Modulatory effects of exercise interventions on the gut microbiota in older adults: a systematic review and meta-analysis

Graphical abstract

Abstract

Background

Structural alterations in the gut microbiota (GM) are closely linked to aging and age-related diseases. Although exercise may improve health in older adults by modulating gut microecology, its specific effects on GM lack systematic evaluation.

Objective

To systematically assess the effects of exercise interventions on GM diversity and characteristic microbial genera in older adults.

Methods

PubMed, Embase, the Cochrane Library, and Web of Science were searched from inception to December 2025 and updated in February 2026. Eligible studies included trials evaluating the effects of exercise interventions on gut microbiota in older adults. Risk of bias and evidence certainty were assessed using the Cochrane Risk of Bias 2.0 tool and the GRADE framework, respectively. Random-effects models were used to calculate Hedges’ g. Subgroup analyses, meta-regression, and restricted cubic spline dose–response analyses were performed.

Results

Sixteen studies were included. Exercise significantly increased α-diversity, reflected by improvements in the Shannon index (g = 0.22, 95% CI: 0.06–0.38, P = 0.007) and Chao1 index (g = 0.22, 95% CI: 0.08–0.35, P = 0.002). The Observed index showed marginal significance (P = 0.052), whereas the Simpson index was unchanged. Moderate-intensity exercise demonstrated a stronger effect (g = 0.38, 95% CI: 0.18–0.58), with greater benefits in males and individuals with higher BMI. Meta-regression suggested a negative age–effect trend (β = −0.0238, P = 0.079). A significant inverted U-shaped dose–response relationship was observed for the Shannon index (non-linear P = 0.0226), with optimal effects at 700–900 METs/week. No significant changes were found in the two dominant phyla or Bifidobacterium; however, Akkermansia increased (g = 0.60, 95% CI: 0.29–0.92, P < 0.001) and Escherichia decreased (g = −0.64, 95% CI: −1.20 to −0.08, P = 0.026).

Conclusion

Exercise enhances α-diversity and modulates specific genera while maintaining core GM stability in older adults, with dose-dependent effects requiring confirmation in longitudinal studies.

1. Introduction

With the accelerating global population aging, achieving "healthy aging" has emerged as a core priority in public health. Older adults often experience a decline in physiological reserve capacity and face a high risk of sarcopenia, frailty, cognitive impairment, and metabolic diseases [1]. However, these pathophysiological processes do not solely stem from the host's own cellular senescence; mounting evidence points to a crucial underlying trigger: the gut microbiota [2]. As the human body's "second genome," the gut microbiota is not merely a collection of commensal microorganisms but a core hub maintaining host metabolic homeostasis and immune defense [3]. Studies indicate that compared to healthy adults, the gut microecology of older adults often exhibits reduced stability, a loss of diversity, and an increased abundance of opportunistic pathogens (e.g., Enterobacteriaceae) [4]. This state, termed "gut dysbiosis," is considered a major driver inducing "inflammaging"—a low-grade, chronic, systemic inflammatory state—thereby accelerating age-related pathological processes [5].

Exercise intervention, as an economical and safe non-pharmacological approach, is garnering widespread attention for its ameliorative effects on dysbiosis in older adults. Unlike probiotics or prebiotics that rely on exogenous supplementation, regular physical activity (PA) can systematically shape the microecology through the endogenous bidirectional regulatory mechanism of the "gut-muscle axis" [6], and enhance α-diversity and the enrichment of SCFA-associated and gut barrier-related taxa (e.g., Faecalibacterium prausnitzii, Roseburia, and Akkermansia muciniphila) [8,9]. These adaptations may enhance butyrate production, reinforce intestinal barrier function, and alleviate chronic low-grade inflammation, ultimately supporting immune homeostasis and metabolic health in older adults

The potential benefits of exercise on the older adults gut microbiota have been explored in multiple studies. However, it still remain some inconsistent. For instance, some studies indicate that aerobic exercise can significantly alter the microbiota structure, whereas the effects of mind-body and short-term exercise are negligible [7,8]. Furthermore, previous individual studies generally suffer from limitations such as small sample sizes, high heterogeneity in intervention protocols, and complex participant health backgrounds, making it difficult to accurately determine the overall regulatory effect of exercise interventions on the gut microbiota in older adults. Therefore, this systematic review and meta-analysis aimed to quantify the effects of exercise interventions on gut microbiota diversity and key bacterial genera in older adults, providing evidence-based insights into exercise-related microecological mechanisms and targeted exercise prescription for this population.

2. Methods

2.1. Study design

This systematic review and meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [9]. The protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) [ID: CRD420261328950].

2.2. Inclusion and exclusion criteria

The inclusion and exclusion criteria were formulated based on the Population, Intervention, Comparison, Outcomes, and Study design (PICOS) framework. This study included randomized or non-randomized parallel-group or crossover trials investigating the effects of exercise interventions on the gut microbiota in older adults. Specific inclusion criteria were: (1) the study population consisted of older adults (age ≥ 55 years); (2) the intervention involved structured exercise or physical activity; (3) the study reported changes in gut microbiota diversity indices or taxonomic abundance and provided extractable data; (4) the study design was a randomized controlled trial (RCT) or non-RCT; and (5) the publication language was English.

Exclusion criteria included: (1) meta-analyses, systematic reviews, conference abstracts, dissertations, or commentaries; (2) non-exercise intervention studies or interventions combined with diet, medication, supplements, probiotics, or antibiotics where the exercise effect could not be isolated; (3) studies with inaccessible or incomplete outcome data; and (4) animal or in vitro studies.

2.3. Search strategy

Two independent researchers systematically searched four electronic databases (PubMed, Web of Science, the Cochrane Library, and Embase) for original studies published from database inception up to December 2025. The search was updated in February 2026 to include the most recently published literature. The search strategy was constructed using Boolean logic, combining Medical Subject Headings (MeSH) and free-text terms to enhance comprehensiveness and accuracy. Specific search strings are detailed in Supplementary Table S1. Furthermore, we manually screened the reference lists of the included literature and reviewed previous research and reviews in related fields to ensure maximum coverage of all eligible studies.

2.4. Data extraction

The literature retrieved from the included studies was first imported into EndNote X9 software to remove duplicates. Subsequently, two independent reviewers screened the titles and abstracts, followed by a full-text assessment of the eligible articles. During the screening and evaluation process, relevant data were independently extracted using a standardized data extraction form (Microsoft Excel). Any discrepancies were resolved through discussion with a third reviewer until a consensus was reached.

Extracted data included study characteristics (first author, publication year, country), participant demographics (age, sex, sample size, body mass index [BMI]), intervention details (modality and duration), clinical measures (triglycerides [TG] and systolic blood pressure [SBP]), and both primary and secondary outcome data.

2.4.1. Primary outcome measures

The primary outcomes were changes in gut microbiota α-diversity, including the Observed, Chao1, Shannon, and Simpson indices. Because α-diversity estimates are highly sensitive to upstream sequence-processing procedures—including feature inference, clustering strategy, rarefaction depth, and normalization—the pooled estimates in this meta-analysis should be interpreted as relative rather than absolute ecological or functional changes. To enhance comparability across studies, quantitative synthesis was limited to OTU-based workflows, and ASV-level results were not combined in the main analysis. This restriction does not imply methodological superiority of OTU over ASV approaches, but rather provides a consistent analytical framework across included studies. OTU workflows themselves differ between studies, particularly regarding clustering thresholds, reference databases, and bioinformatic pipelines. Accordingly, information on sequencing and preprocessing procedures was extracted to provide contextual background for interpreting pooled α-diversity results.

2.4.2. Secondary outcome measures

Secondary outcomes included changes in the abundance of characteristic microbiota, assessed at both the phylum level (Bacteroidetes, Firmicutes, Verrucomicrobia, Proteobacteria) and the genus level (Akkermansia, Prevotella, Bifidobacterium, Escherichia).

To address methodological heterogeneity, study-level information on sequencing region, platform, sequence-processing approaches, rarefaction depth (if reported), and normalization or preprocessing procedures applied prior to diversity calculation was extracted. These variables were documented to facilitate interpretation of cross-study variability in α-diversity outcomes, acknowledging that differences in sequencing platforms, bioinformatic pipelines, and upstream processing may contribute to observed heterogeneity.

For the meta-analysis, the means and standard deviations (SDs) of the outcome indicators before and after the intervention were extracted. When studies presented data graphically rather than as specific numerical values, the WebPlotDigitizer tool was utilized for numerical extraction. For unreported or inaccessible data, attempts were made to contact the original authors to obtain the raw datasets.

2.5. Risk of bias assessment

The risk of bias of the included studies was evaluated independently by two researchers using the Cochrane risk-of-bias assessment tool to appraise the methodological quality of the included literature. The researchers evaluated the risk of bias across six domains (selection bias, performance bias, detection bias, attrition bias, reporting bias, and other biases), categorizing each domain as "low risk," "high risk," or "unclear risk." A comprehensive analysis of each assessment domain was conducted, and any disagreements were resolved by consulting a third researcher [10].

2.6. Evaluation of the quality of circumstantial evidence

The quality of evidence for the outcomes was assessed using GRADEpro software. Five evaluation domains were considered: limitations, inconsistency, indirectness, imprecision, and publication bias. Each domain was independently evaluated and categorized as none (not downgraded), serious (downgraded by 1), or very serious (downgraded by 2). Based on this evaluation, the overall quality of evidence was classified into one of four levels: high, moderate, low, or very low

2.7. Statistical analysis

This study conducted the meta-analysis following standardized procedures. Statistical analyses and graphical plotting were performed using R software, while the quality and risk of bias assessments of the included studies were conducted using RevMan 5.3, as recommended by the Cochrane Collaboration. To pool effect sizes, the standardized mean difference (SMD) between groups was utilized, and Hedges' g was calculated to correct for small sample sizes, expressed with a 95% confidence interval (95% CI) [11]. All statistical tests were two-tailed, and P < 0.05 was considered statistically significant. To ensure comparability across different study data, standardized conversion methods were applied to handle the various data reporting formats found in the original studies. For studies reporting standard error (SE), this was converted to standard deviation using the formula SD = SE × √n (where n is the sample size); for data presented as medians and dispersion indicators (such as range or interquartile range), means and SDs were estimated based on established methodological recommendations [12].

Between-study heterogeneity was quantitatively assessed using the I² statistic and graded according to the Cochrane Handbook criteria: I² < 40% represents low heterogeneity, 30–60% moderate heterogeneity, 50–90% substantial heterogeneity, and >75% considerable heterogeneity. Given the potential clinical and methodological variations across studies regarding intervention modalities, sample characteristics, and measurement methods, a random-effects model was employed for all primary analyses, with between-group variance estimated using the restricted maximum likelihood (REML) method [13]. When moderate or higher heterogeneity was observed, subgroup analyses and meta-regression were further conducted to explore potential sources. Subgroup analyses compared pre-defined stratification factors, reporting within-group effect sizes and between-group difference test results [14]. In the linear meta-regression, the effect size served as the dependent variable, and pre-defined continuous covariates were incorporated into a mixed-effects model for analysis, reporting regression coefficients, standard errors, and the model's explanatory power (R²). Furthermore, the dose-response relationship between exercise volume and intervention efficacy was estimated using restricted cubic splines (RCS), and the goodness-of-fit between linear and non-linear models was compared via a likelihood ratio test (LRT) [15]. Publication bias was visually assessed using funnel plots and quantitatively analyzed using Egger's regression asymmetry test; an Egger's test P < 0.05 indicated potential small-study effects or publication bias [16].

3. Results

3.1. Literature search results

Through searching the Web of Science, PubMed, Embase, and Cochrane databases, a total of 1,515 relevant records were identified. After removing 68 duplicates, the remaining 1,447 articles proceeded to the title and abstract screening phase, during which 1,126 articles were excluded due to lack of relevance to the research topic. The remaining 321 articles underwent full-text evaluation. Based on the pre-established inclusion and exclusion criteria, 305 articles were excluded. The primary reasons for exclusion were: mismatched outcome indicators (48 articles), mismatched participant age (102 articles), interventions falling outside the study scope (28 articles), conference abstracts or review articles (29 articles), and unextractable data (79 articles). Ultimately, a total of 16 studies were included in this systematic review and meta-analysis (See Fig. 1).

Fig. 1.

Flowchart of literature screening.

3.2. Basic characteristics of the included studies

This study included a total of 16 studies focusing on older adult populations. The studies originated from Iran [17], China [7,[18], [19], [20], [21], [22]], Spain [23,24], the USA [[25], [26], [27]], the UK [28], Canada [29], France [30]and Japan [31], comprising a total sample size of 3,241 individuals. The average age of the participants was approximately 67.0 ± 8.5 years. All studies reported gender composition; some studies exclusively enrolled females, while others included both males and females. The mean BMI reported across the studies ranged from 21.16 to 33.0 kg/m². The interventions or exposure formats employed included aerobic exercise, resistance training, combined aerobic and resistance training, high-intensity interval training (HIIT), and daily physical activity assessments, with intervention durations ranging from 6 to 48 weeks. The primary outcome measures were indicators of gut microbiota α-diversity and changes in the abundance of specific microbial taxa. Detailed characteristics are presented in Table 1.

Table 1.

Baseline characteristics table.

Study ID	Country	N	Intervention	Female (%)	Age	BMI	Session Duration	Frequency	Intervention Period	Sequencing	Method	Outcomes
												α-diversity	Taxa
Agyin-Birikorang 2025	American	12	Whole-body resistance training (10 bodyweight squats, 25 jumping jacks + 3 sets progressive warm-up)	66.70%	59 ± 5	NA	60min	2	10	16S rRNA V4, Illumina MiSeq	ASV (DADA2, QIIME2)	Observed OTU	—
ANNAËLLE 2024	France	16	Cycling HIIT (power adjusted by HRmax, cadence 50–70 rpm) or treadmill HIIT (1% incline, speed adjusted by HRmax)	0	54.2 ± 9.6	Cycling: 30.7 ± 2.8; Running: 29.2 ± 1.5	30 min	3	12	16S rRNA V4, Illumina MiSeq	ASV (DADA2, QIIME2)	Shannon	—
Cheng 2022	China	85	Progressive aerobic exercise (60–75% VO₂max) or exercise + high-fiber low-carbohydrate diet (30–40% total energy; 37–40% carbohydrate, 35–37% fat, 25–27% protein)	79%	50−65	NA	30−60min	2–3	32	16S V3–V4, Illumina MiSeq	ASV (DADA2, QIIME2)	Shannon	—
García-Gavilan 2024	Spain	627	Intensive lifestyle intervention (Mediterranean diet + moderate-to-vigorous physical activity)	44%	64.7 ± 5.0	33.0 ± 3.5	45min	7	48	16S rRNA + LC-MS/MS Metabolome	Not described	Chao1, Shannon, Simpson	—
LANGSETMO 2019	American	373	SenseWear® Pro3 armband monitoring	0	84 ± 3.9	26.9 ± 3.8	Continuous wear	Continuous wear	1	16S rRNA V4, Illumina MiSeq	OTU (97% clustering)	Shannon, Simpson	—
Moore 2022	American	14	Resistance training (intensity adjusted by RPE 7–9)	0	65 ± 9	28.1 ± 3.1	45min	2	6	16S rRNA(V4)+ MiSeq	ASV (DADA2, QIIME2)	Observed OTU, Shannon, Simpson	—
Morita 2019	Japan	18	Brisk walking (≥3 METs)	100%	70 ± 6.67	21.7 ± 3.11	60min	7	12	16S rRNA V4, Illumina MiSeq	OTU (97% clustering)	—	Bifidobacterium
Ruiz-Limón 2024	Spain	198	Mediterranean diet + moderate exercise (Δ28.96 ± 23.33%) or Mediterranean diet + intensive exercise (Δ273.64 ± 221.42%)	Moderate: 51.5%; Intensive: 56.6%	64.45 ± 4.64	32.8	Moderate:>30 min; Intensive: >60 min	NA	48	16S + Ion Torrent	ASV (DADA2, QIIME2)	—	Verrucomicrobia, Akkermansia
Shah 2023	Canada	350	Moderate-intensity physical activity at home, work, and leisure	71.80%	57.90 ± 6.16	26.86 ± 1.45	150−500min	Total PA assessed	1	16S V3–V4, MiSeq	ASV (DADA2, QIIME2)	Shannon	—
Shi 2021	China	2151	Regular leisure-time physical activity (moderate 3.0–5.9 METs or vigorous ≥6.0 METs)	26%	64 ± 8	24.4 ± 3.4	NA	＞1	＞9	16S rRNA V4, Illumina HiSeq PE250	OTU (97% clustering)	Observed OTU, Chao1, Shannon, Simpson	Bacteroidetes, Bifidobacterium
Torquati 2022	Britain	12	High-intensity aerobic (85–95% HR) + high-intensity resistance (RPE ≥ 17)	NA	64.3 ± 6.4	NA	26min	3	8	16S rRNA, Illumina NextSeq500	OTU (97% clustering)	Shannon, Simpson	—
Vahed 2025	Iran	50	Combined exercise (40–60% HRR)	100%	60.66 ± 6.61	29.71 ± 4.04	30 min	5	12	Not described	Specific qPCR, 5 target species	—	Akkermansia, Prevotella, Escherichia
Wu 2023	China	21	Aerobic exercise (40–60% age-predicted HRmax) or Liuzijue (breathing–movement exercise)	65%	AT:60.7 ± 8.03; Liuzijue: 65.19 ± 6.48	AT: 24.84 ± 2.53; Liuzijue: 23.88 ± 2.44	60min	3	12	16S rRNA V3–V4, PacBio Sequel II	OTU (97% clustering)	Chao1, Shannon, Simpson	—
Zhong 2021	China	12	Aerobic + resistance training	100%	69.93 ± 4.5	22.69 ± 1.21	60min	4	8	16S V4, Illumina HiSeq 2500 PE250	OTU (97% clustering)	Observed OTU, Chao1, ace, Shannon, Simpson	Verrucomicrobia, Proteobacteria, Akkermansia, Prevotella, Escherichia, Bifidobacterium
Zhong 2021	China	100	activPAL3 micro-monitor to assess habitual PA	44%	69 ± 3.0	27.8 ± 4.0	Continuous wear	Continuous wear	1	16S rRNA V3–V4, Illumina MiSeq	OTU (97% clustering)	Observed OTU, Chao1, ace, Shannon	—
Zhong 2022	China	15	Moderate-intensity aerobic + resistance training (bodyweight or simple equipment)	100%	66.38 ± 4.07	21.16 ± 2.17	60min	4	8	16S V3–V4, Illumina MiSeq	OTU (97% clustering)	Observed OTU, Chao1, Shannon, Simpson	Bacteroidetes, Firmicutes, Proteobacteria

3.3. Risk of bias assessment

The methodological quality of the 16 included studies was evaluated using the Cochrane risk-of-bias assessment tool. Overall, the studies were predominantly rated as having a low risk of bias in the domains of random sequence generation and allocation concealment. However, due to the inherent nature of exercise interventions, blinding of participants and personnel (performance bias) was mostly rated as high risk. The blinding of outcome assessment (detection bias) primarily presented an unclear risk of bias. The majority of studies were evaluated as having a low risk of bias concerning incomplete outcome data, selective reporting, and other biases. (Supplementary Fig. S1). Based on the comprehensive risk of bias judgment, 7 studies (43.8%) were of high quality, 6 (37.5%) were of moderate quality, and 3 (18.7%) were of low quality. The distribution in the funnel plots was generally symmetrical, and Egger's test revealed no obvious signs of publication bias.

3.4. Results of meta-analysis

3.4.1. Primary outcomes

This study systematically evaluated the effects of exercise interventions on the α-diversity of the gut microbiota in older adults. The primary indicators included the Shannon, Chao1, Observed species, and Simpson indices. Under the random-effects model, the analysis for the Shannon index included 17 studies (k = 17, N = 3421). The pooled analysis demonstrated that exercise interventions significantly increased the Shannon index (Hedges’ g = 0.22, 95% CI: 0.06–0.38, P = 0.007), with moderate between-study heterogeneity (I² = 57.2%). The Chao1 index analysis included 9 studies (k = 9, N = 2516) and similarly showed a significant increase following exercise intervention (g = 0.22, 95% CI: 0.08–0.35, P = 0.002), with no obvious heterogeneity observed (I² = 0%). The Observed species index included 8 studies (k = 8, N = 2302) and exhibited a marginally significant increase associated with exercise (g = 0.19, 95% CI: −0.002–0.39, P = 0.052), with low between-study heterogeneity (I² = 0%). In contrast, the Simpson index, which included 10 studies (k = 10, N = 393), showed no significant improvement (g = 0.14, 95% CI: −0.41–0.69, P = 0.62) and exhibited high heterogeneity (I² = 89.1%) (See Fig. 2).

Fig. 2.

Effects of exercise interventions on the gut microbiota in older adults.

Note: The certainty of evidence was assessed using the GRADE approach: ++++ indicates high-certainty evidence; +++ indicates moderate-certainty evidence; ++ indicates low-certainty evidence; + indicates very low-certainty evidence.

To explore potential sources of between-study heterogeneity for the Shannon index, we conducted sensitivity analyses and evaluated publication bias (Supplementary Fig. S2); no obvious publication bias was detected. Further subgroup analyses revealed that exercise intensity may serve as an important moderating factor. The moderate-intensity exercise subgroup showed a significant effect (g = 0.38, 95% CI: 0.18–0.58, P = 0.0002, I² = 10%), whereas the low- and high-intensity subgroups did not reach statistical significance. The difference between groups was statistically significant (P_d = 0.03) (See Table 2).

Table 2.

Subgroup analysis of the effects of exercise interventions on gut microbiota regulation in older adults.

	Moderator	Subgroup	n	Hedges’g	95%CI	Ph	I²%	Pd
Shannon index	BMI	Normal weight	87	0.18	(−0.23,0.58)	0.4	49%	0.06
		Overweight	1191	0.22	(0.07,0.36)	0.004	49%	0.05
	Sex	Female	553	0.24	(−0.07,0.55)	0.02	56%	0.13
		Male	725	0.13	(0.01,0.24)	0.55	0%	0.03
	Exercise intensity	Low intensity	970	0.10	(−0.08,0.29)	0.03	54%	0.27
		Moderate intensity	281	0.38	(0.18,0.58)	0.35	10%	＜0.001
		High intensity	27	0.21	(−0.32,0.73)	0.88	45%	0.45
	TG	Normal TG	128	0.36	(0.12,0.60)	0.73	0%	0.003
		High TG	83	0.46	(0.10,0.83)	0.96	0%	0.01
	SBP	Normal SBP	463	0.03	(−0.10,0.16)	0.92	0%	0.62
		High SBP	76	0.15	(−0.93,1.24)	0.001	85%	0.78
Acetobacteraceae	BMI	Normal weight	12	0	(−0.77,0.77)	1	0%	1
		Over weight	233	0.27	(−0.20,0.75)	0.002	84%	0.26
	Exercise intensity	Low to Moderate intensity	140	0.29	(−0.45,1.04)	0.002	84%	0.44
		High intensity	105	0.11	(−0.16,0.38)	0.84	0%	0.42
	TG	Normal TG	12	0	(−0.77,0.77)	1	0%	1
		High TG	233	0.26	(−0.24,0.75)	0.001	86%	0.31
	SBP	Normal SBP	204	0.02	(−0.18,0.21)	0.61	0%	0.87
		High SBP	41	0.75	(0.33,1.17)	0.01	69%	＜0.001

Note: n represents the number of participants; Hedges’ g is the effect size used to measure the magnitude of differences between two groups; I² is a measure of between-study heterogeneity and is expressed as a percentage; Pₕ represents the P value for the pooled effect size; Pd represents the P value for differences in effect sizes between subgroups; TG denotes triglycerides; and SBP denotes systolic blood pressure. High = High risk of bias (low-moderate quality); Low = Low risk of bias (high quality).

Building upon these findings, we further examined whether a monotonic linear dose-response relationship existed between exercise volume (METs) and the Shannon index. Linear meta-regression revealed no significant linear trend (β = 0.0004, P = 0.33), suggesting that a simple linear model may be insufficient to characterize the correlational structure between exercise dose and intervention efficacy (Fig. 3). Consequently, we employed a restricted cubic spline (RCS) model for a non-linear dose-response analysis. The results demonstrated a significant non-linear relationship between exercise volume and the Shannon index (non-linear term P = 0.0226; LRT P = 0.025). The non-linear model provided a better fit than the linear model (AIC: 18.00 vs. 20.99; BIC: 20.26 vs. 22.91), and the likelihood ratio test confirmed that the model improvement was statistically significant (LRT χ² = 4.99, P = 0.025). Moreover, the proportion of between-study heterogeneity explained by the non-linear model increased (R² = 47.4% vs. 41.7%). The dose-response curve exhibited an inverted U-shaped trend: the effect size increased with exercise volume, peaked within the range of approximately 700–900 METs/week, and gradually declined thereafter. When exercise volume reached ≥1100 METs/week, the effect size approached zero, and the confidence intervals in the high-dose range widened noticeably (Fig. 4).

Fig. 3.

Meta-regression analysis of the regulatory effects of exercise interventions on gut microbiota diversity in older adults.

Fig. 4.

Nonlinear dose–response relationship between exercise volume (METs/week) and the Shannon index.

In other subgroup analyses, stratification by sex indicated a significant intervention effect in the male subgroup (g = 0.13, 95% CI: 0.01–0.24, P = 0.03, I² = 0%), whereas the female subgroup did not reach statistical significance, with a significant between-group difference (P_d = 0.02). Stratification by BMI showed a significant effect in the overweight subgroup (g = 0.22, 95% CI: 0.07–0.36, P = 0.004) but no significant difference in the normal-weight subgroup (P_d = 0.06). Stratification by TG and SBP did not demonstrate a stable moderating effect (Table 2). Meta-regression analysis further assessed the impact of continuous variables on the effect size. Age exhibited a trending negative correlation with the intervention effect (β = −0.0238, SE = 0.0136, P = 0.079), and the model explained approximately 27.7% of the between-study heterogeneity (R² = 27.73%). This suggests that in older adult populations, the ameliorative effect of exercise on α-diversity may gradually diminish with advancing age. Conversely, BMI did not show a significant moderating effect (β = 0.0168, P = 0.652, R² = 0%), indicating no obvious linear association between BMI (as a continuous variable) and intervention efficacy. Similarly, no significant moderating effects were observed for TG (β = 0.0481, P = 0.417) or SBP (β = −0.0118, P = 0.689) (See Fig. 3).

To further assess the robustness of the main findings, we performed restricted subgroup analyses by including only randomized controlled trials (RCTs) and, separately, only studies with low risk of bias and high methodological quality.In the RCT-restricted analyses, significant positive effects on the Shannon index were observed in the overweight subgroup (g = 0.43, 95% CI: 0.09–0.77, P = 0.01), the moderate-intensity exercise subgroup (g = 0.58, 95% CI: 0.27–0.90, P < 0.001), the TG subgroup (g = 0.39, 95% CI: 0.07–0.72, P = 0.02), and the high-TG subgroup (g = 0.46, 95% CI: 0.11–0.81, P = 0.01). In contrast, no statistically significant effects were observed in the normal-weight subgroup, the sex-stratified subgroups, the low- and high-intensity exercise subgroups, or the systolic blood pressure (SBP)-related subgroups (See Supplementary Table S2).

In the analyses restricted to studies with low risk of bias and high methodological quality, the overall direction of the effects was broadly consistent with that of the RCT-restricted analyses. Significant positive effects remained evident in the overweight, moderate-intensity exercise, normal-TG, and high-TG subgroups. In addition, statistically significant effects were observed in the female subgroup (g = 0.34, 95% CI: 0.07–0.61, P = 0.01) and the normal-SBP subgroup (g = 0.46, 95% CI: 0.17–0.74, P = 0.002), whereas the normal-weight, male, low- and high-intensity exercise, and high-SBP subgroups remained non-significant. Overall, the restricted subgroup analyses showed findings that were generally consistent with the primary analyses, further supporting the robustness of the main results (See Supplementary Table S3).

3.4.2. Secondary outcomes

In the secondary outcome analysis, we further evaluated the effects of exercise interventions on the abundance of specific phyla and characteristic genera. At the phylum level, no stable changes were observed following exercise intervention. Specifically, Bacteroidetes (k = 5) did not reach statistical significance (g = 2.68, 95% CI: −1.45–6.80, P = 0.203) and exhibited high heterogeneity (I² = 98.8%); Firmicutes (k = 2) and Verrucomicrobia (k = 4) also showed no significant differences, with low heterogeneity. Proteobacteria (k = 4) displayed a declining trend (g = −1.14, 95% CI: −2.36–0.0768, P = 0.066), though it fell short of statistical significance.

In contrast, more definitive changes were observed at the genus level. The analysis of Akkermansia included 5 studies (k = 5, N = 245), and the pooled results revealed a significant increase in its abundance (g = 0.604, 95% CI: 0.417–0.791, P < 0.001) with low between-study heterogeneity (I² = 4.29%). No significant changes were observed for Prevotella (k = 4) or Bifidobacterium (k = 4). Conversely, Escherichia (k = 3) exhibited a significant decrease (g = −0.640, 95% CI: −1.21 to −0.0746, P = 0.026) with a heterogeneity of 31.3% (Fig. 2).

We conducted sensitivity analyses and evaluated publication bias for the secondary outcomes (Supplementary Fig. S3), finding no distinct publication bias. To further investigate whether the effect of exercise on Akkermansia varied across different study characteristics, we performed subgroup analyses for this genus. Stratification by BMI, intervention intensity, and TG revealed no significant between-group differences. A significant increase in Akkermansia abundance was observed exclusively in the high SBP subgroup (g = 0.75, 95% CI: 0.33–1.17, P = 0.0004) (Table 2). Given the limited number of studies included for Akkermansia, meta-regression analysis was not pursued. For Escherichia (k = 3), subgroup and meta-regression analyses were not performed due to the small number of included studies.

3.5. GRADE quality of evidence assessment

This study utilized the GRADE approach to assess the quality of evidence for the primary outcomes. The results indicated that the evidence quality for the Shannon and Chao1 indices was moderate. The Observed species and Simpson indices were rated as low-quality evidence due to unstable results or high heterogeneity. Regarding characteristic taxa, Akkermansia was supported by moderate-quality evidence, while the remaining taxa were mostly rated as low-quality evidence due to the limited number of studies or high inconsistency (See Fig. 2).

4. Discussion

The findings demonstrate that exercise significantly enhanced gut microbiota α-diversity in older adults, as reflected by increases in the Shannon and Chao1 indices and a marginal improvement in Observed species. Dose-response analysis further showed a significant inverted U-shaped association for the Shannon index, with optimal effects at approximately 700–900 METs/week. In contrast, the Simpson index remained unchanged, likely because it is more sensitive to evenness than to low-abundance taxa. As α-diversity metrics are influenced by sequence-processing methods, including OTU/ASV inference, rarefaction depth, and normalization, these findings should be interpreted as relative effects rather than definitive ecological or functional shifts. At the compositional level, Firmicutes, Bacteroidetes, and Bifidobacterium remained stable, whereas Akkermansia increased and Escherichia decreased. Overall, exercise appears to improve microbial diversity and modulate specific genera while preserving core gut microbiota stability.

4.1. Changes in gut microbiota diversity indices

Gut microbiota diversity, particularly α-diversity indices, is frequently employed as a key metric to gauge the structural complexity and ecological stability of the gut microecology [32]. The present study suggests that exercise interventions may be associated with increases in the Shannon and Chao1 indices in older populations. This may indicate potential improvements in microbial richness and community complexity, rather than necessarily reflecting a systemic rearrangement of dominant taxa.In older adults, declining microbial diversity and compromised ecosystem resilience are common, further exacerbated by immunosenescence, chronic low-grade inflammation, polypharmacy, and monotonous diets [33]. In this context, regular exercise may be associated with trends toward increased colonization or maintenance of low-abundance taxa by ameliorating host-side limiting factors. These include potential improvements in intestinal barrier function, mucosal immune homeostasis, reduced inflammatory burden, and changes in the gut microenvironment, such as altered bile acid metabolism, intestinal motility, and luminal substrate transit times, which may favor the persistence of low-abundance commensals [[34], [35], [36]].

The observed synchronous increase in Shannon and Chao1 indices may reflect trends toward enhanced microbial richness and community complexity, with possible involvement of low-abundance taxa. However, these α-diversity indices do not provide direct evidence for the definitive restoration of low-abundance taxa or specific changes in community composition. The Simpson index did not change significantly, which is consistent with its greater sensitivity to community evenness rather than richness of low-abundance species. The combination of increased Shannon and Chao1 indices with a stable Simpson index suggests that exercise primarily influences richness-related aspects of the aging gut microbiome without substantially altering the overall distribution of dominant taxa. These observations align with Moore’ [26] findings in older adults undergoing resistance training and further support the interpretation that exercise promotes diversity-related trends rather than comprehensive restructuring of dominant taxa [37].

Subgroup analyses revealed notable inter-individual variability in the microecological effects of exercise. Regarding sex differences, the male subgroup demonstrated significant and highly consistent benefits, whereas older females, despite showing a high effect size, exhibited substantial heterogeneity. This discrepancy may be linked to differences in body composition, hormonal profiles, or baseline inflammatory states [38]. Concerning BMI, the enhancement in diversity was more pronounced in overweight older adults. This could largely be attributed to overweight individuals often present with chronic low-grade inflammation and a suppressed microbiota structure [39], and exercise interventions can mitigate this inflammation-related suppression by reducing visceral fat and lowering systemic inflammation, thereby triggering a significant restorative growth [40]. Meta-regression indicated a trending negative correlation between age and effect size, implying that as age advances, declining microbiome plasticity and diminishing host physiological reserves may constrain the regulatory magnitude of exercise signals on the microecology [41,42].

Crucially, the dose-response analysis uncovered a significant "inverted U-shaped" relationship between the Shannon index and exercise volume, peaking at 700–900 METs/week. This suggests the existence of an "optimal adaptation window" for exercise's effect on the gut microecology. Moderate exercise helps optimize the intestinal microenvironment, whereas excessively high loads might attenuate benefits—or even induce transient gastrointestinal ischemia/reperfusion-related responses and permeability alterations—due to stress reactions and visceral blood flow redistribution [43,44]. Given that evidence in the high-dose range remains limited and carries considerable uncertainty, larger-scale studies with standardized exercise dose reporting and long-term follow-ups are necessary to validate the optimal exercise prescription range for different subpopulations of older adults.

Additionally, to further assess the robustness of the findings, we conducted restricted subgroup analyses including only randomized controlled trials (RCTs) and, separately, only studies with low risk of bias. The restricted analyses showed trends broadly consistent with the primary results, and the beneficial effects of exercise on gut microbiota diversity in older adults remained significant in the overweight subgroup and under moderate-intensity exercise interventions. Minor differences in statistical significance across some subgroups may be attributable to the reduced sample size and the exclusion of non-RCTs and high-risk-of-bias studies that may introduce confounding. Overall, these findings further support the robustness of the main results.

4.2. Alterations in characteristic gut microbiota

Compared to macroscopic shifts in overall community diversity, alterations in the abundance of key core genera more accurately reflect the substantive impact of exercise on aging microecological functions. Previous studies have highlighted that a hallmark of the aging gut is the disruption of the "beneficial vs. pathogenic bacteria" balance. Litvak [45] proposed that the expansion of the phylum Proteobacteria can serve as a critical signature of gut dysbiosis and epithelial dysfunction. Concurrently, Gao [46] emphasized that genera associated with mucosal barrier function, such as Akkermansia, may deplete during aging or inflammatory states, thereby undermining the maintenance of intestinal barrier homeostasis. In this context, our findings show that exercise increased Akkermansia and decreased Escherichia, with a trend toward reduced Proteobacteria, suggesting that exercise may help counteract age-related dysbiotic signatures.

From an ecological perspective, Akkermansia and Escherichia represent two distinct functional trajectories within the gut ecosystem. The former is an obligate mucin-degrading bacterium enriched in the intestinal mucus layer and is intimately linked to intestinal barrier homeostasis [47]. Conversely, under conditions of dysbiosis or inflammation, members of Proteobacteria, typified by Escherichia, frequently exhibit relative expansion—a shift regarded as a hallmark of gut microecological imbalance [48]. However, these functions are highly strain-specific and context-dependent, highlighting the complexity of genus-level variation across host conditions. In addition, genus-level findings should be interpreted cautiously because methodological differences across studies—including sequencing strategies, bioinformatic pipelines, and reference databases—may affect taxonomic resolution and classification accuracy. By comparison, higher taxonomic levels such as family and phylum are generally more robust across analytical approaches.

Nonetheless, the relative shifts in the abundance of these two genera may reflect changes in the luminal microenvironment associated with exercise. Existing research suggests that regular exercise can influence intestinal immune and metabolic states and has been linked to alterations in SCFA-related pathways [49]. These alterations may contribute to a microenvironment that is more favorable for the maintenance of Akkermansia's relative abundance [50]. Simultaneously, reduced inflammatory tone and an improved metabolic environment may be associated with a relative reduction in the competitive advantage of certain Proteobacteria members, including Escherichia [51]. In our subgroup analysis, the increase in Akkermansia was primarily observed in individuals with high SBP, with no significant changes noted in normotensive individuals. This suggests that the host's metabolic background may modulate the magnitude of response of specific genera to exercise interventions.

Bifidobacterium and the two dominant phyla (Firmicutes and Bacteroidetes) predominantly fulfill structural maintenance and basal metabolic roles within the gut ecosystem [52]. As long-term colonizing core taxa, they participate in fundamental physiological processes such as energy metabolism, SCFA production, and immune regulation; thus, their relative abundances typically exhibit high homeostatic stability [53]. In the present study, no significant changes were observed in these taxa, which is consistent with the stability indicated by the Simpson index. At the same time, such stability at the phylum level primarily reflects structural consistency rather than functional uniformity. SCFAs represent community-level metabolic outputs arising from complex cross-feeding interactions among diverse microbial populations and are not directly determined by variations in individual taxa or broad taxonomic groups.The absence of significant changes at the phylum level primarily reflects the stability of overall community structure, while functional outputs such as SCFA production may still be influenced by interactions among multiple taxa. Similarly, the unchanged abundance of Bifidobacterium may reflect its reliance on the availability of fermentable substrates, such as dietary fibers [54], suggesting that exercise alone may have limited influence on its ecological niche in the absence of concurrent dietary modifications.

Overall, exercise exerts a selective modulatory effect on the gut microbiota in older adults. The increase in Akkermansia and decrease in Escherichia may reflect shifts related to barrier function and inflammation, whereas the stability of the two dominant phyla and Bifidobacterium suggests that the core microbiota structure remains largely preserved. Larger, long-term studies are needed to clarify these effects across different metabolic backgrounds and their clinical implications.

4.3. Limitations of the study

First, the number of studies available for some outcomes, particularly genus- and subgroup-level analyses, was limited, which may reduce the precision and robustness of the estimates. Second, although some host metabolic markers were reported, functional intestinal indicators such as short-chain fatty acids and barrier-related markers were rarely assessed, limiting mechanistic interpretation. Third, differences in microbiome sequence-processing methods may have introduced technical heterogeneity, so the pooled α-diversity results should be interpreted as relative changes rather than absolute ecological or functional shifts. Finally, variations in diet and lifestyle across countries and regions may limit the generalizability of the findings.

5. Conclusion

This systematic review and meta-analysis showed that exercise significantly improves gut microbiota α-diversity in older adults, as reflected by increases in the Shannon and Chao1 indices. Subgroup analyses suggested that moderate-intensity exercise produced more consistent benefits, with stronger responses in males and individuals with higher BMI. Dose-response analysis further identified a significant inverted U-shaped relationship between exercise volume and the Shannon index, with optimal effects at approximately 700–900 METs/week. In contrast, the Simpson index, the dominant phyla (Firmicutes and Bacteroidetes), and Bifidobacterium remained largely unchanged, suggesting that exercise enhances diversity while preserving core microbiota stability. At the genus level, Akkermansia increased and Escherichia decreased, indicating selective modulation of specific functional genera; the increase in Akkermansia was mainly observed in individuals with high systolic blood pressure. Overall, the effects of exercise on the aging gut microbiota appear dose-dependent and subject to inter-individual variation. Larger, long-term studies are needed to define optimal exercise strategies and confirm their microecological safety across different populations.

CRediT authorship contribution statement

Jiahui Liu: Methodology, database searches, study screening, data extraction, Writing - original draft, Writing - review & editing. Zihan Bao: Methodology, study screening, data extraction, Formal analysis, Writing - original draft, Writing - review & editing. Yifan Zhou: Methodology, study screening, data extraction, Writing - original draft, Writing - review & editing. Weihao Zhang: database searches, study screening, data extraction, Formal analysis, Writing - review & editing. Jian Sun: Funding acquisition, Methodology, study screening, data extraction, Formal analysis, Writing - review & editing.

Ethics approval statement

Not applicable.

Declaration of Generative AI and AI-assisted technologies in the writing process

No generative AI tools were used in manuscript writing, data analysis, figure creation or artwork preparation in this study.

Funding

This study was supported by the Major Project of Philosophical and Social Sciences Research, Department of Education of Hubei Province (Grant No. 17ZD035), as well as the Joint Fund Project for Sports Innovation and Development, Natural Science Foundation of Hubei Province (Grant No. 2025AFD659). The authors gratefully acknowledge the financial support from these funding bodies.

Data statement

The research data are available from the corresponding author on reasonable request.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is Supplementary data to this article: