Acute fasting modulates autonomic nervous system function and ambulatory cardiac interoception

Intermittent fasting has been associated with diverse physical and psychological health benefits. According to previous research, fasting-induced alterations in psychophysiological functioning should facilitate the accurate detection of an internal bodily signal (like the heart), which is referred to as interoceptive accuracy. In two within-subjects studies we aimed to examine whether an intermittent fasting protocol (i) evokes distinct autonomic nervous system changes in the laboratory and (ii) improves (objectifiable) interoceptive accuracy and sensibility (i.e., the subjective belief in perceiving bodily signals) in everyday life. Study 1 (N = 36) found increasing heart rate variability (precisely, the root mean square of successive differences; RMSSD) accompanied by a more vascular than myocardial response following a 16 h fast. Study 2 (N = 40) applied an ecological momentary assessment design including intermittent fasting (8 h normal eating followed by 16 h fasting) and normal eating (24 h normal eating) for three consecutive days each. Findings suggested a tendency toward higher interoceptive accuracy and sensibility during the fasting regimen, which was particularly pronounced in individuals exhibiting lower RMSSD. Together, findings suggest that (short-term) fasting seems to facilitate momentary attention to organismic cues due to alterations in autonomic nervous system function.


Introduction
Intermittent fasting has been hypothesized to benefit health.Specifically, there is increasing evidence that regular fasting for at least 16 h could ameliorate morbidity and prolong life (e.g., Longo et al., 2021;Madeo et al., 2010;Mattson et al., 2017;Weir et al., 2017).Manifold routes have been suggested responsible for the benevolent effect of fasting.On a molecular level, the short-term effects of fasting are well documented and suggest enhanced cellular autophagy and reduced cellular stress (e.g., Alirezaei et al., 2010, Longo & Anderson, 2022;Martinez-Lopez et al., 2017;Stratton et al., 2022).Long-term effects include optimized lipid and glucose metabolism, as well as neurogenesis (e.g., Cabo & Mattson, 2019;Dias et al., 2021).Furthermore, fasting seems to affect the autonomic nervous system (ANS) such that sympathetic activity gets suppressed (e.g., Gonzalez & Cooke, 2022;Young & Landsberg, 1977).In accordance with this, recent research found that avoiding caloric intake for 16 to 24 h (acute fasting) resulted in lower heart rate (HR) and higher heart rate variability (HRV; e.g., Flasbeck et al., 2021;Rominger et al., 2021), thus suggesting a major autonomic shift towards parasympathetic dominance.However, it should be noted that the effects of short-term fasting on cardiac activity are far from being unequivocal.Notably, Herbert et al. (2012) reported rather elevated HR and lower HRV due to a 24 h-fast, while Schulz et al. (2015) found non-significant changes following an 18 h-fast.
Beside cardiac activity, hemodynamic function seems particularly important for adjustment and health (e.g., Gregg, & James, 2002;Sherwood et al., 1990).Specifically, the amount of blood ejected by the heart per minute (cardiac output, CO) and the resistance or elasticity of the arteries (total peripheral resistance, TPR) determine blood pressure such that higher CO and higher TPR result in elevated blood pressure, but each variable also allows for compensatory effects on blood pressure.To the authors' knowledge, only two previous studies assessed the impact of acute fasting on hemodynamic activity.Specifically, Gonzalez and Cooke (2022) found lower blood pressure, elevated stroke volume and higher baroreceptor-reflex sensitivity following acute fasting.Herbert et al. (2012), on the other hand, reported no significant fasting-induced change in stroke volume, but elevated CO, diminished pre-ejection period (a sympathetic cardiac indicator) and elevated cardiac contractility.
Due to the inconsistent cardiovascular effects of a short-term fasting regiment in previous research (e.g., Flasbeck et al., 2021;Herbert et al., 2012;Rominger et al., 2021;Schulz et al., 2015), Study 1 aimed to examine physiological concomitants of a brief (16 h) fasting episode to illustrate changes in cardiovascular activity.Importantly, it has been reported that the ANS changes to fasting could facilitate the perception of bodily processes (Herbert et al., 2012;Rominger et al., 2021; but see Schulz et al., 2015 for nil findings), thus improving interoception, which constitutes a relevant skill related to self-and emotion regulation (Allen & Tsakiris, 2019;Füstös et al., 2013;Strigo & Craig, 2016;Zamariola et al., 2019;Rominger & Schwerdtfeger, 2023), and better health in general (Critchley et al., 2023;Herbert, 2020;Quadt et al., 2018).Thus, Study 1 aimed to lay the foundations for Study 2, which then aimed to unveil changes in interoception during a fasting regimen.Using a within-subject design, Study 1 aimed to analyze cardiac (HR, HRV) and hemodynamic changes (blood pressure, cardiac output [CO] and total peripheral resistance [TPR]) as consequences of fasting.According to our previous research (Rominger et al., 2021) and Flasbeck et al. (2021), we hypothesized (1) that fasting leads to lower HR and higher HRV.Furthermore, (2) we exploratory analyzed CO and TPR to examine whether the hypothesized changes in cardiac activity are accompanied by compensatory increases in vascular resistance.

Participants
An a priori power analysis was conducted to determine required sample size based on a medium-sized effect of d = 0.5 for a withinsubject comparison (t-tests) and a power of.80.The medium effect size was derived from prior research on fasting and cardiovascular activity (e.g., Herbert et al., 2012;Rominger et al., 2021).The required sample size was N = 36.Accordingly, the final sample comprised of N = 36 individuals (72% women) with a mean age of 21.78 years (SD = 2.61) and a mean BMI of 23.13 kg/m 2 (SD = 3.09).The sample mainly comprised of students (97%) and 75% (n = 27) reported no prior experience with intermittent fasting.Of note, participants did not report any psychological or cardiovascular disorders and reported no diabetes and no related medication.Participants were recruited via e-mail newsletters, oral communication and social media.There was no monetary compensation offered, but students could get course credits if applicable.

Design
Study 1 was part of a larger project on psychophysiological effects of acute fasting including cardiovascular measures and cognitive tasks.None of the data presented here has been published previously.For the purpose of the present research, we only focus on baseline cardiovascular measures.Study design was a within-subject repeated measures design with two time points.Hence, data were sampled during two laboratory sessions, one after normal eating and one following a 16 h fasting regimen.During the fasting regimen, no consumption of food or caloric drinks were allowed, while water or non-caloric beverages (e.g., tea, coffee without milk or sugar) were okay per instructions.Conditions were separated by a 1 week-interval and covered approximately the same time of day (all sessions were scheduled between 7 am and 1 pm).The order of conditions was randomized with 50% of the sample starting with the fasting regimen and 50% starting with the normal eating.The study was approved by the local ethics committee (GZ 39/11/63 ex 2019/20).

Psychological measures
Perceived hunger.Perceived hunger was assessed using a single item Likert-type measure asking for feelings of hunger (1 = not at all, 5 = very much).Past eating.Participants were asked when they consumed their last meal (hours since the last meal) and whether they ate anything during the last 16 h.For the fasting condition, the last meal was taken ~17 h ago (SD = 0.91) and for the normal eating condition, participants reported food consumption on average 1.5 h ago (SD = 0.56).Hence, participants were compliant with the study protocol.

Physiological measures
Blood glucose level.Blood glucose was assessed via a commercially available capillary blood sugar test (CALLA Mini; Wellion™, MedTrust HbH, Marz, Austria).The device is EN ISO-certificated and provides reliable assessment.
Cardiovascular variables.Cardiovascular variables were recorded continuously throughout baseline.HR was measured by means of an ECG device (AccuSync® 72, AccuSync Medical Research Corporation, Milford, Connecticut, USA) using a modified Einthoven II-point lead.The ECG was recorded using Ambu BlueSensor® electrodes (Ballerup Sogn, Denmark) with a sampling rate of 1000 Hz.The signal was recorded with the software AcqKnowledge® 4.3 (Biopac Systems Inc., Goleta, California, USA).HR and HRV (RMSSD) were analyzed offline with Kubios premium software [vers. 3.2;Tarvainen et al., 2021;University of Finland, Finland], with nonstationary HRV trends removed via an inbuilt algorithm based on smoothness priors regularization (Tarvainen et al., 2021).Data were visually screened for artifacts (e.g., ectopic beats or muscle artefacts) and corrected if necessary, applying the Kubios automatic correction algorithm.
Continuous blood pressure (i.e., systolic blood pressure [SBP], diastolic blood pressure [DBP]) was measured by non-invasive measurement of arterial finger BP using the Finometer® PRO (Finapres Medical Systems, Amsterdam, The Netherlands).Upon recording, absolute accuracy of the brachial artery blood pressure is calibrated by a wellvalidated (Schutte et al., 2004) procedure (i.e., return-to-flow calibration; e.g., Bos et al., 1996) using an upper arm cuff.Importantly, the Finometer® has an additional on board "Physiocal" calibration feature (Wesseling et al., 1995), which accounts for potential drifts in the data by an automatized and repeated application (max.every 70 beats).The Finometer® finger cuff was attached at participant's middle finger of the non-dominant hand and the upper arm cuff was attached on the same side.After visual inspection, mean SBP and DBP were calculated.CO (via modelflow estimate; Wesseling et al., 1993) and TPR were analyzed offline using Beatscope software (vers.2.10, Finapres Medical Systems, Amsterdam, The Netherlands), by considering individual's sex, age, body mass and weight.CO is quantified as liters per minute and TPR as the ratio of the mean arterial pressure to cardiac output and is expressed as centimeter-gram-second units (dyn.s/cm 5 ).

Procedure
Individuals interested in the study completed a pre-screening to check whether they were suitable for participating in the study (exclusion criteria were cardiovascular and mental disorders, cardiovascular or psychotropic medication).Sociodemographic data were also gathered.Candidates then received information about the study details and were invited to the laboratory study.Each participant agreed on two laboratory visits, separated by one week, which were scheduled between 7 am and 1 pm.Each individual was tested on the same time of day.Prior to one of the visits, participants were instructed to completely abstain from food, including calorie-containing drinks, for at least 16 h.At the other visit, they should appear saturated as usual.Order was randomized and the time between appointments was one week.
At the first appointment, participants gave informed consent to partake voluntarily in the study and were informed that they could withdraw participation at any time.Height, weight, and waist and hip circumference were then measured and capillary blood sugar levels A.R. Schwerdtfeger and C. Rominger were assessed.Subsequently, the physiological devices were attached and baseline recording was initialized for 3 min where participants viewed landscape images on a computer screen.Then, questionnaires on eating behavior and perceived hunger were filled out.

Data parametrization and analyses
Cardiovascular variables were aggregated across the 3 min of the baseline period.In order to quantify difference in CO and TPR, we analyzed hemodynamic profile (HP) and compensation deficit (CD) according to the formulas below, provided by Gregg et al. (2002) and James et al. (2012).
HP aims to assess the compensatory effects of CO and TPR on blood pressure, while CD indicates the cumulative effects.Positive HP values signal a vascular profile in the normal eating condition (meaning that CO is lower than TPR) and a myocardial profile in the fasting condition (meaning that CO is higher than TPR), while negative values indicate a vascular profile in the fasting relative to the normal eating condition.An intention to treat-analysis was applied (Gupta, 2011), which means that participants were analyzed irrespective of a strict adherence to the study protocol.Statistical comparisons were conducted with t-tests for dependent measures and the level of significance was fixed at p < .05(two-tailed).Cohen's d was calculated to quantify effect sizes.

Results
Table 1 shows M and SD for all variables as well as statistical comparisons and effect sizes.

Blood glucose levels and perceived hunger
In a first set of analyses, we were interested to verify attenuated blood sugar levels and increased feelings of hunger in the fasting condition relative to the normal eating condition.Therefore, a series of ttests for dependent samples (fasting vs. non-fasting) were calculated via SPSS (vers.29; IBM SPSS Statistics, Armonk, NY, USA).Blood glucose levels showed a trend towards lower levels in the fasting condition as compared to the normal eating condition (t(35) = 1.75, p = .088,d = 0.29).Furthermore, feelings of hunger were significantly higher in the fasting condition than in the normal eating condition (t(35) = 10.94,p < .001,d = 1.82), indexing a large effect.

Cardiovascular activity
Further t-tests indicated that HR was significantly lower and RMSSD significantly higher in the fasting condition than in the normal eating condition, thus confirming hypothesis 1.Effect sizes were medium (RMSSD: d = 0.63) to large (HR: d = 0.92).There were no significant differences in blood pressure (neither for SBP nor DBP).However, analyses of TPR and CO (exploratory hypothesis 2) revealed compensatory activity on blood pressure.Specifically, while CO was significantly lower in the fasting relative to the normal eating condition, TPR was significantly higher.Both effects were of medium to large size (d ~ 0.80).Testing HP and CD against zero revealed that while CD was very similar in both conditions [M < 0.01, SD = 0.025, t(34) = 0.22, p = .828,d = 0.00], thus aligning with the non-significant effects for blood pressure, HP clearly demonstrated a stronger vascular pattern for the fasting condition as compared to the normal eating condition (t(34) = − 5.10, p < .001,d = − 0.86).

Discussion
The aim of Study 1 was to unveil ANS effects of acute fasting.In accordance with expectations and prior research (e.g., Rominger et al., 2021), a 16 h fast resulted in a trend towards lower blood sugar levels, higher feelings of hunger, lower HR and elevated HRV.Moreover, marked hemodynamic differences could be observed.While blood pressure did not change (note that CD was also comparable between conditions!), fasting led to a strong shift towards higher TPR to compensate for diminished CO (resulting from lower HR) to keep blood pressure constant.These findings highlight the pronounced effects of fasting on autonomic activity, thus confirming previous research (e.g., Flasbeck et al., 2021;Gonzalez & Cooke, 2022;Rominger et al., 2021;Young & Landsberg, 1977).The medium to large-sized changes in ANS activity hint towards a comparably strong and dynamic change in cardiovascular activity, which could increase the salience of bodily signals, ultimately facilitating cardiac interoception.Some limitations warrant further discussion.First, blood glucose Note: RMSSD = root mean square of successive differences, SBP = systolic blood pressure, DBP = diastolic blood pressure, CO = cardiac output, TPR = total peripheral resistance.Schwerdtfeger and C. Rominger levels were not significantly different between conditions, thus questioning adherence to the study protocol.Blood glucose levels usually decline during acute fasting by about 20-30% (Stratton et al., 2022), while in Study 1 the decline was rather modest (~5%) and blood glucose was generally in the range of normal fasting glucose levels.However, previous studies applied a larger time span > 20 h, which seems to boost effects (for a review, see Stratton et al., 2022).It should also be mentioned that glucose release from the liver compensates for a drop in glucose from caloric intake (e.g., Wasserman, 2009).Nonetheless, considering the strong effect for perceived hunger together with the small to medium-sized effects for blood glucose and the ratings of past eating behavior, we are rather confident that fasting was performed as requested.Certainly, a more elaborate, behavioral measure of adherence to the protocol would be desirable in future studies.Second, although the study confirmed acute effects of fasting on ANS functioning, it remains unknown if findings could translate to a prolonged intermittent fasting regimen.For example, it seems plausible that organismic changes somewhat normalize with prolonged fasting episodes.
Irrespective of the above-mentioned limitations, Study 1 suggests that a 16 h fasting episode provoked profound cardiovascular changes.Indeed, fasting constitutes a quite severe (in principle, life threatening) condition, warranting bodily adjustments to ensure survival (e.g., Wang et al., 2006).Hence, from a predictive coding perspective (e.g., Ainley et al., 2016;Petzschner et al., 2021), fasting initiates a strong deviation of sensory information from the prior leading to severe prediction errors, thus initializing motor programs (get some food), feeling more aroused ("hangry"; MacCormack & Lindquist, 2019) and upregulation of (noisy) sensory data.This is the moment when internal bodily signals should get more salient, thus presumably facilitating interoception.Allocating attention to bodily signals could then optimize precision of the sensory input, thus updating the generative model.

Introduction
In Study 2, we aimed to examine cardiac activity (i.e., HR and RMSSD) and cardiac interoception in everyday life in individuals undergoing a fasting regimen for 3 consecutive days.Specifically, based on previous laboratory research (Rominger et al., 2021) and the findings of Study 1 (shift in ANS activity resulting from intermittent fasting), we hypothesized that interoception (i.e., both interoceptive accuracy [IA] as an objectifiable measure of interoception and interoceptive sensibility [IS] as a subjective evaluation of interoception; see e.g., Garfinkel et al., 2015) should increase during a fasting regimen as compared to normal eating.The study was preregistered via OSF (https://osf.io/ymuwj/).Again, we used a within-subjects repeated measures design, during which participants followed an intermittent fasting and a normal eating regimen for 3 days each, one week in-between, and in randomized order.Importantly, in order to ensure ecological validity, we applied an EMA design using the novel Graz Ambulatory Interoception Task (GRAIT; Rominger & Schwerdtfeger, 2023).This task is based on the heartbeat tracking paradigm developed by Schandry (1981) and requests individuals to count their heartbeats following smartphone prompts multiple times a day and to rate how well they perceived their heartbeats.This allows to assess IS (i.e., subjective belief to perceive bodily signals) and IA (i.e., correspondence between recorded and perceived heartbeats; Garfinkel et al., 2015) in diverse everyday life contexts.First, and in accordance with Study 1, we expected that HR would be lower and RMSSD to be higher during the fasting episode (hypothesis 1).Second, we expected that during the fasting regimen both IA (hypothesis 2a) and IS (hypothesis 2b) would be higher as compared to traditional eating days, because of the induced ANS alterations as revealed in Study 1.Moreover, we explored whether individuals benefit differently from fasting by comparing interoception between individuals showing higher and lower RMSSD, respectively.Importantly, lower RMSSD has been associated with generally poorer interoception (e.g., Lischke et al., 2021), and a fasting-induced shift in HRV should be particularly pronounced in individuals with a less flexible, more steady beating heart.Of note, this hypothesis is rooted in the predictive coding framework of interoception (e.g., Barrett & Simmons, 2015;Petzschner et al., 2021), which posits that a fasting-induced increase in RMSSD should particularly challenge priors and increase prediction errors in individuals with low RMSSD.Thus, we aimed to examine whether the effects of fasting on interoception would be moderated by the general (trait) level of RMSSD (see also Rominger et al., 2021) and expected that fasting-induced benefits in interoception would be stronger in individuals showing comparably low RMSSD (hypothesis 3).

Participants
In order to estimate the required sample size, a power analysis was performed using the online tool of Murayama et al. (2022) to derive power for mixed effects models using sample size and summary statistics (e.g., t-values) from prior work.Based on the findings of Study 1 on HR and RMSSD as well as on findings from Rominger et al. ( 2021) on cardiac activity and interoception during fasting and non-fasting, with comparably medium to large-sized effects relying on moderate sample sizes, we arrived at an optimal level 2 sample size between N = 32 and N = 36, applying an alpha level of.05and a power of.80(OSF; htt ps://osf.io/ymuwj/).We strived for the higher bound of the sample size estimate and collected data in 40 individuals of male (n = 10) and female sex (n = 30).The average age of the sample was 21.90 years (SD = 2.21) and the majority were students (95%).There were 4 smokers (10%) and 68% reported regular physical exercise.Similar to Study 1, the majority of the sample (78%) reported no prior experience with intermittent fasting.Participants were recruited via e-mail newsletters, social media and lectures at the university.No monetary compensation was offered, but students received course credits if applicable.Due to missing variables or excessive artefacts, final sample size for analyses varied between N = 36 and N = 40.

Preregistration
The core theme of this preregistered project (OSF; https://osf.io/ymuwj/) was on the effects of a fasting intervention on interoception and perceived stress.Measured variables included IA, IS, HR, HRV (i.e., RMSSD), but also perceived stress and affective wellbeing.In this report, we concentrate on interoception and cardiac function.Notably, hypothesis 3 (the moderating effect of RMSSD on the relationship between fasting and interoception) was not preregistered.

Design
An ambulatory within-subjects repeated measures design was applied using EMA.Data were sampled during 3 consecutive days of a fasting regimen (16 h of fasting followed by 8 h of normal eating; 16:8) and 3 consecutive days during normal eating.During fasting no caloric drinks and no food was allowed.Hence, non-caloric drinks like water, tea or coffee without milk or sugar could be consumed during fasting.Conditions were separated by 1 week and covered the very same days (Tuesday to Thursday) to keep daily routines constant.The order of conditions was randomized (using a random number generator) with 50% of the sample starting with intermittent fasting and 50% starting with normal eating.The study was approved by the local ethics committee (GZ.39/95/63 ex 2021/22) and none of the data presented here has been published previously.

Instruments and measures
We applied the EcgMove4 (movisens GmbH; Karlsruhe, Germany) device to record the ECG and bodily movement.For measuring the ECG, a breast belt was used with disposable electrodes to ensure good signal quality (modified Einthoven-II lead at the right collar bone and low left rib cage; 1024 Hz).We preprocessed the physiological data in adjacent 1-min segments by means of the movisens Analyzer software, which controls for potential artifacts (Rominger & Schwerdtfeger, 2022;Schwerdtfeger & Rominger, 2021).R-peaks were automatically detected (R-R intervals), which allows to count the number of heartbeats for the specific time-intervals delivered (Rominger & Schwerdtfeger, 2023).
Bodily movement together with changes in air pressure allows calculation of metabolic equivalents (METs).Importantly, since HR and HRV are strongly dependent on metabolic changes (e.g., Perini & Veicsteinas, 2003), METs must be treated as important confounds in this research.
For assessing psychological and contextual variables, the movi-sensXS app was used (movisens GmbH; Karlsruhe, Germany).It should be mentioned that movisensXS currently only runs on Android operating system.Hence, in order to enable participation of individuals using other operating systems, smartphones running on Android were handed out for the purpose of this study, if necessary.This applied to 24 participants.Two interoception intervals of the GRAIT as well as several items were presented randomly spaced across each recording day (about 10 prompts within 10 h; 9 am to 7 pm) asking for perceived hunger (single item Likert-type measure; 1 = not at all, 5 = very much), stress and affective wellbeing (not of interest here) and momentary context (location: at home, at work, at the university, at another place, travelling via bike, car, public transport).In order to facilitate interpretation of effects, location was categorized into at home vs. other place vs. in transit prior to analyses.Finally, the confidence of perceived heartbeats during the GRAIT served as a measure of IS (see below).

Interoception
Interoception is a multidimensional concept comprising of IS (subjective belief to perceive bodily signals), IA (objectifiable assessed skill to perceive bodily signals) and interoceptive awareness (meta-cognitive awareness of interoceptive accuracy; Garfinkel & Critchley, 2013;Suksasilp & Garfinkel, 2022).We aimed to assess both IA and IS as the main indicators of interoception on a momentary basis using the GRAIT (see Rominger & Schwerdtfeger, 2023).

Graz Ambulatory Interoception Task (GRAIT)
The GRAIT applies two out of three time intervals at each prompt (15 s, 20 s, 35 s), during which participants are asked to count and report the number of perceived heartbeats (without manually taking pulses).Instructions emphasize that only heartbeats that are actually perceived should be counted (Desmedt et al., 2018).Actual heartbeats are recorded during the same time intervals and IA is calculated as percentage of perceived heartbeats (see Rominger & Schwerdtfeger, 2023;Rominger & Schwerdtfeger, 2024).
Both between-person (R KRn =.99) as well as within-person (R Cn =.68) reliability of IA of the task were good and in accordance with the original study (Rominger & Schwerdtfeger, 2023).Furthermore, we assessed IS after each trial by asking participants to rate how well they performed the task (visual analogue scale from 0 [poor] to 1 [excellent] with 0.01 resolution).This served as a momentary measure of IS (e.g., Garfinkel et al., 2015) and showed good between and within-person reliability (R KRn =.98, R Cn =.77).

Procedure
Participants were pre-screened via an online questionnaire to verify exclusion criteria (cardiovascular and mental disorders, cardiovascular or psychotropic medication).Potential candidates were then invited to partake in the study and informed about the study protocol.After signing informed consent, each participant was introduced in detail to the study conditions and their order.Participants were explicitly informed about the possibility to withdraw participation any time without giving a reason.They underwent a 16:8 fasting regimen for 3 consecutive days and a normal eating condition, separated by one week.
Condition order was randomized.Each day ~10 prompts were delivered and recording was restricted to a time frame between 9 am and 7 pm, covering Tuesday morning to Thursday evening.Nocturnal recordings were not obtained.In the fasting condition, participants were asked to have their last meal at 8 pm the latest and continue fasting for 16 h (~ noon the next day, 12 pm).Thus, in the fasting condition, the last meal was taken already on Monday at 8 pm the latest and normal eating was allowed from 12 pm to 8 pm.Random (acoustic) alarms were emitted for 10 s with the app visible for 50 s on the display.A maximum delay of responses of 20 min was allowed and alarms could be dismissed if responding to the EMA was not suitable in a given moment.On average, 34.24 prompts (SD = 4.78) were emitted per participant during the fasting condition and 33.61 (SD = 3.00) during the normal eating condition (t(37) = 0.90, p = .37), of which M = 7.61 (SD = 6.97) were missed during fasting and M = 7.45 (SD = 5.77) during normal eating (t (37) = 0.18, p = .86).There were no delayed prompts.Hence, adherence to the prompts were comparable between conditions (fasting: 78%, normal eating: 76%).In total, 1906 complete observations were available for perceived hunger, 1798 for HR and 1675 for RMSSD.We further assessed if participants who received a separate smartphone for the study exhibited different adherence rates.Noteworthy, missed alarms were comparable between participants (Android phone for loan: M = 6.08,SD = 5.94, no loan: M = 9.53, SD = 8.17; t(37) = 1.53, p = .14).

Data parametrization and analyses
We applied robust linear mixed effects modeling (package robustlmm; Koller, 2023) using R (R Core Team, 2023) to account for potentially biasing outliers and applied an intention to treat-approach (Gupta, 2011), meaning that all participants were analyzed in both conditions, irrespective of strict adherence to the study's protocol.Several models were calculated to predict perceived hunger, HR, RMSSD, as well as IA and IS, respectively.The first models on hunger and cardiac activity aimed to evaluate the efficacy of the fasting manipulation.Continuous within-person predictor variables were group mean centered and between-person variables grand mean centered.In order to account for the fasting interval within the fasting condition (16 h fasting vs. 8 h eating), a variable "Time" was included with 8 pm to 12 pm the next day set to zero (quantifying the fasting time; non-eating) and 12 pm to 8 pm set to one (during which eating was allowed; i.e., eating window).The level of significance was fixed at p < .05(two-tailed).

Psychological and cardiac changes during fasting
In a first model, we examined whether conditions differed with respect to perceived hunger.In particular, we aimed to predict perceived hunger by condition (fasting intervention vs. normal eating) as well as time (i.e., time between 8 pm and 12 pm as non-eating, and 12 pm to 8 pm as eating window).Precisely, each prompt was coded as "0" when it fell between 8 pm and 12 pm or "1" when it fell between 12 pm and 8 pm.Indeed, perceived hunger was significantly higher during the fasting condition as compared to the normal eating condition (b = 0.76, p < .001)and this effect was further qualified by a condition by time interaction (b = − 0.80, p < .001).Further simple slope analyses confirmed that perceived hunger was meaningfully higher during the non-eating phase (from 8 pm to 12 pm) than during the eating phase (from 12 pm to 8 pm) (b = − 0.96, p < .001),while this time effect was smaller in the normal eating condition (b = − 0.16, p = .020).Importantly, conditions differed significantly during 8 pm to 12 pm (b = 0.76, p < .001),but not during 12 pm to 8 pm (b = − 0.04, p = .540),thus documenting more hunger between 8 pm and 12 pm when fasting was actually performed in the fasting condition than during the normal eating days, thus verifying study compliance (Table 2).
Next, in order to test hypothesis 1, we analyzed whether cardiac activity differed in response to eating (8 h) and no eating (16 h) episodes during the fasting regimen after controlling for several confounds (see Table 3).Hence, we analyzed condition by time effects for cardiac activity, which were significant for both HR (b = 2.61, p = .018)and RMSSD (b = − 0.18, p = .018).HR was lower and RMSSD was higher during non-eating within the intermittent fasting condition, thus supporting the findings of Study 1. Simple slope analyses confirmed that during the fasting days HR dropped from eating (12 pm -8 pm) to noneating (8 pm -12 pm) by about 2 BPM (b = − 2.25, p = .004),while during the normal eating days HR did not change between these time windows (b = − 0.36, p = .644).In a similar vein, RMSSD did not change in time for the normal eating condition (b = − 0.02, p = .567),but increased significantly from eating to non-eating during the fasting regimen (b = 0.08, p = .006),thus documenting a fasting-induced shift in cardiac activity.

Interoception
In the next series of models, both IA and IS were predicted by condition to test whether the fasting regimen in general affected interoception (hypotheses 2a and b).Although during fasting, IA (b = 0.02, p = .074)was not significantly different between conditions, IS (b = 2.10, p = .019)was significantly higher than during the normal eating protocol, thus partially verifying expectations.However, further analyses revealed that effects were dependent on location, indicating that better performance during fasting could be explained by staying more at home (and presumably, having less distractions) as compared to other locations.Thus, neither IA (b = 0.01, p = .262)nor IS (b = 1.50, p = .092)significantly differed between conditions after controlling for location.The effect of location, in turn, suggested that both IA and IS deteriorated when being at another location or in transit as compared to being at home (IA: at home vs. other location: b = − 0.03, p = .040;at home vs. in transit: b = − 0.06, p < .001;IS: at home vs. other location: b = − 5.30, p < .001;at home vs. in transit: b = − 6.95, p < .001),thus conforming previous findings (Rominger et al., 2023).
Importantly, due to our assumption that the boosting effect on RMSSD will particularly benefit individuals with low RMSSD (hypothesis 3), we further analyzed whether a comparably low general level of RMSSD (considered as a trait variable; grand mean centered RMSSD) was associated with a more pronounced increase in interoception due to fasting.In line with the predictive coding framework of interoception (e. g., Ainley et al., 2016;Petzschner et al., 2021) one could suppose that in individuals with a low trait HRV, higher prediction errors occur, because fasting-induced alterations in the ANS challenge priors.To verify this assumption, we first analyzed whether the dynamic shift in RMSSD from not eating to eating episodes within the fasting regimen was more pronounced in individuals showing generally lower trait RMSSD.Mixed effects models controlling for age, sex, location, and METs indeed showed a significant 3-way interaction between trait RMSSD, condition and time (b = 0.23, p = .039).Simple slope analyses documented that individuals with lower RMSSD (− 1 SD) showed a reliable increase in RMSSD from eating to fasting window during the fasting days (b = 0.13, p = .002),while participants with higher RMSSD (+ 1 SD) did not (b = 0.03, p = .469).
In a next step, we aimed to predict IA and IS by an interaction term between condition and trait RMSSD, as well as other important covariates (Table 4).It turned out that for both IA and IS the interaction between condition and trait RMSSD was significant (IA: b = − 0.005, p = .030;IS: b = − 8.40, p = .001).Post hoc-simple slope analyses documented that for individuals showing a generally lower RMSSD at trait level, the condition effect was significant for both dimensions of interoception (IA: b = 0.02, p = .042;IS: b = 4.45, p = .001),thus indicating that interoception increased during fasting as compared to normal eating days.For individuals showing comparably high RMSSD, however, no significant changes between conditions could be observed (IA: b = − 0.01, p = .298;IS: b = − 1.70, p = .192).Importantly, this pattern of findings could not be attributed to momentary RMSSD per se (note that momentary RMSSD predicted IA by b = 0.006, p < .001,but not IS, b = − 0.68, p = .495).

Discussion
The aim of Study 2 was to verify differences in interoception between intermittent fasting and normal eating days, respectively, applying an EMA approach across two weeks.Although we did not obtain blood sugar levels in this study, analysis of perceived hunger and cardiac activity verified that adherence with the study's conditions (fasting vs. normal eating) was likely.Perceived hunger was higher, HR was lower and HRV was higher during the time of acute fasting.Importantly, we found some evidence for better everyday life interoception in the cardiac domain during the intermittent fasting regimen as compared to normal eating, thus supporting earlier laboratory research (Herbert et al., 2012;Rominger et al., 2021).However, this main effect shrinked when controlling for location.Hence, despite the very same recording days, it became evident that participants adjusted to the fasting regimen by narrowing the array of contexts.This demonstrates that fasting not only modulates physiology, cognition, and priors, but also behaviora finding laboratory studies would not have detected.Importantly, and in line with the predictive coding framework of interoception (e.g., Owens et al., 2018;Petzschner et al., 2021), exploratory analyses found that individuals with a generally lower RMSSD seemed to have particularly benefitted from fasting as became evident by increases in IA and IS, respectively.As further analyses revealed, there was evidence for a greater increase in RMSSD to fasting within these participants, thus generally confirming previous findings on a relationship between changes in ANS activity (and presumed vagal efference in particular) and interoception (Lischke et al., 2021;Rominger et al., 2021;Villani et al., 2019).Finally, this study also suggests the validity (and reliability) of the recently introduced GRAIT as an ambulatory tool for assessing interoception in everyday life (Rominger & Schwerdtfeger, 2023), as it confirms the interoception promoting effects of fasting using conventional laboratory protocols (Herbert et al., 2012;Schulz et al., 2015).

Fasting increases prediction error
According to the Embodied Predictive Interoception Coding (EPIC) model (Feldmann Barrett & Simmons, 2015), interoception seems to reflect the brain's expectation of sensory input (interoceptive inference), rather than sensory input itself.Hence, non-precise interoceptive priors could increase the salience of interoceptive cues, which should facilitate interoception (e.g., Ainley et al., 2016;Feldman Barrett, & Simmons, 2015).In line with this reasoning, the findings of Study 2 confirmed the cardiac alterations induced by fasting (see also Rominger et al., 2021 for similar findings in laboratory research), which should shift the interoceptive posterior in the direction of the likelihood (i.e., bodily signals), ultimately increasing interoceptive performance.Notably, however, a fasten-induced interoceptive performance increase was largely restricted to individuals with more rigid cardiac regulation (i.e., lower trait HRV).Thus, the fasting-induced challenge of ANS function in these individuals might have increased prediction errors, thus making organismic signals more salient (e.g., Ainley et al., 2016;Barrett & Simmons, 2015;Petzschner et al., 2021).Accordingly, low trait HRV (i.e., RMSSD) participants showed comparably strong increases in RMSSD, potentially indicating stronger cardiac vagal regulation, which might have challenged priors, thus leading to more accurate interoception.In sum, the findings are compatible with the view that in case of imprecise priors, visceral-sensory data get upregulated and prediction errors could thus be reduced (Ainley et al., 2016).

Limitations
Despite the promising aspects of this study, there are also several limitations of this research that need to be emphasized.First, although the study was preregistered and sample size was determined a priori based on expected effect sizes, replication studies with more diverse, less educated individuals are certainly warranted to evaluate the trustworthiness of the effects.Especially the interaction effect of condition and trait RMSSD needs replication in larger samples.Second, as the GRAIT is based on the well-known heartbeat tracking task (Schandry, 1981), it should be mentioned that this task has been criticized for several reasons (e.g., cognitive strategies, like time estimation could explain performance; see e.g., Brener & Ring, 2016;Corneille et al., 2020;Jones, 1994;Zamariola et al., 2018; but see Ainley et al., 2020;Zimprich et al., 2020).
Although not every criticism applies to the within-person perspective offered by the GRAIT (see Rominger & Schwerdtfeger, 2023 for a more detailed discussion), this task is not immune against non-interoception related cognitive influences.Hence, alternative tasks that could be applied in everyday life are certainly warranted (e.g., Ponzo et al., 2021).Third, although ratings of perceived hunger could be seen as validation of the conditions (together with the alterations in cardiac variables), a direct measure of adherence to the protocol was not applied (e.g., blood glucose measurement, eating protocol).Hence, more direct measures of adherence should be implemented in future research.Nonetheless, we want to emphasize that the study followed an intention to treat-approach, which provides more conservative, less biased effect estimates (Gupta, 2011).Intention to treat-protocols take into account that some individuals do not fully adhere to the protocol, which offers a more naturalistic, ecologically valid view on the effects of interventions.

General discussion and conclusion
Two studies assessing autonomic function and interoception, one of them conducted in the laboratory and one of them in everyday life, provided findings consistent with the predictive coding model of interoception (e.g., Ainley et al., 2016;Barrett & Simmons, 2015;Owens et al., 2018;Petzschner et al., 2021).Fasting-induced changes in ANS activity seem to facilitate attention to organismic cues and individuals showing lower RMSSD seem to particularly benefit from this regimen.Although fasting has considerable health benefits, it also induces considerable physiological (and behavioral) changes.First, as both studies revealed, a fasting episode seems to strongly alter autonomic function (see also Flasbeck et al., 2021).The shift in ANS activity, in turn, seems to have induced a stronger prediction error leading to an increase in cardiac interoception, particularly in individuals showing lower trait RMSSD (possibly indicating a lower cardiac vagal tone; Study 2).Together, both studies support and extend previous findings on a relationship between ANS alterations and interoception (e.g., Herbert et al., 2012;Lischke et al., 2021;Rominger et al., 2021).Furthermore, the behavioral effects of fasting warrant some elaboration.As revealed by Study 2, fasting seemed to have influenced context by showing that participants were more at home as compared to other locations during the fasting condition.This seems reasonable, as going out with colleagues or friends for lunch or dinner etc. might not be that attractive when fasting.While staying at home may reduce environmental distractions, bodily signals might be perceived more accurately.Although this contextual shift could explain differences in cardiac interoception, thus marginalizing the net effects of fasting, behavioral adjustments due to fasting in everyday life could still be considered important catalysts for interoception.Further research is certainly needed to evaluate the long-term stability of these adjustments.
From an applied perspective, the findings of both studies highlight the power of a fasting on psychophysiological functioning.Although the beneficial long-term effects of fasting are beyond doubt (e. g., Longo et al., 2021), starting with fasting obviously challenges the organism and the brain's internal model of the body.Future studies are encouraged to examine long-term adjustments to inform about chronic fasting effects on the brain and the body as well as if and how prior fasting experience modulates the physiological and interoceptive processes found in this research.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 1
Descriptive statistics of the main variables and differences between conditions.

Table 2
Predicting hunger from condition.

Table 3
Predicting heart rate (left) and heart rate variability (RMSSD, right) by condition and time.