Gut Microbiome Composition Is Predictive of Incident Type 2 Diabetes in a Population Cohort of 5,572 Finnish Adults

Abstract

Objective: To examine the previously unknown long-term association between gut microbiome composition and incident type 2 diabetes in a representative population cohort.

Research design and methods: We collected fecal samples from 5,572 Finns (mean age 48.7 years; 54.1% women) in 2002 who were followed up for incident type 2 diabetes until 31 December 2017. The samples were sequenced using shotgun metagenomics. We examined associations between gut microbiome composition and incident diabetes using multivariable-adjusted Cox regression models. We first used the eastern Finland subpopulation to obtain initial findings and validated these in the western Finland subpopulation.

Results: Altogether, 432 cases of incident diabetes occurred over the median follow-up of 15.8 years. We detected four species and two clusters consistently associated with incident diabetes in the validation models. These four species were Clostridium citroniae (hazard ratio [HR] 1.21; 95% CI 1.04-1.42), C. bolteae (HR 1.20; 95% CI 1.04-1.39), Tyzzerella nexilis (HR 1.17; 95% CI 1.01-1.36), and Ruminococcus gnavus (HR 1.17; 95% CI 1.01-1.36). The positively associated clusters, cluster 1 (HR 1.18; 95% CI 1.02-1.38) and cluster 5 (HR 1.18; 95% CI 1.02-1.36), mostly consisted of these same species.

Conclusions: We observed robust species-level taxonomic features predictive of incident type 2 diabetes over long-term follow-up. These findings build on and extend previous mainly cross-sectional evidence and further support links between dietary habits, metabolic diseases, and type 2 diabetes that are modulated by the gut microbiome. The gut microbiome can potentially be used to improve disease prediction and uncover novel therapeutic targets for diabetes.

Figure 1

Proportionality between bacterial taxa significantly associated with incident type 2 diabetes in eastern Finland and western Finland. Annotated HRs and clustering of the taxa were calculated separately in both data groups. Because of identical cluster membership of the taxa, the cluster numbers and their annotations are harmonized.

Figure 2

Comparison of HRs between models for the selected features in eastern and western Finland data. Features with significant associations in the validation (western Finland) data are indicated in bold, and the taxon colors show their membership in a cluster. The information in this figure can be found in numeric format in Supplementary Table 1.

Figure 3

Kaplan-Meier curves for features with significant effect sizes in both data sets, displaying diabetes-free survival times of participants in western Finland. Curves are separated by ranges between quartiles of relative abundance of each feature. Distribution of the participants with the same relative abundance ranges is included as an inlay for each of the features.