Slow breathing modulates brain function and risk behavior

Highlights

•

Prolonged exhalation increases risky decisions by increasing reward sensitivity

•

Prolonged exhalation increases parasympathetic but not sympathetic activity

•

Greater parasympathetic activity predicts reward sensitivity in the vmPFC and precuneus

Summary

Successful decision-making requires that external information be interpreted in the context of the body’s state. Within the framework of body-brain interaction, deliberately modifying one’s autonomic state can shape how we evaluate the world, ultimately influencing choices. Yet, it remains unclear whether and how intentional autonomic regulation affects human decision-making. In this study, we tested instructed use of prolonged exhalation, a slow-breathing technique designed to boost parasympathetic activity during risky decision-making. Participants followed distinct breathing protocols while making risky choices, with neural and physiological activity measured using functional magnetic resonance imaging (fMRI) and multichannel monitoring. Prolonged exhalation increased risky choices by enhancing reward sensitivity and elevating cardiac parasympathetic activity. Importantly, individuals with greater parasympathetic upregulation also showed stronger reward-related responses in the ventromedial prefrontal cortex and precuneus. Our work reveals the transformative role of breathing-based interventions, demonstrating that autonomic regulation via breathing can shape value-based decision-making through neuro-cardiac pathways.

Graphical abstract


Keywords

1. decision-making
2. brain
3. breathing
4. parasympathetic
5. heart
6. vagal signaling
7. vmPFC
8. reward
9. HRV
10. risk taking

Introduction

Imagine you’re running late and rushing to the bank. Upon arrival, you’re immediately faced with an important investment decision. Here, your momentarily unrelated physiological arousal state, such as increased heart rate or rapid breathing, influences your choice, potentially leading to suboptimal decisions you might later regret. However, the same choice made calmly, under a more relaxed physiological state, may lead to a more optimal result. This raises the crucial question: can we regulate our own physiological responses to make better decisions?

Prolonged exhalation is a breathing technique, with a shorter inhalation to longer exhalation ratio (e.g., 2:8 s) that has been shown to effectively shift cardiac activity toward parasympathetic predominance.100339-9#bib1),200339-9#bib2),300339-9#bib3) This is driven by the tight synchronization of breathing with cardiac timing, such that during inhalation, the heart rate increases, whereas during exhalation, it decreases, resulting in periodic fluctuations in heart rate called heart rate variability (HRV).400339-9#bib4),500339-9#bib5) A prolonged exhalation phase enhances these fluctuations and thus leads to an overall increase in HRV, reflecting respiration-coupled cardiac vagal modulation.600339-9#bib6),700339-9#bib7),800339-9#bib8),900339-9#bib9)

This breathing technique becomes particularly crucial in the framework of intentional decision enhancement. A typical decision option consists of both rewards and losses under uncertainty, which guide our choices.1000339-9#bib10),1100339-9#bib11),1200339-9#bib12) Importantly, rewards and losses are interpreted differently, depending on the body’s current internal autonomic state1300339-9#bib13),1400339-9#bib14): states of cardiac parasympathetic predominance (often described as rest-and-digest) have been associated with greater sensitivity to potential rewards, whereas sympathetic predominance (fight-or-flight) has been linked to relatively greater sensitivity to potential losses. Supporting this, individual differences in cardiac parasympathetic activity, reflected by higher resting HRV, are associated with greater reward sensitivity. In the Balloon Analogue Risk Task, individuals with higher HRV (high-frequency HRV [HF-HRV]) are more inclined to take reward-related risk.1500339-9#bib15) Similarly, people with greater resting HF-HRV have a stronger preference for advantageous decks in the Iowa Gambling Task, indicative of increased sensitivity to long-term rewards.1600339-9#bib16) In contrast, enhanced sympathetic activity, as reflected in pupil dilation and skin conductance, has been associated with heightened sensitivity to loss and increased risk-avoidant behavior.1700339-9#bib17),1800339-9#bib18) However, it is unknown whether voluntary upregulation of cardiac parasympathetic activity via regulated breathing enhances momentary risky decisions or reward sensitivity.

At the neural level, the ventromedial prefrontal cortex (vmPFC) is a strong candidate for linking cardiac parasympathetic predominance and reward processing. The human vmPFC not only encodes the subjective value of choice options1200339-9#bib12),1900339-9#bib19) but is also embedded within the central autonomic network,2000339-9#bib20) receiving interoceptive input from the anterior insula2100339-9#bib21) and participating in the integration of interoceptive and affective states.2200339-9#bib22),2300339-9#bib23) In line with this, individual differences in HRV (indexed by the standard deviation of inter-beat intervals [SDNN], a time-domain measure) were shown to predict changes in vmPFC activity in a task requiring self-control over food reward.2400339-9#bib24) Despite this evidence pointing toward prolonged exhalation as a potential modulator of decisions, its feasibility and exact neurovisceral mechanism are unknown.

The current study aims to systematically investigate whether and how prolonged exhalation causally impacts decision-making. We hypothesized that prolonged exhalation (1) increases risky choices, (2) enhances cardiac parasympathetic activity, and (3) changes neural reward representation (pre-registered on Open Science Framework: https://osf.io/4cbfz).

We employed a within-subject experimental procedure, in which participants (_n_ = 41; 24 female, mean ± SD age = 24.78 ± 4.93 years) performed a risky choice task while breathing either according to prolonged exhalation or eupnea (natural respiration pattern) in a counterbalanced manner. Both breathing patterns were instructed visually by dynamically increasing and decreasing bars, during which participants made decisions1100339-9#bib11) (Figure 100339-9#fig1)). In each trial, participants were asked to either accept or reject a gamble. Each gamble had a 50% probability of yielding rewards or losses (Figure 100339-9#fig1)C). During the entire session, participants’ brain activity was measured by means of functional magnetic resonance imaging (fMRI). At the same time, we assessed physiological markers of parasympathetic and sympathetic activity (i.e., respiration, cardiac activity, skin conductance, and pupil responses).


Figure 1 Experimental procedure

(A) First, the individual natural breathing pattern was assessed for individualized eupnea breathing instruction.

(B) Participants practiced instructed breathing for both prolonged exhalation and eupnea breathing, presented by visual cues. Both breathing patterns were instructed, to keep the cognitive load consistent.

(C) Participants either accepted or rejected an option consisting of a certain amount of reward and loss, with a 50% probability of realization, while following the instructed breathing pattern, displayed as a progress bar on the right. Breathing conditions were presented in three-block sets, and the order of these sets was randomly counterbalanced across participants.

(D) Raw respiratory data of the two breathing conditions: eupnea (blue) and prolonged exhalation (orange) for one exemplary participant. Vertical gray solid lines depict the instructed inhalation onset. Vertical dashed lines depict the instructed exhalation onset. During eupnea, participants followed their natural breathing rhythm, while during prolonged exhalation, the ratio was fixed at 2:8 s.

(E) For each trial, the reward and loss values were systematically selected from a predefined reward-loss matrix. Rewards ranged from €10 to €30 (in steps of €2), and losses from €5 to €15 (in steps of €1), with 120 combinations used across both conditions. Red and blue shading indicate positive and negative EVs of a gamble (EV = 0.5 × reward − 0.5 × loss), respectively, and white indicates gambling with an EV close to zero.

Results

Validation of breathing manipulation

We first confirmed that participants followed the breathing instructions by testing whether the breathing conditions modulated exhalation duration and respiration rate on average using paired t-tests. Exhalation duration was significantly longer in the prolonged exhalation condition compared with eupnea (_t_(40) = 34.621, _p_< 0.001, Cohen’s _d_ = 5.407; mean difference = 5.183, SE = 0.150, 95% confidence interval [CI] [4.880, 5.485]; Figure 200339-9#fig2)A). In line with this, the respiration rate was significantly lower in the prolonged exhalation condition compared with eupnea (_t_(40) = −11.616, _p_< 0.001, Cohen’s _d_ = −1.814; mean difference = −6.104, SE = 0.525, 95% CI [−7.166, −5.042]; Figure 200339-9#fig2)B). These results confirm that the prolonged exhalation manipulation successfully slowed respiratory rhythm by extending the duration of exhalation.


Figure 2 Effects of prolonged exhalation on physiological markers of parasympathetic and sympathetic activity

Prolonged exhalation (A) significantly increased exhalation duration and (B) significantly decreased respiration rate, compared with eupnea. In addition, cardiac parasympathetic activity was enhanced as (C) respiratory heart rate variability (RespHRV) and (D) root mean square of successive differences (RMSSDs) were significantly higher during prolonged exhalation compared with eupnea. In contrast, sympathetic readouts, such as (E) tonic skin conductance level and (F) change in pupil size from baseline, showed no significant difference between conditions, indicating prolonged exhalation selectively modulates cardiac parasympathetic activity. Pupil values are baseline-corrected and reported in arbitrary units (a.u.). Points represent participant-level means, densities depict the distribution across participants, and boxplots summarize the median and interquartile range. ∗_p_< 0.05 and ∗∗∗_p_< 0.001; n.s., not significant.

Prolonged exhalation enhances cardiac parasympathetic activity

We next tested whether prolonged exhalation enhances parasympathetic activity, using two well-established cardiac indicators2500339-9#bib25),2600339-9#bib26): respiratory HRV (RespHRV),2700339-9#bib27) also known as respiratory sinus arrhythmia (RSA), and root mean square of successive differences (RMSSDs). RespHRV is a direct index of cardiorespiratory coupling,800339-9#bib8),2800339-9#bib28) whereas RMSSD is a time-domain index of cardiac parasympathetic modulation, which quantifies rapid beat-to-beat variability in heart rate. It is less sensitive to task-related breathing variability and is optimal for stable cardiac estimation in experimental contexts.2600339-9#bib26),2900339-9#bib29)

Prolonged exhalation significantly enhanced cardiac parasympathetic activity, reflected by higher RespHRV (_t_(34) = 8.186, _p_< 0.001, Cohen’s _d_ = 1.384; mean difference = 78.000, SE = 9.529, 95% CI [58.636, 97.364]; Figure 200339-9#fig2)C) and RMSSD (_t_(34) = 2.621, _p_ = 0.013, Cohen’s _d_ = 0.443; mean difference = 7.012, SE = 2.675, 95% CI [1.575, 12.449]; Figure 200339-9#fig2)D) compared with eupnea.

We also examined sympathetic indices, namely tonic skin conductance level and pupil size changes. Interestingly, neither tonic skin conductance level (_t_(25) = 0.762, _p_ = 0.453, Cohen’s _d_ = 0.149; mean difference = 0.273, SE = 0.358, 95% CI [−0.465, 1.011]; Figure 200339-9#fig2)E) nor pupil size (_t_(27) = −0.655, _p_ = 0.518, Cohen’s _d_ = −0.124; mean difference = −53.412, SE = 81.580, 95% CI [−220.799, 113.976]; Figure 200339-9#fig2)F) differed significantly between conditions. These results suggest that prolonged exhalation selectively increased cardiac parasympathetic activity without inducing a general shift in sympathetic arousal.

Prolonged exhalation increases risky choices and selectively enhances reward sensitivity

We next examined whether prolonged exhalation modulates risky choices using a generalized linear mixed-effects model (GLMM) with a logit link function to predict trial-wise binary decisions (accept = 1 versus reject = 0) from breathing condition (prolonged exhalation = 1, eupnea = −1), reward magnitude, loss magnitude, and their interactions with breathing condition. The model included subject-level random intercepts and random slopes for breathing condition, reward, and loss.

Prolonged exhalation significantly increased trial-wise risky choices, as reflected by a main effect of breathing condition on choice behavior (_β_ = 0.168, SE = 0.071, _z_ = 2.37, _p_ = 0.018, 95% CI [0.029, 0.306]; Figure 300339-9#fig3)A). This increase in risky choices was primarily associated with enhanced reward sensitivity: under prolonged exhalation, reward magnitude exerted a stronger influence on decisions compared with eupnea, as indicated by a significant breathing condition × reward interaction (_β_ = 0.176, SE = 0.053, _z_ = 3.32, _p_< 0.001, 95% CI [0.072, 0.279]; Figure 300339-9#fig3)A). Post hoc comparisons confirmed that the reward slope was significantly steeper under prolonged exhalation than under eupnea (_β_ = 0.351, SE = 0.106, _z_ = 3.32, _p_< 0.001; Figure 300339-9#fig3)B), consistent with enhanced reward sensitivity during prolonged exhalation.


Figure 3 Prolonged exhalation increases the impact of reward on decisions

(A) Under prolonged exhalation, participants generally made risky decisions significantly more often and the impact of reward on decisions increased, as evidenced by a significant condition × reward interaction. Points depict odds ratios (OR) with 95% CIs. The dashed line indicates no effect (OR = 1).

(B) Impact of reward on decision-making was significantly greater under prolonged exhalation compared with eupnea, as model-estimated reward slopes were significantly higher under prolonged exhalation than during eupnea. Error bars represent ±1 SE of model-estimated slopes.

(C) Response times (RTs) did not differ significantly between breathing conditions. Points represent participant-level mean RTs, densities depict the distribution across participants, and boxplots summarize the median and interquartile range.

∗_p_< 0.05 and ∗∗∗_p_< 0.001; n.s., not significant.

The breathing condition × loss interaction did not reach statistical significance (_β_ = −0.079, SE = 0.047, _z_ = −1.67, _p_ = 0.095, 95% CI [−0.171, 0.014]; Figures 300339-9#fig3)A and 3B). To directly test whether prolonged exhalation modulated reward versus loss sensitivity differently, we contrasted the condition × reward and condition × loss interaction terms using a likelihood-ratio test. This comparison revealed a significant asymmetry (_χ_ ²(1) = 7.716, _p_ = 0.005), indicating that prolonged exhalation more strongly modulated sensitivity to reward than to loss. We confirmed this finding within a more integrative modeling framework: an additional expected value (EV)-based GLMM showed that prolonged exhalation increased expected-value sensitivity (condition × EV: _β_ = 0.162, SE = 0.053, _z_ = 3.06, _p_ = 0.002, 95% CI [0.058, 0.267]), yielding steeper psychometric curves under prolonged exhalation (Figure S100339-9#mmc1)). Descriptive distributions of the four response categories were comparable across conditions (Figure S200339-9#mmc1)).

We next examined whether the breathing effect on choice was also reflected in the computational parameter within the prospect theory framework. No consistent differences in parameters were observed between breathing conditions (Figure S300339-9#mmc1)), and a non-parametric area under the acceptance threshold curve (AUC) measure of acceptance thresholds likewise did not differ between conditions (_t_(40) = 0.03, _p_ = 0.974, 95% CI [−5.248, 5.419]). To evaluate the reliability of parameter estimates, we conducted a parameter-recovery analysis in which all parameters were jointly sampled from their prior distributions and re-estimated (Table S300339-9#mmc1)). However, recovery of the inverse temperature parameter was limited (eupnea: _r_ = 0.38; prolonged exhalation: _r_ = 0.37), and dependencies among jointly estimated parameters limit the interpretability of condition-specific effects at the parameter level. Accordingly, prospect theory parameters are treated as descriptive, and our main inferences are based on the trial-wise GLMM analyses.

We also examined whether the observed breathing effect on choice could be accounted for by differences in response time (RT). RT did not differ significantly between breathing conditions (_β_ = 0.062, SE = 0.036, _t_(39.96) = 1.72, _p_ = 0.093, 95% CI [−0.011, 0.136]; Figure 300339-9#fig3)C). Neither the breathing condition × reward interaction (_β_ = −0.005, SE = 0.008, _t_(9,362) = −0.54, _p_ = 0.587, 95% CI [−0.021, 0.012]) nor the breathing condition × loss interaction (_β_ = 0.008, SE = 0.008, _t_(9,366) = 0.99, _p_ = 0.323, 95% CI [−0.008, 0.025]) significantly predicted RT. Importantly, including trial-wise RT as a covariate in the choice GLMM did not attenuate the breathing condition × reward effect (_β_ = 0.180, SE = 0.054, _z_ = 3.35, _p_< 0.001, 95% CI [0.074, 0.285]), indicating that the breathing effect on choice was not driven by RT differences.

We further checked whether the main finding could be accounted for by condition differences in overall response randomness or attentional lapses. Choice entropy across central EV bins did not significantly differ between breathing conditions (_t_(40) = −1.67, _p_ = 0.104, 95% CI [−0.073, 0.007]), and lapse-augmented logistic modeling revealed no significant condition difference in lapse rate (_t_(39) = 0.28, _p_ = 0.780, 95% CI [−0.020, 0.026]), suggesting that the breathing effect on reward sensitivity was not attributable to general changes in decision consistency.

We additionally examined whether individual differences in breathing-induced sympathetic changes were associated with breathing-induced changes in risky acceptance. Neither changes in skin conductance level (_r_ = −0.16, _p_ = 0.438, 95% CI [−0.514, 0.244]) nor pupil size (_r_ = −0.094, _p_ = 0.635, 95% CI [−0.451, 0.289]) were associated with changes in risky acceptance, providing no evidence that the behavioral effect covaries with these peripheral autonomic indices. Furthermore, loss sensitivity was not associated with tonic skin conductance (_β_ = −0.008, SE = 0.499, _t_(45) = −0.016, _p_ = 0.988, 95% CI [−1.013, 0.997]), and this relationship did not differ by breathing condition (loss sensitivity × condition: _β_ = −0.023, SE = 0.498, _t_(45) = −0.046, _p_ = 0.963, 95% CI [−1.026, 0.979]), providing no evidence that tonic skin conductance relates to loss sensitivity differently across breathing conditions.

Respiration-driven cardiac parasympathetic shifts amplify neural reward encoding

We next investigated whether and how this cardiac parasympathetic shift under prolonged exhalation reshapes the neural representation of reward during decision-making. To do so, we specified a general linear model (GLM) that predicted blood-oxygen-level-dependent (BOLD) response with parametric modulation of trial-wise reward and loss magnitudes of decision options. Then, at the group level, individual first-level contrast images reflecting reward-related brain activation were entered into a whole-brain regression analysis, with individual differences in breathing-condition-dependent HRV (prolonged exhalation versus eupnea) as a continuous predictor. We found that individuals showing larger cardiac shifts (ΔRMSSD) under prolonged exhalation also exhibited stronger reward-related activity in the vmPFC (Figures 400339-9#fig4)A and 4C; Montreal Neurological Institute (MNI) coordinates: x = −6, y = 39, z = −18; _t_(33) = 4.79, cluster-level family-wise error [FWE]-corrected _p_< 0.05) and the precuneus (Figure 400339-9#fig4)B and 4D; MNI coordinates: x = 0, y = −51, z = 60; _t_(33) = 4.23, cluster-level FWE-corrected _p_< 0.05) during choices. No other regions reached whole-brain cluster-level FWE-corrected significance. Individual differences in interoceptive self-regulation further predicted breathing-related modulation of precuneus activity, whereas no such relationship was observed in the vmPFC (Figure S400339-9#mmc1)).


Figure 4 Parasympathetic modulation of reward-related neural activity under prolonged exhalation

Individuals showing larger cardiac parasympathetic shifts under prolonged exhalation also exhibited stronger reward-related activity in the (A) vmPFC and (B) precuneus. Our whole-brain regression analysis revealed a significantly positive association between changes in cardiac parasympathetic activity (ΔRMSSD = RMSSD [prolonged exhalation − eupnea]) and reward-related activation in these regions. Scatterplots (C and D) show individual relationships between ΔRMSSD and reward-related parameter estimates extracted from the vmPFC and precuneus, respectively. Color bars in (A) and (B) indicate _t_ values.

Discussion

Here, we provide evidence for a neurovisceral pathway through which prolonged exhalation is associated with systematic changes in autonomic state, value-related brain activity, and risky choice. To do so, we applied a multimodal approach, integrating instructed breathing with simultaneous multichannel physiological recordings and functional neuroimaging during decision-making. We demonstrate that prolonged exhalation significantly increases cardiac parasympathetic markers and selectively enhances the impact of reward on decision-making without altering sensitivity to losses. We also observed a significant main effect of prolonged exhalation, which increased the proportion of risky decisions. This increase in risky choice does not reflect a blanket increase in risk tolerance or a shift toward EV-optimal decision-making but is better characterized as a selective up-weighting of reward information without a corresponding change in loss sensitivity. Strikingly, fMRI analyses revealed that, at the group level, individual differences in breathing-induced cardiac parasympathetic enhancement were significantly associated with reward-related BOLD activation in the vmPFC and precuneus. Our multimodal approach provides integrated insights into the neurophysiological and behavioral framework for understanding how bodily changes can modulate the brain’s reward processing and, in turn, shape risky choice behavior.

Whereas previous research focused on individual differences in autonomic nervous system (ANS) variability,1500339-9#bib15),1600339-9#bib16) we demonstrate how changes in the breathing pattern acutely impact parasympathetic activity and thereby neural reward sensitivity during decision-making. Our results not only extend the existing literature on body-brain interaction100339-9#bib1),2400339-9#bib24) but also provide a framework for exploring how breathing can be used to voluntarily regulate autonomic states in relation to decision-making processes. Our investigation uniquely contributes to the current decision neuroscience literature, which mainly focuses on the external situational changes leading to decision modulation. By contrast, these data suggest that external stimuli, such as reward, are interpreted in the light of momentary internal bodily state, in this case cardiac vagal tone, which is a crucial factor that needs to be considered.

In our study, we employed a prolonged exhalation breathing technique. Most existing slow-breathing protocols focus primarily on maintaining a constant slow respiratory rate (e.g., ∼6 breaths/min) rather than explicitly emphasizing extended exhalation phases. In contrast, prolonged exhalation emphasizes a longer exhalation-to-inhalation ratio, which is physiologically meaningful because baroreflex-mediated cardiac vagal activation occurs primarily during exhalation when temporary increases in arterial pressure trigger baroreceptor activity.3000339-9#bib30),3100339-9#bib31) Importantly, compared with other slow-breathing techniques, prolonged exhalation is particularly easy to learn because there is no need to fully standardize the breathing rate, allowing participants to breathe at their own pace. This contributes to restoring cognitive function while benefiting from the parasympathetic effects of prolonged exhalation. Consistent with this, Röttger and colleagues have shown that implementation of prolonged exhalation does not trade off cognitive performance.300339-9#bib3) In direct comparison, other breathing techniques (i.e., tactical breathing) were associated with poorer task performance. In line with this, our prolonged exhalation breathing protocol heightens parasympathetic indices without a downregulation of sympathetic markers (e.g., skin conductance and pupil diameter), suggesting a selective or partial modulation of branches of the ANS,800339-9#bib8),2900339-9#bib29) which potentially also contributes to the preserved cognitive function and decision-making. Notably, breathing-induced changes in skin conductance level and pupil diameter were not associated with the breathing-induced change in risky choice behavior, suggesting that the behavioral shift under prolonged exhalation is not readily attributable to a generalized change in sympathetic arousal. This aspect is particularly important in the framework of decision-making, in which the primary goal is to make effective choices and not simply to promote relaxation.

Contrasting the traditional decision-theoretic view, we show that the evaluation of potential rewards and losses depends on bodily arousal states.3200339-9#bib32),3300339-9#bib33) Specifically, heightened loss sensitivity and more cautious decision-making have been associated with increased sympathetic activation, as reflected in physiological markers such as heart rate and skin conductance during stress or threat exposure.3300339-9#bib33),3400339-9#bib34) Conversely, cardiac parasympathetic states, such as higher baseline HRV or enhanced cardiac vagal activity, have been shown to be associated with more balanced reward-loss evaluation, better emotional regulation, and stronger cognitive control.3500339-9#bib35),3600339-9#bib36),3700339-9#bib37) Empirical findings also highlight that cardiac parasympathetic predominance can foster adaptive decision processes: individuals with elevated HRV show reduced susceptibility to the framing effect.3800339-9#bib38) Our data build upon these insights and provide a neurovisceral account by demonstrating that actively increasing cardiac vagal tone via prolonged exhalation not only promotes a calmer physiological state but also enhances the weight assigned to potential rewards during choice. Importantly, this shift is consistent with prior proposals that autonomic regulation may support more differentiated reward evaluation.1500339-9#bib15),1600339-9#bib16)

At the neural level, individual differences in cardiac parasympathetic enhancement (indexed by ΔRMSSD) predicted reward-related BOLD activation in two key cortical regions: the vmPFC and the precuneus. The vmPFC is well-established as a core hub for subjective-value representation and motivational integration.1900339-9#bib19),3900339-9#bib39) By contrast, the precuneus is implicated in self-referential thought4000339-9#bib40) and acts as a bridge between internal bodily signals and mental simulation processes.4100339-9#bib41) Its integrative role suggests a potential link between physiological states and self-relevant evaluation, consistent with evidence that individual differences in precuneus connectivity with vmPFC or dorsolateral prefrontal cortex (dlPFC) are associated with how people value immediate versus delayed rewards.4200339-9#bib42) Together, these activations suggest a targeted neural mechanism in which prolonged exhalation-driven parasympathetic upregulation aligns with enhanced reward valuation in the vmPFC. The precuneus may integrate bodily signals into higher-order cognitive processes, such as mental simulation and self-relevant evaluation, which shape value-based decision-making. Our exploratory analysis showed that questionnaire-based interoceptive self-regulation scores were associated with task-based precuneus activity specifically, but not with vmPFC activation, further supporting its role in bodily self-integration. Interestingly, our analyses did not reveal correlations between parasympathetic enhancement and the insula or anterior cingulate cortex (ACC), regions commonly associated with interoception.1300339-9#bib13),4300339-9#bib43) However, unlike these studies showing a general link between interoception and activation in the insula and ACC, our analysis specifically focuses on the impact of a respiration-modulated increase in cardiac parasympathetic activity on risky decisions rather than interoception in general.

Although prior work has implicated the striatum/midbrain and insula/amygdala in reward and loss processing,1100339-9#bib11) respectively, we did not observe robust breathing-related modulation in these regions. Instead, breathing-related effects were confined to cortical regions involved in subjective-value representation, most prominently the vmPFC.1200339-9#bib12) We believe this pattern reflects two features of the present design. First, our analyses focused on value computation during the option phase and did not include an outcome phase that typically elicits strong prediction-error or loss-evoked signals in subcortical circuits associated with learning-based value updating.1200339-9#bib12) Second, prolonged exhalation primarily induces a parasympathetic state shift, which may preferentially bias cortical value integration processes rather than engage classical subcortical or aversive systems. Consistent with this interpretation, prolonged exhalation did not alter behavioral loss sensitivity, aligning with the absence of breathing-related effects in loss-related regions such as the insula or amygdala.1100339-9#bib11)

These findings suggest that the observed cortical pattern does not reflect a generalized effect on learning or affective processing. Instead, it can be interpreted within a broader framework of brain-autonomic integration. In particular, the neurovisceral integration model2200339-9#bib22) proposes that prefrontal regions provide a key interface through which autonomic state can influence higher-order cognitive evaluations. Consistent with prior work linking higher cardiac vagal tone to enhanced prefrontal involvement during value-based decision-making,1300339-9#bib13),2400339-9#bib24) respiration-driven parasympathetic modulation may bias value-related processing by modulating prefrontal valuation processes rather than by directly engaging subcortical learning or affective circuits.

Our task was optimized to sensitively detect intervention-dependent differences in decision-making; however, this optimization necessitated selecting reward-loss ranges based on the well-established loss-aversion bias1000339-9#bib10),1100339-9#bib11) rather than using symmetric ones. We therefore view the present study as a proof-of-principle demonstration that prolonged exhalation can modulate choice behavior. Future work employing more balanced reward-loss structures will be needed to determine whether the observed increase in reward sensitivity reflects a universal effect of the intervention or is specific to the current task design. Our present conclusions are therefore drawn within this framework.

The selective impact of prolonged exhalation breathing on reward responsiveness has important implications for clinical contexts, such as anxiety, panic disorder, and depression, given their distinct autonomic signatures and maladaptive reward processing.4400339-9#bib44),4500339-9#bib45),4600339-9#bib46) By enhancing cardiac parasympathetic modulation through prolonged exhalation techniques, individuals may restore reward processing, a valuable pathway for emotional recalibration.4400339-9#bib44),4700339-9#bib47) Prolonged exhalation harbors the potential for a low-cost, low-risk, easily applicable intervention to be incorporated into therapy or rehabilitation programs, especially to support pharmacological treatments.

Although the present study was not conducted in real-world settings, controlled slow-breathing protocols are already in use for arousal regulation in military and law-enforcement training, from which prolonged exhalation is shown to enhance cognitive task performance in soldiers.300339-9#bib3) Our results motivate future investigations in more ecologically valid settings where decisions are made under pressure and uncertainty. Specifically, we speculate that sustaining parasympathetic engagement under acute stress could support value-based decision processes in high-stakes professional contexts, such as emergency response, elite sports, and aviation. Prolonged exhalation has been shown to reinforce emotional resilience and cognitive performance300339-9#bib3),4800339-9#bib48) while enhancing parasympathetic activation. Given the feasibility and simplicity of prolonged exhalation, it may be particularly well-suited for on-the-spot or field applications where real-time arousal regulation is essential. Scientifically, our results have crucial implications for neuroscience, economics, psychology, and psychiatry, fields in which physiological data, such as heart rate and respiration, are often considered as “noise” and neglected. Our findings add to a growing literature on body-brain interactions and interactions between the ANS and decision-making and highlight the value of incorporating physiological measures into decision neuroscience.

Taken together, our findings indicate that prolonged exhalation selectively enhances parasympathetic activity and heightens neural reward representation, thereby increasing reward sensitivity and biasing choice toward accepting gambles. This pattern underscores the essential function of body state in shaping human decisions, an empirical validation for neurovisceral integration.2200339-9#bib22),3500339-9#bib35) Our findings refine current theories of body-brain interaction by revealing effective and easy-to-implement strategies regulating physiological functions, such as breathing and nutrition,4900339-9#bib49),5000339-9#bib50) to intentionally impact the body and brain to change decisions. Our study suggests potential applications for breathing-based interventions, though future work is needed to test their efficacy across diverse populations and real-world situations.

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Soyoung Q. Park ([soyoung.q.park@gmail.com](mailto:soyoung.q.park@gmail.com)).

Materials availability

This study did not generate any new, unique reagents.

Data and code availability

•

The data reported in this paper will be shared by the lead contact00339-9#sec-4-1) upon request.

•

All original code used for the statistical analyses and visualizations reported in this study has been deposited via GitHub and is publicly available at Zenodo: https://doi.org/10.5281/zenodo.19454547.

•

Any additional information required to reanalyze the data reported in this paper is available from the lead contact00339-9#sec-4-1) upon request.

Acknowledgments

This study was funded by the German Federal Ministry of Education and Research (Bundesministerium für Bildung und Forschung [BMBF]; grant 01EE2301E to S.Q.P.) as part of the concept development of the German Center for Mental Health (Deutsches Zentrum für Psychische Gesundheit [DZPG]); BMBF grant 01GP2210C to S.Q.P. (ELSA DecEnt); the German Center for Diabetes Research (Deutsches Zentrum für Diabetesforschung [DZD]; grant 82DZD03D03) and the State of Brandenburg. I.R. was supported by the Marie Skłodowska-Curie Action (MSCA) BRAINSTOM (grant agreement no. 101028203).

Author contributions

S.Q.P., S.R., and W.H. conceptualized the study; W.H., M.S., and L.Y.L. acquired data; S.Q.P., W.H., G.B., M.S., I.R., M.P., F.M., and B.K. analyzed the data; and S.Q.P., W.H., S.R., G.B., M.S., I.R., M.P., F.M., B.K., and P.N.C.M. wrote, reviewed, and edited the manuscript.

Declaration of interests

The authors declare no competing interests.

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

During the preparation of this work, the authors used ChatGPT-4o in order to improve clarity, refine language, and check grammar. After using this tool or service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

STAR★Methods

Key resources table

| REAGENT or RESOURCE | SOURCE | IDENTIFIER | | --- | --- | --- | | Software and algorithms | | MATLAB (versions 2020a and 2023a) | MathWorks | https://www.mathworks.com/products/matlab.html | | Psychtoolbox3 | Brainard5100339-9#bib51) | http://psychtoolbox.org | | R v4.3.0 | R Core Team, 20235200339-9#bib75) | https://www.r-project.org/ | | lme4 1.1.37 | Bates et al.5300339-9#bib52) | https://CRAN.R-project.org/package=lme4 | | emmeans 2.0.1 | Lenth5400339-9#bib53) | https://CRAN.R-project.org/package=emmeans | | lmerTest 3.2.0 | Kuznetsova et al.5500339-9#bib54) | https://CRAN.R-project.org/package=lmerTest | | RStan 2.32.6 | Stan Development Team5600339-9#bib55) | https://mc-stan.org | | FieldTrip toolbox 20240110 | Oostenveld et al.5700339-9#bib56) | https://www.fieldtriptoolbox.org/ | | BreathMetrics toolbox | Noto et al.5800339-9#bib57) | https://github.com/zelanolab/breathmetrics | | EEGLAB v2024.0 | Delorme and Makeig5900339-9#bib58); Niazy et al.6000339-9#bib59) | https://sccn.ucsd.edu/eeglab/ | | Edf2Mat | Adrian Etter and Marc Biedermann at the University of Zurich | https://github.com/uzh/edf-converter | | Ledalab | Benedek and Kaernbach6100339-9#bib60) | http://www.ledalab.de/ | | SPM 12 for fMRI analyses | Functional Imaging Laboratory, London, UK | https://www.fil.ion.ucl.ac.uk/spm/ | | PhysIO toolbox | Kasper et al.6200339-9#bib61); Frässle et al.6300339-9#bib62) | https://www.translationalneuromodeling.org/tapas | | Custom code | This paper | https://doi.org/10.5281/zenodo.19454547 |

• Open table in a new tab

Experimental model and study participant details

Forty-nine healthy adults were recruited from the local university community via flyers and online advertisements. All participants provided written informed consent in accordance with the Declaration of Helsinki, and the study protocol was approved by the Ethics Committee of the University of Potsdam.

Participants were required to be between 18 and 40 years old, fluent in German, right-handed, and to have a body mass index (BMI) between 18 and 25 kg/m². To ensure consistent autonomic and respiratory baselines, individuals were excluded if they reported a history of psychiatric, neurological, cardiovascular, pulmonary, or metabolic disorders; regular medication use; current infection or excessive stress; smoking; or engagement in extreme athletic training. Additional exclusions included abnormal baseline breathing, irregular sleep-wake cycles (e.g., night-shift work). MRI eligibility was assessed using standard institutional safety screening.

Eight participants were excluded from all analyses for the following reasons: one participant completed the task using an incorrect breathing pacing bar due to an incorrect assignment of respiratory settings, one was found to have incidental brain abnormalities on structural MRI scanning, one reported visual impairment that prevented accurate perception of task stimuli, one repeatedly fell asleep during the task, and four failed to follow the instructed breathing rhythm during scanning. The final behavioral sample included 41 participants (n = 41; 24 female, mean ± SD age = 24.78 ± 4.93 years).

Due to modality-specific signal quality criteria, the number of participants included in each analysis varied. For HRV, electrocardiogram (ECG) was used when available. For participants without usable ECG data, high-quality Photoplethysmography (PPG) recordings were used to derive inter-beat intervals and systolic peaks for RMSSD, RespHRV, and cardiac phase estimation used in fMRI preprocessing. The final HRV sample included 35 participants: 19 with ECG and 16 with PPG. fMRI analyses were limited to these 35 participants, who met both imaging quality and cardiac signal criteria.

Method details

#### Preregistration and deviations

This study was preregistered on the Open Science Framework: https://osf.io/4cbfz. Some of the reported analyses deviate from the originally preregistered analysis plan, and the deviations can be seen in Table S100339-9#mmc1).

#### Experimental design

##### Breathing manipulation procedure

We used two breathing conditions to manipulate participants' breathing rhythms: eupnea and prolonged exhalation. Both techniques involved inhaling through the nose and exhaling through pursed lips, with the abdomen rising during inspiration and lowering during expiration. To pace participants' breathing rhythm, a bar was presented on a screen. The green bar filled when participants were asked to inhale and emptied when they were supposed to exhale. The prolonged exhalation and eupnea conditions differed in their duration and ratio of inspiration and expiration. In the prolonged exhalation condition, participants were instructed to inhale and exhale at a fixed ratio of 2:8 seconds, as previously recommended.6400339-9#bib63)

Before the experiment, the individual's baseline breathing pattern was measured, which was then used for the visual instruction of the eupnea condition. Participants were instructed to breathe normally and calmly (Figure 100339-9#fig1)A). Specifically, mean inhalation and exhalation durations were computed from 20 valid respiratory cycles following an initial 5-cycle adaptation period, yielding individualized inhale-to-exhale ratios for the eupnea condition. After this calibration, participants practiced both breathing techniques, eupnea and prolonged exhalation, alternating every 3 minutes to learn to synchronize their breathing with the visual cues used during the task.

##### Decision-making task

Participants performed a decision-making task, namely a risky choice task based on a previously established paradigm.1100339-9#bib11) Each trial began with a fixation cross (1–10 s), followed by a visual display showing a pair of potential monetary rewards and losses (Figure 100339-9#fig1)C). Participants had up to 3 seconds to decide whether to accept or reject the presented gamble. Responses were made using a four-point scale: _strongly accept, weakly accept, weakly reject_, or _strongly reject_.

Task presentation and response collection were implemented using Psychtoolbox5100339-9#bib51) in MATLAB. During each decision, participants followed a designated breathing rhythm, guided by a visual progress bar on the right side of the screen. They were informed that the probability of winning or losing each gamble was 50%, as in a fair coin toss.1100339-9#bib11) Reward amounts ranged from €10 to €30, and loss amounts from €5 to €15. No feedback was provided after each decision, to minimize external interference and maintain consistent task engagement.

Each breathing condition consisted of 120 trials, evenly distributed across three blocks per condition. To ensure incentive-compatible decisions, three trials were randomly selected at the end of the experiment, and participants received or lost 10% of the respective gamble outcomes. Participants could not lose more money than they could gain. Each block lasted approximately 5 minutes. The three blocks of each breathing condition were presented in a continuous sequence, and the order of conditions was counterbalanced across participants (three blocks of prolonged exhalation followed by three of eupnea, or vice versa).

Quantification and statistical analysis

#### Behavioral data analysis

##### Behavioral modeling

To examine the effect of breathing condition on decision-making, we modeled the probability of acceptance as a function of breathing condition, reward magnitude, and loss magnitude. We dichotomized responses into a binary variable: accept (strongly or weakly accept) versus reject (strongly or weakly reject).

The decision variable was analyzed using a generalized linear mixed-effects model (GLMM) with a binomial family and logit link, implemented in the lme4 package5300339-9#bib52) in R (v4.3.0; R Core Team5200339-9#bib75)). The model included fixed effects for breathing conditions (prolonged exhalation = 1, eupnea = −1), reward magnitude, loss magnitude, and their interactions with conditions. Participant-level random intercepts and random slopes for reward, loss, and condition were included to account for within-subject variability. Age, sex, and condition order were entered as covariates.

To further evaluate condition-related differences in reward and loss sensitivity, we performed post hoc comparisons using estimated marginal trends from the fitted mixed-effects model. Specifically, we used the emmeans package5400339-9#bib53) in R to extract estimated slopes of reward and loss separately for each condition and contrasted these using pairwise comparisons. These estimates reflect average slopes adjusted for covariates (sex, age, and condition order) and account for the random subject-level structure of the model.

##### Random-effects structure model comparison

To determine the optimal random-effects structure for our main choice GLMM, we conducted a systematic model comparison starting from the fully maximal structure and progressively simplifying to identify the most complex structure supported by the data. Models ranged from random intercepts only (M1) to full random slopes for all effects and interactions (M8). Model selection was guided by: (1) convergence and singularity status, (2) information criteria (AIC and BIC), and (3) likelihood-ratio tests (Table S200339-9#mmc1)).

#### Reward vs. loss sensitivity comparison

To directly compare the effects of breathing condition on reward versus loss sensitivity, we conducted a likelihood-ratio test (LRT) comparing a full model against a constrained model in which the condition × reward and condition × loss interaction coefficients were forced to be equal. In our main GLMM, reward and loss were z-standardized prior to model fitting, ensuring both predictors are on the same scale and directly comparable.

#### Response time (RT) analysis

To assess RT effects, we conducted three analyses. First, we fitted a linear mixed-effects model (LMM) predicting log-transformed, z-scored RT from condition, reward, loss, and their interactions, using the same random-effects structure as our main choice model (M6). LMMs were fitted using lmer, with p-values computed using the lmerTest package.5500339-9#bib54)

The random-effects structure was matched to the main choice GLMM to ensure comparability of fixed-effect estimates and to avoid overinterpretation of RT differences driven by under-specified random-effects structures. Second, we examined whether condition × reward and condition × loss interactions were significant, which would indicate that prolonged exhalation differentially affects processing time for different stimulus values. Third, we added trial-wise RT (log-transformed and z-scored) as a covariate to our choice GLMM to test whether controlling for RT attenuates the condition × reward effect. Random slopes in the RT-control model were specified without correlation parameters to reduce the number of variance–covariance parameters.

#### Choice entropy analysis

To assess whether prolonged exhalation affects decision noise, we computed Shannon entropy as a model-free measure of choice consistency. Trials were binned into five EV levels (very low, low, mid, high, very high) based on z-scored expected value, using breakpoints at EV = −1.5, −0.5, 0.5, and 1.5. For each participant and condition, we calculated the probability of acceptance within each bin and computed binary entropy:

No alternative text available

Lower entropy indicates more consistent (less noisy) choices. We averaged entropy across central bins (low, mid, high), focusing on the region where choices are most variable. Extreme EV bins were excluded because floor and ceiling effects yield near-zero entropy irrespective of decision noise, limiting interpretability rather than reflecting reduced noise.

#### Lapse rate analysis

To estimate random responding due to inattention, we fitted lapse-augmented logistic models6500339-9#bib64) to each participant's data separately for each condition:

No alternative text available

where y i ∈ {0,1} denotes the binary choice on trial i (1 = accept, 0 = reject), and EV i denotes the expected value of the gamble. ε represents the lapse parameter, capturing the probability of random responding. When ε = 0, the model reduces to standard logistic regression; when ε > 0, a proportion of responses are assumed to be random (50% accept, 50% reject) regardless of stimulus value. Parameters were estimated using maximum likelihood, with ε constrained to [0, 0.5]. One participant's fit did not converge in the eupnea condition and was excluded from the lapse rate analysis.

#### Expected value (EV)-based GLMM

To characterize breathing effects across the full range of expected values, we fitted a GLMM predicting choice from condition, EV (z-scored), and their interaction.

This analysis allows visualization of psychometric curves showing the probability of accepting a gamble as a function of expected value for each breathing condition (Figure S100339-9#mmc1)).

#### Prospect theory parameter estimation and condition effects

To complement the mixed-effects modeling approach, we used Bayesian hierarchical modeling via Stan5600339-9#bib55),6600339-9#bib65) to estimate individual-level prospect theory parameters. These are latent cognitive variables not directly observable on a trial-by-trial basis, and Bayesian methods offer advantages in regularization and uncertainty quantification.

##### Model structure

Binary decisions (accept = 1, reject = 0) were modeled based on the subjective utility of each gamble. For each trial _n_ and participant _j_ under breathing condition _c_, utility was defined as:

No alternative text available

where _x_(_n_) and _y_(_n_) are the magnitudes of potential reward and loss on trial _n_, respectively. The exponent _α_ _j,c_ captures nonlinear valuation of outcomes (risk sensitivity), while _λ_ _j,c_ reflects loss aversion (values > 1 indicating loss aversion). The coefficients 0.5 reflect equal outcome probability. Because outcome probabilities are constant across trials, this scaling factor is absorbed into the inverse temperature parameter during model estimation. The probability of accepting the gamble was modeled via a softmax function:

No alternative text available

where _β_ _j,c_ is the inverse temperature parameter, reflecting choice consistency.

##### Hierarchical parameter estimation

Each parameter (loss aversion parameter (λ), risk sensitivity (α), inverse temperature (β)) was estimated at the individual level for each breathing condition. Condition-specific individual parameters were modeled as drawn from group-level normal distributions:

No alternative text available

No alternative text available

No alternative text available

Group-level means were assigned weakly informative priors:

No alternative text available

No alternative text available

No alternative text available

Standard deviations were modeled with half-Cauchy distributions (scale = 0.5).

##### Model fitting and inference

The model was estimated using four MCMC chains (2,000 iterations each, including 500 warm-up samples). All parameters converged (_R^_< 1.02). Posterior distributions were extracted for all individual and group level parameters under each condition. Sampling was performed using the default No-U-Turn Sampler (NUTS) algorithm in Stan, an adaptive form of Hamiltonian Monte Carlo.

To evaluate condition-related changes in model-estimated decision parameters, we computed posterior differences in group-level parameters between prolonged exhalation and eupnea. We deemed condition differences credible if the 95% credible interval of the posterior distribution of the difference excluded 0.

##### Parameter recovery

To assess the reliability of prospect theory parameter estimates, we conducted a joint parameter-recovery analysis. For each of 100 simulations, all parameters were simultaneously drawn from truncated prior distributions as specified in the hierarchical prospect theory model. Synthetic choice data were then generated from these true parameters using the same prospect theory likelihood as the fitted model, applied to the original trial structure. The model was refitted to each simulated dataset using the same estimation procedure as for the real data. Recoverability was quantified as the Pearson correlation between true and recovered posterior means across participants, separately for each parameter and condition (Table S300339-9#mmc1)).

##### Non-parametric AUC

We also computed the non-parametric area under the acceptance threshold curve (AUC)6700339-9#bib66) as a model-free summary measure. Because the choice sets were identical across breathing conditions, for each participant and condition, we calculated the indifference threshold (reward value at which p(accept) = 0.5) for each loss level using linear interpolation, then integrated across loss levels using the trapezoidal rule. Condition differences in AUC were assessed using a paired t-test.

#### Physiological data collection and analysis

##### Physiological data acquisition

Physiological signals, including ECG, Electrodermal Activity (EDA), and respiration, were recorded simultaneously using a BrainAmp MR-compatible amplifier (Brain Products GmbH, Germany) and BrainVision Recorder software (version 1.21.0303). All signals were sampled at 5,000 Hz with 16-bit resolution and recorded in DC mode. A three-channel configuration was used: ECG was acquired via a bipolar chest setup; EDA was recorded from the non-dominant hand using a GSR-MR module (resolution: 0.006104 μS); and respiration was measured via a thoracic belt (resolution: 0.1526 arbitrary units). PPG signals were simultaneously recorded at 400 Hz using the Siemens physiological monitoring unit integrated with the MRI scanner. Pupil diameter was recorded at 500 Hz using an MRI-compatible EyeLink 1000 eye-tracker (SR Research Ltd., Canada), positioned at the rear of the scanner bore and tracking the participant’s right eye throughout the task.

##### Respiration

Respiration was recorded using a thoracic belt positioned around the upper abdomen. Raw signals were low-pass filtered at 1 Hz using a third-order finite impulse response (FIR) filter via the FieldTrip toolbox.5700339-9#bib56) The data were down-sampled to 100 Hz, detrended, and baseline corrected. Outliers were interpolated using a median-based approach and smoothed with a Savitzky-Golay filter.6800339-9#bib67) Inhalation and exhalation durations were extracted cycle-by-cycle using a peak detection algorithm implemented in the BreathMetrics toolbox.5800339-9#bib57)

As a quality control measure, each block was evaluated based on the proportion of respiratory cycles with inhalation/exhalation ratios falling within an acceptable range: for the prolonged exhalation condition, a fixed target ratio of 2:8 seconds (inhalation/exhalation = 0.25) was used. The eupnea condition used the individually estimated inhalation to exhalation ratios as a target. In both conditions, a cycle was considered valid if its ratio fell within ±60% of the target ratio.

##### Electrocardiogram (ECG)

ECG was recorded via cutaneous electrodes placed below the clavicle in a bipolar chest configuration. Preprocessing focused on artifact removal and the extraction of inter-beat intervals (IBIs) for heart rate variability (HRV) analysis. Gradient artifacts were corrected using EEGLAB functions to implement a volume-locked template subtraction procedure.5900339-9#bib58),6000339-9#bib59),6900339-9#bib68) Scanner volume onset markers were used to extract synchronized ECG segments from each volume, and artifact templates were constructed and subtracted to preserve physiological signals while removing scanner-related noise. Incomplete final volumes were excluded.

Following artifact correction, ECG signals were bandpass filtered between 1 and 100 Hz (fourth-order FIR) to remove baseline drift and high-frequency noise. R-peaks were then identified using a semi-automated MATLAB pipeline that combines z-score normalization, adaptive thresholding, and template matching.

To examine HRV under different breathing conditions, we focused on time-domain measures, specifically the RMSSD and RespHRV, rather than frequency-domain metrics. This decision was based on methodological considerations: time-domain indices are more robust to non-stationary fluctuations and short block-wise recordings and are less confounded by irregular or unbalanced breathing patterns such as those induced by the prolonged exhalation intervention.2600339-9#bib26),2900339-9#bib29)

The final corrected IBI time series were used to compute RMSSD for each breathing condition as a time-domain index of cardiac parasympathetic modulation. In participants where ECG recordings were incomplete or noisy, IBIs were extracted from concurrently recorded PPG data using the same peak detection procedure, and RMSSD was computed accordingly.7000339-9#bib69),7100339-9#bib70)

RespHRV was quantified using a peak-to-trough (P2T) approach based on heart rate fluctuations across the respiratory cycle, following established procedures.500339-9#bib5),7200339-9#bib71) RespHRV was defined as the difference in heart period between inhalation and exhalation phases and was averaged across trials within each breathing condition.

##### Electrodermal Activity (EDA)

EDA was measured via electrodes placed on the index and middle fingers of the non-dominant (left) hand. The raw signal underwent preprocessing to remove scanner-related artifacts and extract tonic components. Gradient artifacts were removed using the FASTR algorithm as implemented in EEGLAB,5900339-9#bib58),6000339-9#bib59) which adaptively subtracts scanner-periodic noise. The corrected signal was then processed in Ledalab6100339-9#bib60) using a standardized batch pipeline. Specifically, the data were low-pass filtered using a first-order Butterworth filter with a cutoff at 5 Hz, and down-sampled from 5,000 Hz to 100 Hz to reduce file size and improve processing efficiency. Adaptive smoothing was applied to further reduce noise. After preprocessing and artifact rejection, EDA data of sufficient quality were retained for 26 participants, who were included in the final analysis.

Continuous Decomposition Analysis was used to separate tonic and phasic components of the EDA signal. Only the tonic component (skin conductance level, SCL) was analyzed in the present study as an index of sympathetic arousal. Tonic SCL values were averaged across blocks for each participant and breathing condition.

##### Photoplethysmography (PPG)

Pulse signals were recorded using the Siemens physiological monitoring unit integrated with the MRI scanner, sampled at 400 Hz. The signal was band-pass filtered (0.5–5 Hz, fourth-order Chebyshev Type II) and linearly detrended to remove baseline drift and isolate cardiac-related fluctuations. Systolic peaks were detected using the same semi-automated MATLAB pipeline used for ECG, which combines z-score normalization, adaptive thresholding, and template matching. All detected peaks were visually inspected and manually corrected as needed, including the adjustment or replacement of implausible IBIs.

PPG was used exclusively in participants for whom ECG data were unavailable or of insufficient quality. In these cases, IBIs derived from the PPG signal were used to compute RMSSD and RespHRV, consistent with the ECG analysis.

##### Pupil diameter

Pupil diameter was measured under identical lighting conditions across all blocks, with the same visual content presented across conditions to ensure comparable visual input.

Raw pupil data were converted using the Edf2Mat MATLAB Toolbox designed and developed by Adrian Etter and Marc Biedermann at the University of Zurich. Blinks were identified using EyeLink event markers and periods of zero-valued pupil size. These segments were excluded, and missing values were linearly interpolated to reconstruct a continuous time series. The signal was then down-sampled to 100 Hz and corrected for blink-edge artifacts by estimating a peak envelope using spline interpolation with a minimum peak separation of 200 ms. The resulting trace was low-pass filtered using a zero-phase Butterworth filter with a cutoff frequency of 8 Hz. Finally, pupil size was baseline-corrected by subtracting the mean value from the first 10 seconds of each block to account for between-subject variability. Pupil data were then visually inspected for quality and retained only if deemed usable based on signal stability and absence of gross artifacts. After preprocessing, pupil data of sufficient quality were retained for 28 participants and included in subsequent analyses.

Average pupil size was computed per condition and used as an index of sympathetic arousal, and is reported in arbitrary units (a.u.) following baseline correction.

##### Physiological data analyses

To assess condition differences in physiological signals, we used two-tailed paired t-tests on subject-level condition means. Each subject contributed one average value per condition, enabling a direct within-subject comparison. Effect size was calculated as Cohen’s d for paired samples (mean difference divided by the standard deviation of differences).

##### Correlations between SCL/pupil changes and behavioral changes

To examine whether individual differences in physiological responses to the breathing manipulation predicted behavioral effects, we computed Pearson correlations between SCL and pupil diameter change scores and acceptance rate change scores. For each measure, we calculated the difference between conditions (Δ = Prolonged exhalation − Eupnea) at the participant level.

#### Loss sensitivity and arousal indices

To examine whether losses were accompanied by increased autonomic arousal, we estimated condition-specific loss sensitivity for each participant using a logistic regression of trial-wise choices on z-scored reward and loss magnitudes. We then tested whether arousal indices were predicted by loss sensitivity and whether this association differed by breathing condition. All models controlled for age, sex, and task order, and inference used HC3 heteroskedasticity-robust standard errors.

##### fMRI data acquisition, preprocessing, and statistical analysis

Whole-brain functional MRI data were collected on a Siemens MAGNETOM Prisma 3T scanner equipped with a 32-channel head coil at the Center for Cognitive Neuroscience Berlin of the Freie Universität Berlin. Functional images were acquired using a T2∗-weighted echo-planar imaging (EPI) sequence with simultaneous multi-slice acquisition (multi-band factor = 4). The acquisition parameters were as follows: repetition time (TR) = 750 ms, echo time (TE) = 30 ms, flip angle = 65°, 40 axial slices, voxel size = 3 × 3 × 3 mm³, field of view (FOV) = 192 × 192 mm, and phase encoding direction posterior-to-anterior. Multiband acceleration was used to increase temporal resolution. High-resolution anatomical images were obtained using a T1-weighted MP-RAGE sequence (TR = 1900 ms; TE = 2.52 ms; flip angle = 9°; 176 slices; voxel size = 1 × 1 × 1 mm³; FOV = 256 × 256 mm) with GRAPPA acceleration.

Blocks were excluded from first-level analysis if head motion exceeded 3 mm translation or 3° rotation on any axis. Based on this criterion, block-level exclusions occurred in 8 participants, of whom 2 had already been excluded due to missing physiological data. The remaining 6 participants were retained in the analysis with partial block-level data (8 blocks total). After combining exclusions from head motion and physiological signal quality, the final fMRI analysis sample comprised 35 participants.

All image preprocessing and analyses were performed using SPM12 (The Wellcome Department of Imaging Neuroscience, Institute of Neurology, London, UK; https://www.fil.ion.ucl.ac.uk/spm/software/). Functional volumes were first slice-time corrected using slice acquisition times provided by the scanner, with the middle slice as the reference. Images were then realigned to the mean image of each run to correct for head motion. Each participant’s T1-weighted anatomical image was co-registered to the mean functional image and segmented into gray matter, white matter, and cerebrospinal fluid using SPM’s unified segmentation procedure. Spatial normalization to the Montreal Neurological Institute (MNI) standard space was performed using the deformation fields generated during segmentation. Functional images were resampled to 3 × 3 × 3 mm³ voxel size and spatially smoothed with an 8 mm full width at half-maximum Gaussian kernel.

To account for the effect of general physiological noise, we performed physiological noise correction using RETROICOR7300339-9#bib72) based on concurrently recorded cardiac and respiratory signals. Physiological regressors were generated using the PhysIO toolbox,6200339-9#bib61),6300339-9#bib62) modeling cardiac phase (third-order Fourier expansion), respiratory phase (fourth order), and their first-order interaction. This resulted in 18 physiological regressors per block. In addition, the six motion parameters estimated during realignment (three translations and three rotations) were included as nuisance covariates in the first-level GLM to account for non-neural sources of BOLD signal variability.

To investigate neural responses to rewards and losses, we specified a GLM in which the onset times of options were modeled as events and parametrically modulated by two regressors: the magnitude of rewards and the magnitude of losses. All regressors were convolved with the canonical hemodynamic response function. First-level contrast images were computed to capture condition-specific neural sensitivity to reward and loss magnitude. Reward and loss were entered as separate parametric modulators of the option-onset regressors without orthogonalization, so that each regressor captures variance uniquely associated with its respective magnitude.

At the group level, contrast images were entered into a second-level random-effects analysis. To assess whether individual differences in cardiac parasympathetic regulation predicted changes in neural sensitivity to prospective reward, we conducted a second-level whole-brain regression analysis. The input for this analysis was the first-level contrast images representing the parametric reward effect difference between prolonged exhalation and eupnea. For each participant, the within-subject difference in heart rate variability (ΔRMSSD (prolonged exhalation − eupnea)) was computed and entered as a continuous covariate. Whole-brain regression was performed using a voxel-level threshold of _p_< 0.001 (uncorrected), followed by cluster-level family-wise error (FWE) correction at _p_< 0.05.

To explore the role of individual differences in interoceptive regulation, we tested whether participants’ self-regulation predicted the neural impact of the breathing pattern. Participants also completed the Multidimensional Assessment of Interoceptive Awareness (MAIA)7400339-9#bib73),7500339-9#bib74) after scanning. Specifically, we extracted the condition-related difference in reward-related activation in the precuneus and vmPFC (prolonged exhalation − eupnea) and calculated mean values, then regressed them on self-regulation as measured by the MAIA subscale scores.

Supplemental information (2)

Document S1. Figures S1–S4 and Tables S1–S3

Document S2. Article plus supplemental information

References

1.

Zaccaro, A. ∙ Piarulli, A. ∙ Laurino, M. ...

How breath-control can change your life: a systematic review on psycho-physiological correlates of slow breathing

_Front. Hum. Neurosci._ 2018; 12:353

2.

Bae, D. ∙ Matthews, J.J.L. ∙ Chen, J.J. ...

Increased exhalation to inhalation ratio during breathing enhances high-frequency heart rate variability in healthy adults

_Psychophysiology._ 2021; 58, e13905

3.

Röttger, S. ∙ Theobald, D.A. ∙ Abendroth, J. ...

The effectiveness of combat tactical breathing as compared with prolonged exhalation

_Appl. Psychophysiol. Biofeedback._ 2021; 46:19-28

4.

Yasuma, F. ∙ Hayano, J.

Respiratory sinus arrhythmia: why does the heartbeat synchronize with respiratory rhythm?

_Chest._ 2004; 125:683-690

5.

Lewis, G.F. ∙ Furman, S.A. ∙ McCool, M.F. ...

Statistical strategies to quantify respiratory sinus arrhythmia: Are commonly used metrics equivalent?

_Biol. Psychol._ 2012; 89:349-364

6.

Gerritsen, R.J.S. ∙ Band, G.P.H.

Breath of life: The respiratory vagal stimulation model of contemplative activity

_Front. Hum. Neurosci._ 2018; 12:397

7.

Laborde, S. ∙ Iskra, M. ∙ Zammit, N. ...

Slow-paced breathing: Influence of inhalation/exhalation ratio and of respiratory pauses on cardiac vagal activity

_Sustainability._ 2021; 13:7775

8.

Grossman, P. ∙ Taylor, E.W.

Toward understanding respiratory sinus arrhythmia: Relations to cardiac vagal tone, evolution and biobehavioral functions

_Biol. Psychol._ 2007; 74:263-285

9.

Ritz, T.

Studying noninvasive indices of vagal control: The need for respiratory control and the problem of target specificity

_Biol. Psychol._ 2009; 80:158-168

10.

Kahneman, D. ∙ Tversky, A.

Prospect theory: An analysis of decision under risk

_Econometrica._ 1979; 47:263-291

11.

Tom, S.M. ∙ Fox, C.R. ∙ Trepel, C. ...

The neural basis of loss aversion in decision-making under risk

_Science._ 2007; 315:515-518

12.

Bartra, O. ∙ McGuire, J.T. ∙ Kable, J.W.

The valuation system: A coordinate-based meta-analysis of BOLD fMRI experiments examining neural correlates of subjective value

_NeuroImage._ 2013; 76:412-427

13.

Critchley, H.D. ∙ Harrison, N.A.

Visceral influences on brain and behavior

_Neuron._ 2013; 77:624-638

14.

Prell, R. ∙ Starcke, K. ∙ Maggioni, M.A.

Following your heart: Cardiac autonomic modulation as a feature of decision-making under ambiguity and risk

_Pers. Individ. Differ._ 2024; 226, 112686

15.

Ramírez, E. ∙ Ortega, A.R. ∙ Reyes Del Paso, G.A.

Anxiety, attention, and decision making: The moderating role of heart rate variability

_Int. J. Psychophysiol._ 2015; 98:490-496

16.

Forte, G. ∙ Morelli, M. ∙ Casagrande, M.

Heart rate variability and decision-making: Autonomic responses in making decisions

_Brain Sci._ 2021; 11:243

17.

Sheng, F. ∙ Ramakrishnan, A. ∙ Seok, D. ...

Decomposing loss aversion from gaze allocation and pupil dilation

_Proc. Natl. Acad. Sci. USA._ 2020; 117:11356-11363

18.

Henn, A.T. ∙ Hüpen, P. ∙ Boccadoro, S. ...

Context effects, skin conductance responses and personality traits – Influencing variables on risk-taking within a modified version of the balloon analog risk task

_Biol. Psychol._ 2023; 177, 108498

19.

Levy, D.J. ∙ Glimcher, P.W.

The root of all value: a neural common currency for choice

_Curr. Opin. Neurobiol._ 2012; 22:1027-1038

20.

Beissner, F. ∙ Meissner, K. ∙ Bär, K.-J. ...

The autonomic brain: an activation likelihood estimation meta-analysis for central processing of autonomic function

_J. Neurosci._ 2013; 33:10503-10511

21.

Kirk, U. ∙ Gu, X. ∙ Harvey, A.H. ...

Mindfulness training modulates value signals in ventromedial prefrontal cortex through input from insular cortex

_NeuroImage._ 2014; 100:254-262

22.

Thayer, J.F. ∙ Lane, R.D.

Claude Bernard and the heart–brain connection: Further elaboration of a model of neurovisceral integration

_Neurosci. Biobehav. Rev._ 2009; 33:81-88

23.

Kim, J. ∙ Kim, H.

Neural Processes Linking Interoception to Moral Preferences Aligned with Group Consensus

_J. Neurosci._ 2025; 45, e1114242025

24.

Maier, S.U. ∙ Hare, T.A.

Higher heart-rate variability is associated with ventromedial prefrontal cortex activity and increased resistance to temptation in dietary self-control challenges

_J. Neurosci._ 2017; 37:446-455

25.

Berntson, G.G. ∙ Thomas Bigger, J. ∙ Eckberg, D.L. ...

Heart rate variability: Origins, methods, and interpretive caveats

_Psychophysiology._ 1997; 34:623-648

26.

Laborde, S. ∙ Mosley, E. ∙ Thayer, J.F.

Heart rate variability and cardiac vagal tone in psychophysiological research – recommendations for experiment planning, data analysis, and data reporting

_Front. Psychol._ 2017; 8:213

27.

Menuet, C. ∙ Ben-Tal, A. ∙ Linossier, A. ...

Redefining respiratory sinus arrhythmia as respiratory heart rate variability: an international Expert Recommendation for terminological clarity

_Nat. Rev. Cardiol._ 2025; 22:978-984

28.

Porges, S.W.

The polyvagal perspective

_Biol. Psychol._ 2007; 74:116-143

29.

Shaffer, F. ∙ Ginsberg, J.P.

An overview of heart rate variability metrics and norms

_Front. Public Health._ 2017; 5:258

30.

Lehrer, P.M. ∙ Gevirtz, R.

Heart rate variability biofeedback: how and why does it work?

_Front. Psychol._ 2014; 5:756

31.

Vaschillo, E. ∙ Lehrer, P. ∙ Rishe, N. ...

Heart rate variability biofeedback as a method for assessing baroreflex function: a preliminary study of resonance in the cardiovascular system

_Appl. Psychophysiol. Biofeedback._ 2002; 27:1-27

32.

FeldmanHall, O. ∙ Glimcher, P. ∙ Baker, A.L. ...

Emotion and decision-making under uncertainty: Physiological arousal predicts increased gambling during ambiguity but not risk

_J. Exp. Psychol.: Gen._ 2016; 145:1255-1262

33.

Molins, F. ∙ Serrano, M.Á. ∙ Alacreu-Crespo, A.

Early stages of the acute physical stress response increase loss aversion and learning on decision making: A Bayesian approach

_Physiol. Behav._ 2021; 237, 113459

34.

Yamakawa, K. ∙ Ohira, H. ∙ Matsunaga, M. ...

Prolonged effects of acute stress on decision-making under risk: A human psychophysiological study

_Front. Hum. Neurosci._ 2016; 10:444

35.

Thayer, J.F. ∙ Lane, R.D.

A model of neurovisceral integration in emotion regulation and dysregulation

_J. Affect. Disord._ 2000; 61:201-216

36.

Williams, D.P. ∙ Cash, C. ∙ Rankin, C. ...

Resting heart rate variability predicts self-reported difficulties in emotion regulation: a focus on different facets of emotion regulation

_Front. Psychol._ 2015; 6:261

37.

Forte, G. ∙ Morelli, M. ∙ Grässler, B. ...

Decision making and heart rate variability: A systematic review

_Appl. Cogn. Psychol._ 2022; 36:100-110

38.

Sütterlin, S. ∙ Herbert, C. ∙ Schmitt, M. ...

Frames, decisions, and cardiac–autonomic control

_Soc. Neurosci._ 2011; 6:169-177

39.

Kable, J.W. ∙ Glimcher, P.W.

The neural correlates of subjective value during intertemporal choice

_Nat. Neurosci._ 2007; 10:1625-1633

40.

Cavanna, A.E. ∙ Trimble, M.R.

The precuneus: a review of its functional anatomy and behavioural correlates

_Brain._ 2006; 129:564-583

41.

Babo-Rebelo, M. ∙ Richter, C.G. ∙ Tallon-Baudry, C.

Neural responses to heartbeats in the default network encode the self in spontaneous thoughts

_J. Neurosci._ 2016; 36:7829-7840

42.

Xu, T. ∙ Chen, Z. ∙ Feng, T.

The preference for future outcomes correlates with the temporal variability of functional connectivity among brain regions

_Behav. Brain Res._ 2019; 375, 112111

43.

Khalsa, S.S. ∙ Adolphs, R. ∙ Cameron, O.G. ...

Interoception and mental health: a roadmap

_Biol. Psychiatry Cogn. Neurosci. Neuroimaging._ 2018; 3:501-513

44.

Chalmers, J.A. ∙ Quintana, D.S. ∙ Abbott, M.J.-A. ...

Anxiety disorders are associated with reduced heart rate variability: a meta-analysis

_Front. Psychiatry._ 2014; 5:80

45.

Hein, T.P. ∙ De Fockert, J. ∙ Ruiz, M.H.

State anxiety biases estimates of uncertainty and impairs reward learning in volatile environments

_NeuroImage._ 2021; 224, 117424

46.

Grabowska, A. ∙ Zabielski, J. ∙ Senderecka, M.

Machine learning reveals differential effects of depression and anxiety on reward and punishment processing

_Sci. Rep._ 2024; 14, 8422

47.

Kemp, A.H. ∙ Quintana, D.S. ∙ Gray, M.A. ...

Impact of depression and antidepressant treatment on heart rate variability: a review and meta-analysis

_Biol. Psychiatry._ 2010; 67:1067-1074

48.

De Couck, M. ∙ Caers, R. ∙ Musch, L. ...

How breathing can help you make better decisions: Two studies on the effects of breathing patterns on heart rate variability and decision-making in business cases

_Int. J. Psychophysiol._ 2019; 139:1-9

49.

Strang, S. ∙ Hoeber, C. ∙ Uhl, O. ...

Impact of nutrition on social decision making

_Proc. Natl. Acad. Sci. USA._ 2017; 114:6510-6514

50.

Liu, L. ∙ Artigas, S.O. ∙ Ulrich, A. ...

Eating to dare - Nutrition impacts human risky decision and related brain function

_NeuroImage._ 2021; 233, 117951

51.

Brainard, D.H.

The Psychophysics Toolbox

_Spat. Vis._ 1997; 10:433-436

52.

R Core Team

R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing

53.

Bates, D. ∙ Mächler, M. ∙ Bolker, B. ...

Fitting linear mixed-effects models using lme4

_J. Stat. Softw._ 2015; 67:1-48

54.

Lenth, R.

emmeans: Estimated Marginal Means, aka Least-Squares Means

55.

Kuznetsova, A. ∙ Brockhoff, P.B. ∙ Christensen, R.H.B.

lmerTest package: tests in linear mixed effects models

_J. Stat. Softw._ 2017; 82:1-26

56.

Stan Development Team

RStan: the R interface to Stan

57.

Oostenveld, R. ∙ Fries, P. ∙ Maris, E. ...

FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data

_Comput. Intell. Neurosci._ 2011; 2011:1-9

58.

Noto, T. ∙ Zhou, G. ∙ Schuele, S. ...

Automated analysis of breathing waveforms using BreathMetrics: a respiratory signal processing toolbox

_Chem. Senses._ 2018; 43:583-597

59.

Delorme, A. ∙ Makeig, S.

EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis

_J. Neurosci. Methods._ 2004; 134:9-21

60.

Niazy, R.K. ∙ Beckmann, C.F. ∙ Iannetti, G.D. ...

Removal of FMRI environment artifacts from EEG data using optimal basis sets

_Neuroimage._ 2005; 28:720-737

61.

Benedek, M. ∙ Kaernbach, C.

A continuous measure of phasic electrodermal activity

_J. Neurosci. Methods._ 2010; 190:80-91

62.

Kasper, L. ∙ Bollmann, S. ∙ Diaconescu, A.O. ...

The PhysIO toolbox for modeling physiological noise in fMRI data

_J. Neurosci. Methods._ 2017; 276:56-72

63.

Frässle, S. ∙ Aponte, E.A. ∙ Bollmann, S. ...

TAPAS: an open-source software package for translational neuromodeling and computational psychiatry

_Front. Psychiatry._ 2021; 12, 680811

64.

Cappo, B.M. ∙ Holmes, D.S.

The utility of prolonged respiratory exhalation for reducing physiological and psychological arousal in non-threatening and threatening situations

_J. Psychosom. Res._ 1984; 28:265-273

65.

Pisupati, S. ∙ Chartarifsky-Lynn, L. ∙ Khanal, A. ...

Lapses in perceptual decisions reflect exploration

_eLife._ 2021; 10, e55490

66.

Nilsson, H. ∙ Rieskamp, J. ∙ Wagenmakers, E.-J.

Hierarchical Bayesian parameter estimation for cumulative prospect theory

_J. Math. Psychol._ 2011; 55:84-93

67.

Walasek, L. ∙ Stewart, N.

How to make loss aversion disappear and reverse: tests of the decision by sampling origin of loss aversion

_J. Exp. Psychol.: Gen._ 2015; 144:7-11

68.

Savitzky, A. ∙ Golay, M.J.E.

Smoothing and differentiation of data by simplified least squares procedures

_Anal. Chem._ 1964; 36:1627-1639

69.

Nierhaus, T. ∙ Gundlach, C. ∙ Goltz, D. ...

Internal ventilation system of MR scanners induces specific EEG artifact during simultaneous EEG-fMRI

_Neuroimage._ 2013; 74:70-76

70.

Lu, G. ∙ Yang, F. ∙ Taylor, J.A. ...

A comparison of photoplethysmography and ECG recording to analyse heart rate variability in healthy subjects

_J. Med. Eng. Technol._ 2009; 33:634-641

71.

Schäfer, A. ∙ Vagedes, J.

How accurate is pulse rate variability as an estimate of heart rate variability? A review on studies comparing photoplethysmographic technology with an electrocardiogram

_Int. J. Cardiol._ 2013; 166:15-29

72.

Grossman, P. ∙ Van Beek, J. ∙ Wientjes, C.

A comparison of three quantification methods for estimation of respiratory sinus arrhythmia

_Psychophysiology._ 1990; 27:702-714

73.

Glover, G.H. ∙ Li, T.-Q. ∙ Ress, D.

Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR

_Magn. Reson. Med._ 2000; 44:162-167

74.

Mehling, W.E. ∙ Price, C. ∙ Daubenmier, J.J. ...

The multidimensional assessment of interoceptive awareness (MAIA)

_PLoS One._ 2012; 7, e48230

75.

Bornemann, B. ∙ Herbert, B.M. ∙ Mehling, W.E. ...

Differential changes in self-reported aspects of interoceptive awareness through 3 months of contemplative training

_Front. Psychol._ 2015; 5:1504