How Gut Bacteria Linked to Obesity Affect Diet-Related Metabolism and Thinking in Older Adults

In this randomized parallel-arm pilot clinical trial, a total of 31 participants (mean age 69.7 ± 5.6 y; 24 females, 7 males) were recruited from diverse rural communities through collaborative efforts with local organizations and institutions, including medical offices, senior centers, and nursing homes. Participants were randomly assigned to receive dietary education on either a MD or a Modified Mediterranean Ketogenic Diet (MMKD). Participants were classified as cognitively normal (CN) or having possible MCI by a trained research assistant, supervised by a licensed clinical psychologist using the Montreal Cognitive Assessment (MoCA). Achieving a MoCA score between 17 and 26 was indicative of MCI; score of >27 as CN. In this sub-group analysis of a larger, ongoing registered clinical trial (NCT06121986↗), we have restructured our analytic approach to investigate the impact of obesity status, rather than diet group, on gut microbiome, cognition, and related biomarkers. Study design is shown in Figure 1a. Details of participant recruitment, randomization, and retention throughout the 10-week dietary education intervention are provided in Figure S1. This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki, the Belmont Report, and the guidelines of the International Committee of Medical Journal Editors (ICMJE) on the protection of research participants. The protocol was reviewed and approved by the Institutional Review Boards of Florida State University (IRB protocol #STUDY00003781). Written informed consent was obtained from all participants prior to their enrollment in the study. Participants were informed of the study objectives, procedures, potential risks, and their right to withdraw at any time without penalty. All data were anonymized to ensure confidentiality. The study was conducted at the Center for Translational Behavioral Science, housed within Department of Behavioral Sciences and Social Medicine in the Florida State University College of Medicine.

Participants were primarily recruited through collaborative efforts with local community organizations, as well as word-of-mouth and flyers. A standardized 30-minute telephone screening was completed with all potential participants, which included questions related to the eligibility criteria listed below: (1) participants aged between 55 and 85 y; (2) a MoCA score ranging from 17−26, indicative of possible MCI, or a score of 27 or higher for CN individuals; (3) expressed interest in the program; (4) stability in medical conditions; (5) stability in medication regimens as evaluated by the study clinician; and (6) the capacity to complete the required assessments throughout the duration of the study. The exclusion criteria included (1) a MoCA score of 16 or lower; (2) any diagnosis of a neurodegenerative disorder; (3) a history of head injury accompanied by loss of consciousness or presence of a brain lesion; (4) major psychiatric disorders, including psychosis, bipolar disorder, or issues related to substance abuse; (5) sensory impairments that may prevent the participant from engaging in the intervention; (6) serious medical risks as determined by the study clinician; (7) use of prescribed warfarin, insulin, antibiotics, or immunosuppressant medications; (8) utilization of anticonvulsant medications or other substances potentially affecting the central nervous system; (9) dietary restrictions, such as allergies to nuts or fish; (10) adherence to restrictive or specialized dietary regiments; (11) significant digestive disorders; 12) previous surgical interventions that could complicate dietary interventions; and (13) an inability to communicate effectively in English. Upon completion of the initial screening call, individuals identified as eligible participants were contacted to engage in the informed consent process.

Participants engaged in a 10-week program of weekly 1-hour group sessions via HIPAA-compliant Zoom, led by a Registered Dietitian and trained research assistant using Motivational Interviewing (MI) and Cognitive Behavioral Therapy (CBT) techniques. The initial 90-minute session was conducted in person and included a psychoeducational video, discussion, and nutrition education. The topics covered included: Nutrition fundamentals, MD or MMKD principles, strategies for addressing factors related to eating, physical activity, rest, and behavioral strategies for healthy lifestyle changes. All participants received a workbook containing meal plans, grocery lists, handouts, and tracking tools for food, and macronutrient intake, as well as access to an online forum designed to facilitate recipe sharing and provide additional social support beyond live group meetings, with pertinent information on accessing this forum disseminated through email correspondence from the laboratory. Overall, education in both groups emphasized high-quality, nutrient-dense foods and promoted fiber intake as key components to fostering adherence and long-term behavior change. These diets were not designed to illicit weight loss, rather promote whole-food dietary improvements and standardize nutrient quality across participants. Given this, rather than evaluate diet-specific mechanisms, the current analysis focused on obesity status and microbiome–cognition relationships, wherein outcomes were examined collectively across diet groups. Prior to the intervention, participants were asked to download a food tracking application (Nutritionix). All participants were required to document their dietary intake using the app over three days each week (two weekdays and one weekend) during the 10-week intervention period. Participants who had trouble with the app were provided 3-d food logs in their workbooks to use instead. Although participants were encouraged to log dietary intake on 3 d/week, inclusion in final dietary analyzes required at least one usable day per week. Participants without at least one usable day of dietary data each week were excluded from analysis. Dietary intake data were ultimately available for 22 participants.

Body weight and measures of body composition were measured at baseline and the 10-week endpoint. Body weight was measured in kilograms, and body composition was assessed using a multi-frequency bioelectrical impedance analysis (BIA) device (Mediana i55 Premium Body Composition Analyzer, Mediana Co., Ltd., South Korea), which is validated for research applications. The analyzer provided estimates of fat mass, lean body mass, skeletal muscle mass, and total body water. Measurements were performed according to manufacturer guidelines, with participants standing barefoot on the analyzer platform and holding the hand electrodes. Prior to assessment, participants were instructed to avoid strenuous exercise, food, and fluid intake for at least 2 hours to minimize variability. We employed BIA due to its portability, lower cost, and feasibility for repeated measures in community-dwelling older adults. We acknowledge that dual-energy X-ray absorptiometry (DEXA) provides more precise estimates of fat distribution and lean mass but was not feasible within the scope of this pilot study. Participants were asked to stand on the BIA machine barefoot, grasping the hand electrodes and ensuring proper alignment of their feet with the foot electrodes. The participant held this position for approximately 1 to 1.5 minutes while the machine computed the relevant data, including fat mass (kg) or expressed as a percentage (body fat percentage), muscle mass (kg), skeletal mass (kg), total body water (L), visceral fat, and subcutaneous fat. Body mass index was calculated as body weight (kg)/height in meters squared. Obesity was defined as having a BMI greater than or equal to 30.0 kg/m². Percent-based metrics (e.g., % skeletal muscle mass) were not available from the device output and could were not calculated post hoc.

Participants completed a series of selected cognitive assessments from the NIH Toolbox Cognition Battery, administered on iPads by a trained research assistant. Selected, standardized tests were administered at baseline and at 10-week follow-up. Tests were selected to evaluate multiple domains of cognitive functioning using validated, age-normed tools suitable for individuals aged 3–85. The battery included eight tests: 1. The Dimensional Change Card Sort Test (DCCS), which assessed cognitive flexibility and attention by having participants match images based on shifting dimensions. 2. The Face Name Associative Memory Exam (FName) Delay tested associative memory by requiring participants to recall 12 face-name pairs after a five-minute delay, measuring memory retention and retrieval. 3. The Flanker Inhibitory Control and Attention Task (Flanker) measured inhibitory control and attention across 20 to 40 trials, where participants must focus on a central target while ignoring peripheral distractions. 4. The List Sorting Working Memory Test (LSWM) evaluated working memory by having participants recall and sequence images of foods and animals by size. 5. The Picture Sequence Memory Test (PSMT) assessed episodic memory through the recall of picture sequences ranging from 6 to 18 images, awarding credit for each correctly ordered pair. 6. The Rey Auditory Verbal Learning Test (RAVLT) and 7. Rey Auditory Verbal Learning Delay (RAVLD) measured immediate memory and verbal learning by asking participants to recall a list of 15 unrelated words across three trials, followed by a five-minute delayed recall. 8. The Speed Matching Test evaluated processing speed by requiring participants to identify the image that differs from the others, with higher scores indicating faster cognitive processing. All cognitive assessments were scored using age-corrected standard scores, where a score of 100 reflects the national average. Scores of 115 or 85 represent one standard deviation above or below the mean, respectively.

Gut microbiome profiles were measured as per our previously described methods.53-58 Briefly, participants were provided with a fecal sample collection kit, along with detailed instructions for sample collection at baseline and the 10-week endpoint. Participants collected freshly voided fecal samples (~1-2 g) as per the kit instructions after each baseline and post visit. Samples were frozen at -80 °C immediately after being delivered to our Gut Biome Lab or picked up by lab research personnel within 8 hours of defecation, while being kept cold from the time of collection to delivery using pre-frozen ice-packs. High-quality genomic DNA was extracted from 200 mg of fecal material using the PowerFecal Pro DNA kit (Qiagen Inc., Valencia, CA), following the manufacturer’s instructions and quantified using a nanodrop spectrophotometer. To target the bacterial community, the V4 hypervariable region of the microbial 16S ribosomal RNA gene was amplified using barcoded universal primers 515F and 806 R, in accordance with the benchmark Earth Microbiome Project standard protocol (http://www.earthmicrobiome.org↗). The resulting amplicons were purified using AMPure® magnetic purification beads (Agencourt) and quantified using a Qubit 4 fluorometer (Invitrogen). Equimolar concentrations of each library were then pooled, and the final amplicon library was subjected to paired-end sequencing (2 × 250 bp) on an Illumina MiSeq platform using the MiSeq Reagent Kit v3 (Illumina Inc., San Diego, United States). The QIIME2 (Quantitative Insights into Microbial Ecology) ver. 2-2024.10 software package (www.qiime2.org↗) was used for microbiome analysis. Raw sequencing reads were demultiplexed and quality-filtered using the q2-demux plugin, followed by trimming and denoising with DADA2.59 All resulting amplicon sequence variants (ASVs) were aligned using MAFFT.60 Taxonomic classification of ASVs was performed using the sklearn classifier and a pre-trained naïve Bayes taxonomy model, trained on the 99% SILVA ver. 138 (data trained: May 30, 2024; Sklearn version 1.4.2).61 ASVs were then collapsed to the genus level, and all subsequent analyzes were performed at the genus level.

Fecal and plasma metabolomes were measured using the global untargeted approach, as per our previously described methods.56-58^,62 Briefly, for plasma metabolome, fasting blood samples were drawn at baseline and the 10-week endpoint visit by a trained phlebotomist using a lithium heparin tube. This tube was inverted 6-8 times post-collection and subsequently spun in a cold centrifuge for 15 minutes at 2800 rpm. Plasma was then pipetted into 0.5 ml microtubes and immediately frozen at –80 °C, and stored without additional freeze–thaw cycles until subsequent metabolomic analysis. Metabolomic profiles were measured using a global untargeted approach via a high-throughput Nuclear Magnetic Resonance (NMR) system. Stool samples were processed according to a previously established protocol,63 with minor modifications. Briefly, samples were extracted by vortexing with deionized water for five minutes. The resulting extracts were mixed with a phosphate buffer (pH 7.4) in D₂O to yield a final solution containing 10% D₂O, 0.1 M phosphate buffer, and 0.1 mM trimethylsilyl propionate (TSP). After centrifugation, the supernatants were transferred into 5 mm NMR tubes and analyzed using a Bruker Ascend 400 MHz high-resolution NMR system (Bruker BioSpin, Germany). For all samples, a 1D NOESY experiment (noesygppr1d) with water suppression was performed, using 64 scans. NMR data were processed in TopSpin 4.06 (Bruker BioSpin) with phasing and referencing to TSP. Spectral analysis was carried out using Amix 4.0 (Bruker BioSpin), where automated binning was applied to reduce peak overlap and splitting, as previously described.64 Metabolites were quantified using Chenomx 8.6 (Chenomx Inc.), and total intensity normalization was performed prior to downstream data analysis.

Statistical analyzes were performed using the SPSS Statistics 29.0 Software Package (SPSS Inc.). All data were initially assessed for normality using the Shapiro-Wilk test. For variables that violated normality assumptions (p < 0.05), non-parametric tests were employed. Wilcoxon matched-pairs signed-rank tests (two-tailed) were used to assess within-group changes from baseline to endpoint, while Mann–Whitney U tests (two-tailed) were used to compare differences between the obese and non-obese groups at either baseline or endpoint. Relative abundance comparisons of microbial features at the phylum, order, class, family, and genus levels were conducted using non-parametric tests to detect differentially abundant taxa between the obese and non-obese groups. Alpha-diversity metrics, including Chao1 (richness) and Shannon Index (richness and evenness), were calculated based on ASVs. In metabolomic analysis, data were log2-transformed to express fold changes between pre- and post-intervention levels, where the log2 fold change was calculated as log2 (Post/Pre). A positive value indicated increased metabolite levels post-intervention, while a negative value indicated decreases, and a value of 0 indicated no change. The significance of changes in metabolite concentrations was assessed using Wilcoxon signed-rank tests, with Benjamini-Hochberg false discovery rate (FDR) correction applied for multiple comparisons. To assess relationships between continuous variables, Spearman’s rank-order correlation was used, given that the data did not meet the assumptions for parametric correlation techniques. Analyzes compared participants with and without obesity and assessed within-group baseline to post-intervention changes; therefore, Mann Whitney U tests were used for between-group comparisons and Wilcoxon signed-rank tests for paired comparisons. Repeated measures ANOVA or Friedman tests were not applied, as the study did not evaluate diet-group effects or time-by-group interactions and included only two time points. Linear regression was employed to examine the strength and direction of significant associations between continuous outcome variables, while logistic regression was used for categorical outcomes. Additionally, Analysis of Covariance (ANCOVA) was applied to evaluate group differences and control for potential confounding variables, such as age and baseline cognitive performance. Potential covariates were evaluated using a priori based on their established relevance to cognitive performance, body composition, and metabolic outcomes in aging research. Candidate covariates included age, sex, years of education, baseline cognitive scores, diet group assignment, and MCI status. Each covariate was first tested individually using univariate screening: Spearman correlations were used for continuous covariates and Mann Whitney U tests for categorical variables. Covariates showing associations with any outcome at p < 0.10 were retained for further evaluation. Multicollinearity among candidate covariates was assessed, and variables demonstrating high intercorrelation (r > 0.70) were not entered simultaneously to avoid redundancy. Final ANCOVA and regression models included only covariates that were both statistically relevant in univariate. Additionally, to evaluate the longitudinal and group-specific effects of the dietary intervention on metabolite concentrations, we conducted three-way mixed ANOVAs with Diet (MD vs MMKD) and Obesity status (obese vs non-obese) as between-subject factors and Timepoint (baseline vs endpoint) as a repeated within-subject factor. This model generated main effects (Diet, Obesity, Timepoint) and interaction terms (Diet × Timepoint, Obesity × Timepoint, Diet × Obesity, and Diet × Obesity × Timepoint). These analyzes allowed us to assess whether and how metabolite trajectories differed over time, varied by obesity status, or were modified by diet. Dietary intake data was calculated by taking the average reported intake of all nutrients over the course of the entire 10-week intervention. Analyzes were conducted on available data; no data imputation was performed. Participants with missing outcome or covariate data were excluded from the relevant analyzes, and sample sizes are reported accordingly Statistical significance was set at α = 0.05 for all analyzes.

At baseline, demographic characteristics were similar between participants with obesity and those without obesity (Table 1). However, obesity status demonstrated a moderate association with baseline cognitive status. While not reaching statistical significance, there was a trend indicating that participants with obesity were more likely to be classified as MCI than their non-obese counterparts. Using Fisher’s exact test, the association approached significance (p = 0.076). Specifically, 71% (12/17) of obese participants met criteria for MCI, whereas only 36% (5/14) non-obese participants (36%) were classified as such.

As anticipated, participants with obesity exhibited significantly higher values across a range of anthropometric and body composition metrics compared to non-obese participants (Table 2). Specifically, participants with obesity had significantly greater body weight (p < 0.01) and body fat percentage (p < 0.01), as well as elevated levels of visceral fat (p < 0.01) and increased subcutaneous fat (p < 0.001), indicating both central and peripheral adiposity. In addition, those with obesity demonstrated significantly higher total body water (p < 0.05), muscle mass (p < 0.05), and skeletal muscle mass (p < 0.05), likely reflecting their greater overall body size and weight-bearing demands. These findings are consistent with established physiological differences between obese and non-obese individuals and confirm the classification criteria used in this study. In terms of cognitive performance, participants with obesity scored lower than participants without obesity across five out of eight tasks of the NIH Toolbox Cognition Battery (DCCS, FName, Flanker, RAVLT, and RAVLD, all p < 0.05), which assessed a broad spectrum of neurocognitive domains including cognitive flexibility/attention, associative memory, inhibitory control, and immediate memory and verbal learning, respectively (Figure 1b, Table 3). Among these, two tasks revealed statistically significant between-group differences: Flanker (p < 0.05), which assesses selective attention and inhibitory control, and RAVLT (p < 0.01), a measure of verbal learning and memory. These results suggest that obesity may be associated with diminished cognitive functioning, particularly in executive function and memory domains.

Obesity status was significantly associated with alterations in gut microbiome. Participants with obesity exhibited reduced alpha-diversity, as indicated by significantly lower Chao1 richness indices (p < 0.05) and Shannon diversity (p < 0.05) suggesting a less diverse and potentially less resilient gut (Figure 1c,d). In addition, beta-diversity analysis using Bray-Curtis dissimilarity revealed distinct clustering of microbial communities between obese and non-obese cohorts, reflecting significant compositional differences in microbial taxa at the community level (Figure 1e). Taxonomic analysis at the genus level further revealed specific microbial shifts associated with obesity. Relative abundance (%) of genus Bacteroides was significantly higher in obese participants compared to non-obese counterparts (p < 0.01) (Figure 1f), while several beneficial or health-associated taxa were significantly reduced in the obese group. These included Prevotella (p = 0.03), Alistipes (p < 0.05), Christensenellaceae_R-7 group (p < 0.05), Oscillospiraceae;g_UCG-002 (p < 0.05), and Akkermansia (p < 0.01), (Figure 1g,k). Many of these taxa have previously been implicated in maintaining gut barrier integrity, modulating host metabolism, suggesting that their depletion may contribute to the metabolic and cognitive consequences observed in obesity. When participants were stratified by cognitive status, further patterns emerged. Among individuals with MCI, obese-MCI participants had significantly higher levels of Bacteroides (p < 0.05) and significantly lower levels of Akkermansia (p < 0.05) compared to non-obese-MCI participants. In contrast, among CN participants, most taxa differences were attenuated; only the Christensenellaceae_R-7 group remained significantly reduced in the obese-CN group (p < 0.05), suggesting this taxon may be particularly sensitive to obesity status independent of cognitive classification. At baseline, fecal and plasma metabolite concentrations did not differ significantly between obese and non-obese participants. However, several metabolites demonstrated trends toward differential abundance that may warrant further investigation in larger cohorts. Notably, plasma glutamine levels were higher in the obese group compared to the non-obese group, approaching statistical significance (p = 0.085). Conversely, plasma pyruvate levels tended to be lower in the obese group (p = 0.072).

Throughout the 10-week intervention, there were no significant between-group differences in total nutrient intake, indicating similar levels of dietary adherence among participants with and without obesity (Table 4). However, dietary pattern analyzes revealed that participants with obesity reported higher intake of eggs and lower intake of nuts compared to the non-obese group, suggesting minor but potentially meaningful differences in food choices that may reflect dietary preferences or habitual patterns. Mediterranean Diet Adherence Screener (MEDAS) scores were comparable between groups (obese: mean = 5.93; non-obese: mean = 6.00), suggesting similar overall adherence to a mediterranean style diet. A more detailed comparison of nutrient and food group intake by obesity status is provided in Supplementary Table 1 (S1).

At the end of the 10-week intervention, the obese group continued to exhibit significantly higher values across a range of anthropometric and body composition measures compared to their non-obese counterparts. These included body weight (p < 0.01), muscle mass (p < 0.01), body fat mass (p < 0.05), total body water (p < 0.01), visceral fat (p < 0.01), skeletal muscle mass (p = 0.01), and subcutaneous fat (p < 0.01). In addition, post-intervention assessments revealed new between-group differences in protein content (p < 0.05) and mineral content (p < 0.05), with higher values in the obese group. These findings are consistent with baseline differences and reflect the persistence of obesity-related body composition characteristics following the intervention period. Both groups experienced significant within-group weight loss over the course of the study. Participants with obesity lost an average of 5.95 kg (pre vs. post, p < 0.05), whereas non-obese participants lost an average of 1.22 kg (pre vs. post, p < 0.01) (Figure 2a). A new variable was created to represent weight change over the 10- week intervention (post weight (kg)—pre weight (kg)). Using non-parametric Mann Whitney U test, grouping by obesity status, it was found that the obese group numerically lost more weight compared to the non-obese group, although this result was not statistically significant (p = 0.057). Despite this difference in absolute weight change, within-group comparisons revealed no statistically significant pre vs. post changes in specific body composition metrics for either group, including fat mass and skeletal muscle mass. This suggests that while modest weight loss occurred, it was not accompanied by significant shifts in any specific compartmental body composition over the 10-week period. It is worthy to note that weight loss was not a goal or hypothesized outcome of the study intervention in either group.

Depicted in Table 3, cognitive performance improved over the intervention period in both groups, with within-group gains observed in specific cognitive domains. The Wilcoxon signed-rank test was used to assess pre- to post-intervention changes in test scores for the obese and non-obese groups separately. Among obese participants, significant improvements were observed on the FName (p < 0.01) and the RAVLT (p < 0.05), indicating enhancements in associative and verbal memory, respectively. A trend toward improved processing speed was noted on the Speed Matching task (p = 0.099). In the non-obese group, significant gains were found on the FName task (p < 0.05) and the PSMT (p < 0.05), suggesting improvements in episodic memory and associative learning. Between-group comparisons revealed that the Flanker (p < 0.05) and RAVLT (p < 0.05) scores remained significantly lower in the obese group compared to the non-obese group, indicating that obesity may have a persistent impact on these cognitive measures (Figure 2b). Notably, both tests showed lower scores at baseline in the obese group, suggesting that obesity could have an ongoing detrimental effect on attention, inhibition, and memory performance, which are assessed by the Flanker and RAVLT tasks, respectively. To further explore this, a MANCOVA was conducted to examine whether obesity status was associated with post-intervention cognitive performance (Flanker and RAVLT scores), controlling for education level, MCI status, and baseline test performance. After adjusting for covariates, obesity status did not have a significant effect on Flanker performance, F(1, 20) = 0.090, p = .768, η² = .004, or RAVLT performance, F(1, 20) = 0.043, p = .838, η² = .002. Baseline performance was a significant predictor for both Flanker (F(1, 20) = 20.595, p < .001, η² = .507) and RAVLT (F(1, 20) = 6.593, p = .018, η² = .248) outcomes. No other covariates were significant.

Among obese participants, body composition changes were significantly associated with improvements in specific cognitive domains. Consistent with the Spearman correlation results, which indicated that greater weight loss was associated with larger improvements in PSMT performance (negative ρ reflects scoring direction) (ρ = –0.642, p < 0.05; Figure 2c), the ANCOVA analysis further supported this association. After adjusting for age, education, and baseline PSMT scores, weight loss remained a significant predictor of improved episodic memory in the obese group (F(1,7) = 8.776, p < 0.05), but not in the non-obese group. These findings suggest that larger reductions in body weight were linked to greater memory gains specifically among participants with obesity. Reductions in body fat percentage were significantly associated with improved performance on the DCCS task, which assesses cognitive flexibility and executive function (ρ = –0.676, p < 0.05) (Figure 2d). To expand upon these correlational results, a linear regression analysis was conducted to explore the relationship between changes in body fat mass and performance on the DCCS task separately for obese and non-obese participants. In obese participants, reductions in body fat mass were significantly associated with improvements in DCCS performance (β = -0.660, p < 0.05). Specifically, for each one-unit decrease in body fat mass, DCCS scores were predicted to increase by 0.951 points. This negative relationship suggests that reductions in body fat mass were linked to enhanced cognitive flexibility in this group. These findings suggest that improvements in body composition may be linked to enhanced cognitive outcomes in metabolically compromised individuals. In contrast, no such associations were observed in non-obese participants. Paradoxically, in this group, increased body fat was positively correlated with improved PSMT scores (ρ = 0.567, p < 0.05) (Figure 2e), and a non-significant trend was noted with FName performance (ρ = 0.512, p = 0.089) (Figure 2f). Age-stratified analyzes revealed differential effects of body composition on total cognitive change. Among older participants (aged 73–80), increases in skeletal muscle mass significantly predicted total cognitive change (B = 19.544, SE = 3.872, t = 5.047, p < 0.01) (Figure S2), supporting the hypothesis that preservation or gain of lean mass in advanced age may have neurocognitive benefits. In contrast, this relationship was not observed in the younger age stratum (aged 58–72), where skeletal muscle change was unrelated to total cognitive change (p = 0.851). These findings suggest that the link between muscle and memory may become more pronounced with aging, possibly reflecting the compounding impact of sarcopenia and metabolic frailty on brain health. Although, these patterns represent correlational relationships only and do not imply that changes in adiposity directly caused changes in cognitive performance.

After the 10-week nutrition education intervention, differences in gut microbiome between participants with and without obesity diminished for Shannon diversity (Figure 3a). In contrast, microbial richness measured by the Chao1 index remained significantly lower in the obese group (p < 0.05), indicating a persistent deficit in species richness (Figure 3b). Differences in beta-diversity based on Bray–Curtis dissimilarity also diminished, showing reduced separation between groups after the intervention (Figure 3c). This suggests that obesity may exert lasting effects on specific aspects of gut microbial composition, particularly species richness, despite dietary modulation. At baseline, the Firmicute/Bacteroidota (F/B) phylum ratio for the obese group was 2.18 (F/B: 65.1%/29.9%), while the non-obese group had a slightly higher ratio of 2.94 (F/B: 69.9%/23.8%). After a 10-week dietary education intervention, the obese group showed a notable increase in the F/B ratio, which rose to 3.52 (F/B: 73.6%/20.9%). This increase reflects a reduction in the relative abundance of Bacteroidota and a corresponding increase in Firmicutes. Conversely, the non-obese group showed a slight decrease in the ratio, which dropped to 2.75 (F/B: 66.4%/24.1%), suggesting a modest shift toward greater Bacteroidota relative to Firmicutes.

To assess whether metabolite trajectories differed by diet and obesity status over time, we conducted three-way mixed ANOVAs with diet (MD vs MMKD) and obesity status (obese vs non-obese) as between-subject factors and Timepoint (baseline vs endpoint) as the repeated within-subject factor. Full ANOVA outputs for all metabolites are provided in Supplementary Table 2 (S2). Only two metabolites, fecal lactate and fecal choline, showed a significant diet × obesity × timepoint interaction, indicating that baseline-to-endpoint changes in these metabolites depended jointly on diet assignment and obesity status. No other metabolites demonstrated this three-way interaction.

Changes in the gut microbiome at the genus level in participants with and without obesity are shown in Figure 3d and 3e, respectively (phylum level changes are depicted in Figure S3 and S4). Associations between gut microbial taxa and both body composition and cognitive function were also observed and varied by obesity status. Among individuals with obesity, increases in Christensenellaceae_R-7_group abundance showed a trend with improved PSMT scores (ρ = 0.584, p = 0.059) (Figure 3f). Reductions in Bacteroides were associated with both lower carbohydrate intake (ρ = -0.733, p < 0.05) and greater loss of body fat (ρ = -0.721, p < 0.05) (Figure 3g). Furthermore, a binary logistic regression was conducted to examine whether the relative abundance of Bacteroides could predict obesity status while controlling for MCI status. The model was statistically significant for Bacteroides abundance, indicating that it was a significant predictor of obesity (B = –10.301, SE = 4.598, Wald = 5.019, p < 0.05), suggesting that higher Bacteroides abundance was associated with substantially reduced odds of being classified as obese. MCI status was not a significant predictor (B = 1.115, SE = 0.925, Wald = 1.456, p = .228), though the odds ratio (OR = 3.051, 95% CI [0.498, 18.680]) indicated a possible trend toward increased obesity risk among participants with MCI. In the non-obese group, increased energy intake was inversely associated with levels of Akkermansia (ρ = –0.709, p < 0.05) and Christensenellaceae_R-7_group (ρ = –0.624, p = 0.054). Notably, increased abundance of Oscillospiraceae;g_UCG-002 was strongly correlated with improved PSMT performance (ρ = 0.765, p < 0.01) (Figure 3h), highlighting a potential role for this taxon in modulating cognitive function. To explore whether changes in key gut microbial taxa could predict cognitive improvements, multiple linear regression models were constructed within the obese subgroup. A model incorporating changes in Bacteroides, Alistipes, Akkermansia, and Christensenellaceae_R-7_group as predictors of episodic memory (as measured by the PSMT) identified Christensenellaceae_R-7_group as a marginal, but not significant contributor (B = 620.239, t = 2.093, p = 0.081).

Distinct relationships between metabolic, microbial, and cognitive changes also emerged when stratifying participants by obesity status. To evaluate overall cognitive change, we computed a composite “total cognitive change” score for each participant by summing age-corrected standard scores from all eight NIH Toolbox cognitive tests at baseline and again at week 10. The difference between the post-intervention and baseline totals (Week 10 total—Baseline total) was used as the primary indicator of global cognitive change, with positive values indicating improvement. In the obese group overall, cognitive improvement was significantly and inversely correlated with fecal acetate (ρ = –0.857, p < 0.01) (Figure 4a). Beyond global cognitive change, specific metabolites showed domain-specific associations. For example, fecal 1,3-dihydroxyacetone was positively associated with performance on the Rey Auditory Verbal Learning Test (RAVLT; ρ = 0.75, p < 0.05) (Figure 4b), implicating it in verbal memory.

Similarly, fecal methanol was associated with improved scores on both the Flanker task (ρ = 0.686, p < 0.05) (Figure 4c). Plasma citric acid was another metabolite positively associated with Flanker (ρ = 0.901, p < 0.01) and FName performance (ρ = 0.764, p < 0.05), linking systemic energy metabolism with cognitive function (Figure 4d,e). Plasma pyruvate also showed strong correlations to improved performance on those two tests (p < 0.05). Fecal propionate was also tied to better cognitive flexibility as measured by the DCCS test (ρ = 0.745, p < 0.05) (Figure 4f). Finally, plasma acetate, alanine, and isoleucine were all tied to improved RAVLD scores (p < 0.05). In the non-obese group, a distinct pattern emerged, with improved total cognitive change being positively associated with fecal methionine (ρ = 0.764, p < 0.01) (Figure 4g), with a similar association between plasma pyruvate and Flanker scores seen again, as previously found with the obese group (ρ = 0.765, p < 0.05). A comprehensive heatmap depicting the correlational patterns between Log2 fold changes in fecal and plasma metabolites versus change total cognitive scores and domain specific scores is shown by Figure 4h. For completeness, we also generated an unstratified heatmap depicting the correlational patterns between Log2 fold changes in fecal and plasma metabolites and all outcome domains across the entire cohort (Figure S5). This figure is provided to complement the obesity-stratified analyzes presented in the main results.

In obese participants, several metabolites demonstrated significant associations with changes in body composition following the intervention (Figure 5). Plasma acetate emerged as a strong predictor of weight loss (B = 27.439, SE = 4.833, 95% CI [6.643, 48.236], t = 5.677, p < 0.05) (Figure 5a). The standardized beta (β = 1.049) indicates a strong positive association between plasma acetate and weight loss. Additionally, reductions in plasma lactate (ρ = –0.893, p < 0.01), and increases in plasma alanine (ρ = 0.821, p < 0.05), and fecal histamine (ρ = 0.717, p < 0.05) were each significantly correlated with weight loss, suggesting a multifaceted metabolic response involving both systemic and gut-derived compounds. A suite of amino acids—including fecal alanine, isoleucine, leucine, lysine, valine and methionine—also positively correlated with increases in skeletal muscle mass (ρ = 0.700 to 0.767) and inversely correlated with changes in body fat percentage (ρ = –0.617 to –0.770) (Figure 5b,e). Notably, leucine, isoleucine and lysine further exhibited positive associations with mineral content, suggesting broader implications for musculoskeletal integrity. Fecal propionate demonstrated an inverse association with body fat (ρ = –0.770, p < 0.05) (Figure 5f). In contrast, within the non-obese group, no significant relationships were observed between amino acid changes and body composition besides one opposite association between elevated fecal lysine and increased body fat percentage, which contrasts to obese participants. However, weight loss in this group was linked to a reduction in plasma acetate (ρ = 0.685, p < 0.05), mirroring findings in the obese cohort and reinforcing acetate’s relevance to metabolic shifts across phenotypes (Figure 5g). A heatmap depicting the correlational patterns between Log2 fold changes in fecal and plasma metabolites versus change in body composition metrics is shown by Figure 5h.

Several key microbial-metabolite relationships emerged, revealing coordinated shifts in the gut ecosystem associated with functional changes in host metabolism and cognition. Among obese participants, the genus Agathobacter consistently displayed positive correlations with fecal acetate (ρ = 0.778, p < 0.05) (Figure 6a), and inverse relationships with plasma lactate (ρ = −0.795, p < 0.05) (Figure 6b), suggesting that this SCFA-producing genus may serve as a central hub in microbiome-metabolome interactions and associated metabolite production and signaling. In non-obese participants, trimethylamine (TMA) showed opposing associations with the bacterial genera Bacteroides and Oscillospiraceae;g_UCG-002, being strongly positively correlated with Bacteroides abundance (ρ = 0.967, p < 0.001) (Figure 6c) and negatively correlated with Oscillospiraceae;g_UCG-002 (ρ = –0.683, p < 0.05) (Figure 6d). A heatmap depicting the correlational patterns between Log2 fold changes in fecal and plasma metabolites versus change in taxa (relative abundance %) is shown by Figure 6e.

Our data showed that participants with obesity demonstrated lower performance on multiple cognitive tasks, with significant impairments observed in executive function (Flanker) and verbal memory (RAVLD). These results align with existing literature linking obesity to diminished cognitive control and memory performance, particularly in aging populations.66^,67 Interestingly, when participants with MCI were excluded, the observed cognitive deficits in the obese group were no longer significant. This attenuation suggests that the impact of obesity on cognitive performance may be mediated or amplified by MCI status, indicating that obese individuals may be more vulnerable to cognitive decline and that obesity could be a contributing risk factor in the progression from normal cognition to MCI. Future longitudinal studies should investigate whether obesity accelerates cognitive decline over time, and whether early interventions in obese but cognitively normal individuals can prevent or delay MCI onset. Obesity was associated with reduced gut microbial alpha-diversity and distinct community structures, consistent with prior findings.68 These changes reflect a less diverse and potentially less resilient microbiome, which may contribute to impaired gut barrier integrity and increased systemic inflammation, both of which are implicated in the pathogenesis of cognitive decline, although not directly measured in this study. Specifically, the obese group showed reduced abundance of several taxa associated with metabolic and neurological health, including Prevotella, Alistipes, Christensenellaceae R-7 group, Oscillospiraceae UCG-002, and Akkermansia. A 2022 meta-analysis found strikingly similar results to our study, showing that obese individuals have significantly altered microbiomes, with a reduced abundance of 23 genera, such as Christensenellaceae_R-7_group, Alistipes, and Akkermansia.69Akkermansia muciniphila is a mucin-degrading bacterium known for its role in maintaining gut barrier integrity and regulating host metabolism. It has emerged as a promising biomarker and therapeutic target in obesity and metabolic syndrome.70 By consuming mucin, Akkermansia stimulates mucus layer renewal, reinforcing gut barrier function and preventing endotoxemia and influences glucose and lipid metabolism through pathways involving GLP-1, adiponectin, and AMPK activation.70 Although once considered a next-generation probiotic, recent literature suggests its impact on host health depends on host genetics, strain-specific properties, and interactions with other microbes.71^,72 Reductions in Prevotella, Oscillospiraceae (g_UCG-002), and Alistipes suggest diminished microbial functionality relevant to metabolic balance and gut-brain signaling. Prevotella efficiently ferments complex polysaccharides into propionate, which has been associated with improved insulin sensitivity and reduced inflammation.73 Lower Prevotella abundance may reflect decreased dietary fiber intake or impaired microbial capacity to metabolize plant-based nutrients, both of which can contribute to metabolic dysfunction.74 However, its role appears context dependent. Prevotella copri has been positively correlated with obesity and fasting glucose in children,73 and linked to poor glycemic outcomes and low vegetable intake in adults with type 2 diabetes.75 These contradictory findings highlight the complex, diet- and host-dependent nature of Prevotella, which may be beneficial in high-fiber diets but detrimental under metabolic stress. Alistipes is generally considered beneficial in obesity contexts, with roles in bile acid metabolism, anti-inflammatory signaling, and tryptophan pathways tied to serotonin and gut-brain communication.76 Collectively, the diminished presence of these taxa in obese individuals may impair microbial resilience and compromise essential signaling pathways that protect against both metabolic disease and cognitive decline. A recent meta-analysis of 16 studies also found that Alistipes was significantly reduced in the guts of obese individuals.77 In contrast, Bacteroides was elevated in obese participants, particularly those with MCI. While species within this genus can have beneficial or pathogenic roles, elevated levels have been associated with Western diets rich in fat and animal protein and may contribute to inflammation and reduced SCFA production. The overrepresentation of Bacteroides in the obese-MCI group suggests a potential interaction between metabolic and microbial dysregulation in cognitive decline.78 Previous studies have linked Bacteroides to MCI and AD: one RCT found an association between B. thetaiotaomicron and MCI in older adults,79 while another linked B. fragilis to neuroinflammation in an AD mouse model.80 Conversely, a 2023 study using imaging and metabolomics linked Bacteroides-related pathways (e.g., phenylalanine and unsaturated fatty acid metabolism) to improved cognition.81 Their research is one of the first to suggest a positive linkage of Bacteroides to cognition. Among CN participants, microbial differences were largely attenuated, except for persistent reductions in Christensenellaceae_R-7_group among those with obesity. This family is consistently associated with leanness, lower inflammation, and improved metabolic profiles,82 and may promote microbiome stability by limiting energy harvest. Notably, lower Christensenellaceae_R-7_group abundance has been associated with higher amyloid positivity in AD patients.82 Notably, recent research has shown that lower abundance of Christensenellaceae_R-7 group is associated with higher odds of amyloid positivity in AD patients,83 suggesting a link between this SCFA-producing taxon and neurodegeneration.

Despite both groups losing weight over the 10-week intervention, significant between-group differences in body composition persisted. Obese participants began with greater fat mass, visceral and subcutaneous adiposity, and muscle mass, which were not substantially altered, likely due to the short timeframe. The obese group lost more weight on average (~6 kg vs. ~1.2 kg in non-obese), likely reflecting higher baseline body mass and energy expenditure. However, specific compartments such as fat mass and skeletal muscle did not change significantly in either group, suggesting that weight loss was modest relative to total mass and distributed across tissues. Increases in protein and mineral content among the obese group may reflect relative gains in lean mass or measurement artifacts related to larger body size. Alternatively, poor dietary adherence may have limited measurable changes. Notably, weight loss was not a primary goal of the intervention, which emphasized caloric adequacy within dietary pattern guidelines rather than restriction. Both groups demonstrated similar adherence to Mediterranean components (via MEDAS score), and comparable carbohydrate intake was observed, likely due to gradual titration in the MMKD group. These findings highlight the difficulty of eliciting significant body composition shifts, especially in fat distribution, over short interventions. More intensive protocols, including resistance training or stricter dietary regimens, may be needed to produce meaningful changes in fat and muscle compartments.

Post-intervention, both groups exhibited shifts in gut microbiota composition, with alpha- and beta-diversity measures indicating partial convergence in microbial community structure. This suggests that dietary education interventions can beneficially modulate the microbiome, even in the context of obesity. However, the obese group maintained lower microbial richness (Chao1), implying persistent barriers to microbiome recovery, potentially due to dysbiosis, or long-term dietary patterns. Notably, the intervention differentially influenced the Firmicutes:Bacteroidota (F/B) ratio between groups. Obese participants began with an elevated F/B ratio (a trait often associated with enhanced energy harvest and fat storage) and this ratio increased further (from 2.18 to 3.52) after the intervention. This was primarily due to a decline in Bacteroidota rather than a rise in Firmicutes, suggesting a limited or altered response to dietary fiber, possibly reflecting microbiome rigidity linked to chronic metabolic stress. A meta-analysis from 2022 identified that at the phylum level, obese subjects do have elevated F/B ratios (largely driven by elevated firmicutes) compared to non-obese counterparts,68 though a more recent paper revealed that elevated F/B ratio in obese participants was not correlated to increases in body mass.84 Thus, F/B ratio may reflect host–microbe adaptations to an obesogenic environment rather than directly causing adiposity. In contrast, the non-obese group experienced a modest reduction in F/B ratio (from 2.94 to 2.75), driven by an increase in Bacteroidota. These taxa are proficient in fermenting dietary fiber into beneficial SCFAs like butyrate, which support gut health and may influence brain function.85 The observed enrichment of Bacteroidota in the non-obese group may reflect greater microbiome plasticity and responsiveness to the fiber-rich MD. Together, these findings underscore both the microbiome’s capacity to shift in response to diet and the modifying role of obesity in shaping that response, with potential downstream implications for metabolic regulation and cognitive health.

The current findings offer nuanced insights into the so-called “obesity paradox”, the observation that in certain contexts, especially among older adults, higher adiposity may correlate with better health outcomes.86 In our study, obesity was associated with lower baseline cognitive performance across several domains. However, over the 10-week dietary education intervention, individuals with obesity showed notable cognitive gains, particularly in episodic memory and executive function (PSMT, DCCS), which were correlated with reductions in body weight and fat mass. This suggests that fat loss, especially visceral fat, known for its inflammatory and metabolic burden, may support cognitive improvement.87^,88 Despite lower cognitive scores at baseline and post-intervention (e.g., Flanker, RAVLT), obese participants demonstrated improvements in verbal, associative, and episodic memory, suggesting that cognitive plasticity remains intact. These data demonstrate that cognitive gains remain possible in obesity, especially when modest weight loss helps reduce metabolic load.

In contrast, the non-obese group showed a different pattern, one that aligns with the obesity paradox. The counterintuitive association in adults without obesity, where small increases in body fat correlated with improved memory, is unlikely to reflect a causal benefit of adiposity. Several factors may explain this pattern. Individuals with lower baseline fat may depend more on weight stability for metabolic homeostasis, so minor increases may represent a return toward a physiological set point rather than unhealthy gain. A similar directional pattern has been observed in other populations where low baseline adiposity impairs neurocognitive function. For example, systematic reviews in patients with anorexia nervosa show that cognitive performance improves following weight restoration, suggesting that inadequate fat reserves can compromise cognitive processes and that increases in adiposity may normalize metabolic and neuroendocrine function.89 Although our cohort is metabolically healthier, this reinforces the concept that modest fat gain in individuals starting with lower adiposity may support, rather than impair cognitive performance.

Furthermore, the current study's age-stratified results emphasize the complexity of the obesity paradox. Age-stratified analyzes further clarified the complexity of the obesity paradox. Among older adults (73–80 y), gains in skeletal muscle mass were strongly associated with cognitive improvements, a relationship absent in younger participants (58–72 y). This suggests that muscle preservation becomes increasingly critical with age, potentially mitigating sarcopenia and its contribution to cognitive decline. Skeletal muscle may exert neuroprotective effects via myokine signaling (e.g., irisin, BDNF), improved insulin sensitivity, and reduced systemic inflammation.90-92 As lean mass diminishes with age, maintaining or increasing skeletal muscle may buffer against both physical and cognitive decline. These findings echo prior research showing that resistance training and muscle-preserving interventions improve cognitive outcomes in older adults with excess adiposity.93

Overall, the relationship between body composition and cognition appears context dependent. Excess fat and low muscle mass may each contribute to cognitive risk, while interventions targeting muscle retention and metabolic balance may offer neurocognitive benefits. Future research should further explore the interplay between fat distribution, lean mass, and metabolic health to clarify the mechanisms underlying the obesity paradox in aging populations.

Acetate, a SCFA, is produced primarily by gut microbiota during the fermentation of dietary fibers, such as those found in plant-based foods, which are a hallmark of the MD. It plays an important role in the gut-brain axis, influencing both metabolic and cognitive functions.94 In this study, fecal acetate levels inversely correlated with global cognitive improvement in the obese group, suggesting higher fecal acetate is linked to poorer cognition. Elevated fecal acetate may indicate impaired absorption or utilization, reflecting gut dysfunction or dysbiosis, which can increase inflammation via gut permeability and bacterial metabolites like LPS, thereby impairing cognition.95 Excess acetate may also disrupt beneficial SCFA balances, notably reducing protective butyrate, and limit acetate availability for brain functions such as energy metabolism and neuroplasticity mediated by G-protein coupled receptors (GPR41/43).96^,97 Reduced receptor activation could impair memory and neuroplasticity. Plasma acetate was associated with weight loss in obese participants, highlighting its role in metabolic shifts during body composition changes. Absorbed from the colon into circulation, acetate influences lipid metabolism, glucose regulation, and inflammation.96 In obesity, where chronic inflammation prevails, acetate may exacerbate inflammatory pathways, impairing cognition. Evidence from murine models shows that gut microbiota depletion reduces muscle fiber size and strength, effects reversed by acetate supplementation. Mice lacking acetyl-CoA synthase 2, necessary for acetate utilization, exhibit reduced muscle mass and lifespan, indicating acetate’s role as a metabolic substrate for muscle.98 A large cohort study found circulating acetate positively associated with gut microbiome diversity and inversely with visceral fat, suggesting a protective cardio-metabolic role.99 Conversely, circulating acetate has been reported to stimulate the parasympathetic nervous system, enhancing insulin and ghrelin secretion, which may increase appetite and promote obesity—contrasting with weight loss observed in our cohort.100 These findings underscore acetate’s complex role as a metabolic mediator: fecal acetate likely reflects microbiome fermentation and gut function, whereas plasma acetate reflects systemic metabolic responses influencing body composition. The inverse relationship between fecal acetate and cognition warrants further research into gut-derived metabolites’ effects on brain health. Future studies should clarify whether acetate directly contributes to cognitive decline or serves as a marker of metabolic disturbances. Interventions modulating acetate levels via diet, prebiotics, probiotics, or fecal microbiota transplantation may offer therapeutic potential for metabolic and cognitive dysfunction in obesity.

Several key metabolites, including lactate, citrate, pyruvate, and propionate, were associated with specific domains of cognitive performance. In the obese group, fecal lactate was inversely associated with weight, and higher lactate levels correlated with less cognitive improvement. As a byproduct of anaerobic metabolism, elevated lactate has been shown to reflect metabolic dysregulation and systemic inflammation, potentially impairing brain function.101 This supports the notion that lactate could serve as a marker of poor metabolic health, where high lactate concentrations may reflect the body's struggle to efficiently utilize energy, ultimately affecting brain function. Future research could explore whether reducing lactate levels through targeted metabolic interventions or modulating lactate dehydrogenase activity could improve cognitive outcomes. Plasma citric acid, a central intermediate of the tricarboxylic acid (TCA) cycle, was positively associated with Flanker and FName performance, suggesting that efficient energy metabolism supports cognitive function.102 Citric acid fuels adenosine triphosphate (ATP) production, and its link to cognition reinforces the importance of mitochondrial health for tasks requiring sustained attention. Dietary or metabolic interventions that enhance citric acid production may support cognitive performance, especially in those with impaired energy metabolism. Similarly, plasma pyruvate, a key metabolite in aerobic respiration, was positively associated with both Flanker and FName scores. Pyruvate's role as a precursor in ATP generation underscores its importance in supporting executive function and cognitive processing.103^,104 Elevated pyruvate levels may reflect better cellular respiration and more efficient energy production, facilitating cognitive tasks that require rapid processing and executive control. Future research could examine how pyruvate metabolism and its associated pathways influence cognitive resilience. Fecal propionate, a SCFA with known anti-inflammatory properties, was linked to better cognitive flexibility (DCCS performance). Propionate promotes GLP-1 and PYY secretion from enteroendocrine cells, potentially contributing to reduced food intake and lower body fat, as observed in this study.105 Elevated fecal propionate was associated with lower adiposity in our cohort. It has also shown neuroprotective effects in Parkinson’s models through free fatty acid receptor 3 (FFAR3) signaling on enteric neurons.106 Gut-derived propionate can bind to the FFAR3 receptor on enteric neurons, activating a signaling pathway that can influence nerve function and reduce inflammation.106 However, propionate has also been linked to adverse outcomes such as polycystic ovarian syndrome (PCOS) and insulin resistance.107 These conflicting results suggest that its cognitive effects depend on host metabolism and microbial context. Future studies should distinguish between circulating and fecal propionate and examine whether targeting FFAR3 signaling can improve cognitive outcomes. These metabolite findings also align with emerging evidence supporting a gut–muscle axis that may contribute to the cognitive and body-composition effects observed in our cohort. SCFAs and other microbial metabolites can influence skeletal muscle metabolism by regulating AMP-activated protein kinase (AMPK) signaling and mitochondrial function, thereby affecting muscle quality.51 Recent clinical data further support this connection. In the PROMOTe randomized controlled trial, targeted modulation of the gut microbiome led to measurable improvements in both muscle function and cognition in older adults, suggesting that microbial metabolites may exert coordinated effects across muscle and brain systems.52