n = 9), high-fat diet-fed (HFD, n = 9), ND-fed triple transgenic (3xtg, n = 8) and HFD-fed 3xtg mice (3xtg-HFD, n = 7). Diet was administered up to 8 months of age, when body weight, cognitive performance by Y-maze test, and brain glucose metabolism by 18F-2-fluoro-2-deoxyglucose positron emission tomography/computed tomography ([18F]FDG PET/CT) were assessed. At the end of the dieting period, 8-month-old HFD mice were heavier than ND mice (51.8 ± 2.6 vs 41.7 ± 1.5, p 

Table 1 Regional cerebral glucose metabolism as determined by [18F]FDG-PET.

Impact on the metabolome

We quantified 77 and 99 metabolites in all serum and faecal (caecum and colon) samples, respectively (Tables S1-2), and built chemometric models for detecting global and individual metabolic trends (Fig. 1, Figs S2–5). Serum analysis by Principal Component Analysis (PCA) showed distinct clusters in 3xtg compared to control mice, indicating genotype-induced changes in metabolome (Fig. 1a). More specifically, we observed a clear separation along the first Principal Component (PC1, 46.54% of variance) between 3xtg and control mice, regardless of diet, suggesting a stronger metabolic impact of 3xtg genotype than HFD. Serum metabolites explaining this separation pattern are given in the PCA loading plot (Fig. S2). Metabolites and samples segregation in response to diet and disease were attenuated at faecal level, where a greater effect of HFD compared to 3xtg genotype was observed in the determination of samples segregation (Fig. 1b,c). To further explore the metabolomic relationships between 3xtg genotype and diet, we built separated partial least-squares discriminant analysis (PLS-DA) models discriminating 3xtg and control mice, under ND or HFD (Figs S3–5). These plots highlighted a strong separation between 3xtg and controls in both serum and fecal extracts, regardless of diet, and revealed greater HFD-mediated separation in 3xtg compared to control mice. For serum and faecal metabolites, we quantified relative fold-changes in all possible binary comparisons to evaluate the extent of metabolomic variations in the different groups (Figs 2–3), thus observing that both 3xtg genotype and HFD affect some metabolites in the same direction, but to a different extent, prevailing for the 3xtg background.

Figure 1

The metabolome of 3xtg mice differs more from control mice than mice under HFD. PCA score plots based on the NMR metabolomic profile of serum (a), and fecal extracts from colon (b) and caecum (c). Each symbol represents a single sample shaped and coloured according to the respective group.

Figure 2

The serum metabolome of 3xtg mice differs from control mice, and HFD exerts additive effects. Relative fold changes for control and 3xtg mice under normal or HFD on NMR integrals for metabolic regions with higher VIP scores in PLS-DA serum analysis. Bars represent fold-change in metabolite content for each comparison (increased content: positive bars; decreased content: negative bars). ND, ND-fed control mice; 3xtg, ND-fed 3xtg mice; HFD, HFD-fed control mice; 3xtgHFD, HFD-fed 3xtg mice. *p 

Figure 3

The fecal metabolome of 3xtg mice is altered with respect to control mice. Relative fold changes for in test groups (HFD, and 3xtg mice under ND or HFD) versus control ND mice, derived from NMR integrals of metabolic regions with higher VIP scores in the PLS-DA analysis in colon and caecum faecal extracts. Bars represent fold-change in metabolite content in each comparison (increased content: positive bars; decreased content: negative bars). ND, ND-fed control mice; 3xtg, ND-fed 3xtg mice; HFD, HFD-fed control mice; 3xtgHFD, HFD-fed 3xtg mice. *p 

Impact of diet

A significant increase in inflammatory marker (glycoprotein acetylation biomarker, GlycA), lactate and albumin lysil serum levels was observed in HFD compared to ND mice (Fig. 2). At faecal level, group comparisons did not show regional differences between colon and caecum tracts (Fig. 3) and a large interindividual variation in metabolic profiles seemed to attenuate group effects. A tendency towards higher ribose, threonine, threonine-ornithyne, glycerol and fatty acid levels, and lower proline, acetate and ethanol amounts was observed in HFD compared to control mice (especially in colon), but only colon ribose levels achieved statistical significance.

Impact of disease

The 3xtg genotype induced strong metabolic changes in host (serum) metabolome, but minor changes in host-microbial co-metabolome (faeces) (Figs 2–3). Differences in serum metabolome of 3xtg compared to control mice, regardless of diet, included elevation in inflammatory markers (GlycA), lactate, albumin lysil, leucine and isoleucine, and deficient levels of unsaturated fatty acids, very low density lipoprotein (VLDL), and choline. In addition, higher levels of trimethylamine (TMA) and a tendency to trimethylamine N-oxide (TMAO) elevation occurred in 3xtg mice compared to controls. At faecal level, 3xtg mice were characterized by a pronounced increase in ornithine-tyrosine levels compared to controls, in both colon and caecum.

Combined impact of diet and disease

The combined effect of HFD and 3xtg genotype amplified the changes observed in each component alone. As expected, the comparison of 3xtg genotype under HFD vs control animals under ND showed the largest number of statically significant differences and fold changes (Fig. 2). Similar to 3xtg mice, high aminoacids (leucine, isoleucine, glycine) and albumin lysil serum levels were detected compared to controls. The amount of systemic inflammatory markers (GlycA) was elevated, but not higher than in 3xtg under ND, due to already high inflammatory levels in this group. Instead, a further elevation in glucose, lactate, glycerol and glutamate levels was observed compared to controls and to 3xtg under ND. Likewise, fatty acid (but not VLDL) deficiencies were also amplified in 3xtg-HFD, compared to each condition (3xtg or HFD) alone. Finally, TMA and TMAO levels were progressively greater from ND to 3xtg and to 3xtg-HFD. Important, the combination of HFD and 3xtg genotype affected ketone bodies pathways in an alternative way compared to the above listed metabolites, due to the opposing effects of HFD, lowering and 3xtg, elevating acetoacetate and 3-hydroxibutyrate. The net effect resulted in a significant reduction of ketone bodies in 3xtg-HFD mice compared to controls. In faeces, ornithine-threonine and acetoin levels were significantly higher in the colon of 3xtg-HFD compared to ND mice (Fig. 3).

Impact on the microbiome

A total of 2,895,848 quality-filtered 16 S rDNA sequences were obtained from caecum and colon faecal extracts (52,651.8 ± 34,735.2 seq/sample) and clustered into 23,874 Operational Taxonomic Units (OTUs).

Impact of location

Analysis of Similarity (ANOSIM) test showed significant differences at OTU level between colon and caecum microbiome (R = 0.184, p = 0.039). In group analysis, they were significant in ND (R = 0.184, p = 0.043) but not in 3xtg mice (R = 0.007, p = 0.412).

Higher Bacteroidetes (27.57% vs 13.0%, p = 0.0016, p Bonferroni correction = 0.0096, false discovery rate, FDR = 0.0075) and lower Firmicutes (64.41% vs 81.97%, p = 0.0025, p Bonferroni correction = 0.015, FDR = 0.0075) were present in colon compared to caecum samples in ND mice. However, no differences at phylum level were observed between colon and caecum samples in the 3xtg group.

At family level, caecum samples showed higher levels of Lachnospiraceae (18.02% vs 10.28%, p = 0.036) and lower levels of S24.7 (13% vs 27.56%, p = 0.0016) than colon samples in ND. Conversely, only Enterococcaceae family was significantly different between colon and caecum samples in 3xtg mice (2.31% vs 0.09%, respectively, p = 0.012). In linear discriminant analysis effect size (LEfSe) tests, S24.7 family was significantly enriched in ND colon compared to caecum samples (linear discriminant analysis, LDA, score > 5, p 

Higher Shannon diversity index (p = 0.022), but no differences in microbial richness measured by Chao1 index (p = 0.512), was observed in caecum compared to colon samples in ND. Conversely, no alpha diversity (Shannon and Chao1 indexes), was reported between colon and caecum samples in 3xtg mice (Fig. S6).

Impact of diet

Redundancy Analysis (RDA) showed significant differences in microbial composition between HFD and ND mice, in colon (p = 0.006) and caecum samples (p = 0.001) (Fig. 4b), as confirmed by ANOSIM tests (colon: R = 0.209, p = 0.023; caecum: R = 0.213, p = 0.032).

Figure 4

HFD and 3xtg determine a dramatic rightward shift in the overall composition of gut microbiota. RDA plot at OTU level between animal groups and diet intervention. The analysis revealed a clear-cut separation between HFD and ND mice (a,b), and between 3xtg and control mice (c,d). Each symbol represents a single sample shaped and coloured according to the respective group.

At family level, lower abundance of Bifidobacteriaceae (p = 0.029), Lactobacillaceae (p = 0.071) and S24.7 (p = 0.055) and higher abundance of Rikenellaceae (p = 0.0013, FDR = 0.023), Mogibacteriaceae. (p = 0.0074, FDR = 0.074), Lachnospiraceae (p = 0.012, FDR = 0.072) Enterococcaceae (p = 0.022, FDR = 0.09), were detected in HFD compared to ND colon samples. Similar results were found in caecum samples, where lower levels of Bifidobacteriaceae (p = 0.063) and S24.7 (p = 0.029), and higher abundance of Rikenellaceae (p = 0.00077, FDR = 0.013), Enterococcaceae (p = 0.0088, FDR = 0.052) and Staphylococcaceae (p = 0.0092, FDR = 0.052) were observed in the HFD compared to ND group.

At genus level, lower abundance of Unclassified Peptococcaceae (p = 0.013), Bifidobacterium and Lactobacillus spp (p = 0.029 and 0.072, respectively) and higher abundance of Unclassified Rikenellaceae (p = 0.0013, FDR = 0.042), Unclassified Mogibacteriaceae (p = 0.007), and Unclassified Enterococcaceae (p = 0.022) were observed in HFD compared to ND colon samples.

DESeq 2 test confirmed diet-based differences between groups, specifically lower abundance of Bifidobacterium and Lactobacillus were found in HFD than ND mice (p 

At OTU level, we found significant differences in 37 OTU belonging mostly to Firmicutes, Bacteroidetes and Actinobacteria phyla (Table S3) in colon, and also 37 OTU in caecum when ND and HFD groups were compared (Table S4). No differences in Shannon diversity index and lower Chao1 richness index (p = 0.086, colon; p = 0.0223, caecum) were observed in HFD compared to ND groups (Fig. S6). LEfSe tests revealed that at family level in ND mice, S24.7 was significantly enriched in colon, while Bifidobacteriaceae were highly abundant in caecum (Fig. 5a). In HFD mice, Enterococcaceae and Rikenellaceae were representative of colon, whereas Lachnospiraceae and Staphylococcaecae were abundant in caecum samples. Venn diagram showed a core of 13 families shared between diet groups and locations (Fig. 5c). Bifidobacteriaceae family was exclusively present in ND colon and caecum. Christensenellaceae, Mogibacteriaceae and Rikenellaceae were present in HFD colon and caecum. Staphylococcaceae and Enterococcaceae, were shared between ND colon and HFD colon and caecum.

Figure 5

HFD consumption, 3xtg background and location significantly influence the composition of the gut microbiome. LEfSe test-identified LDA scores showed the significant bacterial difference due to the effect of diet and location (a, ND and HFD, colon and caecum), and disease and location (b, ND and 3xtg, colon and caecum). Venn diagram showed shared families across groups (c, diet and location; d, disease and location).

Impact of disease

RDA showed differences in the microbial composition of colon (p = 0.001) and caecum samples (p = 0.001), between ND and 3xtg (Fig. 4c,d), as confirmed by ANOSIM tests (colon: R = 0.696, p = 0.001; caecum: R = 0.326, p = 0.001).

At family level, lower abundance of S24.7 (p = 0.018) and higher abundance of Enterococcaceae (p = 0.0077, FDR = 0.081) and Turicibacteraceae (p = 0.0081, FDR = 0.081) were detected in 3xtg compared to ND colon samples. Similar results were found in caecum samples, in which lower levels of S24.7 (p = 0.029) and Bifidobacteriaceae (p = 0.045) were detected in 3xtg compared to ND mice. At genus level, lower abundance of Bifidobacterium spp (p = 0.045) and rc4.4 (p = 0.042), and higher abundance of Roseburia (p = 0.049) were detected in 3xtg compared to ND caecum. LEfSe analyses confirmed that, at family level S24.7 and Lachnospiraceae was significantly enriched in ND colon and caecum, respectively, while Enterococcaceae and Turicibacteriaceae were associated to 3xtg colon samples (Fig. 5b). Families belonging to Ruminococcaceae and Anaeroplasmataceae were representative of 3xtg caecum samples. At genus level, Turicibacter and Staphylococcus were significantly enriched in 3xtg colon samples, whereas Anaeroplasma was associated to 3xtg caecum samples. Most significant findings in pooled samples (caecum and colon) are shown in Supplementary Figure 7.

Venn diagram showed a core of 13 families shared between 3xtg and control mice in both locations (Fig. 5d). Turicibacteraceae family was associated to 3xtg colon, while Christensenellaceae to 3xtg caecum. Bifidobacteriaceae was related to ND colon and caecum, whereas Anaeroplasmataceae was linked to 3xtg colon and caecum. Finally, Staphylococcaceae and Enterococcaceae families were shared between ND colon and 3xtg colon and caecum.

At OTU level, we found significant differences in 48 OTU (Table S5) in colon and 62 OTU in caecum (Table S6) when ND and 3xtg groups were compared. Alpha diversity analysis showed no group differences in Shannon diversity index, but lower Chao1 richness index (p = 0.0195) in 3xtg than ND colon samples (Fig. S6). No significant difference was found in caecum, though the tendency was the same.

Impact of combined location, diet and disease

Principal Coordinates Analysis (PCoA) of the samples clearly separated 3xtg from control animals, in both weighted (Fig. 6a) and unweighed (Fig. 6b) plots (PC1 accounting for 8.7% and 30.94% of the variance in unweighted and weighted UniFrac analysis, respectively), indicating that the microbiota between groups is compositionally distinct. In addition, the different diet groups are well defined and clustered. ANOSIM tests confirmed the overall significant difference between groups and intestinal locations (R = 0.477, p = 0.001). Similar results were found when microbiome of ND, HDF, 3xtg and 3xtg-HFD groups in each intestinal location (colon: R = 0.53, p = 0.001; caecum: R = 0.589, p = 0.001) were compared.

Figure 6

Isolated and combined HFD plus 3xtg background determine a clear-cut separation in microbiota composition. Beta-diversity PCA using weighted (a) and unweighted-UniFrac distances (b). RDA plot at OTU level between groups (c, colon; d, caecum). Each symbol represents a single sample shaped and coloured according to the respective group.

Together with UniFrac data, ADONIS (permutational manova, PERMANOVA) showed significant differences at OTU level between diet groups and sample location (p = 0.0006). In addition, RDA highlighted significant differences in the microbial composition between the diets and disease status in colon (p = 0.001) and caecum (p = 0.001) (Fig. 6c,d).

Alpha diversity analysis showed no differences in Shannon diversity index, but higher Chao1 richness index (p 

Microbiome and metabolome associations

Heatmap generated to examine correlations between colon microbiota and serum metabolome in ND and 3xtg mice (Fig. 7) showed three main clusters: I) lactate, malate, acetylcarnitine, glutamate, succinylacetone, glycerol, citrate, etc; II) proline, methylalanine, methionine, creatine phosphate, leucine, isoleucine, glutamine, alanine, methylsuccinate, etc; III) cholesterol, lipoproteins and fatty acids. Bifidobacterium and Unclassified s24.7 levels, which were shown to be defective 3xtg mice, were related to high lactate, malate and aminoacid levels, and to low cholesterol, low density lipoprotein (LDL) and fatty acid amounts. Coprobacillus, Dorea and other bacterial strains were also negatively related to cholesterol and fatty acid levels. Abundance of rc4.4, which was shown to be deficient in 3xtg mice, was associated negatively with circulating aminoacids and positively with cholesterol and fatty acids levels. Both Unclassified s24.7 and rc4.4 were negatively correlated with the amount of circulating choline compounds and inflammatory marker GlycA. Conversely, Staphylococcus and Unclassified Enterococcus (shown to be overabundant in 3xtg) were positively associated with lactate, malate, acetylcarnitine, succinylacetone, pyruvate and aminoacid levels. Overall, bacteria correlating with an adverse metabolic profile, as defined by 3xtg characteristics, were primarily Enterococcaceae, Staphylococcus, followed by Roseburia, Coprobacillus, Dorea, Enterococcus, Christenellaceae. The ones relating to the reverse healthier profile were s24.7, rc4.4, Bifidobacterium, Dehalobacterium, Peptococcaceae, Acinetobacter.

Figure 7

Associations between colon microbiome and serum metabolome. Heatmap shows associations between metabolite profile and relative abundance of specific bacterial families and genera in ND and 3xtg groups. Red to blue scale: positive to negative associations. Pearson’s correlations were employed in agreement with data distribution, verified by Shapiro-Wilk test. *p 

Microbiome and brain associations

Heatmap generated to examine correlations between caecum-colon microbiota and cerebral parameters (Alternation triplets, n and %, and glucose metabolism) across groups (Fig. 8) showed that cerebral hypometabolism was associated with higher abundance of Firmicutes phylum, Anaeroplasmataceae and Erysipelotrichaceae families, and Coprobacillus, Clostridium, Anaeroplasma and Roseburia genera, and lower abundance of Bacteroidetes phylum, Peptococcaceae, Rikenellaceae, Dehalobacteriaceae and s24.7 families, and rc4.4, Dehalobacterium, Unclassified Coriobacteriacea, Unclassified Rikenellaceae, Unclassified s24.7.

Figure 8

Associations between gut microbiome and cerebral parameters. Heatmap generated from Spearman correlation analysis shows associations between brain parameters (cognitive function and glucose metabolism) and relative abundance of specific bacterial phyla, families and genera across groups, in caecum and colon. Red to blue scale: positive to negative associations. Spearman’s correlations were employed in agreement with data distribution, verified by Shapiro-Wilk test. *p 

Furthermore, we observed that higher abundance of Bifidobacteriaceae and Erysipelotrichaceae families, and Bifidobacterium and rc4.4 genera were positively correlated with cognitive function, whereas negative associations occurred with Coprobacillus, Dorea, Roseburia, and Unclassified Peptococcaceae.