A gut microbiome-kidney-heart axis predictive of future cardiovascular diseases

Figures

• 

  Fig. 1. Overview of the study. Using integrative analyses of multi-omics data combined with extensive confounder-control, Mendelian Randomization (MR) and predictive analyses using multiple populations, we identify a gut microbiome-kidney-heart axis early during the dysmetabolic phase that is predictive of future CVD. We observe associations among microbial metabolism of phenylalanine and tyrosine with kidney and heart function, exemplified by eGFR and circulating proANP levels, respectively, in metabolically healthy controls of the MetaCardis study (i.e., free of metabolic syndrome, obesity, type 2 diabetes and IHD; n = 275; Figs. 2, 3), which exhibit a depletion pattern in CMD participants of the MetaCardis study (i.e., individuals with metabolic syndrome, obesity, diabetes and IHD; n = 1602; Fig. 4). Bi-directional MR analyses using genetic instruments derived from publicly available data from the CLSA, CKDGen and deCode studies for metabolites, eGFR and natriuretic peptide A, (NPPA) respectively, reveal a potential crosstalk between gut microbiome and host kidney function (Fig. 5). Finally, baseline levels of microbial metabolites and host co-metabolites derived from phenylalanine…

• 

  Fig. 2. A gut microbial functional signature for circulating proANP levels in healthy individuals. a Principal coordinates analysis of inter-individual differences in microbiome profiles of metabolically healthy individuals of the MetaCardis study (MGS level Bray-Curtis dissimilarity index), coloured by country of recruitment (i.e., France, Germany and Denmark). Arrows and labels represent key host variables explaining variation in the gut microbiome composition (n = 200; MGS-FDRunivariateADONIS < 0.1; Supplementary Data 2). b Scatter plot showing positive association between circulating proANP levels and faecal microbial load in the MetaCardis healthy individuals (n = 255) tested using two-sided Spearman’s rank correlation adjusted for demographics. Data show linear best-fit line ± 95% confidence intervals segregated for country of recruitment for visualization. Scatter plots showing associations among circulating proANP levels and absolute gut microbial functions at the level of (c) GMM and (d) KEGG pathways in the MetaCardis healthy individuals when adjusted for faecal microbial load (y-axes) in addition to demographics (x-axes) (n = 255; multiple linear regression adjusted for…

• 

  Fig. 3. Microbial metabolism of phenylalanine and tyrosine emerges as a key pathway in healthy individuals. a Heatmap representing associations among metabolome and clinical phenotypes restricted to the metabolites exhibiting FDR < 0.1 for circulating proANP levels in the MetaCardis healthy individuals (n = 254–274; exact sample sizes are given as Source data file; multiple linear regression adjusted for demographics; multiple testing correction using BH criteria, + represents FDR < 0.1). b 3-D plot representing top 50 robustly predicted metabolites (goodness-of-prediction, q2) from multivariate CA-PLS models using gut microbiome taxonomic and functional features (i.e., MGS, GMM and KEGG pathways) as predictors and demographics (i.e., age, sex and country) as covariates in the MetaCardis healthy individuals (Supplementary Data 10) where point size represents maximum q2 of all models and colour green represents the metabolites derived from phenylalanine and tyrosine metabolism versus the rest as brown. Key metabolites derived from microbial metabolism of phenylalanine and tyrosine labelled here were among the top metabolites that were explained and predicted by these models, which were…

• 

  Fig. 4. Confounder-controlled gut microbial taxonomic and functional correlates of key phenylalanine and tyrosine metabolites in healthy individuals exhibit a depletion pattern in individuals with CMD. Chord diagrams revealing significant inter-relationships among microbiome-metabolite-kidney-heart axis in (a) healthy individuals (n = 248–273) and (b) those with CMD (n = 1462–1575) (exact sample sizes are given in Supplementary Data 13) in the MetaCardis population. Inter-relationships are segregated for positive (left) and negative associations (right). Gut microbiome features included higher ecological, compositional (i.e., genus and MGS) and functional (i.e., GMM and KEGG pathways) aspects, which served as predictors in models explaining key metabolites and kidney-heart variables (i.e., proANP, eGFR, creatinine), whereas metabolites only served as predictors in models explaining kidney-heart variables. Inner rings of the chord diagrams correspond to the colour of the response variable (i.e., predicted metabolite or clinical variable). Connections of the chord diagrams represent significant inter-relationships coloured by the predictor variable where gut microbial features were grouped…

• 

  Fig. 5. Mediation and Mendelian Randomization analyses reveal the gut microbiome-kidney-heart axis. Alluvial plots showing (a) ACME or the indirect effects of the gut microbiome features on eGFR via key microbial metabolites as mediators and (b) ADE or the direct effects of the gut microbiome features on circulating proANP levels derived from demographics-adjusted mediation analyses using two-sided significance testing in the MetaCardis healthy individuals (n = 230–271, exact sample sizes in Supplementary Datas 14 and 15). While the indirect effects of each microbiome-metabolite association are coloured for the overall ACME on eGFR, microbiome-mediator associations were separated into positive or negative based on their effect sizes shown in Supplementary Data 13. Bi-directional relationships among key microbial metabolites, eGFR and proANP were next investigated using the two-sample MR testing. Forest plots show IVW estimates ± 95% confidence intervals for analyses involving (c) key metabolites as exposures and eGFR as outcome (Supplementary Data 17), d eGFR as exposure and metabolites as outcomes (Supplementary Data 18), e key metabolites as exposures and NPPA as outcome (Supplementary…

• 

  Fig. 6. Key metabolites derived from phenylalanine and tyrosine metabolism associate with incident CVD in a Canadian population. a Forest plot showing associations among key metabolites measured at baseline and eGFR in the CLSA study population at two time points i.e., baseline and FU1 in 8669 individuals. Data are shown as β estimates ± 95% CIs derived from linear mixed models (LMM) adjusted for age, sex and time (days since baseline), using metabolites as exposures (log-transformed and standardized) and an interaction term across two time points. Filled squares represent significance at FDR < 0.1 for ten tested metabolites. b Forest plots representing associations among key metabolites measured at baseline and disease incidence during the FU1 period in the CLSA study and mortality data available till FU2 period. Data are shown as β estimates ± 95% CIs derived from age-, sex- and baseline eGFR- adjusted Cox proportional hazards models with interval-censoring (and right-censoring for mortality data) using baseline metabolite levels (log-transformed and standardized) as exposures and each disease incidence at FU1 as a binary variable (cases at baseline were excluded) considering time…