CO₂ Evidence Guide

Clinical, human and translational research across inhalation, CO₂ preservation, rebreathing, transcutaneous delivery and bathing

Download as PDF

Over the course of the growth and development of Conscious Breathing - through product development, applied research, clinical questions, and the research behind two books - we have assembled a library of more than 500 scientific papers related to carbon dioxide physiology. Just as importantly, years of reading across fields have helped us build a working understanding of how these findings connect.

The pages that follow are a curated snapshot of the studies we currently find most informative: especially experiments that directly manipulate CO₂, isolate it from common confounders, or show unusually compelling clinical, physiological or mechanistic effects. This is not an exhaustive systematic review, and it is not intended to imply that every study in the wider library has been reviewed to the same depth.

It is a living evidence guide. Our understanding will continue to develop as new research appears and as we revisit, test and connect what is already known.

A useful way to read the evidence: researchers repeatedly approach CO₂ from opposite directions - add it, preserve it, remove it, or change its pattern - and ask whether physiology moves with it. The separation experiments are especially informative because they help distinguish CO₂ from breathing effort, oxygen, temperature, cerebral blood flow and other confounders.

Evidence labels describe study type, not a formal quality grade. Human studies are prioritized; animal, tissue and molecular work is clearly separated and labeled.

1. Direct inhaled CO2 and controlled hypercapnia

These studies add CO2 directly or deliberately control inspired/arterial CO2. They include sleep, respiratory control, cortical function, airway behavior, brain-fluid dynamics and dose-context experiments.

Study Design / CO₂ manipulation Key result Why it matters
HUMAN • RANDOMIZED CROSSOVER • SLEEP Liang et al., 2026 — Journal of Applied Physiology 16 healthy adults; four randomized overnight polysomnography conditions: 0%, 2.5%, 3.5% and 5.0% inspired CO₂. Sleep efficiency was highest at 2.5% CO₂ (91.7 ± 5.7%) and lowest at 5.0% (78.5 ± 10.1%; P<0.001). 2.5% increased N3 sleep; 3.5% was similar to room air; 5% reduced REM and increased arousals. Blood gases, BP, HR and catecholamines remained normal at ≤3.5%. One of the clearest modern human inhalation studies. It shows a dose window rather than a simple “more is better” effect: low-dose CO₂ improved normal sleep architecture while a higher dose reversed the direction.
HUMAN • CLINICAL • SLEEP / RESPIRATORY CONTROL Luo et al., 2023 — Journal of Applied Physiology 18 patients with central sleep apnea; inspired CO2 began around 1% and was individually titrated upward in 0.2% steps until central apneas disappeared. Apnea-hypopnea index fell from 31 ± 14 to 6 ± 3 events/h. Arousal index fell 22 ± 8 to 15 ± 8/h; sleep efficiency rose 71 ± 18% to 80 ± 11%; sleep latency fell 23 ± 18 to 10 ± 10 min. Direct low-dose CO2 stabilized an unstable respiratory control system and improved both breathing and sleep in the same night.
TRANSLATIONAL • RAT → MACAQUE → HUMAN • EPILEPSY Tolner et al., 2011 — Epilepsia 5% inhaled CO2 in rat seizure models; hypercapnia produced by hypoventilation in macaques (not inhaled 5%); and 5% carbogen in a 7-patient pilot in drug-resistant epilepsy. 5% CO2 suppressed cortical after-discharges by ~75% in rats. In macaques, raising pCO2 from ~37 to 44–45 mmHg suppressed stimulation-induced after-discharges by ~70%. In the human pilot, CO2 given after electrographic seizure onset produced rapid electrographic termination, including after generalization. An unusually strong translational chain across three species. It suggests that a relatively modest rise in CO2 can rapidly change pathological cortical excitability.
HUMAN • CONTROLLED • CORTICAL ACTIVITY Thesen et al., 2012 — Human Brain Mapping Healthy awake participants breathed 5% CO2 while magnetoencephalography measured visual and auditory cortical responses. End-tidal CO2 rose by ~8 mmHg. Early visual cortical responses fell about 20%; auditory and later cognitive cortical responses also decreased, while task accuracy and reaction time were unchanged. Direct evidence that mild hypercapnia measurably changes activity in the living human cortex without simply switching cognition off.
HUMAN • CONTROLLED • BRAIN METABOLISM / NETWORKS Xu et al., 2011 — Journal of Cerebral Blood Flow & Metabolism Conscious healthy humans breathed 5% CO2 during multimodal measurements of cerebral metabolism, perfusion, resting-state fMRI and EEG. CMRO2 fell 13.4 ± 2.3% (N=14), proportional to end-tidal CO2 change. Cerebral blood flow rose strongly; default-mode-network connectivity measures fell more than visual-task activation; EEG shifted toward lower frequencies. Shows that controlled hypercapnia simultaneously alters cerebral circulation, resting-network activity and brain metabolism in awake humans.
HUMAN • CONTROLLED • AIRWAY Torchio et al., 2022 — Journal of Applied Physiology Healthy subjects and people with mild asthma underwent methacholine bronchoconstriction while breathing air, 2% CO2 or 4% CO2. CO2 mixtures attenuated methacholine-induced changes in airway resistance/reactance by up to ~50% in both groups. Mediation analysis indicated much of the effect was explained by CO2-induced hyperpnea (increased minute ventilation); airway stretch was the authors’ proposed mechanism for that effect rather than a directly measured mediator. Inspired CO2 materially altered airway behavior during a controlled bronchoconstrictor challenge. The mechanism appears largely mechanical through hyperpnea rather than a simple direct smooth-muscle drug effect.
HUMAN • BIDIRECTIONAL • AIRWAY / ASTHMA van den Elshout et al., 1991 — Thorax 15 healthy and 30 asthmatic subjects rebreathed humidified body-temperature gas while end-tidal CO2 was experimentally raised or lowered. A +1 kPa rise in end-tidal CO2 reduced respiratory resistance by 15% in healthy subjects and 9% in asthma. A −1 kPa fall produced no significant change in controls but in asthma resistance rose 13%, reactance fell 45%, and frequency dependence of resistance rose 240%. A clean bidirectional human airway experiment with heat and water-loss confounders controlled. In asthma, airway mechanics worsened specifically as CO2 fell.
HUMAN • CONTROLLED • BRAIN CLEARANCE Erhardt et al., 2025 — npj Parkinson’s Disease Intermittent low-level hypercapnia was used to drive low-frequency vasomotion in awake healthy older adults and people with Parkinson’s disease. CO2 oscillation induced vasomotion-mediated CSF inflow in both groups. After three 10-min sessions, plasma concentrations of several neural-abundant proteins, including α-synuclein, NfL, GFAP, Aβ40/42 and p-tau217, increased. A modern human experiment using CO2 as an external input into the vascular oscillations that move brain fluid, with a concurrent blood signal consistent with increased efflux of brain-derived proteins.
HUMAN • CONTROLLED • TBI / BRAIN CLEARANCE Mayer et al., 2026 — Journal of Cerebral Blood Flow & Metabolism Oscillating hypercapnia applied during MRI in people with chronic traumatic brain injury and matched controls, with serial blood sampling afterward. Multiple neural-abundant proteins rose in blood around 45 min after hypercapnia, fell around 90 min and returned toward baseline later. Both groups also showed reduced cognitive interference after the protocol. Extends the intermittent-hypercapnia clearance paradigm to a second cohort and adds a time course suggestive of efflux followed by depletion/re-equilibration.
HUMAN • CONTROLLED • EEG / COGNITION Bloch-Salisbury et al., 2000 — Psychophysiology 9 healthy adults were mechanically ventilated with respiratory rate and tidal volume fixed; end-tidal CO2 was counterbalanced at ~30, 38 and 47 mmHg while inspired O2 was high enough to prevent tissue hypoxia. Large CO2 shifts changed EEG power/frequency, but produced no significant change in pattern recognition, matching, logical reasoning, search, alertness or evoked-potential measures. An important boundary study: the cortex clearly registered CO2, yet brief large shifts did not automatically impair cognitive output. It argues strongly against simplistic “CO2 concentration = cognition” claims.
HUMAN • CONTROLLED EXPOSURE • DOSE CONTEXT Maniscalco et al., 2022 — Indoor Air 24 adults exposed for 4 h to 770 ppm or 20,000 ppm (2%) CO2, including two 15-min cycling bouts; capillary blood gases and cognitive tests were measured. At 2%, blood pH fell and pCO2 rose significantly but remained within normal ranges; respiratory rate rose slightly. No significant changes occurred in heart rate, critical flicker fusion, task speed/accuracy or acute symptoms. Useful dose context: even a 4-h 2% exposure produced measurable physiological adaptation without cognitive or symptomatic impairment in this protocol.

Swipe horizontally to view all four study columns.

2. CO2 preserved, restored or deliberately depleted

These experiments are especially valuable because they separate CO2 from the act of breathing itself, from oxygen, from cerebral blood flow, or from other common confounders. Several approach the question from the reverse direction by deliberately lowering CO2.

Study Design / CO₂ manipulation Key result Why it matters
HUMAN • CONTROLLED SEPARATION • STRESS Stäubli et al., 1994 — European Journal of Applied Physiology 12 healthy subjects hyperventilated for 20 min at the same very high ventilation (~36 L/min) under two conditions: room air vs air containing 5% CO2. With room air, PaCO2 fell 21.4 mmHg and adrenaline rose 360%, noradrenaline 151%, lymphocytes 42% and platelets 8.4%. Adding CO2 limited the PaCO2 fall to 4.1 mmHg and largely prevented these changes; catecholamines did not rise. One of the cleanest stress experiments in the collection: the work of breathing was essentially the same, but the major stress response tracked the loss of CO2.
HUMAN • CONTROLLED SEPARATION • EXCITABILITY Hartley et al., 2016 — The Journal of Physiology 10 men in three conditions: hypocapnic hyperventilation; isocapnic hyperventilation with CO2 clamped; and indomethacin-induced cerebral blood-flow reduction with CO2 normal. Hypocapnia lowered end-tidal CO2 41.6→30.1 mmHg and reduced cerebral flow 21.7%. Motor-evoked-potential amplitude rose 11.7% vs isocapnic hyperventilation and 13.3% vs indomethacin. A larger 28.8% flow reduction with normal CO2 did not reproduce the excitability effect. A particularly strong human separation experiment: cortical excitability followed low CO2, not the act of hyperventilating and not the accompanying fall in cerebral blood flow.
HUMAN • TMS • EXCITABILITY Sparing et al., 2007 — Journal of Applied Physiology Voluntary hyperventilation produced marked hypocapnia while motor and visual cortex excitability were tested by transcranial magnetic stimulation. Motor threshold fell from ~40% to 35% stimulator output; phosphene threshold fell ~42% to 37%; intracortical inhibition decreased while facilitation did not significantly change. Independent human evidence that hypocapnia lowers cortical firing thresholds and weakens inhibition. The visual-cortex arm is particularly useful because it is less confounded by spinal/peripheral effects.
HUMAN • END-TIDAL GAS CLAMP • COGNITION Friend et al., 2019 — Experimental Physiology Dynamic end-tidal forcing independently manipulated oxygen and CO2, allowing normoxic hypocapnia to be compared with hypoxia at normal CO2. Lowering end-tidal CO2 by ~8 mmHg with oxygen held normal slowed simple reaction time by ~174 ms and choice reaction time by ~167 ms. In the separate hypoxic arm (end-tidal O2 ~45 mmHg) with CO2 maintained, reaction times were unchanged; spatial working memory was unaffected in all conditions. A clean gas-separation experiment: modest hypocapnia altered reaction speed under this paradigm while substantial hypoxia alone did not.
HUMAN • CLASSIC PHYSIOLOGY • CEREBRAL PERFUSION Kety & Schmidt, 1948 — Journal of Clinical Investigation Normal young men underwent voluntary hyperventilation with arterial and jugular sampling plus whole-brain blood-flow measurements. Arterial CO2 fell ~45→26 mmHg and cerebral blood flow fell ~52→34 mL/100 g/min (~35%). Arterial oxygen content rose slightly while jugular oxygen content fell 11.3→7.8 vol%, widening brain oxygen extraction; cerebral O2 consumption was maintained. Classic demonstration that arterial oxygen can look normal or better while low CO2 sharply reduces brain perfusion and forces greater oxygen extraction.
HUMAN • PET • CEREBRAL PERFUSION Ito et al., 2000 — Journal of Cerebral Blood Flow & Metabolism 11 healthy adults; 15O-water PET measured regional cerebral blood flow at rest, hypercapnia and hypocapnia. At hypocapnia (PaCO2 ~28.8 mmHg), cortical flow fell roughly one-third; internal values include frontal flow ~43.5→28.4 and cortical average ~46.8→30.4 mL/100 g/min. Regional hypocapnic reactivity was broadly ~2.8–3.6% per mmHg. Modern regional imaging confirmation that CO2 is a powerful controller of cerebral perfusion across the human brain.
HUMAN • RANDOMIZED • PANIC DISORDER Meuret et al., 2008 — Journal of Psychiatric Research 37 people with panic disorder randomized to brief capnometry-assisted breathing therapy designed to raise end-tidal pCO2 or to delayed treatment. Treated participants showed significant improvements in panic severity, agoraphobic avoidance, anxiety sensitivity, disability and respiratory measures, with moderate-to-large effect sizes; improvements persisted at follow-up. Clinical evidence that deliberately normalizing end-tidal CO2 via a brief breathing therapy that raises end-tidal CO2 improves panic disorder. Because the comparison was a delayed-treatment (wait-list) control, the trial shows the therapy works but does not by itself isolate raising CO2 from the breathing training as the active ingredient.
HUMAN • RCT • ASTHMA Ritz et al., 2014 (CATCH) — Chest 120 adults with asthma randomized to capnometry-assisted training aimed at raising pCO2 or an active slow-breathing/awareness program; blinded follow-up to 6 months. 17 of 21 clinical indices improved across the interventions, including asthma control (d=0.81), peak-flow variability (d=0.54), quality of life, bronchodilator use, lung function and airway hyperreactivity. CO2-targeted training produced a larger sustained pCO2 increase and additional advantages in respiratory impedance, methacholine distress and later symptoms. A comparatively large randomized clinical trial showing sustained clinical and physiological changes when CO2 regulation is deliberately trained; the active slow-breathing control also improved, which makes the comparison informative rather than trivial.
HUMAN • RANDOMIZED SEPARATION • LACTATE Davies et al., 1986 — Journal of Applied Physiology 8 subjects performed progressive cycling in randomized conditions: hyperventilation-induced respiratory alkalosis; the same hyperventilation after acetazolamide restored pH despite similar low PaCO2/ventilation; and metabolic acidosis. Hyperventilation with alkalosis increased blood lactate. With acetazolamide restoring pH despite similar ventilation and PaCO2, lactate returned to control; metabolic acidosis lowered lactate. Elegant separation experiment showing the hyperventilation-associated rise in lactate was driven by alkalosis/pH, not simply by breathing volume or low PaCO2 itself.
HUMAN • MUSCLE BIOPSY • METABOLISM LeBlanc et al., 2002 — The Journal of Physiology Subjects exercised 15 min at 55% VO2max while breathing normally or hyperventilating; muscle biopsies were taken at rest, 1 min and 15 min. During the first minute of respiratory alkalosis, pyruvate dehydrogenase activation was delayed, pyruvate oxidation fell, glycogenolysis rose and more pyruvate was diverted to lactate. Differences largely disappeared by 15 min. Direct human muscle-biopsy evidence that respiratory alkalosis changes fuel handling at the enzyme/pathway level during the transition into exercise.
HUMAN • COUNTERBALANCED • SPRINT METABOLISM Fujii et al., 2015 — European Journal of Applied Physiology 9 subjects performed 30-s Wingate tests after normal rest or 20 min voluntary hyperventilation; end-tidal CO2 fell from ~34.8 to 19.3 mmHg in the hypocapnia condition. Oxygen uptake during the Wingate fell from 1.95 ± 0.44 to 1.55 ± 0.52 L/min (~21%) while postexercise peak lactate rose from 9.6 ± 1.9 to 10.4 ± 1.9 mmol/L. Exercise performance was maintained. Shows an acute metabolic shift under hypocapnia: less measured aerobic contribution and more lactate for the same short all-out task.
HUMAN • NMR / EEG • BRAIN METABOLISM van Rijen et al., 1989 — Magnetic Resonance in Medicine Healthy volunteers underwent voluntary hyperventilation while 1H/31P NMR, EEG, capnography and middle-cerebral-artery Doppler were recorded. Hyperventilation reduced cerebral flow velocity, increased slow EEG activity and raised cerebral lactate. ATP and phosphocreatine peak heights did not fall; tissue shifted alkaline during hyperventilation and lactate fell again during recovery. Important because the lactate rise occurred without depletion of high-energy phosphates: it supports alkalosis-driven metabolic adaptation rather than a simple “brain became hypoxic and failed” story.
HUMAN • CO2 CLAMP • CSF DYNAMICS Kish et al., 2024 — NMR in Biomedicine Participants underwent resting-state MRI during free breathing and with end-tidal CO2 clamped toward its baseline average, which reduced its natural low-frequency fluctuations (mean CO2 also rose). Clamping CO2 significantly reduced low-frequency hemodynamic oscillations in several cortical regions and reduced low-frequency CSF oscillation amplitude at the fourth ventricle by almost one-half. A striking demonstration that the fluctuation in CO2 matters, not only the mean level: reducing the CO2 fluctuation removed a large part of the human CSF oscillation signal. Because mean CO2 also rose during clamping, this was not a pure variability-only manipulation.

Swipe horizontally to view all four study columns.

3. Rebreathing, added dead space and exercise hypercapnia

These studies retain more metabolically produced CO2 during exercise or warm-up by adding respiratory dead space. They are especially relevant to performance physiology because the CO2 shift occurs alongside real work rather than during passive exposure.

Study Design / CO₂ manipulation Key result Why it matters
HUMAN • CROSSOVER • SPRINT PERFORMANCE Danek et al., 2020 — International Journal of Environmental Research and Public Health 11 physically active men completed a crossover sprint-interval protocol (6 × 10-s all-out cycling) with or without ~1200 mL added respiratory dead space. With added dead space, total work was 4.4% higher, mean VO2 31.3% higher and ventilation 13.2% higher. Blood pH was lower (7.26 vs 7.29), while lactate and perceived exertion did not differ. Direct human performance evidence that mild rebreathing-induced respiratory acidosis can coexist with more work and greater oxygen uptake without higher perceived effort.
HUMAN • CROSSOVER • WARM-UP / POWER Danek et al., 2022 — Frontiers in Physiology Healthy physically active men (N=10) completed a 10-min warm-up with or without ~1200 mL added respiratory dead space before 6 sprint bouts. After the dead-space warm-up, end-tidal CO2 was ~45.0 vs 41.6 mmHg. Mean sprint power was ~771.7 vs 749.6 W (P=0.01) and total work ~46.3 vs 45.0 kJ (~2.9% higher), with no meaningful difference in lactate, fatigue index or perceived exertion. The intervention was removed before the sprints: a brief hypercapnic warm-up was followed by a measurable improvement in subsequent power/work.
HUMAN • RANDOMIZED CROSSOVER • SWIMMING Danek et al., 2025 — PLOS ONE 8 male university swimmers completed standard warm-up, hypercapnic warm-up, or hypercapnic re-warm-up during the transition period before a 50-m sprint. pCO2 reached ~49.7 mmHg after the hypercapnic warm-up and ~45.7 mmHg immediately before the sprint in the re-warm-up condition. Re-warm-up produced the fastest swim (27.5 ± 1.6 s), 1.2% faster than hypercapnic warm-up (P=0.01); the comparison with control was a trend rather than significant. Small but useful evidence that timing of mild hypercapnia around warm-up may matter. The strongest statistical comparison was between the two hypercapnic strategies, not versus control.

Swipe horizontally to view all four study columns.

4. Transcutaneous CO2: gas and topical delivery

These studies deliver CO2 across the skin without requiring the lungs to carry the full dose. The strongest human experiments isolate CO2 from temperature/humidity or directly measure local tissue pH and hemoglobin oxygenation.

Study Design / CO₂ manipulation Key result Why it matters
HUMAN • RANDOMIZED DOUBLE-BLIND • VASCULAR Savin et al., 1995 — Angiology 10 patients with stage II lower-limb arteriopathy received 20 min local CO2-rich spa gas or water-vapor-saturated air at the same temperature. CO2 increased femoral blood flow (~24%), posterior tibial artery pressure (~8%) and foot transcutaneous oxygen tension (~12%). Warm humidified air produced no effect; systemic hemodynamics did not change. One of the cleanest transcutaneous studies because the control removes warmth and humidity. The local vascular and oxygen effects followed CO2 itself.
HUMAN • PHYSIOLOGY • SKIN TRANSFER Dervieux et al., 2024 — Frontiers in Physiology 40 healthy adults were measured at skin temperatures from roughly 29°C to 44°C using custom CO2 conductivity sensors and laser-Doppler flowmetry. Skin CO2 conductivity roughly doubled across the temperature range. Skin blood flow rose ~11-fold overall and was already ~4-fold higher by ~38°C. Directly quantifies that transcutaneous CO2 transfer is state-dependent: warmth and perfusion can substantially change transcutaneous CO2 transfer. The measurement was of CO2 passing outward through the skin, so it characterizes skin-transfer properties rather than directly quantifying inward absorption of applied CO2.
HUMAN + IN VITRO • TISSUE pH / BOHR EFFECT Sakai et al., 2011 — PLOS ONE Pure transcutaneous CO2 plus hydrogel; rat-skin permeation validation, then human muscle 31P-MRS and human arm NIRS. The NIRS oxygen-dissociation arm used blood-flow occlusion. Human triceps-surae intracellular pH fell significantly after 10 min. In the occluded-arm experiment, oxyhemoglobin decreased and deoxyhemoglobin increased significantly after CO2 application vs control. Direct human evidence that transcutaneous CO2 reaches deeper tissue, changes muscle pH and can shift hemoglobin oxygenation in the direction expected from the Bohr effect. The flow-occlusion condition should remain explicit.

Swipe horizontally to view all four study columns.

5. CO2-rich water, carbonated baths and immersion

Water-based studies provide some of the strongest controlled clinical and athletic evidence because temperature-matched fresh-water conditions can separate CO2 from ordinary hydrotherapy.

Study Design / CO₂ manipulation Key result Why it matters
HUMAN • RANDOMIZED CONTROLLED • PAD Hartmann et al., 1997 — Angiology 24 patients with stable intermittent claudication randomized to lower-leg immersion in CO2-enriched water (1000 mg/kg) or fresh water, 33°C, 30 min, 5×/week for 4 weeks. Pain-free walking distance rose 224→367 m (+64%) in the CO2 group; fresh water produced no significant change. Foot transcutaneous oxygen tension rose ~24→34 mmHg and reactive-hyperemia measures improved. Arguably the strongest functional clinical-outcome study in the transcutaneous literature: repeated CO2 exposure produced a large objective walking gain against a matched water control.
HUMAN • DOUBLE-BLIND • MUSCLE OXYGENATION Yoshimura et al., 2020 — PeerJ 15 healthy young men; one lower leg immersed for 15 min in CO2-rich or plain water at 20°C, followed by recovery; NIRS separated skin and muscle-layer signals. CO2-rich cold water increased tissue oxygenation/total heme and oxygen saturation in tibialis anterior and gastrocnemius regions despite cold-induced vasoconstrictive conditions. Muscle-layer effects were positive but more moderate than skin/saturation signals. Supports an effect beyond superficial skin and is striking because CO2 produced the signal while cold was pushing vascular tone in the opposite direction.
HUMAN • RANDOMIZED CROSSOVER • ATHLETIC RECOVERY Yoshimura et al., 2022 — Research in Sports Medicine 15 trained athletes completed repeated cycling/Wingate sessions separated by randomized 20-min recovery in CO2-rich cold water, plain cold water or passive rest. CO2-rich immersion produced the largest heart-rate reduction. Blood lactate was significantly lower than passive recovery (but not significantly lower than ordinary cold-water immersion). Post-recovery lactate strongly tracked maintenance of work efficiency in the second bout in the full paper. Controlled athlete evidence linking post-exercise CO2 immersion to measurable recovery physiology; the plain-cold-water comparison prevents overstating what was specifically added by CO2.
HUMAN • MATCHED-TEMPERATURE • THERMOREGULATION Iwata et al., 2025 — Physiological Reports 10 male college baseball players after daily training: 15 min in 40°C CO2 water (1000 ppm), 40°C tap water, or seated rest. Compared with matched-temperature tap water, CO2 immersion produced higher core temperature (d=0.52) and local sweat rate (d=0.80) without greater thermal discomfort. A useful isolation study because heat was matched. CO2 added a measurable thermoregulatory response beyond hot-water immersion alone.
HUMAN • RANDOMIZED GROUPS • REPEATED RECOVERY Fujita et al., 2022 — Journal of Exercise Science & Fitness 32 trained males randomized to CO2-rich cold-water immersion, cold tap water or passive recovery across 4 days of high-intensity cycling; immersions were performed on days 1–3. Heart rate fell after both immersions, with a larger reduction in the CO2 group (P<0.001). Day-2 lactate was lower vs passive recovery (P<0.01). Mean/peak power changes did not differ among groups (P=0.302), although change in peak pedal repetitions correlated very strongly with post-immersion HR and lactate. Repeated-use human data show a consistent recovery-physiology signal, while the absence of a between-group power difference keeps the performance interpretation appropriately bounded.
HUMAN • CROSSOVER • SWIMMING WARM-UP Akamine & Taguchi, 1998 — Journal of Human Ergology 6 competitive swimmers; 20 min in a 36°C carbonated bath (300 ppm CO2) vs 36°C fresh water before swimming. After the carbonated bath, recovery changes in blood lactate and heart rate were significantly smaller (P<0.05), and rectus-femoris EMG during swimming decreased, suggesting more efficient muscle activity. Small early human study, but valuable because temperature was matched and the outcomes were directly relevant to athletic efficiency and recovery.
HUMAN • PERIPHERAL FLOW • BATHING Hasyar et al., 2020 — Malaysian Journal of Medicine and Health Sciences Healthy volunteers underwent CO2-rich water bathing while peripheral blood flow was measured before, during and after immersion; plasma nitric oxide and endothelin-1 were also examined. The design was within-subject before/during/after immersion with no non-CO2 (plain-water) comparator. Peripheral leg blood flow increased during CO2-rich bathing and moved back toward baseline after exposure. The measured NO/ET-1 indices did not show a corresponding significant change. A useful human replication of the immediate peripheral-flow effect and a reminder that the vascular response may not be fully captured by circulating NO/ET-1 measurements.

Swipe horizontally to view all four study columns.

6. Mechanistic and translational highlights

These are not presented as human clinical efficacy trials. They are included because the experiments answer unusually important mechanistic questions: how CO2 changes excitability, how it is sensed directly, how it interacts with oxidative stress and inflammation, and what transcutaneous delivery can do to injured tissue.

Study Design / CO₂ manipulation Key result Why it matters
BRAIN SLICE • BIDIRECTIONAL MECHANISM • ADENOSINE Dulla et al., 2005 — Neuron Hippocampal slices exposed to graded CO2 above and below the 5% control condition while extracellular adenosine/ATP and synaptic transmission were measured. 20% CO2 raised extracellular adenosine and reduced excitatory synaptic potentials ~45.8%; 10% reduced them ~27.3%. At 2% CO2, extracellular adenosine fell and excitatory synaptic activity increased ~18–23%. Blocking adenosine A1 and ATP signaling abolished key effects. Elegant bidirectional mechanism linking CO2/pH to an endogenous inhibitory brake on cortical excitability. The higher-dose arms are not ordinary human physiology, so this is mechanistic rather than dose-equivalent clinical evidence.
ANIMAL • CAUSAL SEPARATION • ABSENCE SEIZURES Salvati et al., 2022 — eLife Seizure-prone WAG/Rij rats underwent gas manipulations and optogenetically induced hyperventilation to separate hypoxia, breathing rate, arterial CO2 and pH. Hyperventilation/respiratory alkalosis provoked spike-wave discharges. Adding 5% CO2 to prevent the alkalosis blocked the hypoxia-associated seizure increase; hypercapnia alone reduced spontaneous spike-wave counts. Optogenetic hyperventilation at normoxia also triggered CO2-sensitive seizures. A particularly clean causal experiment showing that the CO2/pH shift — not merely fast breathing or hypoxia — can control an absence-seizure circuit.
ANIMAL • FEAR CIRCUIT • CHEMOSENSING Ziemann et al., 2009 — Cell Mice inhaled CO2 while brain pH and fear behavior were measured; ASIC1a was deleted, inhibited, restored specifically in the amygdala, or bypassed by direct acidification. Inhaled CO2 lowered brain pH and evoked fear. Eliminating/inhibiting ASIC1a markedly reduced the response; restoring ASIC1a in the amygdala rescued it; buffering pH reduced fear and direct amygdala acidification reproduced it. Iconic evidence that a core mammalian fear circuit directly monitors CO2/acidity. It also reinforces the central dose principle: acute high CO2 can be an alarm signal even when preserving normal CO2 can reduce hypocapnic stress.
MOLECULAR / CELL • DIRECT CO2 SENSING Meigh et al., 2013 — eLife Connexin proteins from human/mouse/rat systems were mutated to identify the structural motif responsible for CO2 sensitivity. CO2 directly carbamylated connexin-26 and altered channel gating. Introducing the carbamylation motif into CO2-insensitive connexin-31 created CO2 sensitivity; mutating the motif in connexin-26 destroyed it. Critical molecular proof that CO2 can act as a signaling molecule in its own right through reversible protein carbamylation, not only indirectly by changing pH.
MOLECULAR / CELL • DIRECT CO2 BINDING / NF-κB Linthwaite et al., 2021 — Science Advances 13C-NMR and biochemical experiments mapped reversible CO2 carbamylation of ubiquitin and tested downstream ubiquitin signaling. CO2 carbamylated the ubiquitin N-terminus and several lysines, including Lys48. Physiologically relevant CO2 changes altered K48-linked ubiquitination and reduced activation of ubiquitin-dependent NF-κB signaling; mutation of a key CO2-responsive site disrupted the effect. Extends direct CO2 sensing from a membrane channel to a central protein-signaling system, linking the gas to protein turnover and inflammatory signaling.
ANIMAL + IN VITRO • OXIDATIVE STRESS Skoumalová et al., 2008 — Cell Biochemistry and Function Rats exposed for 4 days to 10% O2 with or without 4.3% CO2; isolated erythrocyte membranes were also tested under oxidative stress in bicarbonate buffer. Hypoxia increased erythrocyte lipofuscin-like oxidation products up to ~9-fold; adding 4.3% CO2 completely prevented the increase. Bicarbonate strongly inhibited malondialdehyde and lipofuscin-like products in vitro. Striking animal/in-vitro evidence that hypercapnia/bicarbonate can modify oxidative damage under hypoxic stress, pointing beyond circulation alone.
ANIMAL • TRANSCUTANEOUS • MUSCLE REPAIR Akahane et al., 2017 — International Orthopaedics 27 rats with bupivacaine-induced tibialis-anterior injury randomized to repeated transcutaneous CO2 or no treatment, followed to 6 weeks. Injured fibers were completely repaired by week 6 in the CO2-treated group but only partially repaired in controls. Muscle protein-synthesis genes/proteins were higher, and capillary density was significantly greater at 4 weeks. Shows transcutaneous CO2 changing the biology of tissue repair over weeks — not merely causing an acute vasodilatory response.
ANIMAL • TOPICAL CO2 • REGENERATION / SCARRING Hirota et al., 2023 — International Wound Journal 48 rats with bupivacaine muscle injury received daily local transcutaneous CO2 paste or no treatment; inflammatory, fibrotic and myogenic markers were tracked. CO2 treatment reduced inflammatory signaling, increased myogenic markers including MyoD/myogenin, increased muscle-fiber cross-sectional area and reduced fibrous tissue during repair. Independent extension of the muscle-repair signal: local CO2 altered inflammation, regeneration and scar formation in the same injury model family.
ANIMAL / HUMAN-RELEVANT MECHANISM • NEUROVASCULAR COUPLING Hosford et al., 2022 — Nature Communications Multimodal cortical experiments tested whether CO2 produced by active neural tissue participates in the signal that couples neuronal activity to local blood flow. Manipulating CO2 handling strongly altered neurovascular responses; saturating the system with exogenous CO2 blunted activity-driven blood-flow responses, and disrupting astrocytic CO2/bicarbonate transport could abolish or invert coupling. Important because it treats CO2 as an active messenger within neurovascular coupling rather than only a global vasodilator. It also shows why static high CO2 is not equivalent to a physiological CO2 signal.
MICROBIOLOGY • DIRECT GAS EFFECT Persson et al., 2005 — Surgical Endoscopy Staphylococcus aureus cultured at body temperature under pure CO2, anaerobic gas or air. After 24 h, viable S. aureus on blood agar was ~100× lower in CO2 than anaerobic gas and ~1000× lower than air (P=0.001). In broth, growth was also strongly suppressed. A striking non-human example of CO2 acting directly on biology rather than only through mammalian circulation. It establishes a large bacteriostatic effect under controlled conditions.

Swipe horizontally to view all four study columns.

The pattern across routes

What makes this collection interesting is convergence. Direct inhalation, preservation of endogenous CO₂, deliberate hypocapnia, rebreathing, transcutaneous gas and carbonated-water exposure repeatedly change vascular tone, oxygen delivery, airway mechanics, neuronal excitability, respiratory stability, sleep, stress signaling, metabolism, exercise recovery and brain-fluid dynamics. The interventions and doses are not interchangeable, and not every direction is beneficial. The common point is that CO₂ behaves as a regulated physiological input whose level, timing, oscillation and tissue distribution can materially change biological function.

Cancer evidence is intentionally excluded from this guide. The oncology literature - including carbogen/radiotherapy trials, tumor hypoxia/HIF biology and transcutaneous animal tumor studies - warrants its own companion evidence document because the disease-specific claims and evidence boundaries are different.

Source note: This draft draws from the Conscious Breathing research library and working white papers, supplemented by independent verification of key bibliographic details and current publications. Before formal publication, individual rows should receive a final check against the primary paper used for publication.