Impact of dietary interventions on pre-diabetic oral and gut microbiome, metabolites and cytokines

Pre-diabetes, marked by elevated blood glucose below diabetes thresholds, is prevalent and increases risk for type 2 diabetes and cardiovascular/kidney diseases. Diet is a key driver of hyperglycemia; processed meats, refined carbohydrates and sugary beverages foster inflammation and insulin insufficiency, whereas plant-forward diets are protective. Interindividual variability in postprandial glycemic responses limits utility of one-size-fits-all guidance; prior work (Zeevi et al.) developed a machine-learning model integrating clinical, lifestyle and microbiome features to predict personalized postprandial glycemic response. The gut microbiome may mediate diet–metabolism–immunity relationships via energy extraction and metabolite/cytokine production. The oral microbiome is also linked to hyperglycemia and periodontal inflammation with potential systemic effects. Most microbiome studies focus on species composition, which can miss important strain-level genetic heterogeneity. This study tests how a personalized postprandial glucose-targeting diet (PPT) versus a standard Mediterranean diet (MED) affect oral and gut microbiomes, circulating metabolites and cytokines in pre-diabetic adults, and whether microbiome features mediate diet effects on glycemic and immunometabolic outcomes.

- Personalized nutrition approaches using microbiome-informed models can predict postprandial glycemic responses (Zeevi et al.). - The gut microbiome influences obesity, insulin resistance, and immune function, and produces metabolites (e.g., short-chain fatty acids) linked to glucose homeostasis. - Oral microbiome dysbiosis is associated with periodontal disease and diabetes; local inflammation can drive systemic immune activation. - Focusing solely on species composition is limited; strain-level heterogeneity (up to ~5% genomic differences) can yield distinct host effects. Prior work shows high intra-species diversity and functional variability (e.g., Faecalibacterium prausnitzii subtypes). - Mediterranean diet is widely recommended for glycemic control and has known metabolomic signatures; bilirubin associates inversely with diabetes. These prior findings motivate assessing both compositional and strain-level changes across oral and gut microbiomes, and integrating metabolites/cytokines to understand mediation pathways in pre-diabetes.

Design: Biphasic, randomized, controlled, single-blind dietary intervention in adults with pre-diabetes. Phase 1: six-month intervention comparing a personalized postprandial glucose-targeting diet (PPT) versus Mediterranean diet (MED) with 1:1 allocation using covariate-adaptive minimization (balanced on age, sex, weight, BMI, HbA1c, fasting plasma glucose). Phase 2: six-month follow-up. Participants: Recruited in Israel (Jan 2017–Jan 2019); final follow-up Mar 2020. Inclusion: age 18–65; pre-diabetes per ADA 2010 (FPG 100–125 mg/dL; HbA1c 5.7–6.5%); ability to use a smartphone app. Exclusions included diabetes/weight-loss medications, recent antibiotics (≤3 months), chronic diseases or medications affecting glucose metabolism. 225 randomized (PPT n=113, MED n=112); 200 completed (100 each). Baseline characteristics balanced. Diets: MED followed standard guidelines (45–65% carbohydrates, 15–20% protein, <35% fat, <10% saturated fat); menus from a scored meal bank. PPT menus personalized by a machine-learning algorithm predicting individual postprandial glycemic responses using nutrient composition, blood tests, anthropometrics, lifestyle and gut microbiome features. No calorie restriction or added physical activity prescribed. Dietitian support provided monthly (intervention) and twice (follow-up); interim contact available. Monitoring and data: Smartphone food logs (Personalized Nutrition Project app), continuous glucose monitoring (CGM), anthropometrics, and biospecimens (stool, subgingival plaque, serum) collected at baseline and end of intervention; two 2-week follow-up periods. Dietary data filtered for implausible intake days and aggregated to participant means. Laboratory assays: - Metabolomics: Untargeted LC–MS (Metabolon) on serum; 1,095 metabolites; measured as normalized area counts; batch effects corrected via PCA criteria. - Cytokines: Olink PEA qPCR; 76 proteins; NPX units; batch correction via PCA as needed. - Microbiome: Shotgun metagenomics of fecal and subgingival plaque DNA (Illumina NextSeq 75 bp or NovaSeq 100 bp; Trimmomatic for QC; Bowtie2 to hg19 for human read removal). Species composition mapped to Pasolli SGB representative genomes; taxonomic labels via majority vote; relative abundance via URA (unique relative abundance) method on uniquely mapped reads. Functional pathways via HUMAnN3 using MetaCyc. Diversity (Shannon alpha), richness, and human cell shedding calculated pre-processing. Data processing: Log10 transform, robust standardization, outlier clipping (±5 SD), filter features present in ≥20 samples, impute missing values with feature minimum if complementary timepoint exists, and batch-correct if a top PC (≥5% variance) associated with batch (p<0.05). Statistics: Within-diet pre vs post comparisons using two-sided Wilcoxon paired signed-rank tests with Bonferroni correction (alpha 0.05). Between-diet baseline comparisons via Mann–Whitney U. FDR correction applied only for cytokines in a sensitivity analysis (alpha 0.05). Mediation analyses: Conducted on significantly changed features in each group (pingouin; covariates: baseline age, sex, BMI). Predictor: diet change (PPT or MED); mediator: oral/gut microbial species or pathways; outcomes: changes in glycemic metrics, serum metabolites, or cytokines. Significance by two-sided bootstrap requiring significant predictor→mediator and mediator→outcome (controlling for predictor). Strain analysis: Variant pileups per species; sub-species genetic dissimilarity computed as normalized pairwise distance across ≥20K overlapping positions (≥3 reads per position); dissimilarity threshold clipped to 1/20K. Strain replacement defined when within-person pre/post dissimilarity exceeds lower 5% of inter-person dissimilarity distribution. Assessed per-participant and per-species; diet groups combined for strain mediation analyses. Tested top-10 most replaced species (present in >50 participants) as mediators between diet (absolute change) and outcomes (glycemic, metabolites, cytokines). Model analysis: Applied external model (Bar et al.) trained to predict serum metabolites from gut species composition in an observational cohort to this intervention cohort by predicting metabolites at pre and post and differencing to estimate change; assessed observed vs predicted change correlations and explained variance.

- Scope of change: 166 of 2,803 molecular features (microbial species and pathways, serum metabolites, cytokines) significantly changed in response to PPT or MED. - Diet composition shift: PPT increased lipids (+14.75%±6.21) and decreased carbohydrates (−17.76%±6.22); MED reduced lipids (−4.49%±4.38) and slightly increased carbohydrates (+2.05%±3.96). Both increased protein intake (PPT +3.21%±2.98; MED +1.90%±2.35). PPT constituted a larger macronutrient shift from baseline than MED (Bonferroni p<0.01). - Gut microbiome diversity/richness: PPT increased richness (+11.31±33.43 species, p<0.01) and Shannon diversity (+0.28±0.79, p<0.01) and reduced human cell shedding (−0.11%±5.31, p<0.05). MED increased diversity (+0.12±1.06, p<0.05) without significant richness or shedding changes. - Gut species composition: PPT increased 19 species (notably Ruminococcaceae, Clostridiales, Firmicutes; classified increases included Flavonifractor plautii, Roseburia hominis, Ruthenibacterium lactatiformans, and three Faecalibacterium prausnitzii sub-types). MED increased 4 species (two Ruminococcaceae, two Clostridiaceae) and decreased Eubacterium ventriosum. F. prausnitzii sub-types increased in both diets but involved different sub-types per diet, indicating strain-level specificity. - Gut functional pathways: PPT uniquely increased 7 pathways (e.g., β-(1,4)-mannan degradation, D-fructuronate and β-D-glucuronosides degradation, gluconeogenesis, fermentation, nitrate reduction, putrescine biosynthesis). MED uniquely altered 11 pathways (increased glycogen degradation, Bifidobacterium shunt, sucrose biosynthesis, purine ribonucleoside degradation, molybdopterin and L-glutamine biosynthesis; decreased preQ0 and 6-hydroxymethyl-dihydropterin diphosphate biosynthesis). Seven pathways increased in both diets (four amino acid biosyntheses: two L-arginine, L-cysteine, L-ornithine; two sugar degradation; thiamin biosynthesis). - Oral microbiome: Only Shannon diversity modestly increased in PPT; no significant oral species changes in PPT; MED increased Alistipes putredinis; PPT decreased an oral sucrose degradation pathway. - Serum metabolites: PPT changed 86 metabolites (84 increased, 2 decreased), with 45 lipids (including 13 sphingomyelins), 11 amino acids, multiple fatty acid-related metabolites, and 9 butyrate-containing compounds; MED increased 27 metabolites, including bilirubin and five degradation products, several lipids and amino acids. Six metabolites increased in both arms (guanidino-succinate, 2-hydroxyphenylacetate, cysteine-glutathione disulfide, X-12798, X-23665, and 3-bromo-5-chloro-2,6-dihydroxybenzoic acid). - Cytokines: Strict Bonferroni—PPT increased SCF; MED increased AXIN1 and SIRT2. FDR sensitivity—PPT also increased CCL11, CX3CL1, and TRAIL; MED increased STAMBP and SULT1A1 (ST1A1). - Mediation by microbiome (species level): • Glycemic outcomes: In PPT, four gut species mediated effects of dietary proteins/fibers/vitamin C/cholesterol/potassium/calcium on time-above-140 mg/dL and HbA1c, including F. prausnitzii (SGB_15342) and unclassified SGB_4957 and SGB_15054. In MED, vitamin B6’s effect on time-above-140 was mediated by a different F. prausnitzii sub-type (SGB_15333). • Metabolites: In PPT, two F. prausnitzii sub-types (SGB_15316, SGB_15342) and five unclassified species mediated effects of 10 dietary features on 20 metabolites (25 paths), e.g., SGB_4964 mediated vitamin B12 to multiple sphingomyelins. In MED, unclassified SGB_4714 mediated thiamin’s effect on three bilirubin degradation products. • Cytokines: In PPT only, F. prausnitzii (SGB_15342) mediated diet lipids to CX3CL1 and nuts/seeds to CCL11; Flavonifractor plautii mediated whole grains to SCF; unclassified SGB_15267 mediated fibers/vitamin E to TRAIL; SGB_4957 mediated thiamin to TRAIL. - Microbiome–metabolite change association: Using an external model, changes in gut species composition explained 12.25% of variance in changes of 127 metabolites known to associate with the microbiome (Pearson r=0.35 overall, r=0.41 PPT, r=0.18 MED; all p≤1e−10 except MED p<0.05). For metabolites not well predicted in training, explained variance was 2.99% (r=0.17 overall, r=0.22 PPT, r=0.07 MED). - Strain-level dynamics: Oral microbiome was significantly more genetically dynamic than gut at participant and species levels (p≤1e−23). Oral strain replacement negatively correlated with oral richness (r=−0.49, p<1e−10) and positively with species prevalence (r=0.43, p<1e−10); gut showed a weaker negative richness correlation (r=−0.22, p<0.005) and no prevalence trend. Most replaced gut species: Roseburia intestinalis, Roseburia inulinivorans, unclassified Clostridium (SGB_4910; 30.85–20.45% of participants). Most replaced oral species: Actinomyces naeslundii, Fusobacterium nucleatum, Leptotrichia buccalis (58.59–52.25%). - Strain-level mediation: Oral A. naeslundii mediated the effect of drinks on time>140. Across metabolites/cytokines, 56 gut and 70 oral mediation paths involved gut R. intestinalis, Eubacterium rectale and unclassified SGBs (SGB_5075, SGB_4820, SGB_4914, SGB_15254), and oral A. naeslundii, F. nucleatum sub-types (SGB_6007, SGB_6014), Actinomyces oris, Streptococcus oralis, and unclassified SGB_6055.

The study demonstrates that both personalized (PPT) and Mediterranean (MED) dietary interventions in pre-diabetes induce coordinated changes across the gut/oral microbiomes, serum metabolome, and cytokine milieu. PPT, entailing a larger macronutrient shift and individually tailored PPGR predictions, yielded broader microbiome and metabolite alterations than MED. Key beneficial signatures included increased F. prausnitzii sub-types and butyrate-containing compounds (PPT), and increased bilirubin and degradation products (MED), while some changes (e.g., putrescine biosynthesis) may reflect disease-associated pathways. Mediation analyses reveal that gut microbial species and pathways causally mediate diet effects on glycemic control, metabolites, and cytokines, with strain-specific roles (distinct F. prausnitzii sub-types in PPT vs MED). Cross-domain modeling indicates that changes in gut species composition explain a meaningful portion of interventional metabolite changes, underscoring functional coupling between microbiome dynamics and host metabolomics. Although the gut microbiome showed larger compositional shifts, the oral microbiome exhibited higher strain-level turnover, linked to environmental richness and species prevalence; strain-level mediations implicated oral taxa (e.g., A. naeslundii, F. nucleatum) in connecting diet to metabolic and inflammatory outputs. Collectively, findings support microbiome-mediated pathways linking diet to improved glycemic and immunometabolic status in pre-diabetes and highlight strain-level specificity as a critical axis for mechanistic and therapeutic exploration.

Dietary interventions in pre-diabetes modulate gut and oral microbiomes, circulating metabolites, and cytokines, with extensive cross-associations and microbiome-mediated effects on glycemic and immunometabolic outcomes. The personalized PPT diet produced broader beneficial changes than the standard MED diet, including increased gut diversity/richness, F. prausnitzii sub-types, and butyrate-related metabolites; MED uniquely increased bilirubin and SIRT2. Mediation by specific species/pathways and strain-level dynamics indicates that taxonomic resolution beyond species is important for understanding and targeting host–microbiome interactions. Future work should: (1) validate causal roles of identified species/pathways and uncharacterized metabolites; (2) test targeted microbiome modulation (e.g., specific strains, probiotics, or FMT) in conjunction with tailored diets; (3) refine predictive models linking microbiome change to metabolomic and clinical outcomes; and (4) evaluate long-term durability and generalizability across diverse populations.

- Cytokine dataset had the smallest sample size, reducing power; several cytokine findings required FDR-based sensitivity analysis. - The study applied an external model for metabolite prediction due to limited sample size for training, which may affect predictive alignment with this interventional cohort. - A large fraction of microbial species were unclassified at the species level (77% gut, 59% oral), and many biochemicals were uncharacterized, limiting biological interpretability. - Compositional analyses can induce dependencies among features; although strain-level analyses were performed, residual compositional biases may remain. - One metabolite (3-bromo-5-chloro-2,6-dihydroxybenzoic acid) warrants caution as halogenated molecules are uncommon in humans. - Randomized single-blind design and recruitment from a single country may limit generalizability; exploratory analyses were not pre-specified in the protocol.