Training the nervous system: exercise, autonomic function and mental health

A review article by Trevor Rizun, B.A. Psychology

Abstract. Exercise is now supported by trial-level evidence as an intervention for depression and anxiety, in some comparisons matching psychological and pharmacological treatment [1, 2]. Parallel work in psychophysiology shows that the same interventions move measurable autonomic markers, particularly heart-rate variability (HRV), and that those markers themselves track psychological state [8, 9]. This review draws together five bodies of evidence — exercise as treatment, cardiorespiratory fitness as long-term protection, autonomic regulation and sleep, nature-based activity, and the gut-brain axis — and argues that they describe one system rather than five. It also examines what degrades that system, including anabolic steroids, GLP-1 receptor agonists and alcohol [19, 21, 25], and reviews the instruments used to quantify psychological outcomes [14, 15, 16]. The practical implication is that individualised, physiologically guided training is now testable: the measurements exist, they are inexpensive, and they can be repeated often enough to detect change within a single season.

Introduction

For most of the twentieth century, physical training and mental health treatment were separate disciplines with separate literatures. That separation has closed rapidly. A network meta-analysis of 218 randomised trials covering more than 14,000 participants found that strength training, walking or jogging, and yoga each reduced depressive symptoms about as effectively as cognitive behavioural therapy, with strength training among the most acceptable modalities to participants [1]. An umbrella review two years later reported that exercise matched or exceeded pharmacological and psychological interventions for both depression and anxiety, across every age group examined [2].

Two developments make this more than an interesting finding. The first is consumer-grade measurement: wrist-worn devices now record resting heart rate, HRV, respiratory rate and sleep architecture continuously, at a cost that makes large observational cohorts feasible. The second is the maturation of brief, validated psychological screening instruments, which allow symptom severity to be quantified repeatedly rather than assessed once [14, 15, 16]. Together they permit a question the field could not previously ask at scale: not whether exercise helps, but which physiological state predicts which response to which intervention.

This review synthesises the evidence relevant to that question. It is deliberately broad, because the argument advanced here is that the components interact.

Exercise as an intervention for depression and anxiety

The trial evidence is now substantial enough to support comparative claims. In the 218-trial network meta-analysis, several modalities produced clinically meaningful reductions in depressive symptoms, and the ranking among them was closer than advocacy for any single modality would suggest [1]. The later umbrella review extended the finding to anxiety and across age groups [2].

The same pattern appears in substance use disorder. A network meta-analysis of 57 trials with nearly 3,000 patients found that aerobic and mind-body exercise improved substance-use outcomes, craving, sleep and mental health [6], and an earlier meta-analysis reported higher abstinence rates and reduced withdrawal symptoms [7]. A 2026 trial went further, reporting that eight weeks of walking, jogging and resistance training was associated with recovery of dopamine receptor availability in methamphetamine users [26]. If that result replicates, it describes structural recovery rather than symptomatic relief, which is a different order of claim.

Cardiorespiratory fitness as long-term protection

Where trials establish that training changes symptoms over weeks, epidemiology establishes what fitness predicts over decades. A meta-analysis covering more than four million people reported that high cardiorespiratory fitness was associated with roughly a thirty-six percent lower risk of depression and a thirty-nine percent lower risk of dementia [3]. An overview of 26 meta-analyses, spanning more than twenty million observations, found high fitness strongly predictive of lower risk of heart failure, chronic disease and premature death [4].

Fitness is not only a predictor but also a target. A network meta-analysis in cardiac populations found exercise training improved quality of life, depression and anxiety in people with established heart disease, with larger effects where coaching was delivered in person [5]. That last detail matters for programme design: the supervision context appears to be part of the intervention, not merely its packaging.

Autonomic regulation: HRV as signal and as target

Heart-rate variability — beat-to-beat variation in cardiac interval — indexes the balance between sympathetic and parasympathetic influence on the heart. Higher HRV is associated with better emotional regulation and stress resilience; lower HRV with anxiety and depression [8, 9]. A longitudinal wearable study of more than 180,000 people found that lower resting heart rate, higher HRV and better sleep characteristics consistently tracked with lower depression, anxiety and stress [10]. Related work reports that depression is associated with roughly twenty percent higher risk of developing hypertension [11], and that slower resting respiratory rate is associated with higher psychological well-being and lower distress [12].

The more interesting use of HRV is prescriptive rather than descriptive. In HRV-guided training, the day’s session is determined by the morning’s reading: intensity is added when autonomic recovery is complete and withheld when it is not. Meta-analytic evidence indicates that this approach outperforms fixed periodised plans on VO2 max, maximal aerobic power and ventilatory thresholds, with fewer non-responders, and that it raises vagally mediated HRV indices more reliably [13]. Several of the underlying trials report the counter-intuitive result that guided groups performed fewer hard sessions and still improved more — which suggests the mechanism is avoided accumulated fatigue rather than added stimulus.

A necessary caveat on polyvagal theory

Much popular writing about the vagus nerve derives from polyvagal theory, which describes the autonomic nervous system in three states rather than two. In February 2026 an evaluation signed by Paul Grossman and thirty-eight co-signatories concluded that the theory is untenable, disputing both its evolutionary framing and its treatment of respiratory sinus arrhythmia as a clean index of vagal tone; Porges replied in the same issue that the critique targets a version of the theory he did not propose [18]. The dispute is unresolved.

It is worth stating what does and does not depend on the outcome. The association between HRV and affective state rests on independent observational and trial evidence [8, 9, 10]. Slow breathing, aerobic base training, sleep and time outdoors move those measurements whatever the correct mechanistic account turns out to be. The defensible position is therefore to rely on the measurements and hold the model lightly.

Sleep

Sleep is frequently treated as a downstream consequence of mental health rather than an input to it. Trial evidence argues otherwise: a meta-analysis of randomised trials found that improving sleep quality produced medium-sized improvements in depression, anxiety and stress [17]. Because sleep duration, timing and continuity are among the measurements wearables capture most reliably, this is also the component of the system that is easiest to monitor at scale — and a plausible mediator of several effects attributed directly to exercise.

Nature-based activity

Exercise performed outdoors is not simply exercise with better scenery. A second-order meta-analysis covering more than ten million participants found that nature-based interventions reduced anxiety and depressive symptoms with large effect sizes, and increased positive affect and relaxation [27]. A network meta-analysis ranked green exercise as the most effective intervention for mental health among those compared, significantly above indoor exercise [28], and a meta-analysis focused specifically on trail running reported clinically meaningful reductions in anxiety and depression [30].

Dose and format appear to matter. Nature-based retreats produced larger and more durable reductions in depression and anxiety than single-session interventions [29], which suggests that immersion has properties a weekly appointment does not. Work on forest exposure and health outcomes points to additional pathways, including microbial ones [31].

The gut-brain axis and nutrition

The gut microbiome communicates with the central nervous system through neural, endocrine and immune routes, with the vagus nerve as a principal conduit [32, 34]. This provides a mechanistic bridge between diet and the same autonomic measures discussed above: dietary composition influences microbial populations, which influence vagal signalling, which is what HRV partly reflects. Work on whole foods, ultra-processed foods and appetite regulation supplies the other half of the picture [33].

The candour appropriate here is that this literature is younger and noisier than the exercise literature. Mechanistic plausibility is not efficacy, and the field’s translation to specific dietary prescriptions remains ahead of its evidence.

What degrades autonomic function

A model of adaptation is incomplete without the inputs that impair it, and here the evidence is unusually concrete.

Anabolic steroids. Recreational lifters using anabolic steroids showed lower resting HRV across SDNN, RMSSD, pNN50 and high-frequency power than non-users, with poorer parasympathetic recovery after exercise [19]. A controlled comparison of forty steroid-using strength athletes against forty non-users found reduced baroreflex sensitivity, elevated sympathetic activity and left-ventricular diastolic changes that correlated with years of use [20].

GLP-1 receptor agonists. In a randomised, double-blind, placebo-controlled crossover trial, liraglutide raised mean heart rate by 8.1 beats per minute and reduced SDNN by 33.9 ms [21]. A wearable-based study of sixty-six participants found resting heart rate rose by 3.2 beats per minute over twelve weeks on a GLP-1 drug, statistically mediated by a 6.2 ms fall in HRV [22]. These are not arguments against clinically indicated use; they are a reminder that the same measurement used to guide training will register pharmacological effects, and must be interpreted accordingly.

Unregulated peptides. A year-long open-label extension of the bremelanotide phase 3 trials reported nausea in 40.4 percent of users, flushing in 20.6 percent and headache in 12.0 percent [23]. BPC-157 was placed in Category 2 of the United States Food and Drug Administration’s 503A bulk drug substances list, indicating substances that may present significant safety risks [24], and Health Canada advises that peptides are generally regulated as prescription drugs, with unauthorised products unassessed for safety, efficacy or quality [35].

Alcohol. Among 4,098 employees monitored with heart rate recording, alcohol reduced HRV-derived overnight recovery in a dose-dependent manner: by 9.3, 24.0 and 39.2 percentage units at low, moderate and high intake respectively [25]. The effect concentrates in the first three hours of sleep — the window in which most physiological recovery occurs.

Measuring the psychological outcome

Physiological measurement has outpaced psychological measurement in frequency, not in rigour. Three instruments carry most of the load: the PHQ-9, nine items scoring depression severity over two weeks with a cutoff of ten for moderate-to-severe symptoms [14]; the GAD-7, seven items for generalised anxiety at the same cutoff [15]; and the DASS-21, which adds a stress dimension the other two do not capture [16].

All three share a limitation under repeated administration: respondents encounter identical items, and practice effects accumulate. Computerised adaptive testing addresses this by drawing a different small subset of items from a large bank, targeted to current severity. The CAT-MH has been validated against structured diagnostic interviews in primary care [36], and a graded response model analysis of the DASS-21 found that a simulated adaptive version could cut item count by seventeen to forty-eight percent at equivalent reliability [37]. The significance for this field is straightforward: a questionnaire that survives weekly repetition can be sampled at a frequency approaching that of the physiological data, rather than once at intake and once at discharge.

Limitations

Three constraints bound the argument. First, most of the wearable evidence is observational; associations between HRV and affect do not establish direction, and reverse causation is plausible in both directions. Second, effect sizes in exercise trials vary widely with control condition, adherence measurement and blinding, which is intrinsically difficult in exercise research. Third, the components reviewed here are correlated in practice — people who train outdoors also sleep and eat differently — so the independent contribution of any one component is rarely isolated. Factorial designs, not larger observational cohorts, are what would resolve this.

Conclusion

The literature reviewed here is usually read as several separate findings: that exercise treats depression, that fitness predicts longevity, that sleep matters, that nature helps, that the gut talks to the brain. Read together, they describe a single regulatory system with several inputs and a small number of measurable outputs — heart rate, its variability, respiratory rate, sleep architecture, aerobic capacity — that respond to all of them.

That reframing has a practical consequence. If the inputs act on a common system, then the useful question is no longer which intervention is best on average, but which combination suits a particular person’s current physiological state, and how quickly that can be detected. The instruments to answer it already exist and are inexpensive: continuous autonomic measurement from consumer devices, periodic laboratory measures of aerobic capacity and body composition, and adaptive psychological screening repeatable without practice effects. What remains is the study design — prospective, pre-registered, with psychological outcomes specified in advance — to convert an accumulating set of associations into individualised guidance.

References

Numbered as cited above. The same studies are listed by topic on the references page.

1. Noetel et al., The BMJ, 2024.
2. Dimmock et al., British Journal of Sports Medicine, 2026.
3. Nature Mental Health, 2026.
4. British Journal of Sports Medicine, 2024.
5. European Heart Journal, 2025.
6. Frontiers in Psychiatry, 2025.
7. Wang et al., 2014.
8. Wendt & Thayer, 2024.
9. Umbrella review, Translational Psychiatry, 2025.
10. Inter- and Intrapersonal Associations Between Physiology and Mental Health, 2025.
11. Satapathy et al., 2025.
12. Scientific Reports, 2023.
13. Manresa-Rocamora et al., Applied Sciences, 2020, and IJERPH, 2021.
14. Kroenke et al., 2001 and 2010.
15. Spitzer et al., 2006.
16. Lovibond & Lovibond, 1995; Henry & Crawford, 2005.
17. Scott et al., 2021.
18. Grossman et al., Clinical Neuropsychiatry, February 2026, with Porges’s reply in the same issue.
19. Maior et al., Scandinavian Journal of Medicine & Science in Sports, 2013.
20. Kouidi et al., IJERPH, 2021.
21. Kumarathurai et al., Diabetes Care, 2017.
22. Grosicki et al., American Journal of Physiology – Heart and Circulatory Physiology, 2025.
23. Simon et al., Obstetrics & Gynecology, 2019.
24. U.S. Food and Drug Administration, 2023.
25. Pietilä et al., JMIR Mental Health, 2018.
26. UCLA trial, 2026.
27. Nature Human Behaviour, 2026.
28. Frontiers in Public Health, 2026.
29. Nature Mental Health, 2025.
30. Trail Running as a Therapeutic Modality, 2025.
31. Environment International, 2024, and EFUF, 2025.
32. Nature Reviews Microbiology, 2025; Gut Microbes, 2026; Frontiers in Microbiology, 2025.
33. British Journal of Nutrition, 2024; Dicken et al., 2025; Brunstrom et al., 2026.
34. International Journal of Molecular Sciences, 2025; Clinical Science, 2025; Calvillo et al., 2025; Kumar et al., 2026; Liang et al., 2026; Zhou et al., 2024.
35. Health Canada, “Think twice before injecting peptides bought online”, 9 April 2026.
36. Gibbons et al., Annals of Family Medicine, 2019.
37. JMIR, 2023;25:e45334.