Gut microbiota regulates exercise-induced hormetic modulation of cognitive function

Summary

Background

Lifestyle factors, particularly physical exercise, significantly influence brain structure and cognitive function through a hormetic effect —a phenomenon where low to moderate doses of a stimulus (in this case, exercise) induce beneficial adaptations, while excessive doses could lead to detrimental effects. This effect depends on exercise intensity and duration, though the underlying mechanisms remain largely unexplored. Recently, the gut microbiota has emerged as potent modulator of lifestyle-induced changes in brain and behaviour.

Methods

We used a 40-min, 1200 cm/min exercise protocol. We measured cognition through several tests and analysed microbiota composition comparing adult exercised animals to sedentary controls. Finally, we performed fecal microbiota transplantation from exercised to sedentary mice.

Findings

Exercise enhances cognitive abilities related to object recognition and object location memory, as well as increases hippocampal neurogenesis. However, these cognitive and neurogenic benefits vanish when the exercise intensity or duration is increased. Furthermore, we identified significant changes in alpha and beta diversity and distinct bacteria composition profiles in the gut microbiota associated with different exercise regimens. Specific bacterial families showed altered relative abundances depending on exercise intensity and duration, with certain families' quantities significantly correlating with cognitive performance (Angelakisella, Acetatifactor, Erysipelatoclostridium, and Coriobacteriaceae UCG-002.). To explore causal mechanisms, we performed fecal microbiota transplantation from exercised to sedentary mice, which replicated the cognitive and neurogenic changes observed in the donor animals.

Interpretation

These findings suggest that the hormetic effects of physical exercise on cognitive function and neurogenesis are mediated by corresponding changes in the gut microbiota, highlighting a novel mechanistic link between exercise, brain function, and gut microbiota composition.

Funding

E.C. and P.M. were funded by predoctoral fellowship (FPI) grants from the Spanish Ministry of Economy and Competitiveness (BES-2017/080415 E.C.) and the Spanish Ministry of Science and Innovation (PRE2020/093032 P.M.), and P.T. by a predoctoral fellowship (FPU) from the Spanish Ministry of Universities (18/00069). Work was supported by project grants PID2019-110292RB-100 and PID2022-136891NB-I00 (from Spanish Ministry of Science and Innovation), (to J.L.T.).

Research in context.

Evidence before this study

Prior research has established that physical exercise enhances cognitive function, with moderate-intensity exercise providing optimal benefits in a hormetic, dose-dependent manner. However, systematic reviews of the literature identified three main gaps: a limited mechanistic understanding of how exercise intensity thresholds affect adult hippocampal neurogenesis; insufficient evidence linking specific gut microbiota changes to exercise-induced cognitive improvements; and significant variability in exercise protocols, complicating cross-study comparisons. While animal and human studies consistently report that moderate exercise improves cognition and neurogenesis, the involvement of the gut-brain axis in these effects remained largely speculative, with most evidence being correlational rather than causal.

Added value of this study

This study addresses these gaps by systematically investigating how different exercise intensities affect adult hippocampal neurogenesis, cognitive performance, and gut microbiota composition in mice. Our findings demonstrate a clear hormetic relationship between exercise dose and cognitive/neurogenic outcomes. Importantly, using fecal microbiota transplantation, we provide the first causal evidence that exercise-induced changes in the gut microbiota are sufficient to transfer cognitive and neurogenic benefits to sedentary animals. We also identify specific microbial taxa (such as Lachnospiraceae and Coriobacteriaceae UCG-002) and intestinal barrier modifications that mediate these effects, offering new insights into the gut microbiota's role in exercise-induced cognitive benefits.

Implications of all the available evidence

These findings, in the context of existing evidence, have important implications for human health. They suggest that personalised exercise protocols, considering their impact on the gut microbiota, could be developed to optimise cognitive benefits and reduce risks. This research advances our understanding of the mechanisms linking lifestyle, the gut microbiome, and brain health, and supports the development of targeted strategies for preventing or treating cognitive disorders. Future studies should focus on translating these insights into clinical practice and exploring how lifestyle interventions can be tailored to individual microbiota profiles.

Introduction

Physical exercise has been widely shown to exert widespread beneficial effects in both humans and in experimental animals.1, 2, 3 However, these beneficial effects obtained from moderate exercise can become detrimental when the amount, intensity, and/or duration are increased. Importantly, it is known that exercise effects follows a hormetic-like biphasic dose–response, where the biological response is increased with low doses but decreased with higher doses (inverted U-shaped curve response).^4,5 These beneficial effects have been reported in peripheral systems, such as the musculoskeletal or the cardiovascular systems, as well as in the nervous system. Indeed, moderate exercise is well known for its neurogenic properties in the hippocampus, which explains its pro-cognitive effects.^6,7 Interestingly, the increase in adult hippocampal neurogenesis (AHN) promoted by exercise also appears to follow a hormetic curve.⁸ However, comparing different studies on exercise and its effects is challenging due to the diversity in exercise protocols, the parameters studied, and the species used in experiments. Furthermore, clear evidence of this hormetic profile has not been demonstrated so far, despite its relevance to the neurobiology of exercise. This is particularly significant because subtle changes in exercise intensity, might cause either positive or negative effects, and personalised training protocols for human subjects are lacking.⁹

Gut microbiota is one of the key components of the intestinal ecosystem and plays an essential role in health, including protection against pathogens, the configuration and maturation of immunity, regulation of metabolic intake, and absorption of nutrients and drugs.¹⁰ These microorganisms play also a critical role in maintaining optimal gut function and overall body health¹¹ through their ability to produce and release a variety of neurotransmitters, essential vitamins and amino acids, and hormones from specialised enteroendocrine cells distributed in the intestinal epithelium.¹² It is largely reported that an imbalance of the gut microbiota has been associated with several disorders beyond the intestinal system. Certain pathologies are associated with dysbiosis of the gut microbiota, such as inflammatory bowel diseases, diabetes, metabolic diseases, cancer, and cardiovascular diseases.^10,11,13 This has led to increasing investigation of therapeutic approaches, including diet, antibiotics, prebiotics, probiotics, or fecal microbiota transplantation, both to understand its function and to improve certain pathologies.

More recently the microbiota has emerged to play a very important role on brain and behaviour. Indeed, a microbiota-gut-brain axis has been proposed and refers to the bidirectional communication between the central nervous system (CNS) and gut microbes involving a complex network of communication pathways between the endocrine system, the hypothalamic-pituitary-adrenal axis, the enteric nervous system, and the immune system.¹⁴ For example, gut microbes produce metabolites such as short-chain fatty acids (SCFAs), neurotransmitters (e.g., serotonin, GABA), and immune modulators that can influence brain function. Conversely, the brain can alter gut microbiota composition through stress responses, autonomic nervous system signalling, and changes in gut motility and permeability.¹⁵ Gut microbiota have been implicated in a variety of host's CNS activities, such as cognition, AHN and stress response, and similarly, brain activity affects microbial composition.16, 17, 18

In recent years, attention has focused on the relationship between lifestyle and the microbiota.¹⁹ Given that the composition and function of the gut microbiota is highly plastic and sensitive to a wide variety of environmental factors, it makes the microbiota an attractive target for disease prevention and treatment, as mentioned before. Indeed, many studies have shown that different behaviours modify the microbiota, including sleep patterns, circadian rhythms, physical activity, sedentary lifestyle, environmental enrichment and diet.20, 21, 22, 23

Recently, exercise has emerged as a powerful modulator of both the composition and metabolic activity of the gut microbiota,²⁴ and the effects of chronic physical exercise have been extensively reviewed. Some of these studies have shown that regular exercise is associated with a more diverse and stable gut microbiota, which result in better gut health.^25,26 Specific exercises, such as running or cycling, have been shown to increase the abundance of beneficial bacterial species in humans.²⁷ Exercise also exerts anti-inflammatory effects on the gut by reducing levels of pro-inflammatory cytokines²⁸ and appears to improve gut permeability, further supporting gut health.²⁹ Additionally, recent studies highlight a dynamic interaction between the gut microbiota, neurodegeneration, and the role of physical activity, suggesting that exercise may play a protective role in brain health through its effects on the gut.²⁸ However, excessive exercise (in terms of intensity and/or duration) can have detrimental effects, not only negatively impacting the gastrointestinal system but also altering the composition and function of the gut microbiota.^30,31

In the present work, we aimed to investigate how varying exercise intensity protocols influence AHN, cognition, and intestinal microbial composition. Our hypothesis was that different exercise regimens would lead to distinct changes in these parameters and that the gut microbiota might play a mediating role in the observed effects. We demonstrate that varying exercise intensity protocols lead to distinct changes in these parameters, showing a hormetic profile in AHN, cognition, and microbial composition. Furthermore, we showed that these changes in the microbiota are sufficient to evoke similar changes in neurogenesis and cognition in sedentary animals.

Methods

Animals

A total of 100 male C57BL/6J mice (Mus musculus L.) from Envigo Laboratories, aged 11 weeks at the start of the experiments, were used. As the animals were randomly transported in cages from Envigo and all animals are equal, no more randomisation was required for the experiment. They were all kept under hygienic conditions with a 12-h light/dark cycle, stable temperature (20-22 °C), and food and water ad libitum in accordance with the current European regulations (2010/63/EU). All experiments were performed according to the European Community Guidelines (Directive 2010/63/EU) and Spanish Guidelines (Royal Decree 53/2013) and related rules, and they were first validated by the Committee of Ethics and Animal Experimentation of the Cajal Institute (20/05/2016), subsequently favourably evaluated by the CSIC Ethics Committee (Subcommittee of Ethics) of the Spanish National Research Council (07/27/2016) and authorised by the competent authority, the Animal Protection Area of the Department of Environment of the Community of Madrid (10/26/2016 and 06/19/2020). Sample size for experiments were determined based on prior publication and power analysis using the G∗Power software.

Role of funders

Spanish Ministry of Economy and Competitiveness, Spanish Ministry of Science and Innovation, and Spanish Ministry of Universities provided predoctoral fellowships, and Spanish Ministry of Science and Innovation provided grants to support the project. No funders had any role in study design, data collection, data analyses, interpretation, or writing of report.

Study design

All experiments involve a 32-days exercise protocol (6 weeks +2 days) or home cage control. On the final day of the protocol, some behavioural test (Activity cage, EpM, NOL, or NOR) was conducted, and on day + 44, animals were sacrificed. All animals were randomly assigned to one of the following groups: SED: sedentary, RUN: moderate runner, RUNtime: intense time runners, RUNvel: intense velocity runners. All animals are 11 weeks old at the start of the experiment and are euthanised at 17 weeks of age. 10 animals were allocated to each group, except for the microbiota experiments with 5 animals to each group in composition analysis and 6 in FMT experiment.

Treadmill running protocols

All animals underwent daily 30-min acclimation before treadmill sessions. On Day 1, groups completed speed familiarisation. From Day 2: SED remained sedentary; RUN performed 40-min sessions at 1200 cm/min; RUNtime extended to 60-min; RUNvel reached 1800 cm/min for high-intensity training, all ending with 30-sec cooldown (600 cm/min). For detailed protocols, see Supplementary Materials.

Behavioural assessment

All tests were conducted between 9 am and 3 pm during the light phase, either recorded on video or automatically analysed, with the experimenter blind to groups when analysing videos.

Activity cage

This test was conducted using a Cibertec Actimeter XYZ-8 (MUX_XYZ16L-8 software) to assess locomotor activity. Animals underwent a two-day protocol, spending 5 min in the activity cage each day. On day 1, spontaneous activity was assessed for 5 min in a novel arena, followed by another 5 min on day 2. Measures included horizontal and vertical activity and total distance moved.

Elevated plus maze (EpM)

This test was conducted to assess anxiety, each animal was placed in a maze with two open arms and two closed arms, all elevated at a height of 0.5 m for 5 min. The animals were allowed to move freely around the maze for 5 min. Variables recorded included time and frequency in each arm.

Novel Object Recognition (NOR)

To assess memory enhancement, difficult protocol was designed by modifying the time spent during the training phase (Supplementary Materials, SF.3a and SF.4a and b).

Novel Object Location test (NOL)

A modified version of the object location test was used to study pattern separation performance enhancement (Supplementary Materials, SF.3b and SF.4c).

For a more detailed protocol and related bibliography, please refer to the Supplementary Materials.

Fecal microbiota transplant (FMT) protocol

Preparation of the inoculum

Feces were collected from each group over three days and placed in a sterile 15 mL Falcon tube. The inoculum was prepared daily under a laminar flow hood, with autoclaved PBS-Glycerol (100 mg stool/1 ml PBS-glycerol). The mixture was homogenised, filtered (70 μm; ThermoFisher; REF: 15370801), and frozen at −80 °C.

Procedure for FMT via enema

Animals were treated with a mixture of antibiotics (Ampicillin 1 g/L; Gentamicin 1 g/L; Vancomycin 0.5 g/L and Imipenem 0.25 g/L) in water for 4 days. On day 5, antibiotics were stopped, and FMT was administered. The FMT was performed via enemas under isoflurane anaesthesia (IsoFlo, Zoetis Inc). Once the animal is in the correct position (lateral decubitus), 200 μL of the fecal bacteria suspension was drawn from the inoculum. After applying paraffin oil to the tube's surface, the fecal bacteria suspension was slowly injected (detailed description of the method can be found in the Supplementary Materials).

Tissue collection

For brain extraction, mice were anaesthetised with sodium pentobarbital, perfused intracardially with saline, and the brain was divided into hemispheres, with the left fixed in 4% PFA and the right dissected for further analysis.

Histology

Serial coronal brain sections with a thickness of 50 μm, including the hippocampal formation, were obtained from one hemisphere. This process was carried out using a Leica VT1000S vibratome. Each brain section was individually collected in a 96-multiwell plate filled with PB (Phosphate-Buffer) 0.1M. These plates were then stored at 4 °C until they were ready for further analysis or examination. Slices were incubated for single or double staining (Supplementary Materials, Tables S1 and S2). The Cavalieri method was used as described in Supplementary Materials. The experimenter was blinded to the groups during image analysis.

Image analysis of fluorescence microscopy

PH3-labelled cells were counted by the optical fractionator method. The physical-dissector method, adapted to confocal microscopy, was used to estimate the total number of SOX2⁺GFAP⁺ cells, DCX⁺, and DCX⁺CLR⁺ cells.

Vessels (podocalyxin) and tight junctions (ZO1) were measure in three hippocampal sections. GFAP and IBA1 expression was analysed in the hippocampal zone (hilus (H), Granule Cell Layer (GCL) and molecular layer (ML)). For more information about the quantification see Supplementary Materials.

Corticosterone competitive ELISA kit

Corticosterone was extracted from dried mouse feces using a methanol-based protocol and quantified using a competitive ELISA kit (Invitrogen, Catalogue #EIACORT). Briefly, fecal samples were homogenised in methanol, centrifuged, and the supernatant was evaporated to dryness and stored at −20 °C until analysis. Corticosterone levels were measured in duplicate, normalised to fecal weight (ng/g), and samples outside the standard curve range were excluded to ensure accuracy. For further details, refer to the Supplementary Material.

RNA isolation and quantitative PCR (qPCR)

Brain tissue from euthanised animals was collected, with half used for immunofluorescence and the hippocampus dissected from the other half for RNA extraction. Total RNA was isolated from the frozen hippocampus, quantified, and stored at −80 °C. cDNA was synthesised via reverse transcription, and gene expression was analysed using TaqMan probes and real-time PCR. Several genes related to tight junctions, apoptosis, inflammation, and oxidative stress were measured, with ACTb as a reference gene. Quantification was performed using the 2-ΔΔCt method.

16S amplicon microbiota composition

DNA extraction and 16S library from the caecum

We extracted the caecum from the animals after behavioural testing (Fig. 1) during perfusion. DNA extraction from the caecum was performed using the QIAamp PowerFecal Pro DNA Kit (QIAGEN) following the manufacturer's instructions; with V3–V4 16S rRNA amplification (30 cycles) and MiSeq sequencing (2 × 300 bp).

Fig. 1.

Moderate exercise improves cognition without affecting motor activity or anxiety-like behaviour. (a–c) Diagram of the Activity cage protocol and variables. (a) Total horizontal activity, one-way Kruskall-wallis (Day 1 (H (3, n = 39) = 8.71, p-value = 0.0334); (Day 2 (H (3, n = 39) = 2.01, p-value = 0.57). (b) Total vertical activity, Kruskall-wallis (Day 1 (H (3, n = 39) = 4.59, p-value = 0.204); (Day 2 (H (3, n = 39) = 1.88 p-value = 0.598). (c) Total distance travelled, one-way ANOVA (Day 1, F (1.769), p-value = 0.171), (Day 2, (F (0.587), p-value = 0.628). (d–f) Diagram of the EPM protocol and variables. (d) Time in arms, one-way ANOVA (Open; (F (0.446), p-value = 0.722), Center; (F (2.075), p-value = 0.122), Closed; (F (2.105), p-value = 0.119). (e) Reaching, one-way ANOVA (F (0.564), p-value = 0.643). (f) Total frequency, one-way ANOVA (F (1.033), p-value = 0.39). (g–h) Diagram of the NOR protocol and variables. (g) Discrimination Index. Mixed ANOVA. Group effect (F (3,35) = 17.181, p < 0.05, η2 = 0.344), phase effect (F (2,70) = 34.358, p < 0.05, η2 = 0.387). One-way ANOVA in each phase. (TNG: F (0.74), p-value = 0.532; STM: F (11.6), p-value = 0.00005, η2 = 0.5; LTM: F (10.9), p-value = 0.00006, η2 = 0.48). (h) Total time exploration of both objects. Group differences in each phase, one-way ANOVA and Kruskall-walis. (TNG: H (3, n = 39) = 11.5, p-value = 0.009, STM: F (1.11), p-value = 0.716; LTM: F (0.956), p-value = 0.716). (i–j) Diagram of the NOL protocol and variables. (i) Discrimination Index. Mixed ANOVA. Group effect (F (3,33) = 11.990, p < 0.05, η2 = 0.363), phase effect (F (1,33) = 29.995, p < 0.05, η2 = 0.303). One-way ANOVA in each phase. (TNG: F (0.686), p-value = 0.567; STM: F (18.5), p-value = 0.0000006, η2 = 0.628). (j) Total time exploration of both columns. Group differences in each phase, one-way ANOVA. (TNG: F (3.35), p-value = 0.06; STM: F (2.68), p-value = 0.062). (k) Cognitive Index. One-way ANOVA, (F (34.42), p-value = 1.52e-10, η2 = 0.746). Within-group comparisons, related samples t-test. Results represent the mean ± SEM (n = 10, SED group: n = 10, RUN group; n = 10, RUNtime group; n = 9, RUNvel group). Group comparisons p-value ∗<0.05, ∗∗<0.01, ∗∗∗<0.001. Within-group comparisons p-value +<0.05, ++<0.01, +++<0.001. Ns, not significant; TNG, training; STM, short term memory; LTM, long term memory.

Bioinformatic analysis of caecum microbiota in mice

Bioinformatics analysis employed DADA2 (v1.30.0) in R (v4.3.2) for ASV generation and SILVA v138 for taxonomic annotation. PICRUSt2 (v2.4.1) inferred genomic content using KEGG, with OmixerRpm (v0.3.3) computing gut-brain (GBMs) and gut metabolic modules (GMMs). Alpha-diversity indices (Chao1/Simpson/Shannon) were analysed with ANOVA, while beta-diversity used PERMANOVA (Aitchison distance, 10,000 permutations) after CLR transformation. Differential analysis (Tjazi package, FDR q < 0.2) identified 8 altered genera (4 behaviourally correlated, |ρ|>0.51) and significant modules. All visualizations used ggplot2 with Z-score standardisation. For detailed protocols and additional methodological specifications, please refer to the Supplementary Information section.

Fecal short-chain fatty acid analysis (GS-FID)

Acetic, propionic, butyric, isobutyric, valeric, isovaleric, caproic, and heptanoic acids were quantified in fecal extracts by gas chromatography with flame ionisation detection (GC-FID; Agilent 6890A system). Samples were acidified with 0.5% phosphoric acid, spiked with methyl valerate (internal standard), and extracted with n-butanol. Separation was achieved using a DB-WAXtr column (60 m × 0.25 mm) with helium carrier gas (1.5 mL/min). Data were processed using MSD ChemStation software (v.E.02.00.493). Duplicate analyses of randomly selected samples confirmed reproducibility. All procedures were performed at the ICTAN-CSIC Chromatography and Mass Spectrometry Facility.

Statistical analysis

Data were analysed using SPSS (v26.0) and R (v4.3.3). The experimental unit is the single animal. Group comparisons employed ANOVA (normal distribution) or Kruskal–Wallis tests (non-normal), with Tukey's HSD post-hoc corrections. Mixed ANOVA or Friedman tests (with Wilcoxon post-hoc) analysed behavioural data with repeated measures. Correlations used Pearson/Spearman tests. Data presented as mean ± SEM (Shapiro–Wilk normality test; outliers excluded). Significance thresholds: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; trends: ∗'p < 0.099. Graphs generated in GraphPad Prism 8/RStudio. Metadata will be available in NIH SRA (BioProject PRJNA1222849). Detailed methods in Supplementary Materials. All the data points excluded during the analysis were excluded with a priori criteria.

Results

Study design and behaviour before the physical exercise protocol

To prove whether different types of physical exercise have the same pro-cognitive effect, we implemented a longitudinal and between-groups design (SF.1a), assessing animal behaviour both before and after the exercise interventions. The main difference between exercise protocols lies in the daily running time and speed achieved, which influences the total distance covered both daily (SF.1d) and over the entire treatment period (SF.1e). The RUNtime group covered the greatest distance daily and overall, followed by RUNvel, RUN (moderate protocol), and finally SED (sedentary controls), which remained in their home cages for the duration of the protocol.

To evaluate the effect of different intensities of physical exercise on behaviour, we conducted a battery of behavioural tests before and after the treatment. As expected, baseline conditions were similar across groups, as indicated by results from the activity cage test (SF. 2a–c) and the elevated plus maze (SF. 2d–f). Memory performance was evaluated using the Novel Object Recognition (NOR) and Novel Object Location (NOL) tests, both of which are hippocampus-dependent. Pre-treatment analysis confirmed that none of the groups could successfully pass these tests, as the protocols were intentionally designed to ensure control mice could not pass (SF.2g–k). For further details, please refer to the Supplementary Materials section.

Moderate exercise improves cognition without affecting motor activity or anxiety-like behaviour

Motor activity was evaluated over two days: on the first day in a novel environment and on the second day in the same, now-familiar environment. As shown in Fig. 1, although a difference in horizontal movement was observed between the SED and RUNvel groups on the first day, this difference was not maintained on the second day (Fig. 1a). Furthermore, no significant differences were observed in any of the other measured variables—vertical movement (Fig. 1b), or distance travelled (Fig. 1c)—across the groups. Anxiety-like behaviour was assessed using the elevated plus maze test after exercise treatment. Similarly, no statistically significant differences were found between the groups in terms of time spent in either arm, the frequency of arm entries, or interest in exploring the open arms (Fig. 1d–f).

In the NOR test, after 6 weeks of exercise, the RUN group exhibited clear cognitive facilitation (Fig. 1g). This group demonstrated a higher Discrimination Index (DI) compared to the SED, RUNtime, and RUNvel groups in both the short-term memory (STM) phase (F (11.6), p-value <0.001) and the long-term memory (LTM) phase (F (10.9), p-value <0.001). Moreover, the RUN group was the only one to show significant differences between the training (TNG) and STM phases (p-value <0.001) and between the TNG and LTM phases (p-value <0.001), providing evidence of learning. Total exploration time in the NOR test, which is crucial for DI calculation, also showed some variation. Differences were observed in the training phase (TNG) (H (3, n = 39) = 11.5, p-value <0.001), though no differences were found in other phases (Fig. 1h).

In the NOL test, designed as a more challenging spatial memory task due to the small separation between objects, only the moderate exercise group (RUN) successfully discriminated the change in column position (Fig. 1i). The RUN group had higher DI scores in the STM phase (F (18.5), p-value <0.001) compared to the other exercise groups, while no group differences were seen in the TNG phase. Additionally, the DI for the RUN group in the STM phase was significantly higher than in the TNG phase (t (9) = 14.9, p-value <0.001). No significant differences were found between the two phases for the other groups. Regarding exploration time, no differences were observed between groups or phases (Fig. 1j), confirming that the differences in the NOL test were due to enhanced discrimination ability, not exploration time.

To summarise the memory test differences and correlate them with other variables, a cognitive index (CI) was calculated, this was also intended to reduce variability, as animals had to perform well in all three phases to achieve a high CI (Supplementary Materials). As shown in Fig. 1k, the RUN group had the highest CI, significantly differing from the other groups (F (34.42), p-value <0.001). In conclusion, the moderate exercise protocol (RUN) for 6 weeks was the only protocol that induced cognitive facilitation as measured by hippocampus-dependent tasks.

Moderate exercise promotes neurogenesis and increases neural precursors and progenitors

We analysed the volume of the hippocampal granule cell layer (GCL) and the area of the subgranular zone (SGZ), finding no significant differences between groups (Fig. 2b and c). Thus, the exercise protocols did not affect the size of these structures. Given that the positive effects of exercise on learning and memory are mediated, among other factors, by AHN we considered it essential to analyse this process in our study. To that end, we examined different markers of the neurogenic niche following the different exercise protocols.

Fig. 2.

Moderate exercise promotes neurogenesis and increases neural precursors and progenitors. (a–c) No significant differences were found in hippocampal volumetric analysis. (b) SGZ area, Kruskal–Wallis test, H (3, n = 40) = 1.19, p-value = 0.755). (c) GCL volume, one-way ANOVA, F (0.25), p-value = 0.861). (d–e) Neural precursor numbers were significantly higher in the RUN group. (d) Image of double IF of SOX2 (red) and GFAP (green); DAPI (blue). (e) Total number of SOX2⁺/GFAP⁺ cells in the SGZ, one-way ANOVA, (F (11.52), p-value = 2.78e-05, η² = 0.519). (f–g) No significant differences were observed in cell proliferation. (f) Image of PH3+ (red) and DAPI (blue). (g) PH3⁺ cells, one-way ANOVA, (F (2.248), p-value = 0.101). (h–k) Exercise affected specific populations of neural progenitors. (h) Image of double IF of DCX (red) and CLR (green); DAPI (blue). (i) DCX⁺/CLR⁻ cells in the GCL, one-way ANOVA, (F (1.375), p-value = 0.265). (j) DCX⁺/CLR⁺ cells in the GCL, one-way ANOVA with White's adjustment, (F (6.104), p-value = 0.002, η² = 0.376). (k) Proportion of DCX populations. (l–m) Strong correlations were found between cognitive performance and neural precursors/progenitors. (l) DCX⁺/CLR⁺ population correlated with the cognitive index (r = 0.56, p-value< 0.001). (m) SOX2⁺/GFAP⁺ population correlated with the cognitive index (r = 0.68, p-value <0.001). (n) Summary of significant Pearson correlations between neurogenic markers and cognitive index. Red boxes represent significant correlations. (o) Overview of studied cell populations and markers. Statistical outliers were removed from the analysis. Within-group comparisons, paired t-test or Wilcoxon test. White scale bars, 20 μm. The results represent the mean ± SEM (n = 10, SED group: n = 10, RUN group; n = 10, RUNtime group; n = 9, RUNvel group). MOL: molecular layer of the DG; SGZ: subgranular zone of the DG; H: hilus. Between-group comparisons p-value ∗<0.05, ∗∗<0.01, ∗∗∗<0.001. Ns, not significant.

First, we analysed whether potential changes in the neurogenic rate could be associated with variations in the number of neural precursors by performing double immunofluorescence (IF) for SOX2 and GFAP (Fig. 2d). As shown in Fig. 2e, the number of neural precursors differed significantly among the groups (F (11.52), p-value < 0.001), with the RUN group having a significantly higher number of neural precursors compared to the other groups. However, no differences were found in cell proliferation (PH3⁺, Fig. 2f–g). Next, we performed double IF labelling for DCX and CLR to investigate different populations of neural progenitors (Fig. 2o). This double labelling revealed an increase in differentiating immature neurons (DCX⁺/CLR⁺ population (Fig. 2j) in the RUN group (F (6.104), p-value <0.01), while the increase in proliferative progenitors (DCX⁺/CLR⁻ population (Fig. 2i)) was not significant.

Correlations between cognitive index (CI, see previous section; Fig. 2n) and neurogenic markers showed significant associations with both precursors (Fig. 2m) and progenitors (Fig. 2l). In summary, six weeks of moderate exercise (RUN group) increased the number of neural precursors and progenitors in the hippocampus, which correlated with cognitive improvement, whereas other types of exercise did not.

Study of tight junctions and glial cells in the hippocampus of exercised mice

Maintaining the integrity of the brain's vascular and glial barriers is essential for CNS homoeostasis and can be modulated by physical activity. To investigate this, we performed double IF labelling of brain vessels with anti-podocalyxin and tight junctions with anti-ZO-1 (Fig. 3a). The total vessel area did not differ between groups (Fig. 3b), nor did the total ZO-1-positive area (Fig. 3c). However, when focussing on ZO-1 signal specifically within brain vessels, we detected a significant effect of group (one-way ANOVA: F = 5.089, p < 0.01). The RUN group exhibited a trend toward higher vascular ZO-1 expression compared to both SED and RUNtime groups, and a significant increase compared to the RUNvel group. To further assess tight junction integrity, we analysed gene expression by qPCR in the hippocampus (SF.5g) and ileum (SF.5h) after six weeks of treatment. In the hippocampus, ZO-1 mRNA levels appeared elevated in the RUN group, although the difference did not reach statistical significance. In contrast, Claudin-5 expression showed a significant main effect of group (one-way ANOVA: F = 9.391, p < 0.05), with the highest levels observed in sedentary animals. Both RUN and RUNtime groups showed significantly reduced Claudin-5 mRNA levels compared to the SED group (SED vs. RUN: p = 0.01; SED vs. RUNtime: p = 0.0031). No significant differences were detected in the expression of tight junction-related genes in the ileum.

Fig. 3.

Study of Tight Junctions and Glial cells in the Hippocampus. (a–d) ZO1 measurement by immunofluorescence. (a) Image of Pdcx (red) and ZO1 (green). (b) Area occupied by vessels. One-way ANOVA: F (1.404), p-value = 0.271. (c) Area occupied by ZO1. One-way ANOVA: F (1.774), p-value = 0.184. (d) Area occupied by ZO1 within the vessels. One-way ANOVA: F (5.089), p-value = 0.009, η2 = 0.432. (e–g) GFAP measurement. (e) Image of GFAP (green). (f) Area occupied by astrocytes (GFAP). GCL; One-way ANOVA: F (0.667), p-value = 0.583. Hilus; One-way ANOVA: F (0.67), p-value = 0.581. ML; One-way ANOVA: F (2.428), p-value = 0.099. (g) Fractal dimension (GFAP). GCL; One-way ANOVA: F (0.667), p-value = 0.583. Hilus; One-way ANOVA: F (0.67), p-value = 0.581. ML; One-way ANOVA: F (0.896), p-value = 0.462. (h–k) Iba1 measurement. (h) Image of Iba1 (red). (i) Area occupied by microglia (Iba1). GCL; One-way ANOVA: F (9.488), p-value = 0.000482, η2 = 0.599. Hilus; One-way ANOVA: F (5.388), p-value = 0.0074, η2 = 0.459. ML; One-way ANOVA with White's correction: F (25.47), p-value = 7.19e-7. (j) Fractal dimension (Iba1). GCL; One-way ANOVA: F (8.376), p-value = 0.000942, η2 = 0.569. Hilus; One-way ANOVA: F (3.788), p-value = 0.027, η2 = 0.374. ML; One-way ANOVA with White's correction: F (6.129), p-value = 0.0099. (k) Microglial cell profiles. Kruskal–Wallis, H (3,24) = 6.08, p = 0.108. White scale bars, 20 μm. Results are presented as mean ± SEM (n = 6 per group). Group comparisons p-value ∗<0.05, ∗∗<0.01, ∗∗∗<0.001. Ns, not significant.

To complement the analysis of barrier integrity, we evaluated oxidative stress and apoptosis by quantifying the expression of antioxidant and apoptotic genes (SOD1, SOD2, BAX, and Bcl2) via qPCR in both hippocampus (SF.5c, d) and ileum (SF.5e, f). No significant group differences were found in the hippocampus for SOD1 or SOD2 (SF.5c). However, in the ileum, SOD1 expression was significantly elevated in the RUNtime group (Kruskal–Wallis: H (2,18) = 7.74, p < 0.05), and SOD2 expression were significantly increased in both RUN and RUNtime groups compared to SED (H (2,18) = 9.18, p < 0.01 (SF.5e), suggesting an exercise-induced increase in antioxidant demand in the gut. To determine whether this oxidative response was accompanied by changes in apoptosis, we analysed BAX and Bcl2 expression. No significant group differences were found in either region (SF.5d, 5f), suggesting that the antioxidant response may be sufficient to maintain oxidative balance and prevent apoptotic signalling under the experimental conditions tested.

We next investigated glial cell responses in the hippocampus. Astrocytes labelled with GFAP showed no group differences in area coverage or fractal dimension, an index of morphological complexity (Fig. 3e–g). In contrast, Iba1-positive microglia exhibited significantly increased signal intensity and fractal complexity in the RUNvel and RUNtime groups, without changes in total cell number, indicating mild activation of microglia in response to high-intensity exercise. Lastly, we analysed inflammatory cytokine expression (IL-1β, IL-6, TNF-α, and TGF-β) by qPCR in both hippocampus and ileum. No significant differences were found for IL-1β, IL-6, or TNF-α in either region (SF.5a, b). However, the ileum showed a trend toward reduced levels of these pro-inflammatory cytokines in the RUN group. In the hippocampus, IL-1β appeared to decrease with exercise, while IL-6 showed a non-significant trend toward elevation in both exercised groups. Notably, TGF-β expression in the hippocampus was significantly modulated by exercise intensity (ANOVA: F = 14.342, p < 0.001), with lower levels in the RUN group compared to RUNtime (p = 0.003), and a trend toward reduction vs. SED (p = 0.054).

Moderate exercise is associated with increased microbiota diversity

This study shows that exercise alters gut microbial composition in ways that may influence behaviour and mental health. Fig. 4a–c illustrate alpha diversity indexes, including the Chao1 index (species richness), the Shannon index (richness and evenness), and the Simpson index (species dominance). Results indicate that the RUN group had a higher abundance of genera (F (5.283), p < 0.01, Fig. 4a), with the Chao1 index significantly higher compared to the RUNvel group (p = 0.01) and showing trends toward significance with the sedentary (p = 0.059) and RUNtime (p = 0.054) groups. Similar patterns were observed in the Shannon index (F (3.541), p < 0.05; Fig. 4b), particularly between the RUN and RUNvel groups (p = 0.03). No significant differences were found in the Simpson index (Fig. 4c). Beta diversity, analysed by Euclidean distance and presented via Principal Component Analysis (PCA), also showed significant differences (PERMANOVA: F (2.8), p = 0.001; Fig. 4d). Taxonomic composition is shown as stacked bar plots for the top 20 families and the 33 most abundant genera (Fig. 4e and f), including relevant taxa such as Bacteroidaceae, Prevotellaceae, Lachnospiraceae, and Rikenellaceae.

Fig. 4.

Moderate exercise is associated with increased microbiota diversity. (a–c) Alpha Diversity. (a) Chao1. One-way ANOVA: F (5.283), p-value = 0.011, η2 = 0.514. (b) Simpson Index. One-way ANOVA: F (3.16), p-value = 0.056, η2 = 0.387. (c) Shannon Index. One-way ANOVA: F (3.541), p-value = 0.041, η2 = 0.415. (d) Beta Diversity. PERMANOVA: F (2.8), p-value = 0.001. (e–f) Taxonomy proportion by groups. Rare taxa include all families or genera representing <1% of the total for each sample or animal. (e) Proportion of Families. (f) Proportion of Genera. (g) UPSET plot of the genera present in each group out of the 96 identified genera. Intersection size represents the number of common genera among groups. (h–o) Differential expression of the 15 bacterial genera with fdr ≤ 0.1. (h) Eubacterium ruminantium, One-way ANOVA: F (4.66), p-value = 0.017. (i) Clostridium sensu stricto1. One-way ANOVA: F (13.08), p-value = 0.00018. (j) Candidatus Saccharimonas. One-way ANOVA: F (4.67), p-value = 0.016. (k) Alloprevotella. One-way ANOVA: F (5.425), p-value = 0.009. (l) Helicobacter. One-way ANOVA: F (231.28), p-value = 8.87e-13. (m) Rikenellaceae RC9. One-way ANOVA: F (10.78), p-value = 0.0004. (n) Peptococcus. One-way ANOVA: F (6.317), p-value = 0.005. (o) Enterorhabdus. One-way ANOVA: F (27.893), p-value = 2.36e-6. (p–r) Pearson correlation of differential expression with the Cognitive Index. (p) Angelakisella. One-way ANOVA: F (8.88), p-value = 0.001. Correlation with CI, rho = −0.51, p = 0.026. (q) Erysipelatoclostridium. One-way ANOVA: F (5.046), p-value = 0.01. Correlation with CI, rho = −0.55, p = 0.015. (r) Acetatifactor. One-way ANOVA: F (6.91), p-value = 0.003. Correlation with CI, rho = 0.56, p = 0.12. (s) Coriobacteriaceae UCG-002. One-way ANOVA: F (16.96), p-value = 4.37e-5. Correlation with CI, rho = −0.67, p = 0.0018. n = 5, SED group; n = 5, RUN group; n = 5, RUNtime group; n = 4, RUNvel group. Between-group comparisons p-value ∗<0.05, ∗∗<0.01, ∗∗∗<0.001. Ns, not significant.

Across all groups, 96 genera were identified, 63 of which were shared by all groups, as shown in Fig. 4g, where the intersection indicates the number of shared genera. The RUN group has the highest number of unique genera (represented by the blue bar), followed by the SED group (grey bar), RUNtime group (light green bar), and lastly, the RUNvel group (dark green bar). The specific unique genera for each group can be observed in the corresponding coloured boxes in the figure. Notably, the RUN group had eight unique genera, with four passing the filtering criteria: Lachnospiraceae UCG-008, Lachnospiraceae UCG-010, Defluvitaleaceae UCG-011, and Staphylococcus.

Differential expression analysis identified 15 genera with a false discovery rate (FDR) of less than 0.1, represented by the Centred Log-Ratio (CLR) (Fig. 4h–s). These genera include [Eubacterium] ruminantium group, Clostridium sensu stricto 1, Candidatus Saccharimonas, Alloprevotella, Helicobacter, Rikenellaceae RC9 gut group, Peptococcus, Enterorhabdus, Angelakisella, Erysipelatoclostridium, Acetatifactor, Parasutterella (not shown), Faecalibaculum (not shown), Parabacteroides (not shown), and Coriobacteriaceae UCG-002. We then examined correlations between these genera and cognitive performance. Four genera, Angelakisella, Acetatifactor, Erysipelatoclostridium, and Coriobacteriaceae UCG-002, were significantly associated with cognitive outcomes. In all cases, the correlations were moderate or strong (rho values of −0.51, −0.55, 0.56, and −0.63, respectively), and all were statistically significant (p < 0.05). Specifically, Angelakisella, Erysipelatoclostridium, and Coriobacteriaceae showed negative correlations, suggesting a link with poorer cognitive outcomes, while Acetatifactor was positively correlated. Although no significant group differences in Acetatifactor abundance were observed, a trend was found between RUN and RUNtime (p = 0.07), but not between RUN and RUNvel (p = 0.17).

Given the association of several genera with cognitive performance, many of which are established SCFAs producers, we quantified SCFAs levels in fecal samples. However, no significant differences between groups were detected (SF.6). To further explore microbial functionality, we performed a bioinformatic analysis of gut metabolic modules. Using 16S amplicon data, we also inferred Gut-Brain Modules (GBMs) and Gut Metabolic Modules (GMMs) to predict microbial metabolite activity. For the GBMs, the only significantly altered pathway was propionate degradation (SF.7b), which showed statistically significant differences when comparing the SED vs. RunTime groups (p < 0.005, adjusted p = 0.01925167) and SED vs. RunVel (p < 0.005, adjusted p = 0.0176307). The PERMANOVA analysis for the GBMs was also significant, suggesting changes in the composition of GBMs (i.e., neuroactive potential) among the different groups (R² = 0.29, p < 0.05). Similarly, in the GMMs, the only altered pathway was also propionate degradation (SF.7d), as identified by both analytical tools. The PERMANOVA for the GMMs was also significant, indicating global changes in the microbiota's potential to metabolise dietary products (R² = 0.314, p < 0.05).

In summary, moderate exercise (RUN group) was associated with increased microbial diversity and a distinct microbiota profile linked to improved cognitive function. Microbiota composition varied depending not just on exercise, but on exercise intensity. A pro-cognitive microbial signature was observed in the RUN group, involving genera from Lachnospiraceae and Defluvitaleaceae. Fifteen genera were differentially expressed across groups, with four significantly correlated with cognitive performance.

Fecal microbiota transplant from moderate runners induces cognitive improvement and promotes neurogenesis

To explore the role of microbiota in the cognitive enhancements observed after moderate exercise, a fecal microbiota transplant (FMT) was performed. The study involved three groups: sedentary mice (SED) that received FMT from sedentary donors, and sedentary mice that received FMT from mice that had undergone moderate exercise for 6 weeks (RUN) or extended exercise (RUNtime). After four days of antibiotics, FMT was administered in week 2, followed by behavioural testing during weeks 3 and 4 (Fig. 5a).

Fig. 5.

Fecal microbiota transplant from moderate runners induces cognitive improvement and promotes neurogenesis. (a) Experimental design. (b) Novel Object Recognition. Discrimination index. Linear model, F (8,45) = 14.1, p-value = 5.190e-10. One-way ANOVA in each phase. (TNG: F (2.43), p-value = 0.122; STM: F (21.8), p-value = 3.65e-4, η2 = 0.744; LTM: F (11.5), p-value = 9.4e-4, η2 = 0.605. (c) Novel Object Location. Discrimination index. Linear model, F (5,45) = 3.786, p-value = 9.59e-3. One-way ANOVA in each phase. (TNG: F (1.75), p-value = 0.21; STM: F (5.88), p-value = 0.014, η2 = 0.456). (d) Cognitive Index. One-way ANOVA, (F (31.852), p-value = 3.99e-6, η2 = 0.809). Within-group comparisons, paired t-test or Wilcoxon test. (e) Image of SOX2 (red) and GFAP (green); DAPI (blue). (f) SOX2+/GFAP + cells in the SGZ, Kruskal–Wallis, (H (2,18) = 9.27, p-value = 0.009). (g) Image of DCX (red) and CLR (green). (h) DCX⁺/CLR⁻ cells in the GCL, one-way ANOVA, (F (2.133), p-value = 0.153). (i) DCX⁺/CLR⁺ cells in the GCL, one-way ANOVA, (F (5.366), p-value = 0.01, η2 = 0.433). (j) DCX⁺/CLR⁺ population correlate with the CI (r = 0.6, p-value = 0.007). (k) Pearson correlations of NHA parameters with the CI. Red box represents significant correlation. White scale bars, 20 μm. Results represent mean ± SEM (n = 6 per group). Between-group comparisons p-value ∗<0.05, ∗∗<0.01, ∗∗∗<0.001. Ns, not significant. Within-group comparisons p-value +<0.05, ++<0.01, +++<0.001.

There were no differences in motor activity or anxiety-related behaviours across the groups (not shown). However, mice that received FMT from the RUN group showed memory facilitation, both short-term (STM: F (21.8), p < 0.001) and long-term (LTM: F (11.5), p < 0.001) in NOR test compared to animals receiving FMT from SED or RUNtime mice (Fig. 5b). No differences were found in exploration time. In NOL test, only the FMT RUN group showed increased DI in STM (F (5.88), p = 0.01) and in the cognitive index (CI: F (31.852), p < 0.001; Fig. 5c and d). These findings demonstrate that the features of the microbiota of RUN mice confer cognitive facilitation to otherwise sedentary mice lacking that advantage. Next, we examined the AHN variables of those mice submitted to FMT Recipients of FMT from RUN mice exhibited increased SOX2⁺/GFAP⁺ (H (2,18) = 9.27, p-value <0.01; Fig. 5e and f) and DCX⁺/CLR⁺ (F (5.366), p-value <0.01; Fig. 5g–i) cells compared to those receiving FMT from SED or RUNtime donors. No differences were observed in DCX⁺/CLR⁻ cells (Fig. 5h). Moreover, a positive correlation was found between the CI and the number of DCX⁺/CLR⁺ cells (Fig. 5j and k).

In conclusion, the FMT experiments demonstrated that the pro-neurogenic effects of moderate exercise could be transferred to sedentary mice through microbiota, replicating the cognitive and neurogenic benefits observed in their donor mice.

Discussion

In the present work, we investigated how different exercise intensities and durations affect brain function and cognition, with a particular focus on the mediating role of the gut microbiota. Our findings support the hormetic effects of exercise in male mice, demonstrating that moderate exercise (40 min daily at 1200 cm/min) significantly enhances cognitive function and AHN compared to sedentary controls, and these benefits are not present with longer or more intense exercise. This biphasic response underscores the importance of exercise dose in optimising brain health. Additionally, we observed that both alpha and beta diversity of the gut microbiota varied according to exercise regimen, with specific compositional profiles closely related to the volume of physical activity performed, ranging from sedentary to high-intensity or long-duration exercise. Notably, the relative abundance of certain bacterial taxa correlated with cognitive performance across groups. Through fecal microbiota transplant (FMT) experiments, we further established a possible causal relationship, showing that the cognitive and neurogenesis benefits observed in runner donors could be transferred to sedentary animals. These findings highlight the role of the gut microbiota in the cognitive effects of moderate exercise, suggesting it as a key mediator in the hormetic response.

Our study focused exclusively on male mice because our primary research question was to determine how different exercise intensities modulate cognitive responses and to elucidate the mechanisms underlying these effects. It is well established that both exercise physiology³² and gut microbiome responses display significant sexual dimorphism,^33,34 which would introduce considerable complexity to the experimental design. Addressing these sex differences as a potential caveat would likely require adjustments to exercise protocols and analytical approaches, making it difficult to directly compare mechanistic outcomes across sexes within a single study framework.

The exercise protocols used in our experiments were carefully selected and optimised based on established literature in mice, including our own previous work demonstrating cognitive benefits,^7,35 as well as widely accepted rodent exercise parameters near or above the supra-lactate threshold⁵ (typically 60 min/day at 1800–2200 cm/min). We systematically varied both duration and intensity to characterise potential hormetic responses. Importantly, while our moderate protocol (40 min/day at 1200 cm/min) seems comparable with light jogging in humans and our intense protocols resemble high-intensity interval training (HIIT), these comparisons are only illustrative. Our protocols were specifically designed for mice and are not intended to be directly transferable to humans, due to fundamental physiological and metabolic differences between species, which might be considered a limitation of our study. Moreover, the lack of standardised definitions for exercise intensities across studies further complicates any direct interspecies comparisons.^36,37 Despite these limitations, cognitive outcomes in our rodent models provide a robust framework for investigating the mechanisms underlying hormetic responses to exercise, and we think our results illustrate an exercise response curve that may be similar in humans even when the specific exercise intensities and durations are quite different.

Regarding the mechanisms mediating the effects of physical exercise on the brain, substantial evidence has accumulated over the past two decades. These studies have identified key mediating factors both outside and inside the brain, ^38,39 including enhanced vascular niche blood circulation,^40,41 AHN and recent genetic analyses of exercise's influence on brain function and memory.⁴² While some of these factors have been examined in the context of their hormetic profiles,^4,5 the underlying mechanisms remain poorly understood. Our findings demonstrate a clear hormetic profile where moderate exercise optimally enhanced recognition memory and AHN, consistent with prior work from our group^7,35 and others.^39,43 Behavioural tests were designed with difficulty thresholds that sedentary animals could not achieve, revealing that only moderate exercise improved cognitive performance, a novel finding in adult laboratory mice. Although we did not conduct a full dose–response curve, our protocols represent distinct points along the hormetic curve: moderate exercise near the peak response, longer duration and higher speed protocols near the non-observable effect level (NOEL), and the sedentary group represents the zero end-point (ZEP) level.⁴⁴ Importantly, our protocols avoided confounding stressors (e.g., electric shocks), as confirmed by absence of anxiety-like behaviours, and consistent stable corticosterone levels (SF.1h), adhering to established exercise study design principles.

This hormetic curve was also evident when analysing potential mediating mechanisms, both within the brain (AHN) and outside (microbiota). Specifically, moderate exercise significantly increased neural precursors and immature neurons in the hippocampal granule cell layer, effects not seen with more intense or prolonged exercise. These results align with well-established role of AHN in cognition, learning, and memory^45,46 and support previous evidence linking cognitive benefits of exercise to increased AHN, particularly the number of immature neurons.⁴⁷ Recent studies have also highlighted the microbiome's role in regulating AHN in rodents.^48,49 To explore whether the microbiota mediates exercise-induced changes in AHN and cognition, we analysed neurogenesis and performed hippocampal-dependent tasks in sedentary recipients of fecal microbiota transplants (FMT) from exercised donors. Notably, we found that both cognitive improvements and increased AHN were transferrable via FMT, implicating a microbiota-AHN pathway in the cognitive benefits of exercise. Our findings demonstrate that while forced intense exercise fails to promote either AHN or cognitive enhancement, moderate exercise effectively stimulates both outcomes. This is consistent with previous research associating these benefits with specific physiological markers, such as lactate and corticosterone.^5,8 Interestingly, comparative studies of exercise modalities have shown that while intense aerobic exercise can enhance neurogenesis, it may not produce corresponding cognitive improvements.⁵⁰ Additionally, microbiota changes induced by moderate exercise may act through the vagus nerve,⁵¹ or through microbial metabolites,⁵² though these mechanisms requires further investigation.

Regarding microbiota composition, our analysis aligns with previous research on aerobic exercise, particularly in comparison with sedentary controls. While the current literature on exercise–microbiota interactions is complicated by methodological variations, including differences in animal models, diets, and experimental conditions, our findings support the established association between moderate exercise and enhanced microbial diversity.^24,28 Previous work has identified distinct microbial profiles associated with different exercise modalities (e.g., voluntary vs. forced exercise⁵³ or strength vs. endurance training.²³ Importantly, we provide new evidence that varying intensity within the same exercise paradigm produces distinct microbial responses that correlate with cognitive outcomes. Moderate-intensity exercise was associated with significant modulation of 15 bacterial genera, with four of them (Angelakisella, Erysipelatoclostridium, Coriobacteriaceae UCG-002, and Acetatifactor) showing strong correlation with cognitive performance. We also identified moderate-exercise-specific changes in families previously linked to cognitive function, such as Lachnospiraceae and Oscillospiraceae^54,55 In contrast, reductions in Prevotellaceae UCG-001 and Lachnospiraceae UCG-066 negatively correlated with poorer cognitive performance.⁵⁶

The observed microbial changes align with several established patterns in the exercise-microbiome literature. For example, we confirmed a consistent increase in Eubacterium ruminantium across different exercise protocols,⁵⁷ as well as a reduction in Alloprevotella, mirroring previously reported decreases in the broader Prevotella taxon.⁵⁸ Additionally, the duration-dependent increase we observed in Clostridium sensu stricto (strain 1) has also been documented in both chronic fatigue models²⁴ and in healthy subjects, in both murine models⁵³ and humans.⁵⁹ In our data, the moderate-exercise group showed several unique microbial signatures, like exclusive presence of Streptococcus, with several studies associating this genus with beneficial effects.⁶⁰ We also observed an increase in several Lachnospiraceae species unique to the moderate exercise group, consistent with other findings linking this increase to improved memory.⁶¹

While our findings generally agree with existing literature, we identified important taxon-specific variations. Most notably, we observed a significant decrease in Rikenellaceae RC9 abundance following high-intensity/long-duration exercise protocols in our mouse model, whereas some human studies have reported different patterns.⁶² Similarly, our findings regarding Prevotella abundance differed from some previous reports,⁶³ potentially reflecting differences in training duration⁶⁴ or other methodological variations. These observations underscore the challenges of cross-study comparisons, particularly given differences in experimental protocols, diet, and analytical approaches between animal and human research.

Our results demonstrate that exercise intensity differentially shapes gut microbiota composition and functionality, with significant implications for cognitive outcomes. Moderate exercise was associated with enhanced cognitive performance and a distinct microbial profile enriched in genera from the Lachnospiraceae and Defluvitaleaceae families, both known producers of neuroactive SCFAs such as butyrate and propionate, which have been implicated in promoting synaptic plasticity and reducing neuroinflammation.^65,66 Although fecal SCFAs levels (acetic, valeric, propionic, butyric, and others; SF.6) showed no significant differences between groups, this may reflect transient changes at earlier time points not captured in our sampling, more pronounced effects in the caecum or systemic circulation compared to feces, or increased local absorption/utilization rather than production changes.⁶⁷

Bioinformatic analyses further supported intensity-dependent microbial adaptations. Strenuous exercise protocols not only failed to improve cognition but also upregulated propionate degradation pathways (SF.7b), suggesting a metabolic shift toward SCFAs catabolism, likely to meet increased energy demands.^68,69 In contrast, moderate exercise was linked to increased pectin degradation activity (SF.7d). Notably, we observed divergent patterns in pectin metabolism across exercise protocols. The RUNtime group showed higher pathway abundance than sedentary controls, suggesting compensatory fibre fermentation during prolonged low-intensity activity, while the RUNvel group exhibited reduced pectin degradation, indicating potential exercise-induced dysbiosis.^70,71 Together, these findings suggest that moderate exercise fosters a neuroprotective gut environment, while excessive or suboptimal exercise may disrupt this balance. Such disruption could divert microbial metabolism away from the production of beneficial neuroactive compounds, ultimately attenuating the cognitive benefits of physical activity.

The integrity of biological barriers (brain and gut) serves as a sensitive indicator of physiological state, with regulation influenced by oxidative stress, cytokines, and growth factors.^72,73 While reduced expression of tight junction proteins is often interpreted as indicating increased barrier permeability, this relationship requires careful consideration given the complex regulation of barrier function.⁷⁴ Our investigation of exercise-induced changes in both the blood–brain barrier (BBB) and intestinal epithelium revealed distinct tissue-specific responses (SF.5g, h). At the mRNA level, intestinal tight junctions remained relatively stable across all exercise protocols, while hippocampal tight junctions showed intensity-dependent modulation. Specifically, we observed fluctuations in hippocampal Claudin-5 and selective upregulation of Zo1 in moderate exercised animals, patterns that may represent adaptive mechanisms facilitating dynamic junction formation. These findings align with previous reports of exercise-mediated restoration of BBB components (Zo1, occludin, claudin-5) in disease models,^73,75 though our study provides novel characterisation of these effects across exercise intensities in healthy animals.

Importantly, our intense exercise protocols elicited distinct physiological responses without inducing generalised stress pathology. We observed microglial activation (increased Iba1⁺ area and structural complexity) specifically in high-intensity groups, which was notably unaccompanied by elevation of pro-inflammatory cytokines (IL-1β, IL-6, TNF-α; SF.5a) or oxidative stress markers (hippocampal SOD1/2; SF.5c). The only inflammatory mediator showing alteration was TGF-β, which showed a hormetic response pattern (RUNtime > SED > RUN) consistent with its known concentration-dependent roles in cellular regulation⁷⁶ and prior associations with exercise-induced fatigue.^77,78 Supporting this interpretation, we found no changes in fecal corticosterone levels after 6 weeks (SF.1h), and intestinal oxidative stress in RUNtime animals lacked apoptotic correlates (BAX/Bcl-2; SF.5f), suggesting effective compensatory adaptation.

The observed microglial priming may reflect non-classical activation, as quiescent microglia typically support neuronal health through trophic factor secretion while activated populations respond to immune challenges. This functional dichotomy helps explain apparent discrepancies in the literature: while exercise generally suppresses microglial activation in neuroinflammatory contexts⁷⁹ and promotes AHN when inflammation is blocked,⁸⁰ some protocols increase cortical microglial proliferation,⁸¹ others show no effect after a moderate protocol,^82,83 and some reports indicate an inverse correlation between AHN and microglial density.⁸⁴ Our findings suggest that microglial responses follow an intensity-dependent pattern, potentially mediated through microbiota-derived signals. This mechanism may help explain the biphasic cognitive effects observed with different exercise intensities. Specifically, we propose that the lack of AHN enhancement following intense exercise could be attributed to microglial activation. This interpretation aligns with established evidence demonstrating the crucial role of gut microbiota in microglial maturation and function⁸⁵ and the altered microglial phenotype observed in germ-free mice compared to those with normal gut microbiota.⁸⁶ Collectively, these observations lead us to hypothesise that the hippocampal microglial activation seen in our high-intensity exercise group likely involves exercise-induced modifications of the gut microbiota.

We specifically focused on aerobic exercise due to its well-documented effects on cognitive function and neurogenesis in both animal models and humans, as demonstrated by extensive studies on hippocampal plasticity and learning.^87,88 Aerobic exercise provides a robust framework for systematically investigating dose–response relationships, allowing isolation of the effects of exercise intensity and duration within a single modality.⁸⁹ This approach is critical given the multifactorial nature of exercise-induced brain plasticity. Our results highlight the challenge of attributing cognitive and neurogenic benefits to a single mechanism, instead supporting the view that aerobic exercise acts through interconnected pathways: (1) modulation of neurotrophic factors (BDNF, IGF-1, VEGF) essential for synaptic plasticity and neuronal survival^90,91; (2) enhancement of synaptic function, potentially mediated by increased PSD-95 expression and stabilisation of synaptic receptors⁵⁰; (3) improvements in mitochondrial function and glucose metabolism, facilitated by increased expression of GLUT transporters⁹²; and (4) cytokine-mediated gut-brain axis communication,⁹³ including exercise-induced myokines like irisin and microbial metabolites such as indole that may regulate neurogenesis through AhR signalling.^94,95 Taken together, our study highlights the complexity of exercise-induced brain adaptations and the importance of considering multiple, interacting mechanisms. By employing a controlled aerobic exercise paradigm, we were able to systematically dissect these dose-dependent effects, providing new insights into how exercise intensity and duration shape cognitive and neurogenic outcomes through diverse pathways.

In conclusion, our results demonstrate that the beneficial effects of moderate exercise on specific learning and memory tasks in mice vanish with small increases in daily training time or increases in running intensity, resembling a hormetic curve profile for these cognitive effects of exercise. This profile is parallelled by the AHN and microbiota composition changes. Moreover, an exercise intensity-specific microbiota signature can be outlined, suggesting a potential mediator of the hormetic effects of physical exercise on brain and cognition. Furthermore, the cognitive benefits induced by fecal microbiota transplant from moderate exercised to sedentary animals are not observed when the transplants come from animals running for longer times, which is consistent with the same profile seen in AHN. Altogether, our results indicate that a microbiota-AHN pathway mediates the hormetic effects of physical exercise on the brain and cognition.

Contributors

Elisa Cintado: study design, data collection, data analysis, data interpretation, validation, writing-original draft, writing-review & editing.

Pablo Muela: data collection, data analysis, writing-original draft.

Lucía Martín: data collection, data analysis.

Ignacio Alcaide: data collection, data analysis.

Patricia Tezanos: data collection, data analysis.

Klara Vlckova: data collection, data analysis.

Benjamín Valderrama: data analysis.

Thomaz F.S. Bastiaanssen: data interpretation, writing-original draft.

María Rodríguez-Muñoz: data collection, data analysis.

María L. de Ceballos: data collection, data analysis, data interpretation, writing-original draft.

María R. Aburto: data interpretation, writing-original draft.

John F. Cryan: data interpretation, writing-original draft.

José Luis Trejo: conceptualisation, study design, funding acquisition, data collection, data analysis, data interpretation, validation, writing-original draft, writing-review & editing.

All authors have read and approved the final version of the manuscript. E.C. and J.L.T have verified the underlying data.

Data sharing statement

All materials, data and associated protocols used in this work will be available to interested readers after publication, upon reasonable request to corresponding authors. Sharing criteria will be based fundamentally on a signed data access agreement.

Declaration of interests

The authors declare that they have no conflict of interest.