Multi-omics signatures in new-onset diabetes predict metabolic response to dietary inulin: findings from an observational study followed by an interventional trial

The study investigates how the gut microbiome and metabolome relate to obesity and type 2 diabetes risk and whether these multi-omics features can predict metabolic responses to a prebiotic (inulin). Prior research links gut microbiota composition to obesity and type 2 diabetes, but findings on specific taxa, diversity, and functional signatures are inconsistent. Diet both shapes the microbiome and influences host metabolism, making dietary fiber a promising intervention, though with high inter-individual variability in response. The authors aim to determine: (i) whether MIME differs among lean healthy, obese healthy, and newly diagnosed pre/diabetes; (ii) whether inulin’s effects on glucose tolerance and insulin sensitivity are reflected in microbiota/metabolome changes; and (iii) whether baseline MIME and clinical characteristics predict individual responses to inulin.

The paper outlines mixed literature on gut microbiota in obesity and type 2 diabetes. While many studies report reduced alpha diversity in T2D, others find no reduction or comparable diversity. Taxonomic shifts are inconsistent across studies: some report increased Firmicutes/Bacteroidetes ratio, variable Proteobacteria changes, and recurring alterations at the genus level (increases in Streptococcus, Escherichia, Veillonella, Lactobacillus, Collinsella; decreases in Akkermansia, Dialister, Haemophilus, Roseburia, Faecalibacterium). Diet is a key determinant of microbiome composition and function. Prebiotics, particularly dietary fiber (e.g., inulin-type fructans), show potential benefits for metabolic diseases, yet clinical responses vary widely, likely due to inter-individual microbial differences. Identifying taxa mediating fiber’s benefits may enable individualized dietary therapy.

Design: Two linked components conducted within the TRIEMA project (NCT03710850). Observational cross-sectional case-control study comparing 49 newly diagnosed pre/diabetes (DM; BMI >25; fasting glycemia >5.6 mM and/or 2hOGTT >7.8 mM), 66 metabolically healthy overweight/obese (OB; BMI >25), and 32 lean healthy (LH; BMI <25). Prospective interventional study: 27 DM received inulin 10 g/day for 3 months (single-arm, non-controlled). All participants gave informed consent; protocol approved by EK-VP/26/0/2017. Clinical assessments: 12-h fasting visit; blood/urine sampling; clinical exam; bioimpedance; 75 g OGTT; 3-day dietary records; stool collection within one week. In the interventional study, baseline and post-intervention assessments included indirect calorimetry and a two-step hyperinsulinemic euglycemic clamp (10 and 80 mIU/m² BSA insulin). Insulin sensitivity indices: adipose tissue IS via NEFA and glycerol suppression in step 1; skeletal muscle IS via space-corrected glucose infusion rate per kg fat-free mass (Mcorr, mg/kg FFM/min) and MCR/I (ml/kg FFM/min per steady-state insulin) in step 2. Omics assays: Gut microbiome by 16S rRNA V4 amplicon sequencing (Illumina MiSeq; DADA2 pipeline). Fecal volatile organic compounds (VOCs) by GC×GC-TOF-MS (Agilent 7890B; LECO Pegasus 4D; ChromaTOF). Serum metabolome by 600 MHz 1H NMR (CPMG; PQN normalization; metabolites identified via HMDB). Plasma short-chain fatty acids (SCFAs) quantified by LC-MS using isotope-labeled derivatization. Statistics: Data treated as compositional where appropriate; centered log-ratio (clr) transformation with multiplicative replacement for zeros. Alpha/beta diversity metrics; PERMANOVA for group differences; univariable analyses with FDR control; effect sizes (Cliff’s delta). Machine learning classification via LASSO using microbiome, fecal VOC, serum metabolome, and integrated datasets. Predictive modeling of intervention response via linear regression models of outcomes Y(A)=β0+βy Y(B)+βx X(A)+ε, screening clinical and omics predictors; variables with high leverage excluded; bootstrapped R² reported. Sample size for intervention powered for glucose disposal difference across tertiles with allowance for dropouts.

Observational study: - Participants: LH (n=32), OB (n=66), DM (n=49). Groups differed in glycemic indices, insulin sensitivity, beta-cell function, and lipid biomarkers (Table 1). - Microbiome: 44,332 ASVs; 13 phyla, 367 genera. Alpha diversity indices were significantly lower in OB and DM vs LH; no difference between OB and DM. At genus level, Shannon index remained lower in OB and DM vs LH. Beta-diversity differed by group (PERMANOVA p ≤ 0.001); pairwise LH vs OB and LH vs DM p<0.001; OB vs DM not significant. Thirty-seven genera differed; 15 core taxa (abundance >0.05% and prevalence >75%). Butyrate producers (Anaerostipes, Eubacterium halii, Faecalibacterium, Christensenellaceae R-7) were enriched in LH; Pseudobutyrivibrio and Lachnoclostridium enriched in DM. Potentially harmful non-core taxa (Fusobacterium, Megasphaera, Desulfovibrio) enriched in OB/DM; Fusobacterium correlated positively with C-peptide. - Fecal VOCs: 185 VOCs identified; beta-diversity differed (p=0.0017); LH differed vs DM (p<0.01) and vs OB (p<0.05); OB vs DM not significant. Ten VOCs differed (FDR ≤0.1): nonanoic acid higher, most others (including SCFA esters) lower in OB/DM vs LH; methyl pentanoate showed DM>LH>OB. Nonanoic acid correlated positively with TyG index. - Serum metabolome: 35 NMR analytes and 9 SCFAs (LC-MS) identified. PERMANOVA showed separation; pairwise differences significant (p ≤ 0.001) for all group pairs. Twenty-one metabolites differed. OB and DM signatures overlapped, with DM farther from LH: increases in glucose, lactate, mannose; AA shifts (glutamine and alanine differences among all three groups; tyrosine increased in OB/DM; histidine and asparagine decreased). SCFAs: propionic and succinic acids higher, valeric lower in OB/DM vs LH. DM-specific increases: BCAA derivatives (2‑oxoisovalerate, 3‑methyl‑2‑oxovalerate, 2‑oxoisocaproate), 2‑hydroxybutyrate, acetone, 2‑propanol, and formate. - Classification performance: Microbiome-only LASSO accuracy 51% (sensitivities 66% LH, 50% OB, 43% DM); improved to 75% accuracy when merging OB+DM. Fecal VOC model accuracy 52% (48% LH, 54% OB, 53% DM), improved to 80.5% when merging OB+DM but with low sensitivity. Serum metabolome model accuracy 74% (sensitivities 90% LH, 72% OB, 65% DM); improved to 89% when merging OB+DM; glucose was not selected as a key discriminant. Integrated model accuracy 77% (sensitivities 88% LH, 79% OB, 66% DM) for three groups; 91% accuracy and 89% sensitivity for LH vs OB+DM. - MIME interaction networks: Rich cross-dataset correlations in LH; reduced complexity in OB and DM. Interventional study (27 DM, 3 months inulin 10 g/day): - Microbiome changes: Significant overall composition shift (PERMANOVA p<0.001) with decreased alpha diversity (Observed species p<0.001; Chao1 p<0.001; Shannon p=0.016; InvSimpson p=0.057). Phyla: Bacteroidetes and Proteobacteria decreased; Actinobacteria and Verrucomicrobia increased. Twenty-eight taxa changed: increases in beneficial/SCFA-associated taxa (Faecalibacterium, Anaerostipes, Eubacterium halii, Lactobacillus, Bifidobacterium, Akkermansia, among others); decreases in Alistipes, Odoribacter, Bacteroides, etc. - Metabolome changes: Serum butyric acid, propionic acid, and asparagine increased; glycerol and 2‑propanol decreased. Fecal VOCs showed nominal changes (two propionic acid esters increased, 1‑hexanol decreased) that did not remain significant after multiple-testing correction. - Glycemic outcomes: Group-level improvement in 2h OGTT glucose; trends toward reduced AUC OGTT glucose and fasting glycemia. Skeletal muscle insulin sensitivity (MCR/I) improved by >10% in 14/27 participants (+11.4% to +62.4%), while 13/27 showed little change or decreases (+4.8% to −48.7%). Similar patterns for Mcorr, AUC OGTT insulin, and fasting insulinemia. Predictors of response (Table 2): - Better improvements associated with more favorable baseline glycemic status (e.g., higher ISI Matsuda, lower fasting insulin/HOMA indices predicted increases in Mcorr and MCR/I). - MIME predictors included baseline bacterial abundances and metabolites: lower baseline Blautia and Eubacterium halii group predicted larger MCR/I gains; lower Dialister associated with Mcorr increase; higher baseline serum asparagine predicted improvements in both Mcorr and MCR/I. - Increased AUC OGTT glucose after inulin associated with higher baseline Lachnospiraceae incertae sedis, Lachnoclostridium, BCAA derivatives (3‑methyl‑2‑oxovalerate), and alanine; decreased with higher Ruminiclostridium and ethanol. - Higher 2h OGTT glucose post-intervention associated with higher baseline AUC OGTT insulin, fasting insulinemia, HOMA-INS, Eubacterium halii group, BCAA derivatives, pyruvate, and fecal VOCs including indole (positive association), tridecanol, δ‑dodecalactone, methyl heptenone, 2‑undecanone, and methyl butanal. Overall, obesity dominated MIME differences; inulin shifted microbiota toward beneficial taxa and increased circulating SCFAs; response prediction was feasible using baseline clinical and multi-omics features.

The findings support that obesity, more than glycemic status, shapes gut microbiome and metabolome profiles, with OB and DM sharing a similar metabolic signature that diverges from LH. Reduced abundance of butyrate-producing taxa in OB/DM coincided with elevated medium-chain fatty acids (e.g., nonanoic acid) in feces, suggesting that factors beyond SCFA deficiency—such as increased MCFA with potential pro-inflammatory and barrier-disrupting properties—may contribute to obesity-related pathophysiology. Serum metabolome patterns, including elevated propionate and succinate and a DM-specific increase in BCAA catabolites and redox-related metabolites (2-propanol, acetone, 2-hydroxybutyrate), align with known insulin resistance and dysglycemia pathways. Inulin intervention modified microbiota composition toward an enrichment of beneficial and butyrate-producing taxa and increased circulating butyrate and propionate, consistent with improved insulin sensitivity observed in a subset of participants. The heterogeneous metabolic response to inulin could be partially explained by baseline clinical and MIME features. Individuals with better baseline insulin sensitivity and favorable MIME profiles (lower BCAA derivatives, lower fecal indole, specific bacterial abundances) were more likely to benefit, indicating that MIME-informed stratification could personalize dietary fiber interventions. These results demonstrate the practical utility of integrating multi-omics with clinical data to predict dietary intervention outcomes in early diabetes.

This study shows that obesity is the primary determinant of gut microbiome and metabolome signatures, with glycemic status exerting additional but smaller effects. In newly diagnosed pre/diabetes, inulin supplementation induces favorable shifts in gut microbiota and circulating SCFAs and can improve glycemic control and insulin sensitivity, albeit with substantial inter-individual variability. Baseline clinical measures and MIME signatures can predict metabolic response to inulin, suggesting a path to personalized nutrition in early diabetes management. Future work should validate the identified predictors in larger, independent, and more diverse cohorts, employ controlled randomized designs, and explore mechanistic links between specific taxa/metabolites (e.g., BCAA derivatives, indole) and insulin sensitivity to refine precision nutrition strategies.

- Small sample sizes, particularly in the interventional arm (n=27), limit generalizability and model complexity; predictors were not validated in an independent cohort (internal validation via permutation only). - Age differences between LH and OB/DM groups could confound microbiome comparisons, although major age-related microbiome changes are typically most pronounced at extremes of age. - Diet during the intervention was not strictly controlled; participants were instructed to maintain habitual intake, introducing potential dietary confounding (BMI remained stable). - Single-arm, non-controlled intervention design precludes causal inference regarding inulin’s effects on metabolic outcomes. - Limited power restricted assessment of interactions and synergies among predictors.