Cardio-audio synchronization elicits neural and cardiac surprise responses in human wakefulness and sleep

  1. Andria Pelentritou
  2. Christian Pfeiffer
  3. Sophie Schwartz
  4. Marzia De Lucia

  • Curated by eLife

    eLife logo

    eLife assessment

    This study presents useful findings on an important question in cognitive neuroscience - whether the brain can form sensory predictions during sleep. The paradigm used is compelling but evidence supporting the claims of the authors is incomplete. Major issues pertaining to large differences between the pattern of brain responses observed here relative to effects reported in the literature previously, and some evidence that the study might be underpowered to make strong conclusions about certain aspects of the data.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

The human brain can encode auditory regularities with fixed sound-to-sound intervals and with sound onsets locked to cardiac inputs. Here, we investigated auditory and cardio-audio regularity encoding during sleep, when bodily and environmental stimulus processing may be altered. Using electroencephalography and electrocardiography in healthy volunteers ( N  = 26) during wakefulness and sleep, we measured the response to unexpected sound omissions within three regularity conditions: synchronous, where sound and heartbeat are temporally coupled, isochronous, with fixed sound-to-sound intervals, and a control condition without regularity. Cardio-audio regularity encoding manifested as a heartbeat deceleration upon omissions across vigilance states. The synchronous and isochronous sequences induced a modulation of the omission-evoked neural response in wakefulness and N2 sleep, the former accompanied by background oscillatory activity reorganization. The violation of cardio-audio and auditory regularity elicits cardiac and neural responses across vigilance states, laying the ground for similar investigations in altered consciousness states such as coma and anaesthesia.

Article activity feed

  1. Version published to 10.1038/s42003-024-05895-2
  2. Version published to 10.1101/2022.03.03.482861v3 on bioRxiv
  3. eLife

    Author Response

    Reviewer #1 (Public Review):

    Pelentritou and colleagues investigated the brain’s ability to infer temporal regularities in sleep. To do so, they measured the effect on brain and cardiac activity to the omission of an expected sound. Participants were presented with three different categories of sounds: fixed sound-to-sound intervals (isochronous), fixed heartbeat-to-sound intervals (synchronous), and a control condition without any regularity (asynchronous). When omitting a sound, they observed a difference in the isochronous and synchronous conditions compared to the control condition, in both wakefulness and sleep (NREM stage 2). Furthermore, in the synchronous condition, sounds were temporally associated with sleep slow waves suggesting that temporal predictions could influence ongoing brain dynamics in sleep. Finally, at the level of cardiac activity, the synchronous condition was associated with a deceleration of cardiac frequency across vigilance states. Overall, this work suggests that the sleeping brain can learn temporal expectations and responds to their violation.

    We thank the reviewer for the very useful and informed comments, to which we carefully reply below.

    Major strengths and weaknesses:

    The paradigm is elegant and robust. It represents a clever way to investigate an important question: whether the sleeping brain can form and maintain predictions during sleep. Previous studies have so far highlighted the lack of evidence for predictive processes during sleep (e.g. (Makov et al., 2017; Strauss et al., 2015; Wilf et al., 2016)). This work shows that at least a certain type of prediction still takes place during sleep.

    However, there are some important aspects of the methodology and interpretations that appear problematic.

    (1) The methodology and how it compares to previous articles would need to be clarified. For example, the Methods section indicates that the authors used a right earlobe electrode as a reference. This is quite different from the nose reference used by SanMiguel et al. (2013) or in Dercksen et al. (2022). This could affect the polarity and topographies of the OEP or AEP and thus represents a very significant difference. Likewise, SOs are typically detected in a montage reference to the mastoids. Perhaps the left/right asymmetries present in many plots (e.g. Figure 3) could be due to the right earlobe reference used.

    We thank the reviewer for raising this important point which has prompted us to clarify the reference choice in the manuscript both for completing the information about data recordings in our experiment and for emphasizing the influence of the reference on the EEG results and how they compare to previous reports.

    First, we would like to clarify that although EEG data is referenced to the right earlobe online, electrophysiological data from both earlobes were acquired and offline re-referencing to paired earlobes was performed. This is now clarified in the Methods section on page 26, lines 648-651 as follows:

    ‘Continuous EEG (g.HIamp, g.tec medical engineering, Graz, Austria) was acquired at 1200 Hz from 63 active ring electrodes (g.LADYbird, g.tec medical engineering) arranged according to the international 10–10 system and referenced online to the right earlobe and offline to the left and right ear lobes.’

    Additionally, after preprocessing, we performed common average re-referencing, as is common practise and recommended in the literature (see e.g. Niso et al., 2022), and hence the initial online referencing is no longer of relevance. Nonetheless, we agree with the reviewer that different online and offline referencing schemes could explain why some results in the literature are not optimally reproducible. We have clarified this point in the discussion on page 17, lines 408-411 as follows:

    ‘Finally, while we used largely similar pre-processing (i.e. filters) and experiment implementation (i.e. online and offline reference) as in Chennu et al. (2016), this was not the case for other studies with which direct comparisons are unwarranted.’

    For the SO analysis chosen reference (linked earlobes online and common average offline in our case) we acknowledge that - as the reviewer mentioned - many groups indeed employ mastoid re-referencing for SO detection (e.g. Siclari et al., 2018; Schneider et al., 2020; Ameen et al., 2022). However, to the best of our knowledge, this is not a standard choice, as many other groups choose a linked earlobe reference for online SO detection and the mastoids only for offline SO detection (Ngo et al., 2013; Besedovsky et al., 2017; Ngo and Staresina, 2022). In addition, other recent studies used linked earlobe referencing (Bouchard et al., 2021) or common average re-referencing (Züst et al., 2019) for offline SO detection. In our study we opted for using the same average reference for SO detection and evoked potential analysis in order to be able to relate the results of the omission evoked response comparison to that of the SO analysis.

    Also, the authors did not use the same filters in wakefulness and sleep, which could introduce an important bias when comparing sleep and wake results or sleep results with previous wake papers.

    We fully agree with the reviewer and thank him/her for this suggestion. We have now re-analysed the wakefulness data using a bandpass filter of 0.5-30 Hz as used for the sleep data. The chosen filtering range is commonly used in sleep research. Moreover, Chennu et al. (2016) employed a very similar filtering range (0.5-25 Hz) in an omission EEG study, whose results are similar to ours (Chennu et al., 2016). This new preprocessing resulted in a higher number of valid trials (average trial number: before N=245, now N=286) in wakefulness. Hence, the data from more participants could be used (before N=21, now N=23) and the statistical power of observed differences in our comparisons was improved. The Methods section has been updated accordingly on page 31, lines 763-764 as follows:

    ‘Continuous raw EEG data were band-pass filtered using second-order Butterworth filters between 0.5 and 30 Hz for the wakefulness and sleep session.’

    (2) The ERP to sound omission shows significant differences between the isochronous and asynchronous conditions in wakefulness (Figure 3A and Supp. Fig.) but this difference is very different from previous reports in wakefulness. Topographies are also markedly different, which questions whether the same phenomenon is observed. For example, SanMiguel and colleagues observed an N1 in response to omitted but expected sounds. The authors argue that they observe a similar phenomenon in the iso vs baseline contrast, but the timing and topography of their effect are very different from the typical N1. The authors also mention that, within their study, wake and N2 OEPs were "largely similar" but they differ in terms of latencies and topographies (Figure 3A-B). It would be better to have a more objective way to explore differences and similarities across the different analyses of the paper or with the literature.

    We concur with the reviewer and reviewing editor, who both pointed that the way we previously analysed (see our reply to the reviewer’s previous comment) and reported our data was sub-optimal. The new analysis of the wake data reveals more similarities with the MMN and to some extend with the omission literature (Figure 4). As requested, we also improved the description of the comparison of our results to those from the literature, in the Discussion section (pages 17-19, lines 391-458).

    (3) The authors applied a cluster permutation to identify clusters of significant time points. However, some aspects of this analysis are puzzling. Indeed, the authors restricted the cluster permutation to a temporal window of 0 to 350ms in wake (vs. -100 to 500ms in sleep). This can be misleading since the graphs show a larger temporal window (-100 to 500ms). Consequently, portions of this time window could show no cluster because the analysis revealed an absence of significant clusters but because the cluster permutation was not applied there. Besides, some of the reported clusters are extremely brief (e.g. l. 195, cluster's duration: 62ms), which could question their physiological relevance or raise the possibility that some of these clusters could be false positives (there was no correction for multiple comparisons across the many cluster permutations performed). Finally, there seems to be a duplication of the bar graphs showing the number of significant electrodes in the positive and first negative cluster for Figure 2 Supp. Fig. 1.

    We thank the reviewer for raising this point. We have now performed cluster permutation statistical analysis over the entire -100 to 500 ms window in wakefulness, thus matching the temporal window used for the sleep data (Methods, page 34, lines 843-846). Please note that this modified temporal window was applied to the wake data for which the pre-processing had also been modified (see our reply to comment #1 above). With matching analysis for wakefulness and sleep, we now identify clusters of higher or similar significance compared to our earlier results (Cohen’s d for isoch vs asynch = 0.92 now and 0.67 before; for synch vs asynch = 0.91 now and 1.06 before). In addition, for the isoch vs asynch omission response comparisons, overlapping cluster periods are identified in wakefulness (114-159 ms) and sleep (85-223 ms). The relevant results are thoroughly described on pages 9-10, lines 202-210; page 11, lines 238-251, pages 38-39, lines 970-985.

    We would like to also mention that while multiple comparisons correction is performed across channels and electrodes in the EEG using cluster permutation statistics, it is true that we do not perform multiple comparisons correction across the many comparisons. We now explicitly mention the lack of this correction for multiple comparisons in the Methods section page 34, lines 840-843 as follows:

    ‘Of note, the cluster permutation based multiple comparisons correction only applied across channels and latencies when comparing two experimental conditions, however no multiple comparisons correction was applied across the number of comparisons made in this study.’

    (4) More generally, regarding statistics, the absence of exact p-values can render the interpretation of statistical outputs difficult. For example, the authors report a significant modulation of the sound-to-SO latency across conditions (p<0.05) but no significant effect of heartbeat peak-to-SO latency (p>0.05). They interpret this pattern of results rather strongly as evidence that the "readjustment of SOs was specific to auditory regularities and not to cardiac input". Yet, examining the reported chi-square values show very close values between the two analyses (7.9 vs. 7.4). It seems thus difficult to argue for a real dissociation between the two effects. Providing exact p-values for all statistical tests could help avoid this pitfall.

    To assist the interpretation of statistical analysis results, we have now included exact p-values.

    Specifically, for SOs, we agree with the reviewer on the highly similar chi-squared values for the two analyses of Sound onset to SO peak and R peak onset to SO peak and have now included a comment in the discussion to reflect this on page 20, lines 478-480 as follows:

    ‘However, it should be noted that although not significant, we observed a trend of lower R peak to SO peak latencies during cardio-audio regularity compared to the other auditory conditions, possibly driven by the fixed relationship between heartbeat and sound in the synch condition.’

    Reviewer #2 (Public Review):

    This study was designed to study the cortical response to violations in auditory temporal sequences during wakefulness and sleep. To this end, the study had three levels of temporal sequence, a regular temporal sequence, an auditory tone that was yoked to the cardiac signal, and an irregular tone. The authors show significant EEG differences to an omitted tone when the auditory tone was predictable both during wakefulness and sleep.

    The authors analyze the ERP to the omitted tone as well as when aligned to the R-peak of the HEP. The analysis was comprehensive and the effects reported align with the interpretation given. Of particular interest was the fact that a deceleration of the heart rate was present for omissions when the auditory tone was yoked to the R-peak (synch) in all stages of wakefulness and sleep.

    We thank the reviewer for his/her positive judgment.

    However, one weakness was the rationale for the current study and how the results link to current theoretical frameworks for the role of interoception in perception and cognition. This was in contrast to the clear background and explanation to study the response to omissions for a predictable auditory sequence in wakefulness and sleep. It was unclear why the authors selected the cardiac signal to yoke their auditory stimuli. What is the specific motivation for the cardiac signal rather than the respiratory signal? This was not clear.

    In the revised Introduction section, we improved our description of these aspects, including the interaction between interoception and external stimulus processing. We hypothesized that cardiac signals would be more relevant than respiratory signals in coordinating temporal expectation because of existing prior experimental evidence thereof, as well as data showing a modulation of the neural response to heartbeat by levels of vigilance/consciousness, and the sharp cardiac R peak offering an ideal candidate for online temporal locking to administered sounds (see our detailed reply to the reviewer’s comment #2 below). However, we cannot exclude that respiratory signals could also be used by the brain to assist temporal regularities detection.

    Future studies may test for this possibility.

  4. eLife

    eLife assessment

    This study presents useful findings on an important question in cognitive neuroscience - whether the brain can form sensory predictions during sleep. The paradigm used is compelling but evidence supporting the claims of the authors is incomplete. Major issues pertaining to large differences between the pattern of brain responses observed here relative to effects reported in the literature previously, and some evidence that the study might be underpowered to make strong conclusions about certain aspects of the data.

  5. eLife

    Reviewer #1 (Public Review):

    Pelentritou and colleagues investigated the brain's ability to infer temporal regularities in sleep. To do so, they measured the effect on brain and cardiac activity to the omission of an expected sound. Participants were presented with three different categories of sounds: fixed sound-to-sound intervals (isochronous), fixed heartbeat-to-sound intervals (synchronous), and a control condition without any regularity (asynchronous). When omitting a sound, they observed a difference in the isochronous and synchronous conditions compared to the control condition, in both wakefulness and sleep (NREM stage 2). Furthermore, in the synchronous condition, sounds were temporally associated with sleep slow waves suggesting that temporal predictions could influence ongoing brain dynamics in sleep. Finally, at the level of cardiac activity, the synchronous condition was associated with a deceleration of cardiac frequency across vigilance states. Overall, this work suggests that the sleeping brain can learn temporal expectations and responds to their violation.

    Major strengths and weaknesses:
    The paradigm is elegant and robust. It represents a clever way to investigate an important question: whether the sleeping brain can form and maintain predictions during sleep. Previous studies have so far highlighted the lack of evidence for predictive processes during sleep (e.g. (Makov et al., 2017; Strauss et al., 2015; Wilf et al., 2016)). This work shows that at least a certain type of prediction still takes place during sleep.

    However, there are some important aspects of the methodology and interpretations that appear problematic.
    (1) The methodology and how it compares to previous articles would need to be clarified. For example, the Methods section indicates that the authors used a right earlobe electrode as a reference. This is quite different from the nose reference used by SanMiguel et al. (2013) or in Dercksen et al. (2022). This could affect the polarity and topographies of the OEP or AEP and thus represents a very significant difference. Likewise, SOs are typically detected in a montage reference to the mastoids. Perhaps the left/right asymmetries present in many plots (e.g. Figure 3) could be due to the right earlobe reference used. Also, the authors did not use the same filters in wakefulness and sleep, which could introduce an important bias when comparing sleep and wake results or sleep results with previous wake papers.
    (2) The ERP to sound omission shows significant differences between the isochronous and asynchronous conditions in wakefulness (Figure 3A and Supp. Fig.) but this difference is very different from previous reports in wakefulness. Topographies are also markedly different, which questions whether the same phenomenon is observed. For example, SanMiguel and colleagues observed an N1 in response to omitted but expected sounds. The authors argue that they observe a similar phenomenon in the iso vs baseline contrast, but the timing and topography of their effect are very different from the typical N1. The authors also mention that, within their study, wake and N2 OEPs were "largely similar" but they differ in terms of latencies and topographies (Figure 3A-B). It would be better to have a more objective way to explore differences and similarities across the different analyses of the paper or with the literature.
    (3) The authors applied a cluster permutation to identify clusters of significant time points. However, some aspects of this analysis are puzzling. Indeed, the authors restricted the cluster permutation to a temporal window of 0 to 350ms in wake (vs. -100 to 500ms in sleep). This can be misleading since the graphs show a larger temporal window (-100 to 500ms). Consequently, portions of this time window could show no cluster because the analysis revealed an absence of significant clusters but because the cluster permutation was not applied there. Besides, some of the reported clusters are extremely brief (e.g. l. 195, cluster's duration: 62ms), which could question their physiological relevance or raise the possibility that some of these clusters could be false positives (there was no correction for multiple comparisons across the many cluster permutations performed). Finally, there seems to be a duplication of the bar graphs showing the number of significant electrodes in the positive and first negative cluster for Figure 2 Supp. Fig. 1.
    (4) More generally, regarding statistics, the absence of exact p-values can render the interpretation of statistical outputs difficult. For example, the authors report a significant modulation of the sound-to-SO latency across conditions (p<0.05) but no significant effect of heartbeat peak-to-SO latency (p>0.05). They interpret this pattern of results rather strongly as evidence that the "readjustment of SOs was specific to auditory regularities and not to cardiac input". Yet, examining the reported chi-square values show very close values between the two analyses (7.9 vs. 7.4). It seems thus difficult to argue for a real dissociation between the two effects. Providing exact p-values for all statistical tests could help avoid this pitfall.

  6. eLife

    Reviewer #2 (Public Review):

    This study was designed to study the cortical response to violations in auditory temporal sequences during wakefulness and sleep. To this end, the study had three levels of temporal sequence, a regular temporal sequence, an auditory tone that was yoked to the cardiac signal, and an irregular tone. The authors show significant EEG differences to an omitted tone when the auditory tone was predictable both during wakefulness and sleep.

    The authors analyze the ERP to the omitted tone as well as when aligned to the R-peak of the HEP. The analysis was comprehensive and the effects reported align with the interpretation given. Of particular interest was the fact that a deceleration of the heart rate was present for omissions when the auditory tone was yoked to the R-peak (synch) in all stages of wakefulness and sleep.

    However, one weakness was the rationale for the current study and how the results link to current theoretical frameworks for the role of interoception in perception and cognition. This was in contrast to the clear background and explanation to study the response to omissions for a predictable auditory sequence in wakefulness and sleep. It was unclear why the authors selected the cardiac signal to yoke their auditory stimuli. What is the specific motivation for the cardiac signal rather than the respiratory signal? This was not clear.

  7. Version published to 10.1101/2022.03.03.482861v2 on bioRxiv
  8. Version published to 10.1101/2022.03.03.482861v1 on bioRxiv