Main

Cardiometabolic diseases (CMDs) are the leading causes of morbidity and mortality in Western countries and constitute a heavy burden on global healthcare systems1. The most predominant CMDs are cardiovascular disease (CVD) and type 2 diabetes (T2D)3, which are connected with the increased consumption of calorie-dense, high-risk processed foods observed over the past few decades4. Habitual diet is not only among the known risk factors associated with CMDs, but also the primary modifiable target for prevention and treatment5. Well established anthropometric and intermediary measures of CMDs, ranging from clinical measurements (for example, blood pressure) to lipid profiles (such as triglycerides, cholesterol and lipoproteins), glucose levels (for example, fasting and postprandial glucose, and haemoglobin A1c (HbA1c)), inflammatory markers (for example, glycosylated proteins, the systemic inflammation marker GlycA21 and high-sensitivity C-reactive protein), and known risk factors such as body mass index (BMI), can be used to study the diet–CMD axis6,7,8 but do not consider the biochemical mechanisms occurring in the human gut.

The human gut microbiome has emerged as a cofactor on the same axis as it is associated with diet and cardiometabolic conditions9,10,11,12 and is a modifiable element13,14,15. A change in dietary patterns can shift the species-level composition of the microbiome, with knock-on effects on host health16. However, individual responses to dietary interventions vary, and precision nutrition aims at identifying host-specific factors that modulate the interaction between diet and host health17, but it is currently not possible to disentangle the effects diet plays to improve cardiometabolic health via the microbiome. Furthermore, the composition of the gut microbiome displays high individuality and variation depending on different demographics, ethnicity, sex and age; hence, defining or identifying universal biomarkers of a healthy gut microbiome has proven difficult18,19,20.

Nutritional intervention studies usually involve low sample-size cohorts at the population level and are often limited by their statistical power and specificity to local lifestyle and dietary habits, which are all critical aspects, especially given the microbiome’s complexity and variability. Large-scale comprehensive studies with multi-national populations can help disentangle some of the complex interplays between dietary patterns and the gut microbiome to develop personalized interventions to prevent and treat CMDs. Accordingly, we collated, sampled and analysed five of the largest metagenomic cohorts available to date, comprising more than 34,000 people and spanning two continents, paired with dietary data, detailed anthropometric and health markers. We identified microbiome species consistently associated with more favourable and (inversely) unfavourable health markers across continents. These species were organized into two microbiome rankings, representing host health and diet quality, respectively, that can be the basis for future causal and mechanistic studies.

Metagenomics of the ZOE PREDICT cohorts

We used four large-scale microbiome cohorts from the ZOE PREDICT studies (n = 33,596; Fig. 1a, Supplementary Fig. 1, Supplementary Table 1 and Methods) to assemble an extensive microbiome dataset of people with detailed dietary records along with anthropometric measures. Together with the previously available ZOE PREDICT 1 cohort9 (n = 1,098), the PREDICT cohorts comprise 34,694 participants from both the USA (n = 21,340) and the UK (n = 13,354; Methods). Collected data comprise common health risk factors such as BMI, triglycerides, blood glucose and HbA1c, as well as several dietary indices and clinical markers that are intermediary measures of cardiometabolic health, such as the atherosclerotic CVD (ASCVD) risk, high-density lipoproteins (HDL) and GlycA21 (Supplementary Table 2 and Methods).

Fig. 1: The ZOE PREDICT studies comprise over 34,000 healthy people from five cross-sectional studies from the UK and the USA with gut microbiome samples, detailed individual information and dietary habits.

a, In this study, we considered and harmonized five cross-sectional ZOE PREDICT cohorts with participants from the UK and the USA (Supplementary Fig. 1). For each cohort, sample size and the percentage of female participants (% F) are reported in the upper bar plots, with sequencing depth (left-hand columns, darker colour, average size in gigabases) and the total number of detected species (right-hand columns, lighter colour) are reported in the middle bar plots, showing that cohorts with lower sequencing depths do not have fewer total numbers of detected species. Bottom box plots, distributions of age (left-hand columns, darker colour) and BMI (right-hand columns, lighter colour) in the five PREDICT cohorts (the PREDICT 1 (P1) cohort is split into its UK and US parts, but considered as a single cohort). Box plots show first and third quartiles (boxes) and the median (middle line); whiskers extend up to 1.5 × interquartile range (IQR). b, Random forest classification (discriminating the first three quartiles against the fourth quartile) and regression machine learning models (Methods) trained on the whole microbiome SGB-level relative abundance values with a cross-validation approach, show moderately strong and consistent associations with different categories of clinical data available across the five cross-sectional ZOE PREDICT cohorts (full machine learning results are reported in Extended Data Fig. 1 and Supplementary Tables 2 and 3). HEI, healthy eating index; PDI, plant-based diet index; hPDI, healthful PDI; oPDI, overall PDI.

A systematic machine learning validated approach9,22 (Methods) revealed strong associations consistent across the five ZOE PREDICT cohorts between the microbiome and surrogate health markers and nutrition (Fig. 1b, Supplementary Table 2 and Methods). Markers that were classified accurately by the gut microbiome included glycemia, blood cholesterol, triglycerides and inflammation (both fasting and postprandial; Extended Data Fig. 1), with age, BMI, the healthy eating index23 and the healthful plant-based diet index (PDI)24 also correlated with microbiome machine learning regression estimates (Spearman’s correlation > 0.4; Fig. 1b and Supplementary Table 2). The top predicted markers from both machine learning regression and classification showed consistent associations across PREDICT cohorts, with average area under the receiver operating characteristic curve (AUC) ranging from 0.64 to 0.73, and an average Spearman’s correlation ranging from 0.30 to 0.46 for regression (Fig. 1b and Supplementary Table 3).

Ranking gut species to host and health

We next set out to identify which gut microbial species were most responsible for the microbiome’s associations with host markers. To do so, we grouped the 37 markers into three categories: (1) anthropometric-derived and accessible health-related measures (hereafter called ‘personal’ and including, for example, ASCVD and blood pressure), (2) fasting (for example, GlycA, triglycerides, HDL, cholesterol and glucose) and (3) postprandial markers, which are surrogate measures of cardiometabolic health. As expected, some markers tended to correlate quantitatively (Supplementary Table 4 and Methods).

We considered 661 non-rare microbial species (greater than 20% prevalence; Methods) according to the definition of species-level genome bins (SGBs)19,20, and computed the partial Spearman’s correlations (corrected for sex, age and BMI) between the relative abundance of each micro-organism and the value of each marker. Correlations were ranked, and correlations’ ranks were averaged within each category and then averaged among the three categories in each cohort (Methods). The five resulting cohort-level average rankings were averaged to derive a single ranking that we called the ‘ZOE Microbiome Health Ranking 2025’ (ZOE MB health-rank). This resulted in a ranking for 661 microbial species in which the lowest ranking (closer to 0) species are the most positively associated with the considered panel of host markers and vice versa for the highest ranking (closer to 1) species (Fig. 2, Extended Data Figs. 2 and 3, Supplementary Fig. 2 and Supplementary Tables 5 and 6).

Fig. 2: The 15 top and bottom health-ranked SGBs show consistent associations across the five PREDICT cohorts.

a, Average percentiles for the 15 most favourably (top) and unfavourably (bottom) ranked SGBs (selected for visualization purposes) across all five PREDICT cohorts. Percentiles are computed from the ranking of the correlations between SGBs and the different markers in each clinical data category. Percentiles close to 0 reflect SGBs consistently correlated positively with positive markers and negatively with negative markers, and vice versa for percentiles close to 1. For each cohort, the average percentiles for three clinical data categories are shown (personal, fasting and postprandial) and the cohort-level average. The rightmost column of the heatmap reports the ZOE MB health-ranks with the distribution of their relative abundance values when present (derived from n = 34,694 participants spanning the five PREDICT cohorts). Box plots as in Fig. 1. b,c, Detailed percentiles for the 15 most favourably and unfavourably ranked SGBs against the markers of the three clinical data categories of the PREDICT 1 (UK) (b) and PREDICT 3 US22A (US) (c) cohorts. Detailed panels of the percentiles for the other three cohorts can be found in Extended Data Fig. 2. iAUC, incremental area under the curve; PUFA, polyunsaturated fatty acid; QUICKI, quantitative insulin sensitivity check index; THR, total-cholesterol-to-HDL ratio; VLDL, very-low-density lipoprotein.

Most SGBs ranked within the 50 most favourably or unfavourably linked to host anthropometry belong to the Firmicutes phylum (92 out of 100) and, in particular, to the Clostridia class (n = 80; Supplementary Table 7). Within this class, in the ZOE MB health-ranks, most SGBs belonged to the Clostridiales order, with 32 unfavourably ranked SGBs (of which n = 27 Lachnospiraceae out of 50) and 31 favourably ranked SGBs (n = 13 Lachnospiraceae and n = 12 Ruminococcaceae) assigned to this order. Collectively, the average total relative abundance of the 50 most favourably ranked SGBs is 5.98%, whereas the 50 most unfavourably ranked SGBs account for 13.64% (Supplementary Table 7).

Uncharacterized health-linked bacteria

A large portion of the 50 most favourably ZOE MB health-ranked SGBs are unknown (n = 22), meaning that these are uncultured species represented solely by microbial genomes reconstructed from metagenomic data. Of the 28 known SGBs (with available isolate genomes), 24 are still uncharacterized species without phenotypic descriptions and recognized taxonomic names (Supplementary Table 7). Eubacterium siraeum (SGB4198) and Faecalibacterium prausnitzii (SGB15317) are among the few exceptions with previous support for their favourable role9,25.

By contrast, the 50 unfavourably ZOE MB health-ranked SGBs are generally species with cultured isolates and established taxonomic labels (Supplementary Table 7). Among the 44 known SGBs, several species were already linked with detrimental effects on the host, including Ruminococcus gnavus26, Flavonifractor plautii27, Ruminococcus torques28,29 and Enterocloster bolteae30. Overall, the most prevalent favourably ranked health-associated micro-organisms in the human gut belong to under-investigated species, highlighting gaps in our knowledge of the potential beneficial role of the human microbiome in promoting and maintaining non-pathogenic conditions.

Gut species ranked by diet quality

Similarly to the ZOE MB health-ranks, we defined a species ranking on the basis only of dietary markers across all five PREDICT cohorts, which we called the ‘ZOE Microbiome Diet Ranking 2025’ (ZOE MB diet-rank; Supplementary Table 5). As markers of a generally healthier diet, we adopted five validated indices (Methods) computed starting from validated food frequency questionnaires (FFQs) or logged diet data (logged using a mobile phone app), reflecting long- and short-term dietary habits, respectively (Extended Data Figs. 4 and 5 and Methods).

The ZOE MB health- and diet-rankings showed, as expected, general concordance (Spearman’s ρ = 0.72; Extended Data Fig. 6a and Supplementary Table 5). Although the large majority of the SGBs highlighted by high or low ZOE MB health-ranks and diet-ranks belong to unknown taxa, reported phenotypic characteristics of known species agree with our analysis. For example, R. torques (SGB4608) and F. plautii (SGB15132), discussed previously as unfavourable species according to the ZOE MB health-ranks, were also concordantly unfavourably ranked in the ZOE MB diet-ranks (0.991–0.904 and 0.981–0.901, respectively). On the other hand, the favourably ranked Blautia glucerasea (SGB4816) was described to reduce visceral fat accumulation, blood glucose and triglycerides in mice31 (ZOE MB health-ranks and diet-ranks of 0.267 and 0.062, respectively). As another example, in a dietary fibre supplementation trial involving individuals with T2D, Lachnospira eligens (SGB5082) was increased selectively and associated negatively with postprandial glucose and insulin, body weight and waist circumference32 (ZOE MB health-ranks and diet-ranks of 0.276 and 0.115, respectively), indicating that precise dietary interventions aimed at stimulating beneficial bacterial growth can contribute to treating or managing metabolic disorders symptoms.

Despite the overall agreement between the ZOE MB health- and diet-rankings, 65 out of the 661 ranked SGBs showed discordant rankings (absolute rank difference at least 0.3; Extended Data Fig. 6a and Supplementary Table 8). Generally, the different trends may be due to the different capacities of certain bacteria (for example, generalists) to use a variety of substrates, including those derived from unhealthy diets, while releasing functional metabolites with protective or health-promoting effects. Among these, for example, Harryflintia acetispora (SGB14838) was found associated with favourable cardiometabolic markers and unfavourable diets (ZOE MB health-rank = 0.363 and ZOE MB diet-rank = 0.879) in this study. This strict anaerobe can use readily available monosaccharides such as maltose, glucose and fructose, but can also produce short-chain fatty acids33, which are regulatory and anti-inflammatory mediators34.

Across the US and UK populations, the ZOE MB health-rankings showed high consistency (Spearman’s ρ = 0.61; Extended Data Fig. 6b), whereas country-specific ZOE MB diet-rankings were more heterogeneous (Spearman’s ρ = 0.26; Extended Data Fig. 6c). The intraclass correlation coefficients (ICC)35 also suggest that the ZOE MB health-ranks are more consistent across countries than the ZOE MB diet-ranks (ICC = 0.5929 and 0.2623, respectively; Extended Data Fig. 6b,c). Across cohorts, we obtained an ICC = 0.63 and 0.46 for the ZOE MB health-ranks and diet-ranks, respectively, indicating that health rankings were more able to capture cohorts and countries differences, whereas the most favourably ranked species appeared to match across populations with similar levels of industrialization and lifestyle.

Species rankings stratify by BMI

BMI is an imperfect but widely adopted and easy-to-obtain anthropometric marker of health risk. As BMI was not included among the markers of the ZOE MB health- and diet-rankings, and we corrected for it in the partial correlation analysis, we set out to evaluate how the two rankings can stratify people according to their BMI to assess how health signatures in the gut microbiome are reflected in body mass.

We correlated the 661 ZOE MB health-ranked species with BMI (corrected for sex and age), in each PREDICT cohort, and found that, overall, the ranks were associated positively with BMI (Spearman’s ρ = 0.72), with the favourably ranked SGBs correlated negatively with BMI, whereas unfavourably ranked SGBs correlated positively with BMI (Fig. 3a). These results were confirmed when considering the ZOE MB diet-ranks and discrete BMI categories (Extended Data Fig. 7a–c; all intra-dataset comparisons statistically significant at Q < 0.2 and all 30 except 7 at Q < 0.01) as well as the cumulative abundance of the species in the two 50-species sets (Fig. 3b,c; all intra-dataset comparisons statistically significant at Q < 0.2 and all 30 except 5 at Q < 0.01).

Fig. 3: ZOE MB health- and diet-ranked species show significant and reproducible associations with BMI and diseases.

a, Concordance of ZOE MB health-ranks with partial Spearman’s correlations against BMI (corrected for sex and age) across PREDICT cohorts. Favourably ranked SGBs correlate negatively with BMI; unfavourably ranked SGBs correlate positively (ZOE MB diet-ranks in Extended Data Fig. 7a). Shading represents 95% confidence interval. b,c, Cumulative relative abundance of favourably (b) and unfavourably (c) ranked SGBs across BMI categories. As BMI increases, reflecting higher health risks, the abundance of favourable SGBs decreases whereas that of unfavourable SGBs increases. Similar patterns were seen for SGB richness (Extended Data Fig. 7b,c). Only non-significant (NS) false discovery rate (FDR)-corrected P values (Q > 0.01, two-sided Mann–Whitney U-test) are annotated. Box plots as in Fig. 1. d, Meta-analysis of the 50 most favourable and unfavourable SGBs comparing participants of healthy weight with those with obesity from public cohorts. Lower BMI is associated with more favourable SGBs; people with higher BMI carry more unfavourable SGBs. Meta-analysis on ranks defined on UK and US participants shows reproducibility across countries. Other comparisons are in Extended Data Fig. 8 and the diet-ranked SGBs meta-analysis in Extended Data Fig. 9. Country codes: ARG, Argentina; AUT, Austria; DEU, Germany; DNK, Denmark; FRA, France; GBR, United Kingdom of Great Britain and Northern Ireland; IRL, Ireland; ISR, Israel; KAZ, Kazakhstan; NLD, Netherlands; USA, United States of America. e, Meta-analysis of disease group (adjusted by sex, age and BMI) on standardized mean differences (SMD) of cumulative relative abundance of the 50 most favourable (left) and unfavourable (right) SGBs from both rankings (meta-analysis on SGB richness in Supplementary Fig. 5a; Methods). f, Meta-analysis of normalized ZOE MB health-ranks and diet-ranks, weighted by arcsin square-root of relative abundance values (right, weighted score sum; left, score sum (unweighted)). Horizontal lines in meta-analysis plots represent 95% confidence intervals. CRC, colorectal cancer; IBD, inflammatory bowel disease; IGT, impaired glucose tolerance.

To generalize these associations, we leveraged a total of 5,348 healthy individuals from 27 public cohorts divided into three BMI categories, healthy weight (n = 2,837), overweight (n = 1,562) and obese (n = 949; Supplementary Table 9 and Methods). In 47 pairwise comparisons, 34 had a higher median richness for the 50 most favourably ranked ZOE MB health SGBs in lower BMI groups versus higher BMI groups (binomial P = 0.003; Supplementary Table 10 and Supplementary Fig. 3a), and this was not dependent on country effects or sequencing depth (Supplementary Table 11), highlighting the generalization of the identified ranks. Meta-analysis based on linear regression on single cohorts (Methods) showed that individuals with healthy weight carried, on average, 5.2 more of the 50 favourably ZOE MB health-ranked SGBs than people with obesity (P = 0.0003; Fig. 3d and Supplementary Table 12), which corresponded to a normalized difference in the cumulative abundances of unfavourably and favourably ranked SGBs of Cohen’s d = −0.59 (P < 0.0001; Supplementary Tables 10 and 13 and Methods). Correspondingly, individuals with obesity carried, on average, 1.95 more of the unfavourably ranked SGBs than people of healthy weight (P = 0.0005; Fig. 3d, Supplementary Tables 14 and 15; Cohen’s d on cumulative relative abundances = 0.29; P = 0.0001). Pairwise analysis of the other BMI categories confirmed these results (Extended Data Fig. 8 and Supplementary Tables 10 and 12–15).

Similarly, we tested the association of the 50 most favourably and unfavourably ZOE MB diet-ranked SGBs with BMI, and found similar but milder signals compared with the ZOE MB health-ranks (average Spearman’s correlations between the two ranks and BMI of 0.61 and 0.72, respectively; Fig. 3a and Extended Data Fig. 7a). Using public datasets, 36 intra-dataset comparisons out of 47 showed a higher median cumulative abundance and a higher median richness of the 50 most favourable SGBs in lower BMI classes compared with higher BMIs (binomial P = 0.0003; Supplementary Fig. 3b). Conversely, 36 comparisons showed a higher median count of the least favourable 50 SGBs for the higher BMI classes compared with the lower BMI groups (binomial P = 0.0003; Supplementary Table 10). The contribution of diet-ranked SGBs in different BMI categories similarly showed a decreasing number and cumulative relative abundance of favourably ranked SGBs and an increase in unfavourably ranked SGBs (Extended Data Fig. 7d–g). In meta-analysis, healthy weight and overweight participants carried 3.5 and 1.5 more favourable diet-ranked SGBs, and 1.25 and 0.88 fewer unfavourable ZOE MB diet-ranked SGBs than obesity participants, respectively (Extended Data Fig. 9, Supplementary Fig. 3 and Supplementary Tables 16–19). All these analyses were confirmed when rankings were computed without adjusting for BMI (Extended Data Fig. 7h–k) and, altogether, these results suggest that the ZOE MB health- and diet-ranks can stratify people based on their obesity status regardless of geography.

Species rankings and host diseases

Next, we assessed whether the ZOE MB health-ranked SGBs had a differential presence or abundance in control participants compared with participants with a defined disease condition, exploiting 25 case–control, publicly available microbiome studies (4,816 samples in total with n = 2,707 controls and n = 2,109 cases; Methods) investigating five diseases with variable levels of association with the gut microbiome (Supplementary Table 20). The number of the 50 most favourably ZOE MB health-ranked SGBs was higher in controls than cases for 21 of the 25 cohorts, whereas the count of the 50 most unfavourably ranked SGBs was correspondingly higher in cases for the same number of cohorts (binomial P = 0.0004).

We performed a meta-analysis on the count of the 50 most favourable and unfavourable SGBs from the ZOE MB health- and diet-rankings. Control samples carried, on average, 3.6 more favourably ranked SGBs than participants with disease (random-effect model, P = 0.0002; Methods) and 1.6 fewer unfavourable SGBs (P = 0.0004; Supplementary Fig. 5a and Supplementary Table 21). Similarly, for the ZOE MB diet-ranked SGBs, controls carried, on average, 3.8 more favourable SGBs and 1.3 fewer unfavourable SGBs, P = 9.5×10−6 and P = 0.0006, respectively; Supplementary Fig. 5a and Supplementary Table 21). Furthermore, meta-analyses of the cumulative abundance of the 50 most favourable and unfavourable SGBs confirmed a greater contribution from favourable species in control groups and of unfavourable SGBs in the corresponding disease groups (meta-analysis Cohen’s d = −0.29, P = 7.1 × 10−6 and d = 0.21, P = 0.054 for the ZOE MB health-ranks; d = −0.24, P = 3.1 × 10−6 and d = 0.28,* P* = 0.0002 for the ZOE MB diet-ranks; Fig. 3e and Supplementary Table 22).

To assess how informative the rankings are in summarizing the health-associated status of a single sample, we scored all metagenomes from diseased and control participants by summing the normalized ZOE MB health-ranks of the SGBs present in the sample (Methods). We found a strong separation between diseased and control participants (meta-analysis Cohen’s d = −0.37, P = 8.3 × 10−8), improving over the simple counting of the number of most favourable and unfavourable SGBs (Fig. 3f). Notably, T2D showed the strongest disease-specific association (meta-analysis Cohen’s d = −0.47, *P *= 6.78 × 10−5; Fig. 3f and Supplementary Table 23) with the weighted version of this score showing an even stronger effect for T2D (meta-analysis Cohen’s d = −0.51, *P *= 0.0002). People were also scored using the ZOE MB diet-ranks, and similar links with their health status emerged (Fig. 3f and Supplementary Table 23). Notably, standard alpha diversity measures such as gut SGBs richness and Shannon’s entropy measures showed weaker and less consistent associations, with significant links only in the IBD and T2D comparisons (Supplementary Fig. 5b and Supplementary Table 24).

Although the ranking-based scoring of single samples cannot have the same predictive power for host phenotypes compared with condition-specific supervised learning approaches relying directly on labelled training data, our results showed how embedding the ranking system into a simple one-dimensional microbiome index provides a meaningful evaluation of microbiome health conditions.

Diet changes effects on ranked species

To validate the effect of dietary changes on the presence and abundance of gut microbial species according to their ZOE MB health-rankings, we analysed two dietary intervention studies, namely ZOE METHOD36 and BIOME37 (ClinicalTrials.gov registrations, NCT05273268 and NCT06231706, respectively). In brief, the ZOE METHOD cohort comprised n = 347 people assigned to a personalized dietary intervention programme (PDP; n = 177) arm versus an arm with general diet advice following the US Department of Agriculture recommendations (control, n = 170). People assigned to the PDP group showed lower energy intake and a significant decrease in triglycerides, HbA1c, weight and waist circumference after 18 weeks36. The ZOE BIOME cohort comprised n = 349 healthy adults (intention-to-treat) randomized into the primary intervention group (receiving a defined prebiotic blend, n = 116), the functional control group (receiving bread croutons to match the calories in the control group, n = 120) and the daily probiotic group (supplemented with 15 billion colony-forming units of Lacticaseibacillus rhamnosus per day, n = 113). Overall, weight, waist circumference, metabolites and gastrointestinal symptoms did not differ significantly between groups37.

We identified which microbiome species were impacted significantly by the dietary interventions in the two cohorts. In the ZOE BIOME cohort, 57, 4 and 14 prevalent SGBs showed significant changes at the endpoint (Q < 0.01) for the prebiotic blend, probiotic and control arm, respectively (Fig. 4a). Among the species with a significant change in the prebiotic arm were beneficial fibre-degrading Bifidobacterium adolescentis (SGB17244), Bifidobacterium longum (SGB17248) and Blautia obeum (SGB4811)38,39,40, as well as butyrate-producing Agathobaculum butyriciproducens (SGB14993), Anaerobutyricum hallii (SGB4532) and Coprococcus catus (SGB4670)41,42. By contrast, the species Dysosmobacter welbionis (SGB15078), among the top unfavourably associated SGBs in our study, was decreased significantly by the same dietary intervention (Supplementary Table 25). In the ZOE METHOD cohort, we found 46 SGBs differed significantly in their relative abundance in the PDP arm, and only two in the control arm (Fig. 4b and Supplementary Table 25; Wilcoxon signed-rank test Q < 0.1). Of note, the prominent butyrate producers Roseburia hominis (SGB4936) and A. butyriciproducens (SGB14993) were also found to increase in the PDP intervention.

Fig. 4: Dietary interventions have a large impact on microbiome composition.

a,b, Pre–post dietary intervention variations in prevalent gut microbial SGBs (at least 10% at both time points). The plots show the effect size (log2-transformed ratio of mean relative SGB abundance at endpoint over baseline) against the significance (Q values, FDR–Benjamini–Hochberg-corrected P values). a, BIOME cohort (ClinicalTrials.gov NCT06231706) with n = 321 healthy adults from the UK (n = 106 prebiotic blend, n = 106 probiotic and n = 109 control), significance threshold set to *Q *< 0.01. b, METHOD cohort (ClinicalTrials.gov NCT05273268) with n = 347 US individuals (n = 177 PDP, n = 170 control), and significance threshold set to Q < 0.1. c, Change in relative abundance for the significant SGBs in the intervention arms of BIOME (prebiotic blend, n = 57). d, Change in relative abundance of METHOD (PDP, n = 46), separated into those that increase from those that decrease from baseline (B) to endpoint (E). Extended Data Fig. 10a–d reports the change in relative abundance and prevalence of the control and prebiotic arms. Two-sided Wilcoxon test; box plots as in Fig. 1).

The dietary intervention groups of both clinical trials that aimed at improving diet using different approaches (prebiotic blend for BIOME and PDP for METHOD) showed the highest number of significantly changing SGBs (Fig. 4a and Supplementary Table 25). Focusing on the most significant gut microbial SGBs with largest change in relative abundance after dietary interventions, we found increasing Bifidobacterium animalis (SGB17278)—a bacterium present in dairy-based foods and in the microbiome of people consuming larger amounts of them43,44 (Fig. 5a,b and Supplementary Table 25), an unknown Lachnospiraceae bacterium (SGB4953, BIOME; Fig. 5a) and R. hominis (SGB4936, METHOD; Fig. 5b) both previously associated with a vegan diet[43](https://www.nature.com/articles/s41586-025-09854-7#ref-CR43 “Fackelmann, G. et al. Gut