https://www.cell.com/neuron/fulltext/S0896-6273(26)00339-9
I saw this because of this article which is their news piece:
A new study from the German Institute of Human Nutrition Potsdam–Rehbruecke (DIfE) and Charité – Universitätsmedizin Berlin shows for the first time that targeted control of human breathing rhythm can influence decision behavior by modulating heart and brain function. The research team led by Prof. Soyoung Q Park was able to demonstrate that prolonged exhalation increases heartrate variability and brain’s reward sensitivity, thus enabling us to make bolder decisions. The study was published in the journal Neuron.
chatGPT(5.5paid) on the original:
The uploaded paper is Huang et al. (2026), “Slow breathing impacts inter-organ dynamics modulating brain function and risk behavior,” Neuron. Its central claim is that deliberately prolonging exhalation alters autonomic state, which in turn changes neural reward processing and makes people more willing to accept risky gambles.
1. Summary
Central question
The authors ask whether a person can deliberately alter decision-making by changing their physiological state through breathing. In particular, they hypothesised that prolonged exhalation would:
- increase risky choices;
- increase cardiac parasympathetic activity;
- alter neural representation of rewards.
These hypotheses were preregistered.
The theoretical framework is neurovisceral integration: bodily/autonomic state is not merely a consequence of cognition but contributes to how the brain evaluates external information.
Experimental design
Forty-nine healthy adults were recruited, with eight excluded, leaving 41 participants, 24 female, mean age about 24.8 years. They were unusually tightly selected: age 18–40, BMI 18–25, right-handed, no regular medication, smoking, significant disease, abnormal breathing or major sleep disruption.
Each participant underwent both conditions:
- Eupnea: paced according to their individually measured normal breathing.
- Prolonged exhalation: fixed 2 s inhalation : 8 s exhalation, i.e. about six breaths/minute.
Participants simultaneously performed a gambling task in the MRI scanner. Each gamble offered a 50% chance of a reward of €10–30 and a 50% chance of a loss of €5–15. They accepted or rejected 120 gambles under each breathing condition. Three trials were randomly selected for actual financial consequences.
The authors simultaneously measured:
- respiration;
- ECG/PPG;
- heart-rate variability;
- skin conductance;
- pupil diameter;
- fMRI BOLD responses.
So it is a fairly ambitious multimodal experiment.
2. Main findings
A. The breathing manipulation worked extremely strongly
The prolonged-exhalation condition increased exhalation duration by about 5.18 seconds, and reduced respiratory rate by about 6.1 breaths/minute. These were very large manipulation effects.
This point becomes important for the critique, because the intervention changed both exhalation/inhalation ratio and overall breathing frequency.
B. HRV increased
The authors found increases in two cardiac measures interpreted as increased parasympathetic/vagal activity:
- RespHRV: very large effect, Cohen’s d ≈ 1.38;
- RMSSD: moderate effect, d ≈ 0.44.
Skin conductance and pupil diameter did not change significantly, leading the authors to describe the autonomic effect as relatively selective for parasympathetic rather than sympathetic activity.
This is one of the clearest physiological results in the paper.
C. Participants accepted more gambles
Prolonged exhalation produced a statistically significant increase in acceptance:
β = 0.168, p = 0.018.
More interestingly, the effect appears to come largely from an increased weighting of potential reward:
breathing × reward interaction: β = 0.176, p < 0.001.
There was not a significant corresponding change in sensitivity to losses.
The authors therefore argue that slow/prolonged-exhalation breathing does not simply make people more risk tolerant. Instead, it causes them to assign more weight to prospective rewards.
That distinction is important.
D. This was not simply faster/sloppier responding
Response time did not differ significantly. Choice entropy and estimated lapse rates also did not significantly differ, making a general deterioration in attention or consistency a less plausible explanation.
E. The more HRV changed, the more reward-related fMRI activity changed
Across individuals, a greater breathing-induced increase in RMSSD correlated with stronger reward-related BOLD activity in:
- ventromedial prefrontal cortex (vmPFC);
- precuneus.
The authors interpret the vmPFC as a subjective-value integration area and the precuneus as potentially integrating bodily/interoceptive state with self-related cognition.
Interestingly, they did not find corresponding effects in several regions one might have predicted, including the insula, ACC, striatum or amygdala.
3. What is genuinely novel?
I would distinguish the novelty into three levels.
3.1 Breathing increasing HRV is not novel
The paper itself makes clear that prolonged exhalation and slow breathing have previously been shown to increase HRV. It also cites prior evidence associating higher HRV with reward sensitivity.
So neither:
slow breathing → increased HRV
nor
HRV ↔ decision-making
is especially new.
3.2 The experimental manipulation of reward sensitivity is substantially more novel
The stronger contribution is moving from between-person correlation to an acute within-person manipulation.
Previous studies essentially showed:
People with higher HRV tend to differ in reward-related behaviour.
This experiment instead says:
Change the same person’s breathing → change their autonomic state → change how strongly reward magnitude affects their decisions.
The authors themselves correctly identify this distinction as a major contribution.
That is a meaningful advance.
3.3 The simultaneous breathing–heart–brain–behaviour measurement is probably the paper’s strongest novelty
The most interesting aspect is the attempt to link four levels within one experiment:
controlled respiration → cardiac dynamics → cortical reward representation → choice.
The fMRI association between breathing-induced ΔRMSSD and reward encoding in vmPFC/precuneus gives the study much more mechanistic interest than a simple breathing-versus-behaviour experiment.
Conceptually, the paper moves toward a model such as:
[
\text{breathing}
\rightarrow
\text{cardiac/autonomic state}
\rightarrow
\text{brain value representation}
\rightarrow
\text{choice}
]
This integrated approach is the part I would regard as the paper’s strongest novelty.
4. Critique
There are several important limitations, some of which materially weaken the paper’s causal interpretation.
4.1 The experiment does not isolate prolonged exhalation
This is probably the most important methodological issue.
The intervention is described as “prolonged exhalation”, but compared with eupnea it simultaneously changes:
- inhalation/exhalation ratio;
- exhalation duration;
- total respiratory cycle length;
- respiratory frequency;
- probably tidal volume;
- probably blood CO₂ dynamics.
The authors actually observe a roughly six-breath/minute fall in respiration rate.
Therefore the experiment cannot establish:
prolonged exhalation causes the effect.
It establishes something closer to:
a 2:8 s slow-breathing protocol causes the effect relative to normal breathing.
To demonstrate that exhalation length specifically matters, they would need a rate-matched control, for example:
- 5:5 s breathing versus
- 2:8 s breathing.
Without that control, the title and mechanistic interpretation somewhat overstate what was experimentally isolated.
4.2 HRV during paced slow breathing is not a simple independent measure of “parasympathetic activity”
This is related but potentially even more important physiologically.
RespHRV/RSA is intrinsically generated by the interaction between respiration and heart rate. If you deliberately change breathing frequency and respiratory phase duration, you necessarily change the oscillatory structure from which RespHRV is calculated.
The paper found the enormous RespHRV effect:
d ≈ 1.38
alongside the enormous alteration in respiration.
Consequently, some of the increased RespHRV is not evidence for a separate downstream biological event; it is closely coupled to the breathing manipulation itself.
RMSSD is preferable here, and it also increased, but RMSSD itself remains influenced by respiratory pattern.
Thus the claim:
breathing increased parasympathetic activity
is reasonable, but stronger language such as:
breathing → increased vagal neural output → altered brain reward processing
requires more direct evidence.
Direct vagal measurements are obviously difficult in humans, but blood pressure/baroreflex measurements or respiratory-controlled autonomic modelling would strengthen this interpretation.
4.3 The proposed causal chain is not actually demonstrated
The breathing condition is experimentally manipulated, so:
[
breathing \rightarrow behaviour
]
has reasonably good causal support.
And:
[
breathing \rightarrow HRV
]
also has good support.
But the crucial finding relating HRV to brain activation is between-subject correlational:
people whose RMSSD changed more also showed greater reward-related vmPFC/precuneus activation.
That does not establish:
[
HRV\rightarrow vmPFC \rightarrow reward\ behaviour
]
HRV and fMRI changes could both be consequences of some third breathing-induced variable, including:
- CO₂;
- blood pressure/baroreflex effects;
- attention to breathing;
- respiratory mechanics;
- arousal;
- individual compliance.
The paper therefore provides evidence consistent with a neurovisceral pathway, rather than demonstrating the pathway.
I think “revealing a neurovisceral pathway” in the paper’s summary is somewhat stronger than the experiment warrants.
A formal within-subject mediation model would have helped, although even mediation would not completely establish mechanism.
5. A major missing physiological variable: CO₂
I think this deserves more emphasis than it receives in the paper.
Changing breathing from ordinary breathing to a 10-second cycle can alter:
- alveolar ventilation;
- arterial (P_{CO_2});
- cerebral blood flow;
- cerebral vascular reactivity.
That is particularly important because the neural evidence depends on BOLD fMRI, which is fundamentally haemodynamic.
Therefore respiratory manipulation can change the BOLD signal through vascular physiology, independently of neuronal reward processing.
The authors use sophisticated physiological processing, but unless end-tidal CO₂ was adequately measured/modelled—and the paper does not establish that in the evidence surfaced here—one cannot completely rule out a respiratory/vascular contribution to the fMRI differences.
This is particularly relevant for a paper making mechanistic claims from simultaneous slow breathing and BOLD imaging.
6. The behavioural finding is statistically real but modest
The headline effect that prolonged exhalation increased gamble acceptance has:
[
p=0.018
]
and β = 0.168.
That is much less striking than the very strong respiration and HRV manipulation effects.
The more convincing behavioural result is the breathing × reward interaction, which is p < 0.001.
So I would describe the data as showing fairly persuasive evidence for altered weighting of reward, rather than a dramatic general change in risk-taking.
Indeed, the authors themselves reach essentially this interpretation.
7. Alternative modelling gives a less clean result
This is an important nuance which could easily be missed from the abstract.
When the authors fitted prospect-theory models:
- they did not find consistent differences in prospect-theory parameters;
- their non-parametric acceptance-threshold AUC did not change at all:
p = 0.974; - inverse-temperature parameter recovery was poor (r only ≈0.37–0.38).
They consequently treat the prospect-theory analysis as descriptive and base their conclusions primarily on the trial-wise mixed models.
This does not invalidate the primary GLMM result, but it means the behavioural effect is somewhat model-dependent.
That weakens any strong claim that breathing changes a well-defined underlying psychological parameter such as loss aversion or risk preference.
8. Reward and loss were deliberately asymmetric
Rewards were €10–30 while losses were €5–15.
This was intentional, because the authors designed the task around conventional loss aversion. But it creates a potential interpretive limitation.
They themselves acknowledge that the experiment therefore cannot determine whether the apparent preferential effect on reward would also occur with symmetric reward/loss ranges. They appropriately describe the study as a proof of principle.
This is particularly relevant to their headline result that breathing “selectively” affects reward rather than loss.
It might be a general biological phenomenon—or partly a feature of the particular choice-space sampled.
9. Sample size becomes quite small for the imaging claim
Although behavioural (n=41):
- HRV (n=35);
- fMRI analyses were restricted to those same 35;
- skin conductance (n=26);
- pupillometry (n=28).
Thirty-five subjects is not absurd for within-subject fMRI, and the whole-brain findings were cluster-level FWE corrected, which is a strength.
Nevertheless, a brain–physiology individual-difference correlation with n≈35 should normally be regarded as needing independent replication.
The impressive scatterplots should not be interpreted as establishing a robust population-level brain biomarker yet.
10. Generalisability is limited
This is an extraordinarily healthy and homogeneous population:
- young;
- lean BMI;
- right-handed;
- no regular medication;
- no smoking;
- no cardiovascular/pulmonary/metabolic illness;
- university-community recruitment.
That improves internal physiological control but reduces external validity.
It is particularly relevant because the authors discuss potential applications to:
- anxiety;
- panic disorder;
- depression;
- military settings;
- emergency response;
- elite sport;
- aviation.
Those implications are interesting hypotheses, but the current study doesn’t test those populations or circumstances.
11. Blinding and expectancy are essentially impossible
Participants know whether they are breathing naturally or doing conspicuous 2-second inhalations and 8-second exhalations.
Consequently, expectancy effects cannot be excluded.
A participant might implicitly believe:
“This is the special controlled breathing condition, so I should behave differently.”
The use of instructed visual cues for both conditions helps control task demands, but it does not solve this problem.
An active slow-breathing comparator—again, something such as 5:5 breathing—would greatly improve the design.
12. Block design creates another potential confound
The conditions were delivered as:
three consecutive blocks of one breathing condition followed by three consecutive blocks of the other,
although the order was counterbalanced.
Counterbalancing protects against a simple systematic order effect at group level, but trial-wise randomisation or shorter alternating blocks would have provided stronger control for:
- fatigue;
- habituation;
- learning;
- scanner drift;
- sustained state changes carrying over between conditions.
This is particularly relevant when measuring autonomic physiology.
13. Some preregistered analyses were changed
The study was preregistered, which is a substantial strength.
However, the authors explicitly state:
“Some of the reported analyses deviate from the originally preregistered analysis plan”
with the deviations described in supplementary Table S1.
That does not itself imply a problem—the transparency is commendable—but it means the preregistration should not be treated as making every reported analysis fully confirmatory.
The distinction between preregistered and subsequently developed analyses is important, particularly for the imaging/mechanistic claims.
Overall assessment
I would rate the paper as interesting and methodologically sophisticated, with a credible behavioural effect, but a stronger causal narrative than the data strictly justify.
Its strongest result is not simply that slow breathing “relaxes” people. It is the much more interesting finding that altering breathing appears to selectively increase the weighting of prospective reward while leaving loss sensitivity relatively unchanged.
The central result can reasonably be expressed as:
slow 2:8 breathing→↑HRV+↑rewardweighting+↑gambleacceptance
with an associated relationship:
ΔHRV↔Δreward-related vmPFC/precuneus BOLD.
But the paper has not yet demonstrated the stronger mechanistic sequence:
prolonged exhalation→vagal activation→vmPFC reward encoding→risk behaviour
because respiratory rate, exhalation duration, CO₂/vascular physiology and vagal measures are not sufficiently separated, and the critical heart–brain relationship is correlational.
The experiment I would most like to see next
A particularly decisive follow-up would use at least three conditions:
normal breathing vs 5:5 slow breathing vs 2:8 slow breathing
while measuring end-tidal CO₂, continuous blood pressure/baroreflex activity, ECG and fMRI.
That would allow investigators to distinguish:
[
\text{slow respiration itself}
]
from
[
\text{prolonged exhalation specifically}.
]
If the 2:8 condition still produced greater reward sensitivity and vmPFC effects than rate-matched 5:5 breathing, the paper’s proposed exhalation-specific neurovisceral mechanism would become considerably more compelling.