Received: 19 October 2023 Revised: 18 February 2024 Accepted: 23 March 2024

DOI: 10.1002/hbm.26677

RESEARCH ARTICLE

Intracortical brain-heart interplay: An EEG model source study

of sympathovagal changes

Vincenzo Catrambone 1 | Diego Candia-Rivera 2 | Gaetano Valenza 1

1
Neurocardiovascular Intelligence
Laboratory & Department of Information Abstract
Engineering & Bioengineering and Robotics
The interplay between cerebral and cardiovascular activity, known as the functional
Research Center, E. Piaggio, School of
Engineering, University of Pisa, Pisa, Italy brain-heart interplay (BHI), and its temporal dynamics, have been linked to a plethora
2
Sorbonne Université, Paris Brain Institute of physiological and pathological processes. Various computational models of the
(ICM), INRIA, CNRS, INSERM, AP-HP, Hôpital
Pitié-Salpêtriŕe, Paris, France
brain-heart axis have been proposed to estimate BHI non-invasively by taking advan-
tage of the time resolution offered by electroencephalograph (EEG) signals. However,
Correspondence
Vincenzo Catrambone, Neurocardiovascular
investigations into the specific intracortical sources responsible for this interplay
Intelligence Laboratory & Department of have been limited, which significantly hampers existing BHI studies. This study pro-
Information Engineering & Bioengineering and
Robotics Research Center, E. Piaggio, School
poses an analytical modeling framework for estimating the BHI at the source-brain
of Engineering, University of Pisa, Pisa, Italy. level. This analysis relies on the low-resolution electromagnetic tomography sources
Email: [email protected]
localization from scalp electrophysiological recordings. BHI is then quantified as the
Funding information functional correlation between the intracortical sources and cardiovascular dynamics.
European Commission, Grant/Award Number:
101017727 Using this approach, we aimed to evaluate the reliability of BHI estimates derived
from source-localized EEG signals as compared with prior findings from neuroimaging
methods. The proposed approach is validated using an experimental dataset gathered
from 32 healthy individuals who underwent standard sympathovagal elicitation using
a cold pressor test. Additional resting state data from 34 healthy individuals has been
analysed to assess robustness and reproducibility of the methodology. Experimental
results not only confirmed previous findings on activation of brain structures affect-
ing cardiac dynamics (e.g., insula, amygdala, hippocampus, and anterior and mid-
cingulate cortices) but also provided insights into the anatomical bases of brain-heart
axis. In particular, we show that the bidirectional activity of electrophysiological path-
ways of functional brain-heart communication increases during cold pressure with
respect to resting state, mainly targeting neural oscillations in the δ, β, and γ bands.
The proposed approach offers new perspectives for the investigation of functional
BHI that could also shed light on various pathophysiological conditions.

KEYWORDS
Brain-Heart, EEG, heart rate variability, source analysis

Vincenzo Catrambone and Diego Candia-Rivera contributed equally to this work.

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any
medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2024 The Authors. Human Brain Mapping published by Wiley Periodicals LLC.

Hum Brain Mapp. 2024;45:e26677. wileyonlinelibrary.com/journal/hbm 1 of 10

https://doi.org/10.1002/hbm.26677
2 of 10 CATRAMBONE ET AL.

1 | I N T RO DU CT I O N Source localization of EEG signals aims to solve the inverse

problem of accurately estimating the brain regions that trigger electro-
Characterization of the complex phenomena underlying the functional magnetic changes measured at the scalp level by leveraging spatial
relationship between the central nervous system (CNS) and auto- modeling of electrical propagation in brain tissues. This depends on a
nomic nervous system (ANS) has increasingly gained attention in the variety of factors, including the neurophysiology of EEG sources, bio-
last decade. Although initially believed to pertain solely to central con- physics of electrical propagation, anatomy of conductive tissues, and
trol over peripheral organs, the phenomenon is multi-faceted and can neuronal activity distribution properties (Michel & He, 2019). To this
be studied at different scales, from neural and glial signaling (Marina extent, new source estimation methodologies have been developed
et al., 2016; Truter et al., 2023) to functional psychosomatic reverber- based on the most recent knowledge of EEG signal generation. The
ations (Candia-Rivera, Catrambone, Thayer, et al., 2022; Skora first method for resolving the EEG inverse problem involves localiza-
et al., 2022). tion of a constrained set of equivalent dipoles. According to the tradi-
The study of central-autonomic interplay has placed particular tional method of dipole source localization, the scalp potential field is
emphasis on studying cardiac dynamics because of their ease of mea- produced by only one or a few active brain regions. Nonlinear optimi-
surement and connection to both the sympathetic and parasympa- zation can be used to identify a mathematically ideal solution under
thetic nervous systems (Chen et al., 2021; Liu et al., 2022). In this this restriction. Epileptic foci or primary sensory areas, such as the
context, most functional brain-heart interplay (BHI) estimations have sensorimotor cortex in surgical candidates, can be located with rea-
been performed using noninvasive electrophysiological signals, specif- sonable results using dipole source localization, despite the simplicity
ically EEG and heart rate variability (HRV) series. Notably, these and limitations of this a priori assumption. Distributed-source imaging
approaches offer a sufficiently high time resolution, which is a crucial approaches have largely replaced dipole source localization in experi-
factor, given the time-varying nature of BHI. The variety of physiolog- mental EEG studies. In this case, a large number of equivalent dipoles
ical and clinical contexts in which BHI is involved reflects its impor- (order of magnitude of ≈ 5000) are dispersed in fixed locations across
tance (Benarroch, 1993; Catrambone & Valenza, 2021; Schiecke the entire source space, and the strength of each dipole is estimated
et al., 2016; Silvani et al., 2016; Valenza et al., 2020; Valenza (Michel & He, 2019). The source space is typically limited to the grey
et al., 2024). matter, using an individual's anatomical information or template MRI.
Indeed, several methodologies have emerged for estimating the One of the most commonly used methods is low-resolution electro-
BHI, addressing different aspects, such as directionality, synchroniza- magnetic tomography (LORETA), developed by (Pascual-Marqui
tion, and complexity. These include formulation of an ad hoc synthetic et al., 1994). LORETA and its variations (e.g., sLORETA (Pascual-
data-generation model (Candia-Rivera, 2023; Candia-Rivera, Catram- Marqui et al., 2002)) rely on the minimization of the Laplacian opera-
bone, Barbieri, & Valenza, 2022; Catrambone et al., 2019; tor of the sources, resulting in a smooth (low resolution) distribution
Catrambone, Talebi, et al., 2021), the application of information the- of 3D activity. This constraint is justified by the physiologically tena-
ory to disentangle linear and nonlinear components (Faes et al., 2015) ble hypothesis that the activity in adjacent voxels is correlated. In this
and to quantify information storage (Antonacci et al., 2023; Barà study, we employed sLORETA (Pascual-Marqui et al., 2002) as a
et al., 2023), nervous system—wise functional estimation through source localization technique. While it is a widely utilized method, it
microstate occurrences (Catrambone & Valenza, 2023b), or methods achieves very good performance in localizing electrical activity
exploiting state-space reconstruction to investigate nonlinear and sources, has small computational complexity, provides a closed-form
directed interactions (Schiecke et al., 2016). expression based on an L2-norm-based solution, and is available in
Currently proposed EEG-derived techniques excel in investigat- the form of an open-source toolbox/library. Further information on
ing time-resolved BHI processes with optimal time resolution, but source localization methodologies proposed in the literature can be
they are limited in terms of the spatial domain. Nevertheless, the found in, for example, (Asadzadeh et al., 2020).
efferent and afferent functional pathways of ANS activity involve Previous studies investigated the intracranial activity using inva-
several cortical and subcortical regions of the CNS, which form part sive stereo-EEG-based heartbeat-evoked potentials (HEPs) (Mazzola
of the central autonomic network (CAN) (Benarroch, 1993; et al., 2023). HEPs are calculated by averaging the sEEG signals that
Benarroch, 2012). Indeed, limbic structures, such as the cingulate synchronize with the R peaks of the concurrent ECG signal (Park &
cortex, medial temporal lobe, amygdala, and hippocampus, and Blanke, 2019). These studies indicated that HEPs originate from spe-
regions such as the ventral medial prefrontal, insular, and parietal cific brain regions, including the anterior cingulate, right insula, pre-
cortices are associated with autonomic responses, and these regions frontal cortex, and left secondary somatosensory cortex (Babo-Rebelo
all belong to the CAN (Benarroch, 1993; Catrambone & et al., 2016; Park et al., 2018; Park & Blanke, 2019), which are known
Valenza, 2021; Valenza et al., 2019). The identification of CAN com- to interact with emotional, cognitive, and sensory processing. How-
ponents has primarily been attained through technologies such as ever, this approach comes with limitations (Park & Blanke, 2019).
functional magnetic resonance imaging (fMRI), which allows investi- First, the stereo-EEG recordings are invasive and may not be easily
gation of the activity of deep CNS regions with commendable spatial gathered from the general population. Second, cardiac electric cur-
resolution at the cost of a temporal resolution that is inferior to that rents associated with ventricular contractions might induce artifacts in
of EEG. stereo-EEG signals that can bias the HEP assessment. Third, cardiac
CATRAMBONE ET AL. 3 of 10

cycle is influenced by dynamical variations in sympathetic and para- To better assess robustness and reliability of the experimental
sympathetic autonomic activities, which might be linked to several results, a second dataset (D2) has been considered. It consisted of
EEG oscillations in an afferent and efferent fashion. Therefore, a com- physiological signals recordings (EEG and ECG) gathered employing
prehensive understanding of key cortical and sub-cortical regions the same devices of the first dataset on a group of 34 healthy right-
involved in cardiac autonomic control, that is, the functional brain- handed young subjects (age range, 26.5 years median; 18 males, there
heart axis, using non-invasive techniques is yet to be achieved. were no subject in common between the two groups) performing a
In this study, we investigated the reliability of markers of func- 3 min resting state in a dark room. The same dataset has been already
tional BHI estimated from intracortical source-localized signals and presented and analysed in previous studies (Catrambone, Averta,
cardiac sympathovagal oscillations. To the best of our knowledge, no et al., 2021), and is part of a larger experiment whose data are freely
previous study has endeavored to estimate BHI through a functional available (Averta et al., 2021).
correlation between EEG model sources and cardiovascular activity.
To validate the proposed methodology, we tested it on data gath-
ered during a cold pressor test (CPT), which is a well-known auto- 2.2 | EEG processing
nomic maneuver involving thermal stress that elicits a strong
sympathovagal response (Cui et al., 2002). Physiological responses EEG data were preprocessed using MATLAB R2017a and the Fieldtrip
associated with thermal and nociceptive stress include increased heart Toolbox (Oostenveld et al., 2011). The data were bandpass filtered
rate, which is attributed to higher sympathetic activity and lower vagal with a Butterworth filter of order four, between 0:5 and 45 Hz. Large-
outflow. At the CNS level, brain correlates of the CPT include cortical movement artifacts were removed using wavelet-enhanced indepen-
and subcortical brain activation, such as an increase in power in the dent component analysis (ICA) (Gabard-Durnam et al., 2018), which
delta and gamma bands in the fronto-temporal areas (Chang were identified using automated thresholding over the independent
et al., 2002; Fardo et al., 2017; Shao et al., 2012). Additionally, our component and multiplied by 50 to remove only very large artifacts,
previous studies on functional BHI dynamics during the CPT have as described by Candia-Rivera et al. (2021). Consequently, ICA was
revealed a bidirectional and diffuse interaction (Catrambone & re-run to recognize and reject eye movements and cardiac-field arti-
Valenza, 2023a; Catrambone & Valenza, 2023b) with changes pre- facts from the EEG data (Dirlich et al., 1997). Therefore, a one-lead
dominantly in the EEG gamma and delta bands (Candia-Rivera, ECG was included as an additional input to the ICA to enhance the
Catrambone, Barbieri, & Valenza, 2022; Catrambone et al., 2019), process of identifying cardiac artifacts. Once the ICA components
which were directly connected to ascending communication pathways with eye movements and cardiac artifacts were visually identified,
from autonomic inputs and corroborated by changes in HEPs (Shao they were set to zero to reconstruct the EEG series. This step thus
et al., 2011). yielded eye movement- and cardiac artifact-free EEG data. Individual
EEG channels were analyzed. The channels were marked as contami-
nated if their area under the curve exceeded three standard devia-
2 | MATERIALS AND METHODS tions of the mean for all channels. The remaining channels were
compared with their weighted-by-distance-correlation neighbors
2.1 | Dataset description using the standard field-trip neighbor’ definition. If a channel resulted
in a weighted-by-distance correlation of less than 0.6, it was consid-
The first dataset (D1) came from a group of 32 right-handed young ered contaminated. The contaminated channels were replaced by
healthy adults underwent a CPT while recording physiological signals. neighbor’ interpolation. The channels were referenced using a com-
Data from 26 subjects (age range, 21–41 years; median, 27 years; mon average, which is the most appropriate for a functional BHI
13 males) were considered for further analysis because of data length assessment (Candia-Rivera et al., 2021).
and quality. In particular, data from three subjects were not considered
because of the presence of artifacts in their physiological data (EEG or
ECG), while data from three subjects were discarded because of early 2.3 | EEG source localization
withdrawal of their hands from the cold water. Each subject's recording
comprised of a 128-channel high-density EEG (Electrical Geodesics, Source-level analysis was performed on EEG at 60 to 60 s with
Inc., Eugene, OR, USA), respiratory activity, and one-lead ECG, sampled respect to the cold-pressure onset for the first dataset, and an homo-
at 500 Hz. Before data acquisition, the subjects were asked to sit com- geneous time window of 60 s was taken for dataset 2 as well. The
fortably on a chair to ensure hemodynamic stabilization. Throughout source reconstruction was performed using the Brainstorm toolbox
the protocol, subjects were asked to keep their eyes closed to minimize (Tadel et al., 2011). Default anatomy (ICBM152) was used together
artifacts. The task consisted of a 3-min resting state, followed by a with EGI 128 electrode locations. A three-shell sphere was created
3-min CPT, in which the subjects were guided to submerge their left for each participant using no-noise modeling, and sources were com-
hand up to the write into asked place their hand into a bucket filled puted using default parameters (minimum norm imaging, sLORETA
with ice water (0–4
C), and a recovery phase, in which subjects were method, constrained). The conductivities for surface estimation were
asked to withdraw their hand from the ice-water bucket. set to default (scalp 1, skull 0.0125, brain 1). With the aim of avoiding
4 of 10 CATRAMBONE ET AL.

curse of dimensionality in the inverse problem of EEG-source localiza- 2.5 | Intracortical brain-heart interplay estimation
tion, we choose to employ a standard brain volume partition in which
the brain volume itself is divided into a number of region of interest As the first attempt to estimate BHI at the brain source level, we con-
(ROI) defined according to anatomical and functional criteria. Among ducted a functional correlation analysis using the Spearman method
the different available standard ROI partitions of the brain volume, between the time-resolved PSD from the source ROI and cardiovas-
also called brain atlases, we employed the automated-anatomical cular dynamics.
labelling (AAL) atlas for a total of 95 brain ROIs (Tzourio-Mazoyer To obtain an intersubject BHI measure, each PSD series
et al., 2002). This choice was due to AAL large use and acknowledge- (i.e., related to a specific experimental phase, brain- and HRV-related
ment in the field, as well as for the comparable cardinality of the atlas frequency bands, and ROI) was constructed by concatenating the
(i.e., 95 ROIs) with respect to the number of EEG electrodes (i.e., 128). subject-specific PSD series. To level the intersubject range difference,
The present study does not delve into the anatomical and structural each subject-specific PSD series was z-scored before concatenation.
differences among standard ROI partitions and their functional conse- Eventually, each series counted 60 (time instants)  26 (subject) points
quences, but an interested reader may find material in (Evans for the first dataset, and 60 (time instants)  34 (subject) points for
et al., 2012). Sources were visualized using the ggseg package for R the second dataset.
(https://www.r-project.org/) and AAL implementations (Rolls Finally, Spearman's correlation coefficients were calculated for
et al., 2015). each HRV-PSD time series (i.e., LF, IF, and HF) and each band-specific
The source spectrogram was computed for ROIs by averaging the ROI-PSD series (i.e., δ, θ, α, β, γ). To statistically assess significance of
time-varying vertices using a short-time Fourier transform with a Han- the correlation coefficients, the associated p-values were calculated
ning taper. Calculations were performed through a 2s sliding time through a Z-test, due to the large sample size. The significance thresh-
window with a 50% overlap, resulting in spectrogram resolutions of old was fixed at 0:05, and p-value correction for multiple comparisons
1s and 0:5Hz. The power spectral density (PSD) time series were then was accounted for using the Bonferroni–Holm correction over the
integrated within five canonical frequency bands (delta: δ  ½1  4Hz, 95 source ROIs. Detailed statistical description of the correction
theta: θ  ½4  8Hz, alpha: α  ½8  12Hz, beta: β  ½12  30Hz, and applied can be found at (Abdi, 2010).
gamma: γ  ½30  45Hz). The entire processing procedure pipeline is illustrated in Figure 1.

2.4 | ECG processing 3 | RE SU LT S

The ECG time series was bandpass filtered using a Butterworth filter The experimental results were expressed in terms of Spearman's cor-
of order 4, between 0.5 and 45 Hz. The R-peaks from the QRS waves relation coefficient and associated p-values as calculated per ROI and
were first identified via an automatized process, followed by a visual EEG- and HRV-derived frequency bands. To provide an overview of
inspection of misdetections and final automated correction of the the findings, Figure 2 is a radar chart of the number of significant cor-
remaining misdetections or ectopic heartbeats. This procedure is relations detected in the resting state (upper plot) and CPT phase
based on a template-based method for detecting R peaks (Candia- (lower plot), for both LF (blue), IF (orange), and HF (yellow), in the
Rivera et al., 2021). All detected peaks were visually inspected over analysis of dataset D1. Each vertex of the radar chart represents a
the original ECG along with an interbeat interval histogram. Manual specific brain-associated frequency band (i.e., δ, θ, α,β, γ) for the
corrections were performed where needed and then automatic cor- 95 ROIs.
rections were done using a point-process algorithm (Citi et al., 2012). As shown in the top panel of Figure 2, during the resting state, no
After cubic interpolation of HRV series at 4Hz, the time-resolved significant correlations were detected within the IF and HF cardiovas-
spectogram of the HRV series was estimated using the smoothed cular bands, as well as within a higher frequency range of the EEG-
pseudo-Wigner–Ville distribution integrated into the two classical inter- sources spectrum (i.e., α, β,γ bands), which encompassed frequencies
vals 0:04–0:15Hz (i.e., low-frequency band LF) as a non-specific above 8Hz. Notably, significant correlations were primarily detected
marker of sympathovagal activity and in the range 0:15–0:4Hz in δ  LF combination, comprising a third of the ROIs (32 of 95), and
(i.e., high-frequency band HF) for parasympathetic activity. We also θ  LF relationship with 12 significant ROIs out of 95. Complementary
considered a third interval 0:1  0:2Hz (i.e., intermediate frequency IF) to this, during the CPT phase, as depicted in the lower panel of
to account for the HRV oscillations known to exists centered around Figure 2, no significant correlations were detected when considering
0:15Hz (Klimesch, 2018; Kluger & Gross, 2020; Perlitz et al., 2004; the IF and HF cardiovascular bands, whereas a notably higher number
Pfurtscheller et al., 2022). Such an IF band may be practical for study- of regions showed significant correlations with the LF cardiovascular
ing the BHI (Keller et al., 2020) and its relation to respiratory sinus component as compared to the resting state. Specifically, nearly the
arrhythmia and baroreflex mechanism (Eckberg, 2009; Julien, 2006). entire set of ROIs exhibited a significant correlation with the LF band,
The three power series derived from HRV (i.e., LF, IF, and HF) considering both the δ and γ frequency bands (i.e., 84 and 81 of
were sampled at 1Hz to be homogeneous with those derived from 95 ROIs, respectively), 46 of 95 ROIs were highlighted in the β band,
the source estimation. and 31 of 95 ROIs were highlighted in the θ band.
CATRAMBONE ET AL. 5 of 10

significant correlations were found in γ  IF combination. Considering

the limited amount of data in both datasets, a disagreement of this
range could be expected, and it is encouraging that in an overwhelm-
ing part of combinations of ROIs, and sources- and HRV- frequency
bands there is agreement between the two different datasets.
As a further investigation of the specific locations of the signifi-
cant correlations observed and the quantitative evaluation of this BHI
estimation, Figure 4 graphically represents most of the significant cor-
relation coefficients, particularly related to dataset D1. The three
most enhanced combinations of frequency bands were HRV-LF and
EEG sources in the δ, β, and γ ranges. The other frequency ranges
(sources θ and α, HRV-HF, as well as the significant regions found in
dataset D2) are reported in Figures 1–3 of the Supplementary Mate-
rial, respectively. Specifically, Figure 4 comprises six subpanels orga-
nized as follows: the two columns are associated with two
experimental phases (i.e., the resting state on the left and CPT on the
right), whereas the three rows are associated with the frequency
bands of the three sources. Subpanels were generated using
R-Studio1 software and the ggseg2 package, and represent the right
hemisphere in the first row (the left hemisphere is in the second row),
for both the lateral (left column) and medial (right column) views.
Figure 4 graphically represents the varying numbers of significant
regions reported in Figure 2. We can now appreciate that all the
regions that are significant in the δ  LF interplay during the resting
state hold their significance during CPT, with the addition of various
other ROIs. In the resting state, a significant δ  LF BHI was detected
on a belt spanning the left lateral hemisphere from the frontal to
dorso-parietal area. Additionally, a narrower portion was observed in
the right hemisphere, and a few regions on the medial frontal lobes
and an area in the left dorso-parietal region also presented significant
correlations. During the CPT phase, the δ  LF interplay showed sig-
nificance across the cortex, except for the central-medial region in
both hemispheres, which was more pronounced on the left side.
Conversely, the β  LF and γ  LF BHI did not reveal significant
areas during the resting state, except for a small left temporal region
for the β  LF relationship. However, these became more prominent
during the CPT phase. Specifically, both the lateral and medial frontal
regions in both hemispheres as well as the right temporal lobe and a
region encompassing the left dorso-temporal and occipital areas
FIGURE 1 Graphical representation of the processing pipeline.
exhibited significance in both β and γ frequency bands. While the β
band does not exhibit any other ROI, the γ band increases its signifi-
As previously mentioned, to better assess robustness of the cance by covering nearly the entire left hemisphere, leaving only a
method a second dataset (D2) has been analysed comprising a resting centro-parietal region on the right hemisphere unaffected.
state, and the experimental results are presented in Figure 3. As in the Interestingly, all detected correlations showed a positive sign,
resting state of dataset D1, no significant correlations were detected indicating that an increasing trend in one of the PSDs involved
within the IF and HF cardiovascular bands with all EEG bands excep- (e.g., the sources' γ-band) corresponds to a homogeneous trend in the
tion given for the γ band. Similarly, no significant Spearman correla- HRV-LF series.
tion were detected within the HRV-LF band. The only significant For a detailed list of the specific ROIs correlated with HRV-LF
value reported are related to the HRV-HF (10 of 95 ROIs) and HRV-IF dynamics, please refer to Table 1 in the Supplementary Material. We
(22 of 95 ROIs) with the EEG-γ band. Experimental results from the can appreciate the significant correlations detected between source-δ
analysis of the two different datasets are largely in agreement, even if
they show some differences. The relations that were the only signifi- 1
https://posit.co/download/rstudio-desktop/
cant in dataset D1 were absent in D2, whereas in dataset D2 some 2
https://github.com/ggseg
6 of 10 CATRAMBONE ET AL.

F I G U R E 3 Radar chart representation of the number of ROIs

having significant correlations with HRV-LF (blue lines), HRV-IF
(orange lines), or HRV-HF (yellow lines) during the resting state of
dataset D2. Each vertex of the charts refers to a specific brain-
associated frequency band (i.e., δ, θ, α, β,γ).

canonical frequency bands. To validate the methodological approach,

an experimental dataset comprising recordings from 26 subjects dur-
ing the resting state and a strong sympathovagal elicitation, such as
CPT, was employed. The experimental results revealed that several
brain regions exhibited significant correlations with HRV activity.
Interestingly, the nature of these correlations varies depending on the
experimental phase and frequency band considered.
From a methodological viewpoint, the proposed approach marks
an advancement in the spatial specificity of the BHI estimation. This
F I G U R E 2 Radar chart representation of the number of ROIs represents a noteworthy step away from the spatial constraints typi-
having significant correlations with HRV-LF (blue lines), HRV-IF cally observed in the development of biomarkers based on EEG. This
(orange lines), or HRV-HF (yellow lines) during the resting state (upper approach can pave the way for further developments in the localiza-
panel) and CPT phase (lower panel) of dataset D1. Each vertex of the
tion of BHI while preserving the time-frequency resolution of electro-
charts refers to a specific brain-associated frequency
band (i.e., δ,θ, α, β, γ). physiological analysis. However, it is worth noting that EEG-based
source localization does not reach the spatial resolution provided by
fMRI, and represents an indirect extrapolation of the measure of inner
and -γ and HRV-LF, particularly in specific CAN regions, such as CNS region activity. Nevertheless, source localization techniques have
insula, amygdala, anterior and middle cingulum, hippocampus, and achieved a reliable level of accuracy in the reconstruction of brain
thalamus. For the sake of completeness, Tables 2 and 3 of the Supple- dynamics. The time resolution achieved by these techniques, which
mentary Material report a detailed list of the corrected p-value related we have only partially explored, may help further develop the BHI
to all the sources- and HRV- frequency band combinations for the research field.
ROIs considered. From a physiological viewpoint, the results presented in this study
further deepen our knowledge of BHI under sympathovagal elicita-
tion. Our findings revealed a notable increase in the number of ROIs
4 | DISCUSSION that showed correlations with LF-HRV, a feature often linked to car-
diac sympathetic activity, albeit subject to debate. Furthermore, we
In this study, a novel approach to investigate the BHI phenomenon at identified a significant correlation specifically during CPT. This correla-
the brain source level was introduced. The BHI was estimated using tion has been observed in regions frequently associated with cardiac
Spearman's correlation between the brain source electrical activity autonomic regulation, including the insula, anterior and midcingulate
(derived through sLORETA technique) and cardiovascular dynamics, cortices, amygdala, thalamus, and hippocampus (Mazzola et al., 2023).
both estimated through their PSD representation, separated into The significant correlation detected in the resting state, mainly at low
CATRAMBONE ET AL. 7 of 10

F I G U R E 4 Graphical
representation of the significant
correlation coefficient. The left
column refers to the resting state
period and the right to the CPT
phase. On the left side of each
column, the frequency bands
involved are expressed for both
the brain and heart. Each
subpanel has two rows and two
columns: the first row is the right
hemisphere, the second row
represents the left hemisphere,
whereas left column is the lateral
view and the right column
represents the medial view.

frequencies (i.e., EEG-δ and HRV-LF), can be interpreted within the brain) is the first response to the CPT stimulus, which is followed by
framework of the Mayer wave phenomenon. These are transient the descending flow (Candia-Rivera, Catrambone, Barbieri, &
oscillatory responses to hemodynamic perturbations (Julien, 2006), Valenza, 2022). Interestingly, these results are consistent with
suggesting that a central pacemaker is present at low frequencies those reported in high-arousal conditions, such as emotional elicita-
(Pfurtscheller et al., 2017; Pfurtscheller et al., 2020; Pfurtscheller tion (Candia-Rivera et al., 2023; Candia-Rivera, Catrambone,
et al., 2022). In this framework, heartbeat and breathing dynamics Thayer, et al., 2022), suggesting a leading role for BHI in the physi-
might induce the pressure pulsatility to which the mechano-sensitive ology of arousal. The current study, even without investigating the
pyramidal neurons responds in a frequency range belonging to the δ timing and direction of the interaction, supports the literature on
band (Hamill, 2023; Klimesch, 2023). Moreover, the absence of corre- the brain frequency bands involved in the BHI (Candia-Rivera,
lation with the higher frequency bands of the HRV series may be Catrambone, Barbieri, & Valenza, 2022; Catrambone et al., 2019),
attributed to the specificity of CPT stimuli, which act as physical mainly δ and γ, and bridges fMRI and electrophysiological studies
stressors and may elicit specific changes other than other cognitive pointing at the CAN as the main player in CNS and ANS interaction
stressors (Candia-Rivera et al., 2023; Young & Benton, 2015). (Benarroch, 2012; Valenza et al., 2019).
CPT is an interesting elicitation protocol since it represents a This study had some limitations. First, in this primary study esti-
complex of concurrent stimulations: somatosensory, via the thermal mating BHI in the EEG source space, we utilized a basic model-free
stress and sub-threshold pain; emotional, since the perception analysis relying on Spearman's correlations and predominantly
raises a certain arousal and valence; and motor, by holding the hand employed the general sLORETA technique for solving the source
in an unpleasant state, requiring strong voluntary motor control. localization inverse problem. However, more specific computational
These stimuli arouse concurrent and partially independent CNS and models are required to better catch the BHI dynamics as a whole,
ANS responses, meaning that both systems are affected by their including its temporal dynamics, nonlinearity, and directionality, both
specific activities, as well as by their inter-system communication. in terms of interplay estimation and source localization. For instance,
Among the expected ANS-specific effects elicited by CPT are the sLORETA is known to have relatively coarse spatial resolution, which
baroreceptor modulation of heart rate due to physiological arousal becomes particularly significant when addressing source localization
(Cui et al., 2002) and the decreased heart-beat evoked potentials at a microscopic scale. Nevertheless, this study focused on a partition
under cold stimuli, mainly over the frontal and central scalp loca- of the brain volume in agglomerated ROIs, thus maintaining a macro-
tions (Shao et al., 2011). It is noteworthy that CPT involves physio- scopic scale. Moreover, the proposed methodology does not consider
logical signaling from the skin to the brain. These physiological potential confounding factors related to physiological variables such
dynamics have been captured through physiological modeling of as respiration or blood pressure. Specifically, relationships have been
the BHI, in which the ascending BHI information flow (i.e., heart-to- found between respiration and spectral content of hippocampal
8 of 10 CATRAMBONE ET AL.

activity in mice (Bagur et al., 2021), rodents, and human patients RE FE RE NCE S
(Herrero et al., 2018; Zelano et al., 2016). Furthermore, the proposed Abdi, H. (2010). Holm's sequential bonferroni procedure. Encyclopedia of
processing pipeline is susceptible to specific limitations, especially Research Design, 1(8), 1–8.
Antonacci, Y., Barà, C., Zaccaro, A., Ferri, F., Pernice, R., & Faes, L. (2023).
linked to the nonspecificity of HRV power as well as the source locali-
Time-varying information measures: An adaptive estimation of infor-
zation procedure. More advanced techniques will be employed in mation storage with application to brain-heart interactions. Frontiers in
future studies to develop a source-related BHI framework for applica- Network Physiology, 3, 1242505.
tion in larger cohorts. Finally, actual temperature of the hand was not Asadzadeh, S., Rezaii, T. Y., Beheshti, S., Delpak, A., & Meshgini, S. (2020).
A systematic review of eeg source localization techniques and their
measured before and after the CPT.
applications on diagnosis of brain abnormalities. Journal of Neurosci-
ence Methods, 339, 108740.
Averta, G., Barontini, F., Catrambone, V., Haddadin, S., Handjaras, G.,
5 | C O N CL U S I O N Held, J. P., Hu, T., Jakubowitz, E., Kanzler, C. M., Kühn, J.,
Lambercy, O., Leo, A., Obermeier, A., Ricciardi, E., Schwarz, A.,
Valenza, G., Bicchi, A., & Bianchi, M. (2021). U-limb: A multi-modal,
In conclusion, we here proposed an approach for estimating the func-
multi-center database on arm motion control in healthy and post-
tional relationship between the activity of specific brain regions and stroke conditions. GigaScience, 10(6), giab043.
cardiovascular dynamics during strong sympathovagal elicitation using Babo-Rebelo, M., Wolpert, N., Adam, C., Hasboun, D., & Tallon-Baudry, C.
noninvasive recordings. This approach holds promise for investigating (2016). Is the cardiac monitoring function related to the self in both
the default network and right anterior insula? Philosophical Transac-
the interplay between the brain and heart with high-time resolution in
tions of the Royal Society of London. Series B, Biological Sciences,
flexible ecological setups without renouncing spatial information. Our 371(1708), 20160004.
results indicate that intracortical BHI increases during CPT with Bagur, S., Lefort, J. M., Lacroix, M. M., de Lavilléon, G., Herry, C.,
respect to resting state, mainly targeting neural oscillations in the θ, β, Chouvaeff, M., Billand, C., Geoffroy, H., & Benchenane, K. (2021).
Breathing-driven prefrontal oscillations regulate maintenance of
and γ bands.
conditioned-fear evoked freezing independently of initiation. Nature
Communications, 12(1), 2605.
AUTHOR CONTRIBUTIONS Barà, C., Zaccaro, A., Antonacci, Y., Dalla Riva, M., Busacca, A., Ferri, F.,
V.C., D.C.R., and G.V. designed the research; V.C., D.C.R., and Faes, L., & Pernice, R. (2023). Local and global measures of information
storage for the assessment of heartbeat-evoked cortical responses.
G.V. performed the research; V.C. and D.C.R. analyzed the data;
Biomedical Signal Processing and Control, 86, 105315.
and V.C., D.C.R., and G.V. wrote the paper. Benarroch, E. E. (1993). The central autonomic network: Functional orga-
nization, dysfunction, and perspective. In Mayo clinic proceedings (Vol.
FUND ING INFORMATION 68, pp. 988–1001). Elsevier.
Benarroch, E. E. (2012). Central autonomic control. In Primer on the auto-
This research has received partial funding from the European Com-
nomic nervous system (pp. 9–12). Elsevier.
mission under grant agreement N. 101017727 for the project EXPE- Candia-Rivera, D. (2023). Modeling brain-heart interactions from Poincaré
RIENCE, and from the Italian Ministry of Education and Research plot-derived measures of sympathetic-vagal activity. MethodsX, 10,
(MIUR) in the framework of the FoReLab and CrossLab projects 102116.
Candia-Rivera, D., Catrambone, V., Barbieri, R., & Valenza, G. (2022). Func-
(Departments of Excellence).
tional assessment of bidirectional cortical and peripheral neural control
on heartbeat dynamics: A brain-heart study on thermal stress. Neuro-
CONF LICT OF IN TE RE ST ST AT E MENT Image, 251, 119023.
The authors declare that they have no competing financial interests or Candia-Rivera, D., Catrambone, V., Thayer, J. F., Gentili, C., & Valenza, G.
personal relationships that may have influenced the work reported in (2022). Cardiac sympathetic-vagal activity initiates a functional brain–
body response to emotional arousal. Proceedings of the National Acad-
this study.
emy of Sciences, 119(21), e2119599119.
Candia-Rivera, D., Catrambone, V., & Valenza, G. (2021). The role of elec-
DATA AVAI LAB ILITY S TATEMENT troencephalography electrical reference in the assessment of func-
This study was approved by the local ethics committee of the Area tional brain–heart interplay: From methodology to user guidelines.
Journal of Neuroscience Methods, 360, 109269.
Vasta Nord-Ovest Toscana. The participants signed an informed con-
Candia-Rivera, D., Norouzi, K., Ramsøy, T. Z., & Valenza, G. (2023).
sent form to participate in the study, as required by the Declaration of Dynamic fluctuations in ascending heart-to-brain communication
Helsinki. None of the participants had any history of neurological, car- under mental stress. American Journal of Physiology-Regulatory, Integra-
diovascular, or respiratory diseases. Raw data of dataset 2 are freely tive and Comparative Physiology, 324(4), R513–R525.
Catrambone, V., Averta, G., Bianchi, M., & Valenza, G. (2021). Toward
available (Averta et al., 2021) and raw data of dataset 1 are
brain-heart computer interfaces: A study on the classification of upper
freely available through reasonable mail requests if the ethical require- limb movements using multisystem directional estimates. Journal of
ments are met. Neural Engineering, 18(4), 046002.
Catrambone, V., Greco, A., Vanello, N., Scilingo, E. P., & Valenza, G. (2019).
Time-resolved directional brain-heart interplay measurement through
ORCID
synthetic data generation models. Annals of Biomedical Engineering, 47,
Vincenzo Catrambone https://orcid.org/0000-0001-9030-7601 1479–1489.
Diego Candia-Rivera https://orcid.org/0000-0002-4043-217X Catrambone, V., Talebi, A., Barbieri, R., & Valenza, G. (2021). Time-
Gaetano Valenza https://orcid.org/0000-0001-6574-1879 resolved brain-to-heart probabilistic information transfer estimation
CATRAMBONE ET AL. 9 of 10

using inhomogeneous point-process models. IEEE Transactions on Bio- Liu, W., Zhang, X., Wu, Z., Huang, K., Yang, C., & Yang, L. (2022). Brain–
medical Engineering, 68(11), 3366–3374. heart communication in health and diseases. Brain Research Bulletin,
Catrambone, V., & Valenza, G. (2021). Functional brain-heart interplay. 183, 27–37.
Springer. Marina, N., Teschemacher, A. G., Kasparov, S., & Gourine, A. V. (2016).
Catrambone, V., & Valenza, G. (2023a). Complex brain-heart mapping in Glia, sympathetic activity and cardiovascular disease. Experimental
mental and physical stress. IEEE Journal of Translational Engineering Physiology, 101(5), 565–576.
in Health and Medicine, 11, 495–504. Mazzola, L., Mauguière, F., & Chouchou, F. (2023). Central control of car-
Catrambone, V., & Valenza, G. (2023b). Nervous-system-wise functional diac activity as assessed by intra-cerebral recordings and stimulations.
estimation of directed brain–heart interplay through microstate occur- Neurophysiologie Clinique, 53(2), 102849.
rences. IEEE Transactions on Biomedical Engineering, 70, 2270–2278. Michel, C. M., & He, B. (2019). Eeg source localization. Handbook of Clinical
Chang, P. F., Arendt-Nielsen, L., & Chen, A. C. (2002). Dynamic changes Neurology, 160, 85–101.
and spatial correlation of eeg activities during cold pressor test in man. Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J.-M. (2011). Fieldtrip:
Brain Research Bulletin, 57(5), 667–675. Open source software for advanced analysis of meg, eeg, and invasive
Chen, W. G., Schloesser, D., Arensdorf, A. M., Simmons, J. M., Cui, C., electrophysiological data. Computational Intelligence and Neuroscience,
Valentino, R., Gnadt, J. W., Nielsen, L., Hillaire-Clarke, C. S., 2011, 1–9.
Spruance, V., Horowitz, T. S., Vallejo, Y. F., & Langevin, H. M. (2021). Park, H.-D., Bernasconi, F., Salomon, R., Tallon-Baudry, C., Spinelli, L.,
The emerging science of Interoception: Sensing, integrating, interpret- Seeck, M., Schaller, K., & Blanke, O. (2018). Neural sources and under-
ing, and regulating signals within the self. Trends in Neurosciences, lying mechanisms of neural responses to heartbeats, and their role in
44(1), 3–16. bodily self-consciousness: An intracranial EEG study. Cerebral Cortex,
Citi, L., Brown, E. N., & Barbieri, R. (2012). A real-time automated point- 28(7), 2351–2364.
process method for the detection and correction of erroneous and Park, H.-D., & Blanke, O. (2019). Heartbeat-evoked cortical responses:
ectopic heartbeats. IEEE Transactions on Biomedical Engineering, 59(10), Underlying mechanisms, functional roles, and methodological consid-
2828–2837. erations. NeuroImage, 197, 502–511.
Cui, J., Wilson, T. E., & Crandall, C. G. (2002). Baroreflex modulation of Pascual-Marqui, R. D., Michel, C. M., & Lehmann, D. (1994). Low resolution
muscle sympathetic nerve activity during cold pressor test in humans. electromagnetic tomography: A new method for localizing electrical
American Journal of Physiology-Heart and Circulatory Physiology, 282(5), activity in the brain. International Journal of Psychophysiology, 18(1),
H1717–H1723. 49–65.
Dirlich, G., Vogl, L., Plaschke, M., & Strian, F. (1997). Cardiac field effects Pascual-Marqui, R. D. (2002). Standardized low-resolution brain elec-
on the eeg. Electroencephalography and Clinical Neurophysiology, tromagnetic tomography (sloreta): Technical details. Methods and
102(4), 307–315. Findings in Experimental and Clinical Pharmacology, 24(Suppl D),
Eckberg, D. L. (2009). Point: Counterpoint: Respiratory sinus arrhythmia is 5–12.
due to a central mechanism vs. respiratory sinus arrhythmia is due to Perlitz, V., Lambertz, M., Cotuk, B., Grebe, R., Vandenhouten, R.,
the baroreflex mechanism. Journal of Applied Physiology, 106, 1740– Flatten, G., Petzold, E. R., Schmid-Schönbein, H., & Langhorst, P.
1742. (2004). Cardiovascular rhythms in the 0.15-Hz band: Common origin
Evans, A. C., Janke, A. L., Collins, D. L., & Baillet, S. (2012). Brain templates of identical phenomena in man and dog in the reticular formation of
and atlases. NeuroImage, 62(2), 911–922. the brain stem? Pflügers Archiv, 448, 579–591.
Faes, L., Marinazzo, D., Jurysta, F., & Nollo, G. (2015). Linear and non- Pfurtscheller, G., Blinowska, K. J., Kaminski, M., Rassler, B., & Klimesch, W.
linear brain–heart and brain–brain interactions during sleep. Physiologi- (2022). Processing of fmri-related anxiety and information flow
cal Measurement, 36(4), 683–698. between brain and body revealed a preponderance of oscillations at
Fardo, F., Vinding, M. C., Allen, M., Jensen, T. S., & Finnerup, N. B. (2017). 0.15/0.16 Hz. Scientific Reports, 12(1), 9117.
Delta and gamma oscillations in operculo-insular cortex underlie Pfurtscheller, G., Schwerdtfeger, A. R., Rassler, B., Andrade, A.,
innocuous cold thermosensation. Journal of Neurophysiology, 117(5), Schwarz, G., & Klimesch, W. (2020). Verification of a central pace-
1959–1968. maker in brain stem by phase-coupling analysis between hr interval-
Gabard-Durnam, L. J., Mendez Leal, A. S., Wilkinson, C. L., & Levin, A. R. and bold-oscillations in the 0.10–0.15 Hz frequency band. Frontiers in
(2018). The harvard automated processing pipeline for electroenceph- Neuroscience, 14, 922.
alography (happe): Standardized processing software for developmen- Pfurtscheller, G., Schwerdtfeger, A. R., Seither-Preisler, A., Brunner, C.,
tal and high-artifact data. Frontiers in Neuroscience, 12, 97. Aigner, C. S., Brito, J., Carmo, M. P., & Andrade, A. (2017). Brain–heart
Hamill, O. P. (2023). Pressure pulsatility links cardio-respiratory and brain communication: Evidence for “central pacemaker” oscillations with a
rhythmicity. Journal of Integrative Neuroscience, 22(6), 143. dominant frequency at 0.1 Hz in the cingulum. Clinical Neurophysiol-
Herrero, J. L., Khuvis, S., Yeagle, E., Cerf, M., & Mehta, A. D. (2018). ogy, 128(1), 183–193.
Breathing above the brain stem: Volitional control and attentional Rolls, E. T., Joliot, M., & Tzourio-Mazoyer, N. (2015). Implementation of a
modulation in humans. Journal of Neurophysiology, 119, 145–159. new parcellation of the orbitofrontal cortex in the automated anatomi-
Julien, C. (2006). The enigma of mayer waves: Facts and models. Cardio- cal labeling atlas. NeuroImage, 122, 1–5.
vascular Research, 70(1), 12–21. Schiecke, K., Pester, B., Piper, D., Benninger, F., Feucht, M., Leistritz, L., &
Keller, M., Pelz, H., Perlitz, V., Zweerings, J., Röcher, E., Baqapuri, H. I., & Witte, H. (2016). Nonlinear directed interactions between hrv and eeg
Mathiak, K. (2020). Neural correlates of fluctuations in the intermedi- activity in children with tle. IEEE Transactions on Biomedical Engineering,
ate band for heart rate and respiration are related to interoceptive per- 63(12), 2497–2504.
ception. Psychophysiology, 57(9), e13594. Shao, S., Shen, K., Wilder-Smith, E. P., & Li, X. (2011). Effect of pain per-
Klimesch, W. (2018). The frequency architecture of brain and brain body ception on the heartbeat evoked potential. Clinical Neurophysiology,
oscillations: An analysis. European Journal of Neuroscience, 48(7), 122(9), 1838–1845.
2431–2453. Shao, S., Shen, K., Yu, K., Wilder-Smith, E. P., & Li, X. (2012). Frequency-
Klimesch, W. (2023). Heartbeat, brain oscillations and body awareness: A domain eeg source analysis for acute tonic cold pain perception. Clini-
commentary. Journal of Integrative Neuroscience, 22(6), 155. cal Neurophysiology, 123(10), 2042–2049.
Kluger, D. S., & Gross, J. (2020). Depth and phase of respiration modulate Silvani, A., Calandra-Buonaura, G., Dampney, R. A., & Cortelli, P. (2016).
cortico-muscular communication. NeuroImage, 222, 117272. Brain–heart interactions: Physiology and clinical implications.
10 of 10 CATRAMBONE ET AL.

Philosophical Transactions of the Royal Society A: Mathematical, Physical Valenza, G., Sclocco, R., Duggento, A., Passamonti, L., Napadow, V.,
and Engineering Sciences, 374(2067), 20150181. Barbieri, R., & Toschi, N. (2019). The central autonomic network at
Skora, L. I., Livermore, J. J. A., & Roelofs, K. (2022). The functional role of rest: Uncovering functional mri correlates of time-varying autonomic
cardiac activity in perception and action. Neuroscience & Biobehavioral outflow. NeuroImage, 197, 383–390.
Reviews, 137, 104655. Young, H., & Benton, D. (2015). We should be using nonlinear indices
Tadel, F., Baillet, S., Mosher, J. C., Pantazis, D., & Leahy, R. M. (2011). when relating heart-rate dynamics to cognition and mood. Scientific
Brainstorm: A user-friendly application for meg/eeg analysis. Computa- Reports, 5(1), 16619.
tional Intelligence and Neuroscience, 2011, 1–13. Zelano, C., Jiang, H., Zhou, G., Arora, N., Schuele, S., Rosenow, J., &
Truter, N., Malan, L., & Essop, M. F. (2023). Glial cell activity in cardiovas- Gottfried, J. A. (2016). Nasal respiration entrains human limbic oscilla-
cular diseases and risk of acute myocardial infarction. American Jour- tions and modulates cognitive function. Journal of Neuroscience,
nal of Physiology-Heart and Circulatory Physiology, 324(4), H373– 36(49), 12448–12467.
H390.
Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F.,
Etard, O., Delcroix, N., Mazoyer, B., & Joliot, M. (2002). Automated SUPPORTING INF ORMATION
anatomical labeling of activations in spm using a macroscopic anatomi- Additional supporting information can be found online in the Support-
cal parcellation of the mni mri single-subject brain. NeuroImage, 15(1),
ing Information section at the end of this article.
273–289.
Valenza, G., Di Ciò, F., Toschi, N., & Barbieri, R. (2024). Sympathetic and
parasympathetic central autonomic networks. Imaging Neuroscience, 2,
1–17. How to cite this article: Catrambone, V., Candia-Rivera, D., &
Valenza, G., Passamonti, L., Duggento, A., Toschi, N., & Barbieri, R. Valenza, G. (2024). Intracortical brain-heart interplay: An EEG
(2020). Uncovering complex central autonomic networks at rest: A model source study of sympathovagal changes. Human Brain
functional magnetic resonance imaging study on complex cardiovas-
Mapping, 45(6), e26677. https://doi.org/10.1002/hbm.26677
cular oscillations. Journal of the Royal Society Interface, 17(164),
20190878.