Multiomics signatures of type 1 diabetes with and without albuminuria

Abstract

Aims/hypothesis: To identify novel pathophysiological signatures of longstanding type 1 diabetes (T1D) with and without albuminuria we investigated the gut microbiome and blood metabolome in individuals with T1D and healthy controls (HC). We also mapped the functional underpinnings of the microbiome in relation to its metabolic role.

Methods: One hundred and sixty-one individuals with T1D and 50 HC were recruited at the Steno Diabetes Center Copenhagen, Denmark. T1D cases were stratified based on levels of albuminuria into normoalbuminuria, moderate and severely increased albuminuria. Shotgun sequencing of bacterial and viral microbiome in stool samples and circulating metabolites and lipids profiling using mass spectroscopy in plasma of all participants were performed. Functional mapping of microbiome into Gut Metabolic Modules (GMMs) was done using EggNog and KEGG databases. Multiomics integration was performed using MOFA tool.

Results: Measures of the gut bacterial beta diversity differed significantly between T1D and HC, either with moderately or severely increased albuminuria. Taxonomic analyses of the bacterial microbiota identified 51 species that differed in absolute abundance between T1D and HC (17 higher, 34 lower). Stratified on levels of albuminuria, 10 species were differentially abundant for the moderately increased albuminuria group, 63 for the severely increased albuminuria group while 25 were common and differentially abundant both for moderately and severely increased albuminuria groups, when compared to HC. Functional characterization of the bacteriome identified 23 differentially enriched GMMs between T1D and HC, mostly involved in sugar and amino acid metabolism. No differences in relation to albuminuria stratification was observed. Twenty-five phages were differentially abundant between T1D and HC groups. Six of these varied with albuminuria status. Plasma metabolomics indicated differences in the steroidogenesis and sugar metabolism and circulating sphingolipids in T1D individuals. We identified association between sphingolipid levels and Bacteroides sp. abundances. MOFA revealed reduced interactions between gut microbiome and plasma metabolome profiles albeit polar metabolite, lipids and bacteriome compositions contributed to the variance in albuminuria levels among T1D individuals.

Conclusions: Individuals with T1D and progressive kidney disease stratified on levels of albuminuria show distinct signatures in their gut microbiome and blood metabolome.

Keywords: albuminuria; lipidomics; metabolomics; microbiome; multiomics; phageome; type 1 diabetes.

Figure 1

Study design and analytical overview. The study cohort comprised 50 healthy and 161 individuals with T1D, who stratified on albuminuria levels: normo-albuminuria (n=50, albuminuria levels =<3.39 mg/mmol), moderately increased albuminuria (n=50, albuminuria levels =3.39–33.79 mg/mmol) and severely increased albuminuria (n=61, albuminuria levels =≥33.90 mg/mmol). For each study participant 1) plasma samples were collected to perform non-targeted metabolomics analysis, including both polar metabolites and lipids, and 2) faecal samples collected for metagenomics analysis. For the metagenomic samples, differences between the clinical groups at both taxonomical and functional level were assessed. Lipidomics data was clustered into highly correlated lipid clusters in order to reduce dimensionality. Finally, different omics data types were integrated with the Multi Omics Factor Analysis (MOFA+) tool and metabolite origin screened to identify bacterial- and host-related metabolites through Least Shrinkage Selector Operator (LASSO) regression models. QMP, Quantitative Microbial Profiles; GMM, Gut Metabolic Modules.

Figure 2

Differential bacterial abundance at metagenomic species (MGS) level, between T1D and healthy individuals. (A) Volcano plot showing difference in absolute abundance of metagenomic species (MGS) between T1D individuals (n=161) and the healthy controls (n=50). X-axis indicates Cliff’s Delta effect size; Y-axis represents FDR-corrected (negative log) p–values. MGS that associated positively with T1D have been depicted towards increasing direction of effects (right). MGS circle color depicts the corresponding annotated phylum. Circle size corresponds to the number of individuals in the cohort where the specific MGS was found. Transparency of the circle corresponds to the average relative abundance in which each MGS is found within participants. (B) Significant contrasts in MGS abundance between T1D and healthy controls depicted with the bar length corresponding to Cliff’s Delta effect size: green for higher and red for lower MGS abundances within T1D individuals. (C) Differential absolute MGS abundance between T1D subgroups stratified on levels of albuminuria. Individual distribution of the log10 transformed QMP counts (absolute abundance) is depicted for each MGS in violin and dot plots. Global distribution of the MGS counts for each T1D subgroup is depicted with a boxplot, indicating median value of the distribution with a horizontal line, first and third quartile with the limits of the white rectangle and the upper and lower limits of the distribution with vertical bars. Significance for pairwise comparison between different study groups is indicated with p-value. (D) Significant correlations between T1D duration in years and absolute abundance of MGS (QMP counts) are depicted as a heat map. Positive correlations are shown in red and inverse correlations in blue.

Figure 3

Contrasted relative abundance of bacteriophages in T1D and healthy control individuals. (A) In the left panel, a volcano plot compares the relative phage abundances between T1D and healthy individuals. The X-axis represents Cliff’s Delta effect size while the Y-axis represents the association threshold for the comparison, (negative log) FDR-adjusted p-value. Significantly contrasted phages (above the red dotted line) are annotated. Dot size is relative to the global prevalence of phage in the present study sample, while their transparency is relative to their mean abundance in the cohort. In the right panel, significantly contrasted phages are shown as bar plots corresponding to Cliff’s Delta effect sizes. Phages with significantly lower abundance in T1D are indicated as green bars while red bars represent phages with significantly higher abundance in T1D. (B) Distribution of abundance of selected phages when comparing healthy controls with the three groups of T1D stratified by albuminuria. For each phage, the CLR-transformed abundance distribution is represented in differently colored dots and boxplots showing the global group distribution. Multi-group comparison computed with Kruskal-Wallis test is included in each plot, as well as the pairwise comparisons between all study groups, performed with Wilcoxon test. (C) Comparison between the global composition of the phageome and the bacteriome, performed with Procrustes test. Principle coordinates analysis (PCoA) shows the disposition of individuals in the corresponding scores plot, for both bacteriome (circles) and phageome (triangles) based distance matrix. An arrow has been drawn connecting the same individuals. Individuals have been colored depending on their clinical group. Correlation and significance are indicated in the bottom right corner of the plot.

Figure 4

Profiles of plasma polar metabolites and lipids in T1D and healthy control individuals. (A) Scores plot resulting from the partial least square discriminant analysis (PLS-DA) demonstrating polar metabolites that contribute highly (Variable influence on project (VIP) > 1) to a T1D signature. Colors depict clinical study groups and ellipses demarcate the spread of individuals within each group. Explained variance for first two principal components (PC1 and PC2) is indicated on the respective X- and Y- axis. Global distribution of the participants from each clinical group is shown as a density plot in the secondary Y-axis. (B) Univariate analysis (volcano plot) depicting differential circulating polar metabolite abundances between healthy and T1D individuals. Significantly contrasted (FDR < 10%) polar metabolites are depicted in red, with increasing effects in direction of controls (right). (C) Functional metabolic pathway identification comparing T1D and healthy controls and depicting enriched metabolic pathways: annotations of top polar metabolites using the human metabolome database (HMDB). Bars have been ordered and colored by the enrichment p-value score. (D) Univariate analysis (volcano plot) depicting differential abundances of serum lipid cluster between healthy and T1D individuals. Significantly contrasted (FDR < 10%) lipid clusters are colored in red with increasing effects in direction of controls (right). (E) Triglyceride (TG) species associated with different T1D albuminuria groupings (compared to healthy controls) based on the number of carbon atoms (Y-axis) and the number of double bonds (X.axis) present in the chemical structure. Blue-to-red gradient heatmap suggests directionality of the association (negative to positive). Significant associations are indicated in each cell with the following symbols: + p<0.1, X p<0.05, *p<0.01.

Figure 5

Multi-Omics Factor Analysis (MOFA). Results from multi-omics data integration after combining multiple data types: metagenomics data (taxonomical and functional), plasma metabolomics data (polar metabolites and clustered lipids) and biochemistry data using the MOFA+ tool. (A) Global explained variance for each of the data types (included in the integration step) has been represented as a bar chart. (B) Composition of 15 individual factors generated with the proportion of explained variance by each of the data types, displayed as a white-to-blue gradient in increasing order. (C) Distribution of the eigenvalues obtained for each of the factors for the combined dataset. Within each of those factors the individual, eigenvalues are represented as dot plots. Different colors in the dot plot depict different clinical study groups: orange for healthy controls, dark gray for T1D with normo-albuminuria, light gray for T1D with moderately increased albuminuria and black for T1D with severely increased albuminuria. Each of the study groups are also labelled in the bottom horizontal axis.

Figure 6

Differentially abundant circulating metabolites and microbiome origin and functionality assessment. Summary of the circulating metabolites and lipids data integrated with gut microbiome origin and functionality in the T1D vs healthy controls comparison. The two heatmaps display significantly contrasted (FDR < 10%) polar metabolites (left) and lipid clusters (right) between T1D and healthy individuals. Column 1 (red green of each of the two heatmaps shows the Cliff’s Delta effect size for the clinical group comparison, ordered from higher to lower abundance in T1D individuals and colored from red (higher) to green (lower) gradient depending on its relative abundance in T1D individuals. Column 2 (black-white) shows the usefulness of the metabolite for discrimination between T1D and healthy individuals based on Area Under the Curve (AUC) analyses. The black-to-white-to-black gradient for AUC depicts assessment ability with 50% being uninformative AUC, 0% being the limit for the identification of healthy individuals and 100% being the limit for identification of T1D individuals. Column 3 (purple white) depicts relationship between the metabolites and bacterial QMP counts (absolute MGS abundance). Cells are colored if there is any significant association with metabolite levels (based on LASSO modelling) and the white-to-purple color gradient depicts the explained variance by the microbiome. Column 4 (purple white) depicts the relationship between the circulating metabolites and functional abundance of GMMs. Cells are colored if there is any significant association with metabolite levels (based on LASSO modelling) and the white-to-purple color gradient depicts the explained variance by the microbiome metabolic potential (or functionality).