Progressive Shifts in the Gut Microbiome Reflect Prediabetes and Diabetes Development in a Treatment-Naive Mexican Cohort

Abstract

Type 2 diabetes (T2D) is a global epidemic that affects more than 8% of the world's population and is a leading cause of death in Mexico. Diet and lifestyle are known to contribute to the onset of T2D. However, the role of the gut microbiome in T2D progression remains uncertain. Associations between microbiome composition and diabetes are confounded by medication use, diet, and obesity. Here we present data on a treatment-naive cohort of 405 Mexican individuals across varying stages of T2D severity. Associations between gut bacteria and more than 200 clinical variables revealed a defined set of bacterial genera that were consistent biomarkers of T2D prevalence and risk. Specifically, gradual increases in blood glucose levels, beta cell dysfunction, and the accumulation of measured T2D risk factors were correlated with the relative abundances of four bacterial genera. In a cohort of 25 individuals, T2D treatment-predominantly metformin-reliably returned the microbiome to the normoglycemic community state. Deep clinical characterization allowed us to broadly control for confounding variables, indicating that these microbiome patterns were independent of common T2D comorbidities, like obesity or cardiovascular disease. Our work provides the first solid evidence for a direct link between the gut microbiome and T2D in a critically high-risk population. In particular, we show that increased T2D risk is reflected in gradual changes in the gut microbiome. Whether or not these T2D-associated changes in the gut contribute to the etiology of T2D or its comorbidities remains to be seen.

Keywords: Mexico; deep phenotyping; metformin; microbiome; type 2 diabetes.

Figure 1

Study design. (A) 405 Individuals were recruited from Guanajuato state and classified into normoglycemic (NG), impaired fasting glucose (IFG), impaired glucose tolerance (IGT), impaired fasting glucose, and impaired glucose tolerance (IFG+IGT), and type 2 diabetes (T2D). Twenty-five individuals under treatment for a previous T2D diagnosis or with previous metformin history were added as controls (T2D treated). (B) Correlations (Spearman ρ) between bacterial genera in the study (intra-microbiome) are shown in the left correlation matrix whereas correlations between clinical variables are shown in the right correlation matrix. (C) Blood glucose curves for all individuals in the study colored by classification. (D) Receiver-Operator curves for predictions from a Random Forest model. Individual cross-validation curves are shown along with the mean trend and standard deviations.

Figure 2

Associations between the microbiome and phenotype. (A) The number of significant associations between the microbiome and clinical variables grouped by category (FDR corrected LRT p < 0.05). The positive test rate denotes the significant tests/total tests for the category. (B) Significant tests per genus (FDR corrected LRT p < 0.05). Color denotes the category of the clinical variables the genus associates with. (C) Significant associations (FDR corrected LRT p < 0.05) between bacterial genera and alpha diversity (Shannon). Points denote the log fold change (DESeq2 regression coefficient) of a genus when the diversity increases by one standard deviation. Error bars denote the standard error of the coefficient. Fill color denotes the mean of normalized reads across all samples.

Figure 3

Associations between microbiome composition and disease progression. (A) Significant associations (FDR corrected LRT p < 0.05) between bacterial genera and T2D clinical variables. White boxes denote a lack of significant associations (p > 0.05) and fill denotes regression coefficient between genus and variable (log2 fold change in genus abundance if the variable is increased by one standard deviation). (B) Associations between disease state and selected bacterial genera. Blue lines indicate regression lines and light gray bands denote the standard error of the regression. (C) Overall T2D risk was evaluated by the number of T2D risk factors associated with the same genera as observed earlier. This relationship was gradual across the number of risk factors.

Figure 4

Associations between bacterial genera and the primary T2D-related clinical measurements. (A) The identified genera associated with the area under the glucose curve (AUC glucose). AUC values were rank-transformed in the panel to make the regression independent of outliers. The blue line denotes a linear model between log-transformed normalized counts and rank transformed AUC values. (B) Bacterial abundances stratified by beta cell function (“affected” meaning beta cell function was negatively affected). Normal beta cell function was identified by a beta cell disposition index larger than 2 (see Figure S1B ). Stars denote significance under the likelihood ratio test. (C) T2D treatment restored some of the altered bacterial genera (Anaerostipes, Blautia, Escherichia, Veillonella) to their normal levels but this was not true for all of them (Romboutsia remained at low levels). Stars denote significance under a Mann-Whitney test. For B–C “*” denotes p < 0.05, “**” denotes p < 0.01, and “***” denotes p < 0.001.

Figure 5

Immune/Inflammation makers in the cohort. (A) Association between Veillonella read counts and C-reactive protein in treatment-naive individuals. (CRP). (B) Association between Anaerostipes/Blautia read counts and Interleukin 6 in treatment-naive individuals. (C) C-reactive protein levels were elevated in untreated diabetic individuals (T2D) and this was not altered by treatment. (D) Interleukin 6 levels were elevated in untreated diabetic individuals and this was not altered by treatment. In C–D Stars denote significance under a Mann-Whitney test (* - p < 0.05, ** - p < 0.01, *** - p<0.001).

Figure 6

Adding prominent confounders from other classes of clinical variables did not influence effect size for diabetes-related clinical response variables but did abolish associations in obesity and some cardiovascular responses. Clinical variables are grouped into T2D, obesity, and cardiovascular disease, and association tests between each bacterial genus and variable are either not confounded with additional variables (without confounding) or confounded with all variables from the other groups (with confounding). Points denote the coefficient associated with the response variables under the DESeq2 model (log fold change associated with an increase of one standard deviation in the clinical variable) and error bars denote the standard errors of the model coefficient. Colors denote bacterial genera.