Adherence to Life’s Essential 8 enhances gut microbiota diversity and cognitive performance

This study included participants from the Framingham Heart Study (FHS) who had available data on LE8 metrics, gut microbiome sequencing, and cognitive assessments. The FHS is a longitudinal community-based study initiated in 1948, with multiple offspring cohorts enrolled in subsequent decades (Andersson et al., 2019; Fongang et al., 2023). Participants in our analysis were drawn from the 3^rd Generation cohort, New Offspring Spouses cohort, and the minority oversample Omni Group 2, who attended a clinic examination that included stool sample collection for microbiome analysis and comprehensive clinical evaluation (2016–2019). Eligible individuals were adults who provided a stool sample and underwent cognitive testing in approximate temporal proximity. All participants gave written informed consent, and the study protocol was approved by the Institutional Review Board of Boston University.

LE8 scores for each participant were computed according to the American Heart Association (AHA) guidelines for LE8, with minor adaptations for available FHS data (Lloyd-Jones et al., 2022; Carbonneau et al., 2023). Each of the eight health domains (diet, physical activity, nicotine exposure, sleep health, BMI, blood glucose, blood lipids, blood pressure) was calculated and scored on a scale from 0 to 100, with higher scores indicating better (healthier) status in that domain (Kannel and Sorlie, 1979; Mellen et al., 2008). Briefly, diet score was calculated using Dietary Approaches to Stop Hypertension (DASH) based on components including the intake of vegetables, fruits, nuts, legumes, and whole grains; as well as low-fat dairy, and the intake of red and processed meat, sugar-sweetened beverages, and sodium (Kannel and Sorlie, 1979; Mellen et al., 2008). The physical activity score was computed by considering the duration and strength of activities, including sleep, sedentary, and different strengths of activities (i.e., light, moderate, and vigorous) (Kannel and Sorlie, 1979). The other scores, such as sleep quality, blood sugar levels, BMI, blood lipids, and blood pressure, were derived based on the standard of the AHA (Lloyd-Jones et al., 2022). The overall LE8 score was estimated as the unweighted average of the eight component scores, also yielding a composite range from 0 to 100, with higher scores reflecting overall better cardiovascular health. In our cohort, the LE8 score distribution was approximately normal, with a mean of 76.3. Based on AHA-established cutoffs (Lloyd-Jones et al., 2022), participants were initially categorized into three adherence levels: low (score 0–49), moderate (50–79), and high (80–100). These thresholds align with AHA guidelines for poor, intermediate, and ideal cardiovascular health metrics. In our study sample, very few individuals (n = 13) fell into the low adherence category (LE8< 50). Therefore, for analytical purposes, we combined the low and moderate groups into a single category (“ModLow”) to ensure adequate group sizes. Thus, participants were finally classified into two categories: High LE8 adherence (score 80–100) and ModLow LE8 adherence (score< 80).

To provide insight into the composition of the LE8 score, we also examined how each individual component correlated with the overall score. We performed a Spearman correlation analysis between each of the eight component scores and the total LE8 score across participants. All eight components showed a positive correlation with the overall LE8 score (each correlation p-value< 0.05), indicating that improvements in any single domain contributed to a higher composite score (with correlation coefficients ranging from ρ ≈ 0.2 to 0.6) as displayed in. This supports the interpretation of the LE8 score as an integrated measure of cardiovascular health. 1

Stool samples were collected from FHS participants following standardized procedures to ensure sample integrity, as previously described (Walker et al., 2021; Fongang et al., 2023; Okoro et al., 2023). Briefly, participants were provided stool collection kits and instructed on proper sample collection. Samples were immediately frozen and stored at –80°C until analysis. Microbial DNA was extracted using the Qiagen PowerSoil DNA Isolation Kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. The V4 region of the 16S rRNA gene was amplified from each sample and sequenced on the Illumina MiSeq platform, generating paired-end reads (2 × 250 bp). Sequence data were processed using the DADA2 pipeline (Okoro et al., 2023) to infer Amplicon Sequence Variants (ASVs), followed by taxonomic assignment against the SILVA reference database (version 138). The processing steps included quality filtering, denoising, chimera removal, and merging of paired reads, resulting in high-quality sequences for analysis.

Across the 781 samples, sequencing depth was high: each sample yielded on the order of tens of thousands of reads (median approximately ~18,000 reads per sample). After DADA2 processing and filtering, we identified a rich microbial community comprising over 200 distinct genera, spanning 12 bacterial phyla in total. To focus on prevalent taxa and reduce sparsity, we excluded very rare ASVs/features; in particular, features not present in at least 10% of samples at a relative abundance of ≥0.1% were filtered out (this threshold aligns with the prevalence cutoff used in downstream association analyses). The remaining microbiome data included predominantly gut-associated phyla such as Firmicutes, Bacteroidetes, Actinobacteria, Proteobacteria, Verrucomicrobia, and others, and hundreds of genera, providing a robust foundation for diversity and association analyses.

To assess cognitive performance, participants completed a neuropsychological test battery administered by trained examiners during their FHS clinic visit, as previously described (Au et al., 2004). The current study included the following tests: Trail Making Test Part B (a measure of executive function and processing speed), Logical Memory (immediate and delayed recall, assessing verbal memory), Visual Reproduction (immediate and delayed recall, assessing visual memory), and Similarities (a subtest of the Wechsler Adult Intelligence Scale, measuring verbal abstract reasoning and executive function) (Elwood, 1991; Drozdick et al., 2018). Trail Making B times were log-transformed to normalize the distribution, and the sign was inverted so that higher values on all tests uniformly indicated better performance. We then constructed a Global Cognitive Score (GCS) by applying principal component analysis (PCA) to the battery of test scores (with each test standardized). The PCA yielded a first principal component (PC1) that accounted for the largest share of variance across the tests. This single component explained more than 42% of the total variance in the cognitive test battery, consistent with prior studies that have derived a general cognitive factor in community-based cohorts (e.g., ARIC, CHS, FHS offspring) (Pase et al., 2016; Pase et al., 2023; Yiallourou et al., 2024; Young et al., 2024). We used PC1 as the summary measure of global cognition for each participant; higher GCS values indicate better overall cognitive performance. By construction, the GCS is a weighted linear combination of the individual test scores (with loadings approximately reflecting each test’s correlation with the general factor). Thus, GCS captures broad cognitive function, combining elements of memory, executive function, and processing speed into a single metric. This GCS was treated as a continuous outcome in primary analyses. For secondary analyses (e.g., comparing microbiome features between cognitively “normal” vs. “poor” performers), we dichotomized the GCS: participants in the lowest quintile of GCS (bottom 20%) were classified as having “poor” cognitive performance, and those in the upper 80% were classified as “normal.” This grouping was used to illustrate differences in microbiome diversity and composition between individuals with relatively low vs. higher cognitive function.

Figure 1 illustrates our study framework, investigating how cardiovascular health (LE8 adherence) relates to cognitive performance and the potential mediating role of the gut microbiome. Our research addresses three key objectives: (1) LE8 and cognition: test the association between overall LE8 adherence (and its individual components) and global cognitive performance (GCS); (2) LE8 and gut microbiome: evaluate the relationship between LE8 adherence (overall and component-wise) and gut microbiome composition (diversity and taxa abundance); and (3) Gut microbiome and cognition: assess the associations between gut microbiome features and cognitive performance, and (4) specifically test whether the microbiome mediates the LE8–cognition association. We hypothesize that higher LE8 adherence will correlate with better cognition, and that a more favorable gut microbiome (greater diversity, higher abundance of beneficial taxa) will be associated with both high LE8 adherence and better cognition. Further, we hypothesize that some of the effect of LE8 on cognition is transmitted via changes in the gut microbiome, i.e., specific microbial taxa will show mediation effects in the LE8–cognition pathway.

The study cohort consisted of 781 participants (57.1% female) aged 32 to 78 years (mean age 54.9 ± 8.3 years). The majority (~64.4%) had a college-level education or higher. The overall LE8 score ranged from 27.5 to 100, with a mean of 76.3 (indicating generally intermediate-to-high cardiovascular health). Based on the LE8 score, 322 individuals (~41%) were classified as High LE8 adherence (score 80–100), and 459 (~59%) as ModLow LE8 adherence (score< 80, combining moderate and low categories, though only 13 were truly low). By design, the High adherence group had significantly better profiles on all individual LE8 components (each component score was higher on average, p< 0.001 for all). Participants in the high LE8 group also had a slightly lower mean age than those in the ModLow group (53.2 vs. 56.2 years, p = 1×10⁻⁶). Prevalence of hypertension treatment, diabetes, and cardiovascular disease was lower in the High adherence group, consistent with their healthier profiles (for example, diabetes: 1.7% vs. 98.3%, p = 2.2×10⁻¹⁶). Importantly, GCS was higher among participants with high LE8 adherence compared to those with moderate/low adherence. In unadjusted comparison, the High group’s median GCS was significantly greater than the ModLow group’s (0.71 in High vs. 0.47 in ModLow, p = 2.2×10⁻⁶), indicating better cognitive performance in the high cardiovascular health group. Table 1 summarizes the key demographic and health characteristics by LE8 group.

In terms of gut microbiome data, as noted in Methods, sequencing coverage was robust for all individuals. No sample had fewer than 5,000 reads; the median sequencing depth was around 18k reads per sample (Q1–Q3: ~12k–25k). A total of ~4,500 ASVs were inferred across all samples, clustering into >200 genera across 12 phyla. The most abundant bacterial phyla overall were Firmicutes and Bacteroidetes, followed by Actinobacteria and Proteobacteria (with smaller contributions from Verrucomicrobia, Fusobacteria, Lentisphaerae, etc.). The median Shannon diversity index was 3.7 (inter quartile range 3.3–4.1), indicating a fairly diverse gut microbiome on average. These descriptive statistics provide a context that our cohort’s microbiome diversity and composition are comparable to other adult populations. We next report the associations of LE8 adherence with gut microbiome features and subsequently their links to cognitive outcomes.

We first examined whether overall gut microbiome diversity differed by LE8 adherence. Alpha-diversity indices were significantly higher in the High LE8 adherence group compared to the ModLow group (Figure 2A). Specifically, the median Chao1 richness in the High group was significantly greater than in the ModLow group (Wilcoxon rank-sum p = 0.0014). Similarly, the Shannon diversity index was higher in the High LE8 group (median 3.9 vs. 3.6; p = 0.0071). These results indicate that individuals with better cardiovascular health (as captured by LE8) tend to harbor richer and even microbial communities in their gut. To ensure that these differences were not confounded by age or sex, we also performed a multivariable linear regression of each diversity measure on continuous LE8 score, adjusting for age and sex. The regression confirmed the association: each 10-point increase in LE8 score was associated with an estimated increase of +0.35 in Shannon index (β = 0.0345 per point, 95% CI 0.0208–0.0482; p = 9.88×10^–7) and +4.0 in Chao1 (β = 0.404 per point, 95% CI 0.303–0.505; p = 3.14×10^–9). Thus, both categorical and continuous analyses consistently show that higher LE8 adherence correlates with greater gut microbiome diversity.

We next assessed beta-diversity, which reflects differences in overall microbial community composition between groups. Figure 2B displays PCoA plots of Bray–Curtis dissimilarity, illustrating some separation between the two LE8 groups. Using PERMANOVA, we found that the High vs. ModLow LE8 groups have significantly distinct microbiome profiles (p = 1×10⁻⁴). This indicates that, considering the microbiome as a whole (all taxa abundance profiles), there is a detectable difference associated with cardiovascular health status. When we adjusted for age and sex in the PERMANOVA (essentially comparing residual microbiome differences after accounting for those factors), the group difference remained significant (PERMANOVA p = 1×10⁻³). These results suggest that the composition of the gut microbiome varies with LE8 adherence beyond what would be expected by chance or due to basic demographic differences. Key phylum-level differences included a trend for the High adherence group to have relatively more Firmicutes and Lentisphaerae and slightly less Proteobacteria, although overall, Firmicutes and Bacteroidetes dominated both groups (Supplementary Figure S2). At the family and genus levels, some taxa were noticeably more abundant in the High group, such as Ruminococcaceae and Faecalibacterium (a well-known beneficial genus in the Ruminococcaceae family), whereas certain taxa like Enterobacteriaceae (family containing some opportunistic pathogens) appeared more represented in the ModLow group (Supplementary Figures S2, S3).

As a sensitivity analysis, we repeated the diversity comparisons after excluding the few individuals with very low LE8 scores (< 50). This left a comparison of High vs. moderate adherence only. The results were consistent: alpha-diversity remained significantly higher in the High vs. moderate group (Chao1 p = 0.0078; Shannon p = 0.024; Supplementary Figure S4A), and beta-diversity differences were still significant (PERMANOVA p = 1×10⁻⁴; Supplementary Figure S4B). Thus, even when considering a more stringent contrast, better cardiovascular health was linked to a more diverse and distinct gut microbiome.

We next examined the associations between LE8 adherence and the relative abundances of specific gut microbes. Using multivariable linear models in MaAsLin2, we regressed each taxon’s abundance on the continuous LE8 score, adjusting for age and sex. We considered results significant at FDR q< 0.05. Figure 3 (and Supplementary Figure S5) presents heatmaps of the significant associations at phylum, family, and genus levels.

Our analysis identified several taxa whose abundance was significantly associated with LE8 scores. Higher LE8 scores (better cardiovascular health) showed positive associations with the abundance of several beneficial or commensal genera, including: Barnesiellaceae Barnesiella: β = 0.0263 (95% CI 0.0261–0.0266), p = 1.4×10⁻²⁹⁷. This indicates that individuals with higher LE8 scores tend to have a higher relative abundance of Barnesiella. The extremely low p-value reflects both a consistent direction and relatively small variance due to the large sample size and prevalence of this genus. Lachnospiraceae Ruminococcus: β = 0.0128 (0.0122–0.0134), p = 6.8×10⁻²⁹⁴. Higher LE8 adherence was associated with more Ruminococcus (a genus of fiber-fermenting bacteria in the Lachnospiraceae family). Ruminococcaceae Clostridium (unclassified Clostridium in the Ruminococcaceae family): β = 0.0583 (0.0401–0.0766), p = 2.9×10⁻⁹. Clostridiaceae Clostridium: β = 0.0283 (0.0092–0.0475), p = 0.013. It should be noted that different Clostridium groups appear in different families; some Clostridia can be beneficial, e.g., producers of butyrate. Ruminococcaceae Ruminococcus: β = 0.0091 (0.0018–0.0163), p = 0.041. Victivallaceae Victivallis: β = 0.0430 (0.0153–0.0707), p = 0.0095. These positive associations suggest that people with healthier lifestyles (higher LE8) have more of these potentially beneficial or fiber-degrading bacteria in their gut.

Conversely, higher LE8 scores were associated with lower abundances of several taxa that could be considered less beneficial or linked to poorer diet: Lachnospiraceae Clostridium (a specific genus in Lachnospiraceae, distinct from the Ruminococcaceae Clostridium above): β = –0.0521 (95% CI –0.0673 to –0.0368), p = 1.9×10⁻¹⁰. This genus was less abundant in high LE8 adherers. Ruminococcaceae Subdoligranulum: β = –0.0248 (–0.0378 to –0.0119), p = 9.9×10⁻⁴. Coriobacteriaceae Collinsella: β = –0.0216 (–0.0334 to –0.0009), p = 1.8×10⁻³. Collinsella (family Coriobacteriaceae) is a genus sometimes associated with high-fat diets and adverse metabolic profiles; it was found in lower abundance with better LE8 adherence. Erysipelotrichaceae Clostridium: β = –0.0261 (–0.0424 to –0.0097), p = 7.7×10⁻³. Streptococcaceae Streptococcus: β = –0.0301 (–0.0499 to –0.0102), p = 0.011. Some Streptococcus species can be pathobionts in the gut; their lower levels in high LE8 adherers might reflect healthier diet (lower simple sugars, for instance). Coriobacteriaceae Eggerthella: β = –0.0345 (–0.0601 to –0.0089), p = 0.026. Erysipelotrichaceae Catenibacterium: β = –0.0129 (–0.0136 to –0.0124), p = 7.2×10⁻²⁸². (This extremely significant result likely indicates nearly ubiquitous presence but lower proportion in high LE8; Catenibacterium has been associated with high-fat animal-based diets in some studies (Kilburn et al., 2020). These negative associations highlight that certain taxa were less abundant among those with better cardiovascular health.

In summary, the LE8–microbiome association analysis suggests a pattern where higher LE8 adherence (healthier lifestyle) corresponds to a gut microbiome enriched in families like Barnesiellaceae, Ruminococcaceae, Lachnospiraceae, and Victivallaceae, and depleted in some Erysipelotrichaceae, Coriobacteriaceae, Streptococcaceae, and specific Clostridium groups. Many of the taxa increased with high LE8 are known butyrate producers or associated with fiber-rich diets, whereas those decreased (like Collinsella, Eggerthella, Streptococcus) have been linked to less healthy dietary patterns or inflammation.

It is worth noting that these genera associated with continuous LE8 scores often belong to the same higher-level taxa that differentiate the LE8 groups. For instance, the Methanobacteriaceae, Victivallaceae, Barnesiellaceae, and Clostridiaceae families showed higher abundance in high LE8 adherers (as families), consistent with their member genera listed above. On the other hand, families like Streptococcaceae and Coriobacteriaceae were more abundant in the ModLow LE8 (and their genera Streptococcus, Collinsella mirrored this in the continuous analysis). We also observed that LE8 score was positively associated with the relative abundance of several major phyla: specifically, Firmicutes, Actinobacteria, Proteobacteria, and Lentisphaerae (the latter includes the family Victivallaceae). This indicates that broad compositional shifts (not just individual genera) underlie the differences in microbiome with respect to cardiovascular health.

To disentangle which aspects of the LE8 construct might be driving the microbiome associations, we performed secondary analyses examining each LE8 component in relation to the gut microbiome (Figure 3). Using similar MaAsLin2 models, we tested associations between microbial taxa abundance and each component score (diet, physical activity, smoking status, sleep health, BMI, blood glucose, total cholesterol, blood pressure), adjusting for age and sex in each case. Figure 3 (and Supplementary Figure S5) presents a summary heatmap of significant associations between LE8 component scores and microbial taxa (with effect directions). The results suggested that not all components were equally influential on the gut microbiome; a subset of components stood out: Diet score: A healthier diet (higher Dietary score, based on DASH-like diet adherence) was linked to increased abundance of fiber-fermenting bacteria (Ruminococcaceae, Lachnospiraceae genera) and decreased abundance of some pro-inflammatory taxa (Streptococcaceae, Streptococcus). This is unsurprising since diet is known to be a major determinant of gut microbiota. Physical activity: Showed trends of association with diversity, but relatively fewer individual taxa passed significance after multiple testing correction. Still, higher physical activity tended to align with microbes associated with leanness or metabolic health (e.g., Ruminococcus had a positive trend). Nicotine exposure (smoking): This was one of the top components influencing microbiota. Non-smokers or those with higher nicotine health scores (meaning no tobacco exposure) had higher levels of beneficial genera from Clostridium clusters (e.g., Ruminococcaceae Clostridium, Clostridiaceae Clostridium) and lower levels of pro-inflammatory Clostridium clusters (e.g., Lachnospiraceae Clostridium) and some Eggerthella. Smoking is known to alter gut microbiota, so this is consistent with prior knowledge. Sleep health: Better sleep (longer duration and quality, as per LE8 definition) was associated with higher abundance of certain SCFA producers. BMI (Body mass index): Although BMI is included in LE8, we considered it separately. Higher BMI (poorer score) correlated with microbiome signatures of obesity (e.g., reduced Firmicutes, Bacteroidetes, Actinobacteria, more Ruminococcaceae Clostridium, etc.), while a healthy BMI (ideal range) showed the opposite. These mirrored known obesity–microbiome patterns. Blood glucose, Blood pressure, Blood lipids: These clinical metrics had somewhat weaker individual associations with microbiome composition, though a few links emerged (e.g., higher blood glucose was related to more Collinsella, which aligns with literature on type 2 diabetes microbiome; higher blood pressure had associations with decreased Ruminococcus and Christesenella).

In summary, the top four LE8 components that showed the most prominent relationships with the gut microbiome were nicotine exposure, BMI, diet, and sleep health (Figure 3 and Supplementary Figures S6, S7). These four factors likely drive the bulk of the LE8–microbiome association. For example, the genus Barnesiella was positively associated with most of these healthy behaviors: it was higher with a healthy diet, in non-smokers, with ideal BMI, and good sleep as showed in Figure 4A. Suggesting that Barnesiella enrichment might be a common endpoint of a healthy lifestyle (except it showed no clear link with blood pressure). Similarly, a reduced abundance of Lachnospiraceae Ruminococcus was tied to worse scores in multiple components (especially sleep and smoking), while its higher abundance aligned with healthier behaviors (Figure 4B).

It appears that some bacteria respond to a general “healthy lifestyle” milieu (hence showing up in the composite LE8 analysis), whereas others are more domain-specific. For instance, Barnesiellaceae Barnesiella and Ruminococcaceae Clostridium are mostly influenced by LE8 components but might not respond to physical activity and blood pressure. Our component analysis provides a nuanced view: health behaviors (including diet, smoking, sleep) and obesity largely affect the abundance of common gut commensals linked to metabolic health, smoking influences a different set of taxa, perhaps via inflammation, and sleep could affect gut rhythms and certain microbe abundances.

These findings underscore that while the composite LE8 score captures an aggregate effect, the underlying drivers in relation to the microbiome are heterogeneous. Nevertheless, the overlap of taxa influenced by multiple healthy behaviors explains why we saw a robust overall LE8 effect.

We evaluated the direct relationship between LE8 adherence and cognitive performance in our cohort. Prior studies have suggested a link between better cardiovascular health metrics and better cognition, and our data allowed us to test this within a multivariable framework (Zhou et al., 2023; Clocchiatti-Tuozzo et al., 2024; Liang and Zhang, 2024).

We conducted multivariable linear regression analysis to examine the relationship between LE8 scores and GCS, adjusting for age, sex, and education. The analysis revealed that higher LE8 scores were significantly associated with increased GCS (β = 0.0047, (95% CI 0.0005–0.0089); adjusted p-value = 2.9×10⁻²), as depicted in Figure 5A. In a simpler sense, participants in better cardiovascular health tended to perform better on cognitive testing. Figure 5A illustrates this association, showing the regression line of GCS on LE8 score with confidence bands.

To complement the previous analysis, we also performed the same multivariable linear regression with as predictor the binary LE8 status (High vs. ModLow adherence) and the global cognitive scores as outcome. We noted that findings discriminated score of cognitive performance (GCS) from individuals with high adherence compared to those with moderate/low adherence (β = 0.0971 (0.0013–0.1956; p = 7.4×10⁻³), as displayed in Figure 5B.

Collectively, these results align with the hypothesis that maintaining ideal cardiovascular health (as summarized by LE8) is associated with better cognitive function. They set the stage for investigating whether the gut microbiome plays a role in this link.

We next turned to the connection between gut microbiome features and cognitive outcomes (independent of LE8). First, we evaluated microbiome diversity in relation to cognitive performance. After confirming GCS was approximately normal and dichotomizing as described, we found that participants with poor cognition (lowest quintile GCS) had significantly lower alpha-diversity compared to those with normal cognition. In, the Chao1 and Shannon indices are both lower in the Poor cognition group; for example, median Shannon was 3.5 in Poor vs. 3.7 in Normal (Wilcoxon p = 0.0066). Chao1 richness median was 149.2 vs. 160.1 (p = 0.0012). This suggests that lower cognitive performance was associated with a less diverse gut microbiome. We also performed adjusted linear models for diversity and GCS (with age, sex, BMI, education, time difference as covariates) and found a modest but significant positive association: e.g., each 1-unit higher GCS was associated with +0.07 higher Shannon index (p = 0.025). These results mirror the LE8-diversity findings, now in the context of cognition. 1

Beta-diversity analysis similarly indicated that the overall microbiome composition differed between the Normal vs. Poor cognition groups. PERMANOVA on Bray–Curtis distances gave p = 1×10⁻⁴ (unadjusted) for cognitive group differences, and p = 1×10⁻³ after adjusting for age, sex, BMI, education, and time interval. PCoA plots (Supplementary Figure S8B) show some separation of the two cognition groups, albeit with overlap (which is expected given that many factors influence microbiome beyond cognition). These findings imply a global association where cognitively poorer individuals, as a group, have a shifted gut microbiome structure relative to cognitively normal individuals. Supplementary Figures S9 and S10 provide stacked barplots of the microbiome composition in normal vs. poor cognition groups at broad taxonomic levels, illustrating these differences.

We then identified specific taxa associated with cognitive performance by using MaAsLin2 models as described in the Methods section. Figure 6 (with Supplementary Figures S11–S12) summarizes these associations. Several interesting patterns emerged: Higher (better) cognitive performance (GCS) was positively associated with the abundance of certain genera, many overlapping with those identified in the LE8 analysis: Lachnospiraceae Ruminococcus: Higher GCS correlated with higher abundance of this genus (β = 0.2523 (95%CI 0.2423–0.2623), p = 5.3×10⁻²⁸¹). This genus was also positively linked to LE8; here we see it also aligns with better cognition. Barnesiellaceae Barnesiella: β = 0.1219 (0.1182–0.1255), p = 4.4×10⁻²⁶⁷. Again, Barnesiella appears as a beneficial taxon correlated with both healthy lifestyle and better cognition. Ruminococcaceae Oscillospira: β = 0.0943 (0.0663–0.1223), p = 3.0×10⁻¹⁰. Oscillospira (often associated with leanness and high-fiber diet) was more abundant in those with higher GCS. Victivallaceae Victivallis: β = 0.6558 (0.1840–1.1277), p = 0.029. This genus was at higher level in the good cognition group as well.

In addition, at the family level, Methanobacteriaceae, Christensenellaceae, Rikenellaceae, Ruminococcaceae, and Victivallaceae all had positive associations with GCS (these families are typically considered part of a healthy gut microbiome profile in literature, often linked to low inflammation or positive metabolic traits). At the phylum level, better cognition was associated with higher relative abundances of Lentisphaerae and Firmicutes, and OD1 (Parcubacteria).

On the other hand, lower cognitive performance (poor GCS) was associated with higher abundances of some taxa: Erysipelotrichaceae cc_115: β = –0.0396 (−0.0579 to −0.0213), p = 1.3×10⁻⁴. This indicates higher cc_115 (an uncultured genus of Erysipelotrichaceae) in those with lower GCS. Lachnospiraceae Clostridium: β = –0.3984 (−0.6608 to −0.1361), p = 0.014. This same genus was higher in the ModLow LE8 score and appears again with poorer cognition. Ruminococcaceae Subdoligranulum: β = –0.2950 (−0.5155 to −0.0775), p = 0.037. Higher Subdoligranulum in poor cognition; interestingly this genus also decreased with higher LE8. Ruminococcaceae Anaerotruncus: β = –0.3881 (−0.6889 to −0.0871), p = 0.047. Anaerotruncus higher in poor cognition. Additionally, phyla OD1, Lentisphaerae, and Firmicutes had positive associations with GCS (meaning they were lower in poor cognition).

To summarize, the gut microbiome of individuals with better cognitive performance was enriched in SCFA-producing, fiber-degrading, and putatively anti-inflammatory taxa (like Ruminicoccus, Barnesiella, Oscillospira, Christensenellaceae family) and had fewer pro-inflammatory or dysbiosis-associated taxa (like certain Clostridium clusters). These patterns mirror those seen with LE8 adherence, which is suggestive of a common pathway: the same beneficial bacteria that are promoted by a healthy lifestyle might also promote or reflect better neurological health.

To drill down further, we explored whether certain bacteria were particularly associated with specific cognitive domains or tests (rather than the composite GCS alone). We conducted exploratory analyses relating taxa abundance to each of the four main cognitive test scores (Logical Memory, Visual Reproduction, Similarities, Trail Making B). These were adjusted for the same covariates. While a full description is beyond scope, a few noteworthy observations include: (a) The reduced abundance of Barnesiella (which was strongly linked with poor GCS) was specifically associated with lower scores on Similarities, Visual Reproduction, Logical Memory, and slower Trail Making B (i.e., all components of GCS). In other words, individuals with reduced abundance of Barnesiella tended to perform worse across multiple cognitive domains. (b) Lower abundance of Lachnospiraceae Ruminococcus was significantly associated with lower scores on Logical Memory and Similarities, and higher (worse) time on Trail making part B, though interestingly it was associated with higher Visual Reproduction in this analysis (perhaps reflecting a complex relationship or multiple strains). (c) Lachnospiraceae Clostridium and Ruminococcaceae Subdoligranulum, which were higher in poor performers, each showed negative associations with some cognitive tests (e.g., more Subdoligranulum linked to worse memory and executive function). (d) Some taxa had differential associations: for example, Bifidobacterium (Actinobacteria phylum) had a positive correlation with memory scores but wasn’t a top mediator in overall GCS.

These results (depicted partially in Figure 6) suggest that the microbiome-cognition link is not driven by a single domain but rather a general effect across cognitive functions. Barnesiella, in particular, stands out as consistently associated with better performance on all tests, reinforcing its potential role in overall cognitive health.

Finally, we formally tested whether the gut microbiome mediates the relationship between LE8 adherence and cognitive performance. From the above analyses, we identified overlapping taxa that were associated with both LE8 and GCS. These overlapping taxa are prime candidates for mediation. Notably, Victivallaceae Victivallis, Barnesiellaceae Barnesiella, Lachnospiraceae Ruminococcus, Lachnospiraceae Clostridium, Victivallaceae (unclassified genus), Methanobacteriaceae (family level), Lentisphaerae (phylum level), and Firmicutes (phylum level) were in this overlapping set. For each candidate, we conducted a mediation analysis using the approach described above. We report the mediation effect sizes (indirect effects, IDE) and significance in Figure 7 and Figure 8 (with details in Supplementary Figure S13 for all tested mediators).

The mediation analyses revealed that several bacterial taxa significantly mediated the association between LE8 adherence and GCS: Victivallaceae Victivallis: This genus had one of the strongest mediation effects. High LE8 adherence was associated with increased Victivallis, which in turn was associated with higher GCS. The indirect effect (IDE) was 0.0173 (95% CI 0.0143–0.0200), with p_boot< 0.001, indicating a significant mediation. Victivallaceae (unclassified family): IDE = 0.0166 (95% CI 0.0139–0.0197), p_boot< 0.001. Similar interpretation: part of LE8’s effect on GCS goes through changes in the Victivallaceae family. Barnesiellaceae Barnesiella: IDE = 0.0160 (95% CI 0.0130–0.0190), p_boot< 0.001. This suggests Barnesiella accounts for a notable portion of the LE8–cognition link; i.e., LE8 adherence might improve cognition partly by increasing Barnesiella abundance, which has neuroprotective associations. Lentisphaerae (phylum): IDE = 0.0144 (95% CI 0.0118–0.0170), p_boot< 0.001. Many of the Victivallaceae belong to Lentisphaerae, so this aligns with the above. Methanobacteriaceae (family): IDE also significant (~0.012, p_boot< 0.001); Methanobacteriaceae (includes Methanobrevibacter) were higher with LE8 and with good cognition. Firmicutes (phylum): Showed a smaller but significant mediation, indicating that broad community shifts in Firmicutes abundance could explain part of the effect. Lachnospiraceae Ruminococcus: This genus mediated a portion as well (IDE ~0.0065, p_boot< 0.001). Conversely, Lachnospiraceae Clostridium showed a negative link effect: higher LE8 was associated with lower Lachnospiraceae Clostridium, which was associated with higher cognition (since this genus is harmful), meaning that reducing this taxon is one pathway by which LE8 adherence might benefit cognition.

Figure 8 highlights the top four mediating bacteria by absolute effect size: Victivallis, Victivallaceae, Barnesiella, and Lentisphaerae. These each contributed an indirect effect on the order of 0.014–0.017 in GCS units. Given that the total effect of LE8 on GCS (per 10 points of LE8) was ~0.047, an indirect effect of ~0.016 suggests that roughly one-third of the LE8→cognition relationship could be explained by that single mediator. Cumulatively, considering all significant microbial mediators together, they potentially account for a substantial portion of the association (though they are not independent, so we refrain from simple summing).