Distinct gut microbial species, but not phylum-to-genus composition, associate with insulin resistance: a unique...

Links between gut microbiota and insulin resistance (IR) vary across populations. We profiled the fecal microbiota of Kazakh adults to test whether community composition associates with IR at broad (phylum → genus) and species levels.

In a cross-sectional case control study (N = 200; IR = 183, controls = 17), TyG indexed IR status. 16S rRNA sequencing (two primer pools; nine hypervariable regions) characterized taxa. After CSS normalization, we compared presence/absence across groups (χ²) and modeled species with univariate and multivariable logistic regressions, using absence of each species as the predictor.

High-level composition did not differ between IR and controls (phylum, class, family, genus; all p > 0.05). In contrast, several species differed. In univariate models, absence of Actinomyces odontolyticus (OR = 25.55, p = 0.010), Bifidobacterium kashiwanohense (OR = 12.69, p = 0.015), Lactobacillus sp. (OR = 5.71, p = 0.020), and Streptococcus lactarius (OR = 6.27, p = 0.044) associated with higher IR odds, suggesting protection when present; whereas absence of Alistipes onderdonkii (OR = 0.30, p = 0.044) and Prevotella copri (OR = 0.19, p = 0.003) associated with lower IR odds, suggesting risk when present. In multivariable models, these signals persisted: absence of P. copri (OR = 0.146, p = 0.003) and Roseburia inulinivorans (OR = 0.143, p = 0.011) predicted lower IR odds (risk alignment), while absence of Lactobacillus sp. (OR = 8.29, p = 0.016) and Coprococcus catus (OR = 7.04, p = 0.004) predicted higher IR odds (protective alignment).

In this Kazakh cohort, no broad compositional signal emerged, but species-specific associations were strong and bidirectional. Findings highlight population-specificity and identify candidate species associated with IR that may serve as hypothesis-generating targets for future validation. Any attempt to modulate these taxa for insulin resistance is unproven and requires function-resolved, diet-measured longitudinal studies and randomized trials before clinical application. The IR:control imbalance (183:17) increases uncertainty for low-prevalence taxa; species-level findings are hypothesis-generating and require validation in a more balanced design. Because 16S rRNA profiling does not measure gene functions or metabolites, these species-IR associations are hypothesis-generating and warrant validation using shotgun metagenomics and metabolomics.