Abstract
A growing body of evidence suggests that the link between the cardiac autonomic nervous system (ANS) and the central nervous system (CNS) is crucial to the onset and development of major depressive disorder (MDD), affecting perception, cognition, and emotional processing. The bottom-up heart–brain communication pathway plays a significant role in this process. Previous studies have shown that slow-frequency oscillations of peripheral signals (e.g., respiration, stomach) can influence faster neural activities in the CNS via phase–amplitude coupling (PAC). However, the understanding of heart–brain coupling remains limited. Additionally, while MDD patients exhibit altered brain activity patterns, little is known about how heart rate variability (HRV) affects brain oscillations. Therefore, we used PAC to investigate heart–brain coupling and its association with depression. We recorded MEG and ECG data from 55 MDD patients (35 females) and 52 healthy subjects (28 females) at rest and evaluated heart–brain PAC at a broadband level. The results showed that the low-frequency component of HRV (HRV-LF) significantly modulated MEG alpha power (10 Hz) in humans. Compared with the healthy group, the MDD group exhibited more extensive heart–brain coupling cortical networks, including the pars triangularis. LF-alpha coupling was observed in the bilateral insula in both groups. Notably, results revealed a significantly increased sympathetic-dominated HRV-LF modulation effect on left insula alpha oscillations, along with increased depressive severity. These findings suggest that MDD patients may attempt to regulate their internal state through enhanced heart–brain modulation, striving to restore normal physiological and psychological balance.
- autonomic nervous system
- depression
- heart–brain interplay
- heart rate variability
- MEG
- phase–amplitude coupling
Significance Statement
The afferent pathway from the heart plays a pivotal role in conveying information to the brain. This process involves the transmission of signals related to the physiological state of the heart. Our understanding of this pathway and its association with major depressive disorders (MDDs) remains limited. In this study, the low-frequency component of heart rate variability (HRV-LF) was found to modulate neural activity during rest, revealing a bottom-up information transmission mechanism between the cardiac ANS and the CNS. Alterations in the LF-alpha coupling pattern were observed in patients with MDD, suggesting this as a potential neurobiological mechanism behind their altered interoception, which might affect the perception and emotional processing.
Introduction
Major depressive disorder (MDD) is a prevalent mental health disorder that impacts emotional well-being and can cause physical symptoms like palpitations and fatigue. There is now substantial evidence demonstrating that depression increases the risk of cardiovascular disease occurrences and mortality (Penninx, 2017). Additionally, the prevalence of MDD is higher among individuals with heart disease (Li et al., 2022). These suggest a potential bidirectional relationship between depressive and cardiac disorders, where each condition exacerbates the other, creating a vicious cycle (Catrambone et al., 2021).
A number of existing studies have demonstrated that the interplay between cardiac autonomic nervous system (ANS) and central nervous system (CNS) holds potential as a biomarker for assessing health and emotional disorders (Engelen et al., 2023a). An EEG study reported a positive correlation between HRV and delta power in depressive patients during a working memory task (Lee et al., 2022). The connectivity networks involving the orbitofrontal cortex, which correlated with HRV indices, showed differences between the MDD and healthy control groups during the execution of no-go tasks (Zhou et al., 2020). Such neural substrate findings implied that dysfunctional interactions between the brain and heart could be pivotal in the initiation and development of mental disorders.
Previous studies have primarily used correlation analysis to examine heart–brain interplay, which overlooks its bidirectional nature. On the one hand, the brain mediates cardiac autonomic output through the central autonomic network, providing top-down regulation (Heiss et al., 2021). On the other hand, cardiac signals also reach both cortical and subcortical areas, including anterior cingulate cortex, medial prefrontal cortex, and bilateral insula, which contribute indispensably in processing sensory information and affective reasoning (Haruki and Ogawa, 2023). Therefore, correlation analyses are insufficient for understanding the distinct roles in brain–body communication pathways.
Notably, the afferent pathway from the heart plays a pivotal role in conveying physiological information to the brain. This pathway transmits signals related to the heart's activity, including rhythm and pressure changes (Bonaz et al., 2021). The heart-evoked potential (HEP) is a common approach for studying this pathway, reflecting specific potential changes in the brain that are generated in response to its own heartbeats (Park and Blanke, 2019). In MDD, HEP during the emotional anticipation phase was significantly reduced, indicating that interoceptive dysfunction may affect processing of emotional information (Zhou et al., 2022). However, HEP is limited to specific R-peak periods and does not capture the complete dynamics of ANS phase information.
Recent studies suggest that phase–amplitude coupling (PAC) effectively characterizes brain–body interactions, with low-frequency body rhythms influencing high-frequency brain oscillations (Liu et al., 2022). Kluger and Gross (2021) found respiratory phase modulated the global field signal amplitude in resting-state MEG. Similarly, hypogastric activity phases influenced alpha rhythm amplitudes in the occipito-parietal region (Richter et al., 2017). In fact, cardiac ANS rhythms can be transmitted afferently in real time to the brain via the parasympathetic vagal and sympathetic nervous systems to regulate spontaneous brain activity (Parviainen et al., 2022). Raut et al. (2021). found significant coherence between HRV and blood oxygen level-dependent signals, particularly in the low-frequency range (0.01–0.1 Hz). Candia-Rivera et al. (2022) showed HRV signals significantly influence emotional arousal by integrating experiences in the CNS. Furthermore, research has confirmed that altering the oscillatory amplitude of HRV signals through biological feedback can enhance emotional well-being (Critchley et al., 2011). Thus, studies of how CNS integrates afferent cardiac ANS information outflow may help us gain a deeper understanding of heart–brain interactions in MDD.
In light of the above, we hypothesize that slow HRV oscillation phases modulate high-frequency brain rhythms, reflecting a stable cardiac ANS–CNS connection. Variability in the intensity of awareness regarding changes in body states may inform the development of depression. To test this, we recorded magnetoencephalographic (MEG) and electrocardiogram (ECG) data from 55 MDD patients and 52 healthy subjects (HS) at rest and evaluated heart–brain PAC at a broadband level.
Materials and Methods
Participants
The study initially included 55 participants in the MDD group and 52 in the HS group. After data quality assessment, three participants in the MDD group were excluded due to excessive MEG artifacts and two due to poor ECG quality. In the HS group, two participants were excluded for similar reasons. The final sample consisted of 50 participants in each group (MDD: mean age 24.8 ± 4.31 years, 33 females; HS: mean age 23.92 ± 2.72 years, 26 females; Table 1). There was no significant difference in the age and gender ratio between the two groups (p > 0.05).
Demographic and clinical characteristics of study subjects
Participants in the MDD group were recruited from the Department of Psychiatry at the Affiliated Brain Hospital of Nanjing Medical University. They were enrolled in the study if their Hamilton Depression Rating Scale 17-item (HAMD-17) score was no less than 17. None of these participants had used antidepressant medications or drugs affecting the autonomic or central nervous systems within the last 2 weeks before scanning. Additionally, they refrained from consuming beverages containing caffeine or alcohol for 2 h before data collection.
This study was approved by the Research Ethics Review Board of the Affiliated Brain Hospital of Nanjing Medical University and was in accordance with the Helsinki declaration. All participants were informed about the procedures of the experiment and signed a written informed consent.
Procedure and recordings
MEG data were acquired using a 275-channel whole-head MEG system (Omega 2000, VSM MedTech). Each participant placed himself or herself in a supine position inside the magnetically shielded room, with their eyes open and fixed on a white crosshair in the center of a black background screen for 5 min. All participants were instructed to make efforts to minimize head and body movements. During the recording, the positions of the head-marker coils were measured to determine the position and orientation of the MEG sensors relative to the participant's head, and the maximum acceptable difference in position was 5 mm. ECG measurements were conducted using a dual-channel input electrode configuration per the MEG system guidelines. Electrodes were placed on the radial arteries of both wrists, with a reference electrode on the one hand and a ground electrode on the other to minimize noise. Additionally, we recorded horizontal and vertical electrooculogram (EOG) signals to eliminate blink artifacts. Three pairs of EOG electrodes were used, with one pair positioned ∼2.5 cm lateral to the outer canthi of both eyes, and the other two pairs placed above and below both eyes, aligned with the pupils. MEG data were collected at a sampling rate of 1,200 Hz and a frequency range of 0–600 Hz.
For MEG source localization, high-resolution T1-weighted structural magnetic resonance imaging (MRI) were obtained by using a Siemens Verio 3.0 Tesla MRI scanner. We employed standard Siemens weighted imaging sequences with the following specific scanning parameters: TR, 1,900 ms; TE, 2.48 ms; 250 × 250 mm2 field of view; 192 slices; 1 × 1 × 1 mm3 voxel size. MRI measurement was conducted in supine position. To align MEG data with individual anatomical images, three energizing coils were placed at the nasion as well as left and right preauricular points to determine the participant's head location within the scanner.
MEG data preprocessing
We performed a semiautomatic labeling algorithm to annotate muscle activity. Briefly, we bandpass filtered the raw data at 110–140 Hz and z-scored them (Ferrante et al., 2022). Data segments exceeding a predefined threshold (e.g., the 95th percentile) were marked as muscle artifacts, with a minimum data length of 0.2 s. Afterward, we visually inspected the MEG data to detect remaining segments that might contain environmental or instrumentation artifacts. Channels that malfunctioned or had variances greater than
ECG data preprocessing
We performed offline preprocessing on the ECG data to extract HRV series for subsequent PAC analysis. The raw data were first high-pass filtered at 0.5 Hz using a Butterworth filter of order 5 to eliminate baseline drift and then were applied a notch filter at 50 Hz. R-peaks in continuous QRS waves were detected based on the steepness of the absolute gradient of the ECG signal, as implemented in Neurokit2 toolbox (Makowski et al., 2021). Beat-to-beat intervals (RRIs) were derived by calculating the first-order forward difference of R-peaks. We corrected R-peak misdetections by visually inspecting the detection results. All RRI values constitute the HRV series. Because ECG artifacts often result in missing R-waves and outlier RRI values, we removed these outliers from raw HRV series. Then we evenly resampled HRV series at 4 Hz using quadratic interpolation. The HRV series was analyzed in both the low-frequency (LF: 0.04–0.15 Hz) and high-frequency (HF: 0.15–0.4 Hz) ranges to quantify sympathovagal and parasympathetic activity from the ANS, respectively.
Source reconstruction
For source localization, we first constructed subject-specific anatomical models using FreeSurfer and MNE-Python. Subsequently, we built a source model based on the single-shell boundary element model and the gray-white matter boundary interface at the oct6 space, which consists of 4,098 source points for each hemisphere. The linearly constrained minimum variance (LCMV) beamforming algorithm was applied to estimate cortical point currents for each participant. The Tikhonov regularization parameter for the LCMV filter was set to 0.05 to regularize the data covariance matrix, balancing noise sensitivity and spatial resolution. Additionally, unit-noise gain normalization was applied to reduce power bias toward central brain locations. The optimal set of weights for spatial filters was obtained by minimizing the L2 norm of their output with unit-gain constraint. Source time series at each cortical point for each participant were derived to calculate heart–brain PAC at the source level.
Evaluation of heart–brain phase–amplitude coupling
We used the modulation index (MI) to evaluate the intensity of PAC between HRV and MEG series. MI quantifies the extent to which the amplitude of brain oscillations at different frequencies is modulated by the phase of HRV, as derived from the normalized Kullback–Leibler (KL) distance of the phase–amplitude distribution from a uniform distribution (Tort et al., 2010). To calculate MI for each sensor, frequency, and participant, we first applied the Hilbert transform to extract the phase of HRV series and the amplitude of MEG series, respectively. Subsequently, the low-frequency phases were divided into 18 intervals, and the average high-frequency amplitudes corresponding to each interval were computed. We normalized the mean amplitude by dividing each bin value by the sum over the bins to create the phase–amplitude plot. The frequency of the phase was divided into LF and HF, while the frequency of high-frequency MEG ranged from 1 to 100 Hz with a step of 1 Hz. MI values for each subject on each channel were then calculated for these frequency combinations.
To locate heart–brain PAC in specific brain regions and identify differences in modulation patterns between DP and HS, based on the results obtained from PAC analysis at the sensor level, we filtered the HRV time series to the LF frequency bands due to the significant LF-brain modulation observed in both groups and performed LF-brain PAC analysis in the source space (Fig. 1D). Specifically, we calculated the MI values for each LF-source time series pair. The MI values for each cortical point were then divided into 68 cortical regions from the Desikan–Killiany atlas (DK), and an average value was computed for each cortical region.
The flowchart of heart–brain phase–amplitude coupling analysis. A, Recruitment of subjects, with 55 people in the MDD group and 52 in the HS group. B, Simultaneous collection of resting-state MEG and ECG data for 5 min. C, Preprocessing pipeline of raw MEG and ECG data. Artifact-free MEG and HRV signals were obtained to calculate heart–brain coupling. D, Determination of significant heart–brain PAC clusters at the sensor level. The amplitude (bold dark line) of brain oscillations and the phase (bold red line) of HRV time series were determined via Hilbert transform. MI values for each subject on each channel were then calculated for each phase–amplitude frequency combination. Then, the cluster-based spatial-frequency permutation testing was used to determine the significant frequency bands of PAC for each group. E, LF-brain PAC analysis at the source space. We used LCMV to reconstruct source time series and employed the DK atlas to extract cortex point currents per voxel for ROI brain regions. Then parcel oscillations were filtered to 1–100 Hz to calculate MI values. The significant heart–brain coupling brain regions for each group were determined using the same cluster-based permutation method as at the sensor level. MDD, major depressive disorder; HS, heathy subjects; MEG, magnetoencephalography; ECG, electrocardiogram; HRV, heart rate variability; MI, modulation index; PAC, phase–amplitude coupling; LCMV; linearly constrained minimum variance.
Statistical analysis for determining clusters with significant heart–brain PAC at the sensor level
The significance of sensor-level MI values for each group at each sensor and frequency was determined via surrogate control analysis and cluster-based permutation testing. Surrogate control analysis was performed to determine whether the observed MI value differs from what would be expected by chance. We repeated the random cutting of amplitude signals at a single point and exchanged the two resulting time courses (Aru et al., 2015). Accordingly, the shuffled data preserved the original data distribution while disrupting the inherent dynamics between the two sequences. We generated 500 surrogate datasets to obtain a surrogate MI distribution, reflecting the chance level of heart–brain PAC (Fig. 2). Then the significance of MI values at the group level was determined using the cluster-based spatial-frequency permutation testing implemented in MNE. Specifically, one-tailed paired t tests were conducted on the MI values and the 95th percentile of the surrogate MI distribution for each HRV and MEG frequency combination. The cluster-forming threshold was then set at
Schematic diagram of the surrogate control analysis. Compute surrogates by swapping amplitudes time blocks. The amplitude series was cut at a random time point to calculate the surrogate MI values. We assessed 500 surrogate MI values for each combination of HRV and MEG frequency bands, as well as for each channel and participant. The chance level MI value was determined by the 95th percentile of the surrogate dataset. MEG, magnetoencephalography; HRV, heart rate variability; MI, modulation index.
Statistical analysis for determining regions with significant LF-brain PAC and between-group differences at the source level
Similar to the sensor-level statistical method, surrogate distributions of MI values were computed for each pair of virtual HRV and source time series. We defined an adjacency matrix based on the DK atlas and used cluster-based spatial-frequency permutation testing to identify cortical regions with significant LF-brain PAC in the two groups at a broad band (1–100 Hz). The statistical procedures and parameter settings were consistent with those at the sensor level. To further compare the coupling differences in coupling strength between the two groups in these regions, we calculated the normalized MI value
Granger causality analysis
We employed Granger causality (GC) analysis to investigate the causal relationship between LF and 10 Hz brain alpha power. This analysis assessed whether LF significantly predicts alpha power in brain regions (from LF to brain) and vice versa (from brain to LF), thus determining the directionality of the causal relationship. The maximum lag in the Granger causality test was determined according to the Schwarz information criterion (Schwarz, 1978). For the significant LF-alpha coupling brain regions in the MDD and HS groups, we calculated the GC statistic using the SSR chi-square test. Statistical significance was set at p = 0.05, and we reported the number of participants with significant results in each direction for each group. Additionally, paired t tests were conducted to compare the F statistics between the two directions.
Results
Frequencies with significant modulation of cardiac ANS's rhythms on resting-state brain oscillations
For each participant in the MDD and HS groups, MEG and ECG signals during the resting state were obtained to investigate whether HRV modulates oscillatory brain activity. MI was used to measure the bottom-up phase–amplitude modulation from cardiac ANS to CNS, where sympathetic and parasympathetic activity of the cardiac ANS was represented by HRV-LF band and HRV-HF band and CNS activity was studied from 1 to 100 Hz using MEG signals (Fig. 1D). Then we computed surrogate MI values and compared true MI with surrogate MI to find significant coupling clusters using frequency-space cluster-based permutation testing for both groups. Our statistical analysis revealed significant LF-brain coupling in the MEG low-frequency range (i.e., θ, α) for both groups. Figure 3, A and B, displays the statistical results of significant clusters for LF-brain coupling in the MDD and HS groups, respectively. The left panel shows the group-level statistics topomap of significant frequency bands. The right panel (top) presents the total t values of significant sensors across the broad band, while the bottom shows the mean normalized MI value for these significant sensors across the same frequency range. Two significant clusters were observed for both groups: one in the θ frequency range (6 Hz) and another in the α frequency range (10 Hz). In the MDD group, the significant cluster comprised 20 channels for LF-θ coupling (sum t values = 52.14, p = 0.0253) and 22 channels for LF-α coupling (sum t values = 53.60, p = 0.0162). In the HS group, LF-θ coupling involved 19 channels (sum t values = 48.17, p = 0.0365), while LF-α coupling comprised 24 channels (sum t values = 55.71, p = 0.0245). The statistically significant LF-θ coupling regions were primarily located in the right frontal area of the maps, while significant LF-α coupling occurred in the left frontotemporal area, extending to the central axis. However, no significant HF-brain coupling cluster during the resting state was observed in either MDD or HS group.
Statistical results of heart–brain PAC at the sensor-space for each group. A, The left panel displays the group-level topomap of significant frequency band t values, while the right panel shows the total t values of significant sensors across the 1–100 Hz range at the top and the mean normalized MI value for these sensors at the bottom. In the MDD group, the significant cluster comprised 20 channels for LF-θ coupling and 22 channels for LF-α coupling. B, In the HS group, LF-θ coupling involved 19 channels, while LF-α coupling comprised 24 channels. MDD, major depressive disorder; HS, heathy subjects; MEG, magnetoencephalography; MI, modulation index; PAC, phase–amplitude coupling.
Regions with significant heart–brain modulation in low-frequency band
We further investigated the cortical regions and frequency bands that show LF-brain coupling at the source level. To this end, preprocessed artifact-free MEG epochs were used for source reconstruction. Then, we calculated the MI value for the estimated cortical currents at each vertex and grouped them into 68 cortical areas on the DK atlas. The averaged MI value for each cortical area characterized heart–brain PAC. To detect significant coupling cortical areas within each group, we compared averaged MI value with the chance-level MI values based on cluster-based spatial-frequency permutation testing method. In both groups, there were two significant clusters of LF-modulated brain amplitude at 10 Hz in the left and right hemispheres, respectively. MI values for LF-α PAC were significantly higher than the phase-shuffled MI values in the bilateral insula, entorhinal, pars triangularis, and left pars orbitalis in the MDD group (Fig. 4). In contrast, significant LF-α coupling cortical areas in the HS control group only included the bilateral insula and left entorhinal. Notably, no significant LF-brain coupling areas at the source level were evident for LF-θ PAC.
Cortical areas and frequency bands with significant LF-brain phase–amplitude coupling for each group. A, The total t values and mean normalized MI values of significant clusters for MDD group (left) and HS group (right). B, The statistical t values brain map for the paired t test at 10 Hz in both groups. lh, left hemisphere; rh, right hemisphere; Ins, insula; Tr, pars triangularis; Or, pars orbitalis; En, entorhinal; MDD, major depressive disorder; HS, heathy subjects.
Increased LF-alpha phase–amplitude coupling in the left insula in MDD
To compare the heart–brain coupling strength between two groups of people and identify brain regions with significant LF-α coupling differences, we first calculated the normalized MI values based on LF-α coupling, which were computed by subtracting the mean and then dividing by the deviation of the surrogate dataset for each brain area. We then employed a cluster-based paired t permutation test to compare the normalized MI values across all frequencies and parcels between the two groups. One significant cluster of coupling differences was observed at 10 Hz (Fig. 5A). In MDD patients, the LF-α coupling MI values in the left insula were significantly higher than those of the HS group (sum t values = 3.63, p = 0.0045). The normalized MI difference and t test statistical values brain map at 10 Hz was shown in Figure 5B,C, respectively.
Increasing LF-alpha PAC in the left insula characterizes MDD. A, The group difference statistical t values (top) and normalized MI differences (bottom) between groups in the left insula across all MEG frequencies. B, The normalized MI difference brain map for LF-α PAC at 10 Hz. C, The group difference statistical t values brain map for LF-α PAC at 10 Hz. D, Spearman correlation between HAMD-17 total score and normalized MI values of LF-α PAC at 10 Hz in the left insula in MDD group. MI, modulation index; lh, left hemisphere; L, left; Ins, insula; MDD, major depressive disorder; PAC, phase–amplitude coupling.
Correlation between LF-alpha PAC and severity of symptoms of depression
For the cortical areas with significant LF-α PAC at 10 Hz, we conducted a Spearman correlation analysis to evaluate the correlation between the regional-level normalized MI value and the severity of depressive symptoms, as measured by the 17-item Hamilton Depression Rating Scale (HAMD-17) total score in MDD patients. The results showed that the LF-α coupling MI values of the left insula were significantly positively correlated with the HAMD-17 total scores (Spearman's ρ = 0.41, p = 0.0029; Fig. 5D). We applied the Benjamini–Hochberg (BH) procedure for FDR correction (Benjamini and Hochberg, 1995), and the result remained significant (adjusted p = 0.02). However, no FDR-corrected significant correlations with the HAMD-17 total score were observed in other significantly coupled cortical regions.
Heart to brain Granger causality effect
Granger causality analysis was conducted on the significantly coupled LF-alpha brain regions in both the MDD and HS groups. The results indicated that all brain regions exhibited causal effects from heart to brain in over half of the participants, while only a small proportion demonstrated a reverse effect from brain to heart. The remaining participants did not show significant causal effects. Detailed results of these causal relationships are provided in Table 2. Pairwise t tests comparing the F statistics for these two directions revealed that the causal effect from heart to brain was significantly dominant in both groups (all p values < 0.001).
Granger causality analysis results between HRV-LF and significant LF-alpha coupling brain regions power in both groups
Discussion
In the present study, we evaluated the functional phase–amplitude coupling between the cardiac ANS activities and brain activities in MDD compared with the healthy subjects. Our findings revealed significant modulation of the MEG signals in the theta and alpha bands by the phase of low-frequency band of HRV signals at the channel level. We further reconstructed the source time series of the MEG signals and detected significant LF-alpha coupling in the bilateral insula and left entorhinal in both groups. Compared with the healthy subjects, extra regions were identified in patients with MDD, such as the bilateral pars triangularis and left pars orbitalis, indicating a more extensive cortical network affected by LF HRV signals. Additionally, the MI value for LF-α coupling in the left insula was significantly higher in MDD group than that in HS group. The severity of depression, evaluated by the HAMD-17 total score, significantly correlated with MI values for resting-state LF-α coupling in the left insula. These together suggested that the increased LF-α PAC in the left insula during the resting state is a potential index for depression in the aspect of heart–brain modulation.
The first main finding of this study demonstrated the existence of heart–brain PAC. The low-frequency component of HRV, also known as Mayer waves, typically ranging from 0.04 to 0.15 Hz, represents rhythmic fluctuations associated with blood pressure variation (Hirabayashi and Iwamoto, 2019). Mara Mather and colleagues proposed the hypothesis that slow oscillation frequencies, which overlap with the HRV range, can modulate faster frequencies of neural activity (Mather and Thayer, 2018; Cho et al., 2023). The present study provides experimental evidence for this form of heart–brain connection in humans. A plausible physiological mechanism underlying this coupling effect involves heartbeats inducing blood pressure changes detected by baroreceptors in the aortic arch and carotid sinus, generating low-frequency oscillations. These signals are transmitted to central nervous system regions via the vagus and sympathetic nerves (Parviainen et al., 2022; Corcoran et al., 2023). Through cross-frequency coupling, these low-frequency rhythms may modulate higher-frequency oscillations and influence upstream brain regions, potentially contributing to cognitive processes and behavior. Our results support the existence of an association between the brain and heart activities, suggesting that the LF component of cardiac-brain autonomic activity might act as an external oscillator that restricts the brain's spontaneous fluctuations. This may have significant implications for understanding how psychological states can be influenced through physiological feedback and other intervention measures.
We further detected significant increased LF-α PAC in the left insula of patients with MDD and demonstrated that MI value of the left insula was positively correlated with the severity of depression. Notably, previous work has evidenced that there is asymmetry in brain alpha activity in patients with MDD at rest, with reduced activity in the left frontal lobe and potentially increased activity in the right frontal lobe (Debener et al., 2000). Moreover, sympathetic activity is under tonic inhibitory control by the prefrontal cortex (Shokri-Kojori et al., 2023). Accordingly, we suggest that the tonic inhibitory control of the sympathetic nervous system by frontal lobe is weakened and the HRV-LF activity dominated by the sympathetic nerves is enhanced; thus the bottom-up modulation in the insula is strengthened. This leads to an increase in heart–brain coupling in patients with MDD, which in turn imposes restrictions on insula activity. The left insula transmits internal information to higher-order regions through neural pathways, subsequently restricting activities in the left frontal cortex, which might affect patients’ decision-making and behavior (Fig. 6).
The mechanism of LF-alpha PAC alterations in the left insula related to heart–brain connection in MDD. The tonic inhibitory control of the sympathetic nervous system by frontal lobe is weakened and the HRV-LF activity dominated by the sympathetic nerves is enhanced, thus the bottom-up modulation in the insula is strengthened. This leads to an increase in heart–brain coupling in patients with MDD, which in turn imposes restrictions on insula activity.
The insula plays a crucial role in this process as it processes complicated information related to emotions and bodily sensations, and its dysfunction is closely linked to the onset and progression of depression (Avery et al., 2014; Zhao et al., 2022). First, it directly receives and processes signals from the body's internal signals (Engelen et al., 2023b). Second, the insula is connected to many other brain regions, including connections with the frontal cortex and cingulate cortex, and the network connections between these areas may be affected, resulting in changes in emotion, cognition, and bodily functions (Gogolla, 2017). Some studies have proposed that the insula exhibits a lateralization effect in response to emotional perception and experience, suggesting that the left insula is primarily involved in receiving and processing emotion-related information and plays a crucial role in an individual's lower-order neural responses (Duerden et al., 2013a, 2013b). Therefore, in this context, the function of the left insula may extend beyond emotion and body state perception to include coordination with physiological responses of the body signals, which may underlie pathogenesis behind the abnormal heart–brain connection in MDD. That is to say, the left insular response to cardiac activity is an integration rather than a simple superposition. Additionally, several neuroimaging studies have shown that the decreased interoceptive accuracy of patients with depression during heartbeat detection tasks is associated with hypoactivation of the insula (Avery et al., 2014). Our results may provide a new direction for understanding the neural mechanisms behind this interoceptive dysregulation.
Interestingly, our findings showed that this modulatory effect of cardiac sympathetic activity can be traced to a wider cortical area in MDD, indicating that other regions (i.e., bilateral pars triangularis and pars orbitalis) also participate in the integration of internal signals in addition to the bilateral insula. In fact, the pars triangularis and pars orbitalis play important roles in emotion processing. The former is involved in processing emotional stimuli and distinguishing between positive and negative emotional responses (Van der heiden et al., 2014), while the latter participates in handling painful and pleasurable emotional responses (Rolls et al., 2023; Xi et al., 2023). Allostatic strain, as a source of motivation, drives individuals to seek a balanced and stable psychological state (McEwen, 1998). For example, when an individual faces stress, the body preemptively adjusts hormone levels, heart rate, and energy distribution to best cope with the stress (Candia-Rivera et al., 2023). Long-term or frequent allostatic load can lead to what is called “allostatic overload,” which may negatively impact health. Therefore, we suggest that enhanced coupling between HRV and the brain's alpha rhythms might indicate that, in a state of depression, the body is attempting to regulate internal states through tighter cardio-cerebral coordination. This altered coupling could be a sign of the brain's over-regulation response as the body strives to restore normal physiological and psychological equilibrium. However, accurate interpretation of these findings requires more in-depth research to explore the specific mechanisms behind the increased coupling of HRV and brain alpha rhythms and how this coupling affects the behavior and feelings of patients with MDD.
Despite the important findings of this study, there are several limitations that need to be addressed. Although we observed coupling effects within a narrow frequency band, our findings may be incomplete due to the methodological limitations of PAC in capturing aperiodic signal changes. Additional forms of heart–brain coupling may exist and warrant further exploration. We also need to note that this study primarily conducted measurements in a resting state and did not examine the heart–brain coupling under different tasks or contexts. Different contexts and tasks might influence heart–brain coupling differently, and future research could explore these dynamic changes through task design.
In this study, the low-frequency component of heart rate variability was found to modulate neural activity during rest, revealing a bottom-up information transmission mechanism between the cardiac ANS and the CNS. Alterations in the LF-alpha coupling pattern were observed in patients with MDD, suggesting this as a potential neurobiological mechanism behind their altered interoception, which might affect the perception and emotional processing. Our findings provide a new perspective for future research focused on improving depression treatment, such as psychotherapy, pharmacotherapy, or heart rate variability biofeedback training, which aims at modulating the autonomic nervous system and thereby improving emotional regulation capabilities.
Footnotes
The study was supported by the National Natural Science Foundation of China under Grant No. 82151315 and No. 82271568; Key Project supported by Suzhou Science and Technology Plan under Grant No. 2022SS04; and Jiangsu Medical Innovation Center for Mental Illness under Grant No. CXZX202226.
The authors declare no competing financial interests.
- Correspondence should be addressed to Qing Lu at luq{at}seu.edu.cn or Zhijian Yao at zjyao{at}njmu.edu.cn.
