Epigenetic and microbiome responses to greens supplementation in obese older adults: results from a randomized crossover-controlled trial

This 60-day randomized, crossover-controlled trial was designed to evaluate the effects of a greens-based supplement on epigenetic aging and physiological outcomes in adults with obesity. Body composition, dietary intake, and physical activity were assessed as contextual measures to characterize the study population and help isolate the effects of the intervention. Twenty adults aged 50–65 years with a body mass index (BMI) > 30 kg/m² were randomized to immediate or delayed intervention group (see Figure 1). The study was conducted at Auburn University School of Kinesiology Building and consisted of four total visits. Prior to data collection, this study was approved by the Auburn University Institutional Review Board (IRB; Protocol # STUDY00000013), conformed to standards set by the latest revision of the Declaration of Helsinki, and was pre-registered as a clinical trial (NCT06537232↗).

During Visit 1 (screening), participants provided written informed consent, completed a medical history questionnaire, and received instructions on study procedures. Eligible participants were randomized using a stratification scheme based on age (50–57 vs. 58–65 years) and race/ethnicity (non-Hispanic White vs. Other), with randomization lists generated using the Sealed Envelope Ltd Platform in blocks of four (22). Participants were randomly assigned to one of two sequences: immediate supplementation followed by a free-living phase, or vice versa. Each participant received three stool collection kits with written instructions for home use.

Visit 2 (Day 0) included baseline body composition measurements using a Tanita bioelectrical impedance analyzer Model BC-568 (Tanita Corporation of American, Inc, Arlington Heights, IL), height measurement, fasting blood collection, and breath testing for hydrogen and methane. Participants also completed the DASS-21, International Physical Activity Questionnaire (IPAQ), SF-12 Health Survey, a 24-h dietary recall, and gastrointestinal symptom visual analog scales. Stool samples, collected 24–48 h prior, were submitted during this visit. Participants issued a Garmin Vivofit 4 activity tracker (Garmin International, Olathe, KS), to wear continuously, with data uploaded to Fitabase data collection software (Fitabase, San Diego, CA).

At Visit 3 (Day 30; crossover) participants switched conditions (supplementation or free-living) and repeating all procedures performed at Visit 2, including submission of a second stool sample. Visit 4 (Day 60; endpoint) included final assessments identical to those at Visits 2 and 3, along with submission of the third and final stool sample. Adherence and symptom monitoring were monitored using a HIPPA-compliant text messaging platform, Emitrr (Emitrr Inc., https://emitrr.com↗). Participants received daily reminders to take the supplement during the intervention phase and were encouraged to report any adverse symptoms or missed doses via the platform. Participants experiencing significant GI discomfort were advised to discontinue the supplement and could withdraw.

Sample size was determined a priori to detect a mean difference of 1.5 years in epigenetic age, based on prior nutritional supplementation trials. A recent study calculated that 76 participants would be needed to detect a 1.5-year change in inflammatory age (iAGE) with 80% power and a significance level of 0.05, using a paired design (23). For the present study, a power analysis using G^*Power (version 3.1.9.7) indicated that 10 participants would provide 80% power to detect a 1.5-year within-subject difference (SD = 1.5 years, d = 1.0, α = 0.05). To increase statistical power while also accounting for attrition, 20 participants were enrolled.

Recruitment was conducted in the Auburn and Opelika, AL area through a Qualtrics screening survey. Recruitment was conducted over a one-month period between September and October 2024. Flyers were distributed at community locations (e.g., grocery stores, retail shops, post offices) and across Auburn University departments. The Qualtrics-hosted screening survey collected demographic and health-related information, including age, biological sex, ethnicity, height, weight, highest level of education, and marital status. Additional questions assessed the frequency and quantity of dark green leafy vegetable consumption and current use of antibiotics, immunosuppressants, corticosteroids, or probiotics. Eligible respondents were contacted and invited for an in-person screening visit, where study procedures were explained and informed consent obtained prior to participation.

The greens-based supplement used in this study was a commercially available powdered blend, Field of Greens (BrickHouse Nutrition, Cooper City, FL), composed primarily of dehydrated vegetables, fruits, and botanical extracts (24). Each serving (10.43 grams) contained a proprietary mixture of organic greens including kale, spinach, parsley, chlorella, barley grass, and wheatgrass, alongside functional plant compounds such as beet root, green tea extract, turmeric, and ashwagandha. The supplement also provided B-complex vitamins (B1, B2, B6, B12, and folate), vitamin C, vitamin D3, and vitamin K2, and contained natural flavors and stevia with no added sugars. Participants were instructed to consume one full scoop mixed with water or a beverage of choice each morning during the 30-day intervention period. To maintain consistency, participants were asked to avoid introducing any new supplements, herbal products, or dietary changes for the duration of the study. Notably, this study was open-label, and no placebo supplement was consumed given the potential influence that commonly used ingredients (e.g., maltodextrin, inulin, or cellulose) could potentially affect some of the secondary outcomes (e.g., gut microbial diversity) (25).

At each blood draw, 7 ml of whole blood was collected into a BD Vacutainer^® CPT™ Mononuclear Cell Preparation Tube (Becton, Dickinson and Company, Franklin Lakes, NJ, USA) from the antecubital vein of the arm using aseptic techniques by study staff with phlebotomy training. Tubes were gently inverted after collection and centrifuged within 2 h at 1,500–1,800 RCF for ≥20 min at 18–25 °C. Peripheral blood mononuclear cells (PBMCs) and plasma were separated, aliquoted into 1 ml cryovials, and stored at−80 °C. PBMCs were shipped to TruDiagnostic (Lexington, KY, USA) for DNA methylation profiling and epigenetic clock analysis (Horvath, PCGrimAge, AdaptAge, and DamAge). Plasma samples were transported on ice to East Alabama Health Medical Center (Opelika, AL, USA) and analyzed for a comprehensive metabolic panel (CMP), triglycerides, blood glucose, and lipoproteins as we have described previously (26).

Isolated PBMC samples were shipped to TruDiagnostics on dry ice (Lexington, KY, USA), where DNA methylation profiling was conducted using the Illumina Infinium MethylationEPIC BeadChip (>850,000 CpG sites). TruDiagnostics provided quality-controlled methylation data and epigenetic clock outputs for each participant and timepoint. To evaluate intervention effects, within-subject differences in clock scores were calculated by comparing pre- and post-intervention values for each treatment phase (supplement vs. control) according to group assignment. These difference scores were then analyzed using repeated-measures and correlational approaches to assess treatment effects and individual-level associations.

Four established epigenetic clocks: Horvath, PCGrimAge, AdaptAge, and DamAge. These clocks were selected based on their distinct yet complementary capacities to quantify biological aging processes. The Horvath clock (353 CpG sites) estimates chronological age across tissues and remains a standard measure of biological age acceleration, with higher values indicating more advanced biological aging relative to chronological age and lower values indicating slower aging trajectories (6, 8). PCGrimAge (1,030 CpG sites) incorporates DNA methylation surrogates for plasma proteins and smoking history to predict morbidity and mortality risk, with higher values indicating elevated mortality risk and lower values generally reflecting improved healthspan (27). AdaptAge (1,000 CpG sites) is designed to quantify the biological response to environmental and physiological stress, capturing adaptive resilience mechanisms rather than cumulative damage (28). Directional changes in AdaptAge should therefore be interpreted as shifts in stress-responsive biological domains rather than strictly as changes in chronological aging; decreases may reflect improved stress adaptation capacity, whereas increases may indicate heightened stress burden or reduced resilience. DamAge (1,090 CpG sites) captures the accumulation of irreversible cellular and molecular damage associated with long-term aging burden, with higher values reflecting greater accumulation of methylation signatures linked to damage processes and lower values indicating reduced damage burden (28, 29). These clocks were selected based on evidence that aging is a modular process, with each clock representing distinct biological domains responsive to lifestyle or nutritional interventions (29, 30).

Body composition was assessed at Visits 2, 3, and 4 using a multi-frequency bioelectrical impedance analyzer (Tanita BC-568; Tanita Corporation of America, Arlington Heights, IL, USA). Participants arrived fasted for ≥8 h and well hydrated. Height was measured with a stadiometer, and participants removed shoes and socks before standing on metal footplates while holding hand electrodes. The device provided estimates of body weight, body fat percentage, lean mass, and bone mass. Standardized protocols were used across visits to ensure reliability.

Breath samples were collected in duplicate at 0, 30, 60, and 90 min during Visits 2, 3, and 4 using the QuinTron EasySampler collection kit and BreathTracker SC Analyzer (QuinTron, Milwaukee, WI, USA). The greens-based supplement was consumed immediately prior to the 0-min sample during the supplementation phase (Visit 2 for the immediate group, Visit 4 for the delayed group, and Visit 3 for both groups). Participants also collected duplicate home breath samples on Days 21 and 51, timed to correspond with supplementation phases. All participants were instructed to fast ≥2 h before sampling and received reminders 2 days prior and the morning of each collection. Hydrogen, methane, and carbon dioxide were quantified, with correction factors applied by the analyzer.

Stool samples were collected 24–48 h prior to Visits 2, 3, and 4 using OMNIgene^® GUT OMR-200 tubes (DNA Genotek, Ottawa, Canada) and returned at study visits. Participants recorded stool consistency using the Bristol Stool Chart and noted date and time of collection. Upon receipt, samples were stored at−80 °C until processing. Microbial DNA was extracted with Zymo Fecal/Soil DNA Miniprep kits (Irvine, CA, USA) according to manufacturer instructions. The V4 region of the 16S rRNA gene was amplified and sequenced on an Illumina MiSeq platform. Raw reads were quality-checked with FASTQC and processed in QIIME2, where DADA2 was applied for filtering, denoising, chimera removal, and amplicon sequence variant (ASV) generation (31–34). Taxonomic classification was performed against the SILVA reference database (35). Alpha diversity was calculated using observed species richness, Shannon index, and Faith's phylogenetic diversity, while beta diversity was assessed using Bray–Curtis dissimilarity and UniFrac clustering. Differential abundance testing was performed using ANCOM and DESeq2 to identify microbial taxa significantly associated with the intervention (36).

Participants were provided with Garmin Vivofit 4 devices (Garmin International, Olathe, KS, USA) at Visit 2 and instructed to wear them continuously for the 60-day study. Study staff assisted participants with linking the devices to their smartphones, which were synced weekly through Fitabase (Small Steps Lab, LLC, San Diego, CA, USA). Daily step counts and sleep duration were extracted for analysis. Text reminders were used to encourage compliance and device syncing. Sleep phase data generated by the manufacturer were not analyzed, as the device contained only accelerometer-based sensors.

At Visits 2, 3, and 4, participants completed validated surveys to assess quality of life, psychological symptoms, activity, and gastrointestinal outcomes. The 12-Item Short Form Survey (SF-12) measured health-related quality of life across physical and mental domains (37). Depression, anxiety, and stress were assessed with the 21-item DASS-21 using a 4-point Likert scale (38). Physical activity was captured with the International Physical Activity Questionnaire, Short Form (IPAQ-SF), which was converted to MET-minutes per week (39). Gastrointestinal symptoms were assessed using a 100-mm Visual Analog Scale (GI-VAS). All questionnaires were administered via Qualtrics.

Dietary intake was assessed using three interviewer-administered 24-h recalls at each study visit, including two weekdays and one weekend day. Interviews followed the USDA multiple-pass method and included details on portion size, timing, and preparation. Recalls were analyzed in the Nutrition Data System for Research (NDSR 2024; University of Minnesota), and average daily intake was calculated for energy, macronutrients, sugars, and fiber.

All statistical analyses were conducted using RStudio (2025.05.0) with graphs plotted using GraphPad Prism (Version 10.1; San Diego, CA, USA). Primary outcomes included DNA methylation based epigenetic clock measures, which were evaluated using one-way repeated measures ANOVA to assess within-subject differences over time. Secondary outcomes included immune markers, breath gases, body composition, microbiome alpha diversity, actigraphy data, and scores from the SF-12 and DASS-21 questionnaires, which were analyzed using the same repeated measures framework. ANOVAs were conducted using the ezANOVA function from the ez package, with Greenhouse–Geisser corrections applied when assumptions of sphericity were violated. Post hoc pairwise comparisons were performed using the emmeans package with p-values adjusted for multiple comparisons.

Exploratory analyses included Spearman correlation coefficients with false discovery rate (FDR) correction to examine associations between epigenetic clock outcomes and physiological, microbiome, behavioral, or psychological measures across the treatment period. Categorical variables, including demographic characteristics (e.g., sex, race/ethnicity, education level) and Bristol stool types, were analyzed using Pearson's chi-square tests with Sidak adjustments. Statistical significance was set at α ≤ 0.05 for all analyses, and data are presented as means ± standard deviation values throughout.

The CONSORT diagram (Figure 2) outlines participant accrual and retention throughout the study. Of the 65 individuals assessed for eligibility, 44 were excluded, with 42 not meeting inclusion criteria and two declining to participate. Twenty-one participants were randomized into immediate (n = 11) and delayed (n = 10) intervention groups. Two participants withdrew, one due to time conflicts, and one participant due to illness. In total, 19 participants completed the full protocol and were included in the final analyses (immediate group: n = 10; delayed group: n = 9).

Baseline demographics and characteristics are presented in Table 1. No between-group differences were observed for age (Immediate: M = 58.0 ± 5.0 years; Delayed: M = 55.0 ± 4.3 years), sex (65% female), or ethnicity. Educational attainment differed significantly (p = 0.044), with more college-educated participants in the immediate group. Additionally, BMI classification differed (p = 0.019), with more delayed group participants classified as obese class III (BMI > 40 kg/m²; Immediate: M = 34.8 ± 3.4 kg/m²; Delayed: M = 41.4 ± 9.9 kg/m²). Weekly servings of dark green leafy vegetables also varied slightly between groups (Immediate: M = 1.35 ± 1.43 servings; Delayed: M = 2.04 ± 1.64 servings), though this difference was not statistically significant.

Fifteen participants (immediate group: n = 8; delayed group: n = 7) had viable PBMC samples used to assess epigenetic age across three timepoints. Participants without viable PBMC samples did not differ meaningfully from those included with respect to age, sex, BMI, or baseline clinical characteristics. As shown in Supplemental Table 2, DNA methylation age estimates, Horvath, PCGrimAge, AdaptAge, and DamAge, remained relatively stable over the 60-day study period, with no statistically significant main effects of group, time interactions. AdaptAge displayed greater variability than the other clocks, particularly in the delayed group, but the Group × Time interaction did not reach significance (p = 0.096).

Figure 3 illustrates individual changes in each epigenetic clock during 30-day supplementation versus control. Horvath epigenetic age (Figure 3a) decreased during the control period and an increase during supplementation, with paired t-test result approaching statistical significance (p = 0.059). PCGrimAge, AdaptAge, and DamAge (Figures 3b–d) remained consistent (p = 0.369, p = 0.250, and p = 0.107).

Mean changes in epigenetic age across the 30-day supplementation and control (Supplemental Table 3) revealed that the Horvath clock age increased 1.43 ± 3.61 years during supplementation compared to a decrease of−2.07 ± 4.22 during control, yielding a significant main effect of treatment (p = 0.015). PCGrimAge decreased slightly during supplementation (−0.25 ± 1.59 years) and increased 0.35 ± 2.02 years during control, without significant effects. DamAge also decreased during supplementation (−2.84 ± 9.06 units) and increased 5.30 ± 9.77 units during control, with a significant group effect (p = 0.028) but non-significant treatment interactions. AdaptAge increased during supplementation by 3.70 ± 9.10 units during and decreased during control (−4.50 ± 11.60 units), though these changes were not statistically significant.

Plasma from 18 participants were analyzed for CMP measures (immediate group: n = 10; delayed group: n = 8), with two excluded due to incomplete data. Supplemental Table 3 reports relatively stable estimated marginal means across the three study visits for most markers, with a few notable exceptions. Glucose levels decreased slightly over time in the immediate group (from 92.00 to 84.20 mg/dl) but remained relatively stable in the delayed group, resulting in a marginal group × time interaction (p = 0.090). Potassium and chloride exhibited significant main effects of time (p = 0.026 and p < 0.001, respectively), though changes were minor and not clearly linked to supplementation. ALT increased over time and was consistently higher in the delayed group, contributing to a near-significant interaction (p = 0.062) and a group effect trending toward significance (p = 0.085), while AST followed a similar pattern without significance. CRP was higher in the delayed group across timepoints (p = 0.042), but no time or interaction effects were observed. Lipid markers (HDL-C, LDL-C, total cholesterol, triglycerides) declined slightly over time in both groups, but none reached statistical significance (p > 0.07 for all effects). As shown in Supplemental Figure 1, paired comparisons confirmed that changes in CRP, ALT, AST, glucose, and lipids were small and did not differ between supplementation and control. Together, these findings indicate that the greens supplement did not produce meaningful changes in metabolic or inflammatory markers, although modest shifts in CRP, glucose, ALT, and electrolytes suggest potential physiological responses warranting further study.

Two-way ANOVAs revealed no significant main effects or interactions for body weight or body fat percentage (all p > 0.16). On average, body weight increased by 0.33 ± 3.97 kg, and body fat percentage by 1.33 ± 6.62% during supplementation, though all effect sizes were small and not statistically significant.

Actigraphy data were available for 17 participants. Step counts and sleep duration remained stable across the 8-week study, with no significant effects of group, treatment, or their interaction in either ANOVA or LMM models (all p > 0.13). The immediate group consistently recorded higher step counts across weeks (e.g., Week 8: 8283 vs. 5158 steps/day, Δ = 3126 ± 1890, p = 0.113), but differences did not reach significance. Average sleep duration was ~7.7 h/night in both groups, with no meaningful variation across phases. As illustrated in Figures 4a, b, step counts and sleep duration remained relatively stable across groups and phases. These consistent findings across statistical methods support the conclusion that the greens supplement did not significantly impact physical activity or sleep behavior during the intervention period. Paired t-tests also confirmed no within-group changes in steps or sleep time during supplementation (all p > 0.39).

Breath samples from 19 participants were analyzed for hydrogen (H₂) and methane (CH₄) across three visits. At baseline (visit 2), no significant group differences were observed (H₂: Delayed = 12.7 ± 8.73 ppm, Immediate = 5.99 ± 4.01 ppm, p = 0.058; CH₄: Delayed = 4.41 ± 6.47 ppm, Immediate = 3.74 ± 5.51 ppm, p = 0.812). Two-way repeated measures ANOVAs showed no significant main effects of group, visit, or their interaction for either H₂ (all p > 0.20) or CH₄ (all p > 0.29). Although Mauchly's test indicated sphericity violations, Greenhouse-Geisser corrections did not alter the non-significant results. Short-term gas production at four timepoints (T0, T30, T60, and T90 minutes), also showed no consistent phase-dependent trends. As illustrated in Supplemental Figure 2, H₂ ranged from 8.1 to 14.8 ppm and CH₄ from 2.8 to 10.4 ppm, with standard deviations often exceeding 12–20 ppm, reflecting high inter-individual variability. Overall, breath hydrogen and methane concentrations varied widely between participants but were not significantly influenced by the greens supplement. Although minor visual trends (e.g., elevated CH₄ in the immediate group at later visits) were noted, they did not reach statistical significance, suggesting colonic and small intestinal fermentation remained stable throughout the intervention.

Nineteen participants with valid 24-h recalls across all visits were included in the analysis. Supplemental Table 4 summarizes mean intakes of energy, fat, carbohydrates, protein, cholesterol, fiber, and sugars, across the three study visits for both the immediate and delayed groups. Overall, dietary intake patterns remained stable between groups and across timepoints, with only minor numerical fluctuations. Carbohydrate intake also remained stable (all p > 0.20) across visits. Protein intake varied modestly across time (p = 0.049) but showed no group (p = 0.467) or interaction (p = 0.320) effects. Cholesterol (all p > 0.19), fiber (all p > 0.13), and sugar (all p > 0.40) intake did not differ significantly. Together, these findings indicate that dietary intake remained largely unchanged throughout the study, aside from a modest but statistically significant variation in protein intake over time.

Seventeen participants with complete microbiome data across all visits were included in the analysis. Alpha diversity, assessed by Shannon entropy, Faith's phylogenetic diversity (PD), Pielou's evenness, and observed features, was similar between groups at baseline and remained stable throughout the study. As shown in Figure 5, individual participant statistical comparisons revealed no significant within-subject and group-level changes between supplementation and control periods (Shannon p = 0.268; Faith's PD p = 0.280; Pielou's evenness p = 0.756; observed features p = 0.521), indicating that alpha diversity remained stable. Overall microbial richness and community structure were not altered.

Beyond alpha diversity, exploratory analyses (Supplemental Table 5) showed that most phyla, including Bacteroidota, Actinobacteriota, Proteobacteria, and Firmicutes, were unaffected with supplementation. Desulfobacterota demonstrated a significant Group × Treatment interaction (p = 0.031), and at the genus level, Bilophila exhibited both a main effect of Treatment (p = 0.037) and a Group × Treatment interaction (p = 0.038), indicating modest, taxon-specific responses. These changes occurred in the absence of significant shifts in overall microbiome alpha diversity and therefore should be interpreted as localized microbial responses. Other genera, including Desulfovibrio, Collinsella, Bifidobacterium, Eggerthella, Adlercreutzia, Actinomyces, Escherichia-Shigella, Enterobacter, Bacteroides, and Alistipes, showed no significant effects (all p > 0.13). At the species level, Alistipes onderdonkii, A. putredinis, Bacteroides vulgatus, and B. fragilis did not show significant changes (all p > 0.08), though A. putredinis trended toward a Group × Treatment interaction (p = 0.081). Collectively, these results indicate that 30 days of supplementation did not significantly alter gut microbial alpha diversity or overall composition, though specific taxa such as Bilophila and Desulfobacterota showed modest, group-dependent responses.

Nineteen participants completed all assessments of self-reported quality of life (SF-12) and psychological symptoms (DASS-21). Scores from the SF-12, including both the Physical Component (PCS) and Mental Component (MCS), remained stable across study visits, with no significant effects of group, time, or their interaction (all p > 0.15; Supplemental Figure 5).

Depression, anxiety, and stress scores from the DASS-21 showed no significant group differences at baseline (all p > 0.20) and did not change over time, with no main effects of group, time, or interaction (all p > 0.31; Supplemental Figure 6). Although minor baseline differences were noted, particularly for stress and depression, none reached statistical significance, and no persistent changes were observed. Overall, the greens supplement did not impact participants' quality of life or psychological distress during the study.

Correlation heatmap results (Figure 6) illustrate the strength and direction of relationships between pre-to-post green supplementation changes in epigenetic clocks, microbiome diversity indices, clinical biomarkers, body composition, and psychological outcomes. After FDR correction, no significant change score associations were observed between the primary epigenetic clocks (Horvath, AdaptAge, DamAge, and PCGrimAge) and other study variables, including each other. At the unadjusted level, Horvath age showed moderate correlations with sleep duration (r_s = −0.49, p = 0.069) and anxiety (r_s = 0.48, p = 0.070), while AdaptAge displayed weak associations with Shannon diversity, Observed species richness, and step counts (all p > 0.18), though none reached significance after correction. Among clinical blood biomarkers, strong positive associations were identified between AST and CRP (r_s = 0.71, p = 0.0006) and ALT and AST (r_s = 0.71, p = 0.001), with ALT also moderately associated with CRP (r_s = 0.66, p = 0.001). Faith's PD was inversely correlated with glucose (r_s = −0.81, p < 0.001), suggesting that lower microbial diversity may be linked with impaired glycemic control. Faith's PD also showed moderate inverse trends with depression and stress, though these did not persist after correction.

Microbial correlations with epigenetic clocks were further explored at multiple taxonomic levels. At the phylum level, PCGrimAge was positively associated with Proteobacteria (r = 0.844, q < 0.001), and DamAge with Bacteroidota (r = 0.844, q < 0.001). At the genus level (Supplemental Figure 3), significant associations were observed for Desulfovibrio with DamAge (r = 0.875, q < 0.001), Collinsella with PCGrimAge (r = 0.875, q < 0.001), and Eggerthella with Horvath age (r = 0.671, q = 0.024). Additional trend-level correlations included Enterobacter with Horvath (q = 0.055), Alistipes with AdaptAge (q = 0.060), and Bacteroides with DamAge (q = 0.091). At the species level (Supplemental Figure 4), Alistipes putredinis correlated positively with DamAge (r = 0.844, q < 0.001), while Bacteroides fragilis was associated with PCGrimAge (r = 0.844, q < 0.001).

Together, these results indicate limited evidence for broad associations between epigenetic clocks and health measures, while highlighting robust correlations between specific microbial taxa, glycemic control, and blood biomarkers.

Interpretation of epigenetic clock outcomes requires consideration of the distinct biological domains each clock is designed to capture. The differential responsiveness of epigenetic clocks aligns with prior trials showing clock-specific effects of nutritional and lifestyle interventions. The increase in Horvath age was unexpected given observational studies linking higher fruit and vegetable intake to lower epigenetic age acceleration (42, 43). However, first-generation clocks such as Horvath are less sensitive to short-term interventions and may reflect shifts in cell composition rather than targeted biological pathways (29). This consideration may be particularly relevant in the present cohort, which had a mean BMI of 38.1 kg/m², as obesity is associated with chronic low-grade inflammation, immune cell dysregulation, and altered leukocyte composition that can influence DNA methylation patterns independent of cumulative biological aging (12, 14, 15). Previous work has demonstrated that immune cell activation, differentiation, and turnover are associated with methylation changes that may occur independently of cumulative biological aging processes, reflecting shifts in immune dynamics rather than accelerated aging (44). Such effects may be particularly relevant in short-term interventions, where transient immune or inflammatory responses could influence methylation patterns at CpG sites weighted heavily in first-generation clock algorithms. In contrast, newer generation clocks capture distinct aging related domains. Second generation clocks, including AdaptAge and DamAge, exhibited effect sizes consistent with our hypotheses, though without statistical significance. These clocks capture domains related to stress adaptation and cumulative molecular damage (30), mechanisms plausibly targeted by polyphenols and bioactive compounds in leafy greens. The modest trends observed here are consistent with models proposing that clocks represent distinct biological modules of aging (45, 46).

Comparisons with prior trials further contextualize these findings. Multicomponent interventions combining diet, activity, and stress management have demonstrated reductions in epigenetic age across multiple clocks (40, 41), while nutrient-specific trials have reported more modest or null effects (23). The 30-day, single-component design of the present study may therefore have been insufficient to elicit systemic changes. Given that epigenetic aging measures reflect cumulative biological processes, longer or repeated interventions may be required to induce stable shifts in clock trajectories (19, 47). Accordingly, the 30-day, single-component design of the present study may have captured early or adaptive responses rather than durable epigenetic remodeling. Nonetheless, the moderate effect sizes observed for AdaptAge and DamAge indicate directional trends consistent with the biological domains captured by these clocks; however, given the modest sample size and lack of statistical significance, these findings should be interpreted as exploratory rather than evidence of efficacy.

Clinical markers including glucose, ALT, AST, CRP, lipids, and electrolytes remained largely stable across intervention phases. Small fluctuations were observed, such as higher CRP and ALT in the delayed group and a modest decline in glucose in the immediate group, but none reached statistical significance. These findings suggest limited short-term physiological effects in a relatively healthy population. Exploratory correlations provided additional insight. AST was positively associated with CRP, supporting links between liver function and systemic inflammation. CRP also trended with body fat percentage but not with epigenetic clocks. These results are consistent with prior trials showing that baseline inflammatory status moderates biomarker responsiveness (20, 23). The absence of group-level changes here may reflect low baseline inflammation, short intervention duration, and reliance on supplementation rather than whole-food dietary shifts (20, 48).

No significant changes were detected in microbiome alpha diversity or fermentation gases, diverging from some longer or more intensive fruit and vegetable-based interventions (49, 50). However, exploratory correlations revealed that greater microbial richness (Faith's PD, observed species) was linked to lower glucose, depression, and stress scores, although only the glucose association remained after correction. These findings align with growing evidence reporting associations between specific microbial taxa and metabolic or psychological outcomes, rather than global changes in microbiome diversity, potentially reflecting shared gut–brain and immune-related processes (51–54). Taxonomic correlations suggested that pro-inflammatory taxa, including Proteobacteria, Desulfovibrio, and Eggerthella, were positively associated with accelerated epigenetic age, whereas Bacteroides and Lachnospira were inversely associated with damage related clocks. These patterns are consistent with prior work linking microbial composition to systemic inflammation and aging processes (55–58). Importantly, the absence of changes in alpha diversity indicates that greens supplementation did not induce broad microbiome remodeling over the 30-day intervention, and observed effects should be interpreted as taxon-specific. Such results underscore the potential interplay between diet, the microbiome, and epigenetic regulation.

No significant group-level changes were observed in daily step counts, sleep duration, or self-reported psychological measures. Nonetheless, Horvath age showed a trend toward inverse association with sleep duration and positive association with anxiety, supporting prior studies linking poor sleep and psychological distress with accelerated epigenetic aging (59–62). AdaptAge also trended inversely with daily steps, suggesting potential sensitivity of newer clocks to behavioral inputs. While non-significant, these patterns are consistent with prior literature identifying sleep and activity as potential modifiers of biological aging but should be interpreted as exploratory associations in the context of the present study. However, given the open-label design, any subjective trends observed should be interpreted cautiously, as expectation effects may influence self-reported outcomes independent of physiological change.

This study has several strengths, including a randomized crossover design that minimized interindividual variability and strengthened causal inference for molecular outcomes known to exhibit high between-subject heterogeneity. Additional strengths include the integration of multiple DNA methylation based epigenetic clocks capturing distinct biological domains, the concurrent assessment of epigenetic aging and gut microbiome outcomes, and the focus on an at-risk population with chronically low vegetable intake. The use of a commercially available, mixed greens-based supplement further enhances the translational relevance of the findings by reflecting a pragmatic, real-world dietary intervention.

The absence of a placebo control and the use of an open-label design represent key limitations, introducing the potential for expectation bias, particularly for subjective and behaviorally mediated outcomes such as sleep, quality of life, and psychological measures. Participant awareness of supplementation may have influenced self-reported perceptions or behavioral engagement independent of biological effects. Other limitations include the short intervention duration, modest sample size, and lack of dietary control; in addition, compliance with the greens supplement was not objectively verified using biomarkers such as plasma carotenoids or folate, and adherence was inferred from self-reported dietary recalls rather than biochemical confirmation. Menopausal status was not formally assessed, and the inclusion of women across a transitional reproductive age range may have contributed to variability in epigenetic aging measures. The modest sample size, combined with the known inter-individual variability of DNA methylation based epigenetic clocks, limits statistical power to detect small-to-moderate effects. Accordingly, this study should be viewed as exploratory and hypothesis-generating rather than confirmatory.

Additionally, the reliance on 16S rRNA sequencing without additional shotgun metagenomic or metatranscriptomic testing limits insight into microbial functional capacity or metabolic activity. As such, interpretations related to inflammation, metabolism, or epigenetic regulation should be considered speculative and based on inferred associations rather than direct functional evidence. Lastly, the supplement-only design, without concurrent lifestyle modifications, may have restricted the magnitude of observed effects.

The original contributions presented in the study are publicly available. This data can be found here: https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1403516↗.