Effect of Propofol on Heart Rate and Its Coupling to Cortical Slow Waves in Humans

Thirty-two–channel EEG and single-channel electrocardiogram were collected in n = 16 healthy subjects (8 female; age, 28.6 ± 7 yr; table 1) during slowly increasing intravenous infusion of propofol up to an estimated effect-site concentration of 4 µg/ml using the Marsh pharmacokinetic model.⁸  The experiment was separated into four main periods: 10 min awake, 48 min induction, 10 min peak anesthesia, and 48 min emergence. Informed written consent was obtained from all participants; details of this experiment have been published previously.⁸ 

EEG data preprocessing was carried out with BrainVision Analyzer version 2.1 (BrainProducts GmbH, Germany), custom written MATLAB code (Matlab 2019b, Math Works Inc., USA), and the EEGLAB (v2019.1) analysis toolbox. The EEG and electrocardiographic data were re-referenced to the common average of signals from all EEG channels. This was done as theoretical reasons suggest scalp average to be a robust null reference that decreases volume conduction effects.²¹  Independent component analysis and bad channel rejection were performed to remove EEG data with blinks and ocular movements. EEG data were band-pass–filtered with a phase-preserving third order 0.5- to 45-Hz Butterworth filter. EEG data were downsampled to 100 Hz and electrocardiogram to 500 Hz.

HR, electrocardiogram waveform templates, and R-wave amplitudes were extracted using the BioSPPy toolbox (https://github.com/PIA-Group/BioSPPy/), which uses Hamilton segmentation.²²  This was used on each subject to identify individual R-wave peaks, the HR, and electrocardiogram waveform templates, which were subsequently Spearman-correlated to the propofol effect-site concentration.

In order to explore correlates of autonomic activity, standard HR variability metrics were extracted for 5-min segments in each subject using the pyHRV toolbox²³  and Spearman-correlated with propofol effect-site concentration at the group level. These HR variability metrics included root-mean-square successive difference between R-wave peaks as well as other frequency domain metrics. Specifically, the frequency domain metrics used were the ratio of low-frequency (largely sympathetic, 0.04- to 0.15-Hz) and high-frequency (parasympathetic, 0.15- to 0.4-Hz) HR variability and the peak frequency in the high-frequency band. The low- to high-frequency ratio is thought to index the balance between sympathetic and parasympathetic activity, with a low low-frequency/high-frequency reflecting parasympathetic dominance, although this simple interpretation has been challenged.²⁴ 

Slow-wave activity was defined as the Fourier power in the 0.5- to 1.5-Hz band on the frontal Fz (international 10-20 system) channel, as is common convention.⁸  The correlation between slow-wave activity and HR/effect-site concentration and its significance was found using Spearman correlation in 5-min segments. Individual slow waves were then identified using standard methodology based on amplitude and duration thresholding implemented in the yasa toolbox.^15,16,25  In brief, each slow wave had to have amplitude in the 99th percentile of the 0.5- to 4-Hz amplitude and negative duration between 0.25 and 1.25 s. This wider filter is commonly chosen to capture more details of the nonsinusoidal slow wave shape,²⁶  although proposals for multiple slow-wave types exist.²⁷  Slow-wave onset was defined as the initial downward zero crossing, and slow-wave frequency was extracted as the inverse of the slow-wave period.

Once heartbeats and slow waves were identified, we aimed to test whether heartbeats occurred at preferential times in the slow-wave cycle. For each slow wave detected, the time delay relative to the slow-wave start (initial downward zero crossing) was noted for eight heartbeats closest to it. Eight beats were chosen as this window length fully covers a slow wave. This resulted in eight R-wave to slow-wave intervals, using methodology similar to previous cardiorespiratory analyses.^28,29  We wanted to know if electrocardiogram R-wave to EEG slow-wave timings were distributed randomly or in phase with the slow-wave onset. For robustness, this was tested against a surrogate null distribution in several ways. First, we utilized the same method that has previously been used to study cardiorespiratory coupling.³⁰  This method compares the R-wave to slow-wave_-1 interval (i.e., the time interval between slow-wave start and the preceding R-wave peak) to a uniformly random null distribution. Starting from the beginning of each subject’s set of R-wave to slow-wave_-1 values, we used a moving window of 40 slow waves, and placed the corresponding R-wave to slow-wave_-1 intervals in a 10-bin histogram with outer limits of 0 and the mean heart period for that window. From the histogram, the proportional Shannon entropy is calculated as follows:

where P_b is the histogram probability of bin b and N is the number of histogram bins.

During perfect coupling, all R-wave to slow-wave_-1 intervals fall into one bin, and SH_P = 0. In the absence of coupling, R-wave to slow-wave_-1 intervals are distributed randomly, producing maximum entropy with SH_P = 1. For each subject, the mean SH_P across the whole experiment was computed. To determine a significance threshold, SH_P was computed for n = 10,000 surrogate series of 200 random numbers each, drawn from a uniform distribution between 0 and 1 (mean HR of 60 beats/min). The 0.1st percentile was used to indicate significance at the P = 0.001 level (SH_P = 0.970). Additional methods testing autocorrelation of the R-wave to slow-wave histogram and more complex surrogates were used to further verify robustness of this result (see Supplemental Digital Content, https://links.lww.com/ALN/D329).

Additionally, for each slow wave identified, ± 2 s of EEG and electrocardiogram activity was saved around the slow-wave start. This was then averaged across slow waves and subjects to reveal any coherent electrocardiogram patterns during a slow wave.

In order to carry out a preliminary exploration of whether our HR results could be replicated in clinical data, we performed a post hoc analysis of HR and drug concentrations in n = 96 patients collected as part of the AlphaMax study (median age, 74 yr [range, 61 to 86 yr]; 66 men; ASA Physical Status II/III; variety of procedures; table 1). This dataset contained EEG, drug concentration, HR, and demographic data. Unfortunately, individual electrocardiogram waveforms were not available in this dataset, so we could not determine the HR variability or corticocardiac coupling analyses.

The AlphaMax study patients received a standardized desflurane- and fentanyl-based maintenance general anesthesia that was titrated to maximize the EEG alpha power in the intervention group. For each patient, HR and drug concentrations (propofol, fentanyl, desflurane) were sampled—or estimated using population based pharmacokinetic models—every 5 s. The HR was smoothed with a 2-min moving median window to suppress artifacts, and any HR greater than 250 beats/min or less than 10 beats/min was not used. A large mixed-effects general linear model was constructed with HR, drug concentrations, and demographic variables as regressors. Specifically, the fixed effects of propofol, fentanyl, and desflurane (plus their linear interaction terms), as well as age, body mass index, ASA Physical Status (II or III), and sex were studied. A random effect of each individual’s mean HR was included. In summary, the model equation was

HR [beats/min] = β₀ + β₁ × age + β₂ × body mass index + β₃ × ASA + β₄ × (sex = female) + β₅ × prop + β₆ × fent + β₇ × des + β₈ × prop × fent + β₉ × prop × des + β₁₀ × fent × des + (1 | patient number),

where β_i are the model coefficients, and prop = propofol [µg/ml], des = desflurane [%], and fent = fentanyl [ng/ml] effect-site concentrations. To compare possible corticocardiac interactions with the propofol dataset, individual slow waves were extracted from intraoperative EEG (from first incision to the end of closing up) and mean slow-wave frequency per subject extracted and compared with the mean HR.

As these analyses were all post hoc analyses of previously collected and published data, no power calculation was done. Spearman correlations and their P values were used to test associations between electrocardiogram/EEG parameters (HR, R-wave amplitude, root-mean-square successive difference, low-frequency to high-frequency ratio, peak high frequency, slow-wave power) and propofol concentration. Repeated-measures ANOVA was performed on electrocardiogram/EEG-derived parameter traces in 5-min segments to further test for significant changes. For display purposes, mean ± standard error across participants is shown, except where the data were not normally distributed (as tested with D’Agostino and Pearson’s test). In these nonnormally distributed cases, median ± bootstrapped 95% CI (10,000 iterations) are shown. Mean ± SD (or median [25th, 75th percentile]) are given in the text. Significance was set at the P = 0.05 level unless otherwise specified. All custom code used in this study is available at https://gitlab.com/marcoFabus/fabus2022_brain_heart.

First, we tracked the HR and time series electrocardiogram properties across an ultraslow propofol induction and emergence in n = 16 healthy volunteers (fig. 1). At higher propofol doses, we observed a shortening of the QT segment and decrease in R-wave amplitude (fig. 1A). In every subject, the HR increased and very robustly tracked the propofol dose with Spearman correlation of ρ = 0.923, P < 0.001 (fig. 1B). Mean HR across volunteers increased from 58.2 ± 10 beats/min at baseline to 73.4 ± 8.8 beats/min at peak anesthesia, equivalent to an increase of 4.2 ± 1.5 beats/min/(μg · ml⁻¹). The maximum effect size comparing HR at baseline and peak propofol was Cohen’s d = 1.546. A linear regression showed the HR and propofol relationship to be HR [beats/min] = 56.1 (54.9 to 57.2) + 4.23 (3.75 to 4.80) × propofol [µg/ml], where brackets show 95% CI. Similarly, the R-wave amplitude was also strongly inversely correlated with the propofol effect-site concentration (Spearman ρ = −0.902; P < 0.001; fig. 1C). R-wave amplitude decreased from 966 [707, 1,133] μV at baseline to 742 [627, 1,068] μV at peak anesthesia, equivalent to a decrease of –83 [–245, –28] μV.

Next, we studied autonomic activity through HR variability (fig. 2). The root-mean-square successive difference between heartbeats, which indexes parasympathetic tone, decreased in proportion to propofol concentration, and rebounded on emergence (fig. 2A; Spearman ρ = −0.785; P < 0.001; Cohen’s d = 1.296 for baseline vs. peak concentration). This was confirmed by a repeated-measures ANOVA across subjects with significance P < 0.001.

With regard to the frequency domain metrics, the low- to high-frequency ratio showed higher between-subject variability, but the group average confirmed the shift toward a relative predominance of sympathetic activity with increasing propofol concentration (fig. 2B; Spearman ρ = −0.763; P < 0.001). The associated repeated-measures ANOVA result also showed a significant change (P = 0.003; Cohen’s d = 0.539) between baseline and highest propofol concentration. The peak frequency in the high-frequency parasympathetic HR variability range also tracked with propofol concentrations (supplementary fig. 7, https://links.lww.com/ALN/D329; Spearman ρ = 0.885; P < 0.001; repeated-measures ANOVA P < 0.001).

Cortical activity during propofol anesthesia is known to be associated with non-REM deep sleep–like slow-wave activity (fig. 3). We first confirmed the previous finding of saturation of frontal slow-wave activity with propofol dose at the Fz electrode⁸  (fig. 3A). However, more strikingly, this slow-wave activity increase correlated very strongly with the increasing HR (fig. 3B; Spearman ρ = 0.910; P < 0.001).

The identified correlation between slow-wave activity and HR, as well as previous literature describing their coupling in sleep, led us to focus on quantifying the presence of any time-related coupling between individual slow waves and heartbeats. The methodology is illustrated on the single-subject level in figure 4. After identifying individual slow waves, we studied the distribution of the eight closest heartbeats around each slow-wave onset. If there was no coupling, it would be expected that R-wave to slow-wave intervals should follow a uniform probability distribution, and that a time-averaged electrocardiogram would converge on a horizontal line around zero. As observed in previous work on cardiorespiratory coupling,²⁸  the distribution of time intervals between electrocardiogram R-waves and EEG slow waves (R-wave to slow-wave intervals; fig. 4B) was nonuniform and concentrated around specific phases in the slow-wave cycle. This appeared as a residual low-frequency oscillation in the electrocardiogram, after averaging around the slow-wave onset (fig. 4C), and as peaks in the distribution of heartbeat timings (fig. 4D).

Importantly, this effect was present and significant at the group level (fig. 5). The group-average lag between the electrocardiogram peak and EEG slow-wave onset was 447 [392, 510] ms (fig. 5B). The slow-wave/R-wave coupling, as measured by entropy in relation to a uniform null distribution, was SH_P = 0.866 ± 0.05 (P < 0.001 compared to a uniform null hypothesis). Additional tests to verify this is not a random effect were carried out and can be found in the Supplemental Digital Content (https://links.lww.com/ALN/D329). Furthermore, at the group level, the subjects’ mean HRs and slow-wave frequencies were significantly linearly correlated (Pearson r = 0.519; P = 0.0395).

These analysis results were qualitatively unchanged when EEG data were re-referenced to linked mastoids (supplementary fig. 3, https://links.lww.com/ALN/D329) and when the electrocardiogram was time-locked to slow-wave trough instead of downward zero crossing (supplementary fig. 4, https://links.lww.com/ALN/D329).

In order to explore whether these HR results hold in a clinical setting, we analyzed the association between effect-site drug concentration and HR using a large general linear model with n = 96 older patients collected as part of the AlphaMax trial (table 2; supplementary fig. 5, https://links.lww.com/ALN/D329). After adjusting for age, body mass index, sex, and ASA Physical Status, all agents (propofol, fentanyl, and desflurane) had a significant effect on the HR (P < 0.001). Propofol led to a mild increase in HR, on average, with a coefficient of +1.3 beats/min/(μg · ml⁻¹) (95% CI, 1.1 to 1.5). Fentanyl, however, led to a decrease in the HR, on average −2.6 (95% CI, − 2.7 to −2.5) beats/min/(ng · ml⁻¹), as did desflurane, with average of −1.84 (95% CI, −1.90 to −1.78) beats/min/(1% end-tidal concentration). The interaction terms were also significant, although with smaller coefficients. With mean individual HR included as a regressor, no demographic parameters were significant, suggesting that the drug effects on the HR may be independent of these demographic variables. The effect size comparing HR with propofol less than 0.5 µg/ml and greater than 3 µg/ml was Cohen’s d = 0.796.

Interestingly, at the group level, mean intraoperative slow-wave frequency was not related to the mean HR in this dataset (appendix 1; P = 0.65). Furthermore, desflurane-fentanyl slow-wave frequency was significantly higher than propofol slow-wave frequency (propofol f = 1.01 ± 0.11 Hz; desflurane-fentanyl f = 1.26 ± 0.15Hz (mean ± SD); Welch t test P < 0.0001), as seen before.³¹  Unfortunately, individual electrocardiogram waveforms were not included in these data, so coupling between slow waves and individual heartbeats was not assessed.

In this paper, we have shown that administration of propofol leads to increased mean HR. The ultraslow propofol administration in healthy volunteers led to an increase in the mean HR of roughly +4 beats/min/(μg/ml propofol concentration). In an exploratory analysis of an older patient population, the effect of propofol on heart was about threefold smaller, on average +1.3 beats/min/(µg/ml). These clear and significant mean HR increases confirmed our hypothesis but are surprising in view of the mixed existing literature.

Experimental studies have seen a HR increase across a variety of research paradigms.^7,9–12,32  We contend the lack of HR increase (or HR decrease) with propofol reported in some clinical studies may be due to other drugs given, the patient population, the surgical context, or dose-/rate-dependent effects. Clinically, it is common to administer opioids and other premedication, which can decrease the HR and affect cardiovascular dynamics.^32,33  The smaller HR increase observed in the older clinical population could be due to a previously proposed U-shaped relationship between propofol and HR.¹¹  Older patients have higher anesthetic sensitivity and may be more susceptible to a HR decrease at relatively high propofol concentrations. This is supported by a previous healthy volunteer study where plasma concentrations of about 7.4 µg/ml increased the HR by approximately 30 beats/min, but excessively high concentrations up to mean plasma levels of 18.3 μg/ml reversed the effect and decreased the HR compared to lower concentrations.¹²  Finally, clinical procedures may provide autonomic stimulation, which could affect intrinsic HR increases with propofol.

The ultraslow induction rate used in our healthy volunteer study may also affect cardiac changes. This is supported by previous work finding rate-dependent cardiac effects of propofol with greater HR decrease in fast inductions, perhaps due to a rate-limiting central nervous system distribution process.^19,34  Our HR increase is unlikely to be due to anxiety, as while HR increased from baseline to loss of responsiveness, it continued to increase at the same rate when drug concentration increased beyond the point of loss of consciousness (appendix 2). Finally, while it is plausible some of the HR increase could be due to endothelial irritation, no “pain on propofol injection” phenomenon was reported by volunteers.

The biologic basis for the increase in mean HR may be due to propofol inhibiting cardioinhibitory vagal neurons in the brainstem.¹¹  Studies of propofol’s effect on autonomic cardiac influences have also produced mixed results.

The literature agrees that propofol reduces HR variability,^6,12,35–37  a result also confirmed in our experiment. The distinct sympathetic and parasympathetic contributions to this are less clear. An early study proposed that propofol mostly depresses sympathetic activity and suggested this as a mechanism for propofol bradycardia and hypotension. However, opioids were also used in that study.³⁶  Several studies since, including this one, have concluded that propofol predominantly decreases high-frequency HR variability. This is conventionally thought to reflect a decrease in parasympathetic vagal influences.^6,12,35,37  However, vagal and sympathetic activity tend to be mutually reciprocal, and thus we cannot conclusively show if the vagal decrease is a primary propofol effect or is secondary to sympathetic activation. Notably, however, the low-frequency/high-frequency ratio, traditionally taken as a measure of sympathetic/parasympathetic balance, had a much less consistent relationship with propofol concentration, indicating the sympathetic response was less consistent than the vagal response. We therefore conclude that propofol causes a shift in the autonomic balance to cause tachycardia, probably mainly via the parasympathetic branch, as this relationship was more consistent and more strongly correlated with propofol concentration. This is further supported by our R-wave/slow-wave coupling results. These would not occur through slow sympathetic influences with a lag of up to 10s of seconds.

This increased HR is unlikely to entirely be a reflex tachycardic response to vasodilation, as propofol has been shown to depress the baroreflex.^38,39  Specifically, reflex tachycardic responses to hypertension are reduced by propofol in conditions of both normocapnia and hypercapnia.⁴⁰  In our study, as previously reported,⁸  hypotension was not observed, and subjects had baseline end-tidal CO₂ 39.3 ± 3.3 mmHg and peak end-tidal CO₂ 47.1 ± 5.7 mmHg. A limitation of the current study is the lack of hemodynamic data, as tachycardia associated with decreased stroke volume due to propofol could be part of the HR increase mechanism. Future studies should focus on distinguishing between centrally and hemodynamically mediated mechanisms.⁴¹  A final potential confound is differing anesthetic susceptibility, and future studies could benefit from direct sampling of plasma propofol levels.

We observed dose-dependent changes in electrocardiogram morphology that may reflect both direct cardiac effects and centrally mediated changes of propofol. The R-wave amplitude decrease seen in our study might be related to previously observed propofol effects on ventricular depolarisation.⁴²  Propofol may decrease myocardial contractility, possibly due to a direct propofol effect on myocyte ability to expel intracellular calcium,⁴³  although this may only happen above clinical doses.⁴⁴  A change in the mean electrical axis or direct vagal effect could also explain the R-wave amplitude decrease; findings have been mixed so far.⁴²  Future studies should establish these changes in ventilated patients with more direct cardiac measurements such as cardiac output to assess the clinical significance of this finding.

Low-frequency corticocardiac coupling has been observed in sleep.^16,45  As propofol slow waves show some sleep-like properties,¹⁵  we hypothesized this effect would also be present in anesthesia. We found individual cortical slow waves and cardiac R-waves were coupled as hypothesized. A heartbeat was most likely to precede the slow-wave onset by about 450 ms, a time interval similar to that seen in sleep.^16,45  We did not see any evidence of dose-dependent coupling effects (supplementary fig. 10, https://links.lww.com/ALN/D329).

Importantly, this coupling is nontrivial, as it relates to the phase relationship between individual EEG slow waves and the electrocardiogram, not just two ongoing oscillations that happen to have similar frequencies around 1 Hz. The fact that the mean electrocardiogram line in figures 4D and 5C is not zero indicates a time-linked relationship—analogous to evoked potentials in EEG work. We explored this in detail through a simulation study, showing that our proportional entropy metric is sensitive to genuine coupling (Supplemental Digital Content Section 3, https://links.lww.com/ALN/D329).

Mensen et al. proposed several hypotheses for why this coupling may occur.¹⁶  The first was a possible metabolic constraint. Overall, neurochemical tone favors hyperpolarized down states with heartbeats acting as a stimulus to evoke a down state when neuronal resources are depleted. Lower regional blood flow between heartbeats could have this effect on a few critical neurons, leading to a network change. However, this seems unlikely as the necessary time resolution of changes in metabolic energy demand seems shorter than that of a damped feeder capillary blood flow, coupled with the energy substrate diffusion time and the presence of intrinsic neuronal energy stores. The other possibility is a third generator controlling both the HR and slow-wave genesis. Knowing this effect is present both in sleep and propofol anesthesia suggests a possible nature of this generator. Sleep and anesthesia differ in noradrenaline levels, but both show low acetylcholine levels.⁴⁶  Combined with the brainstem projecting both in a cephalad direction to higher brain areas and caudally to the heart, we propose it as a possible place for a common generator. For instance, the nucleus of the solitary tract or cholinergic pontine nuclei may project both to fast-spiking GABA-mediated interneurons in the thalamus and to medullar regions controlling the HR.^47,48  Given that the thalamus is involved in slow-wave generation in vivo, a brainstem connection could explain corticocardiac coupling, perhaps by weak-coupling synchronization.^49,50  This is supported by subjects with a faster HR also having faster slow-wave frequency. Interestingly, this frequency relationship was not observed during desflurane-fentanyl slow waves (appendix 1), suggesting volatiles may differ in their corticocardiac coupling effects. Further work is needed to explain the relationship between slow waves and cardiac activity, especially as pertains to wider coupling of autonomic and central activity.^17,18  Our proposed common brainstem generator could be ruled out if patients with pacemakers also show this coupling.

In summary, in this observational study, slow propofol administration in healthy subjects robustly led to an increase in mean HR that was strongly proportional to drug concentration, and not influenced by changes in behavioral responsiveness. A preliminary analysis in a larger clinical dataset replicated this result, but with a decreased effect size. The HR increase could be explained with decreased parasympathetic inputs, as indexed by decreased high- frequency HR variability. Similar to sleep, frontal cortical slow waves preferentially occurred coupled to the heart rhythm, perhaps due to a common brainstem generator. The observational nature of the study limits causal inferences that can be made, and more work is needed to elucidate the mechanism and role of these cardiac changes and the clinical significance of their coupling to the cortex.

Thus, in the clinical management of patient hemodynamics, propofol should not be assumed to decrease the HR. In fact, particularly for slow infusions and younger patients, propofol is likely to increase the HR. Ultimately, HR will be a complex result of opioid, hypnotic, and surgical factors.

The authors thank Irene Tracey, M.A., D.Phil., F.R.C.A., F.Med.Sci., and colleagues at the University of Oxford, Nuffield Department of Clinical Neurosciences, Oxford, United Kingdom, who conducted the original propofol study used in this work.

Supported by the Wellcome Trust (London, United Kingdom; grant No. 203139/Z/16/Z; to M.S. Fabus); the Department of Anesthesiology, University of Auckland (Auckland, New Zealand; to Dr. Sleigh); and the Medical Research Council Development Pathway Funding Scheme (London, United Kingdom; award reference No. MR/R006423/1; to Dr. Warnaby).

Dr. Warnaby is listed as an inventor on patents for “Perception Loss Detection” (World Intellectual Property Organization No. 2013/179048 A1) under anesthesia. Dr. Sleigh is a handling editor for Anesthesiology. Dr. Fabus declares no competing interests.

Supplementary figures and control analyses, https://links.lww.com/ALN/D329