Altered Relationship Between Heart Rate Variability and fMRI-Based Functional Connectivity in People With Epilepsy

Abstract

Background: Disruptions in central autonomic processes in people with epilepsy have been studied through evaluation of heart rate variability (HRV). Decreased HRV appears in epilepsy compared to healthy controls, suggesting a shift in autonomic balance toward sympathetic dominance; recent studies have associated HRV changes with seizure severity and outcome of interventions. However, the processes underlying these autonomic changes remain unclear. We examined the nature of these changes by assessing alterations in whole-brain functional connectivity, and relating those alterations to HRV. Methods: We examined regional brain activity and functional organization in 28 drug-resistant epilepsy patients and 16 healthy controls using resting-state functional magnetic resonance imaging (fMRI). We employed an HRV state-dependent functional connectivity (FC) framework with low and high HRV states derived from the following four cardiac-related variables: 1. RR interval, 2. root mean square of successive differences (RMSSD), 4. low-frequency HRV (0.04-0.15 Hz; LF-HRV) and high-frequency HRV (0.15-0.40 Hz; HF-HRV). The effect of group (epilepsy vs. controls), HRV state (low vs. high) and the interactions of group and state were assessed using a mixed analysis of variance (ANOVA). We assessed FC within and between 7 large-scale functional networks consisting of cortical regions and 4 subcortical networks, the amygdala, hippocampus, basal ganglia and thalamus networks. Results: Consistent with previous studies, decreased RR interval (increased heart rate) and decreased HF-HRV appeared in people with epilepsy compared to healthy controls. For both groups, fluctuations in heart rate were positively correlated with BOLD activity in bilateral thalamus and regions of the cerebellum, and negatively correlated with BOLD activity in the insula, putamen, superior temporal gyrus and inferior frontal gyrus. Connectivity strength in patients between right thalamus and ventral attention network (mainly insula) increased in the high LF-HRV state compared to low LF-HRV; the opposite trend appeared in healthy controls. A similar pattern emerged for connectivity between the thalamus and basal ganglia. Conclusion: The findings suggest that resting connectivity patterns between the thalamus and other structures underlying HRV expression are modified in people with drug-resistant epilepsy compared to healthy controls.

Keywords: SUDEP; insula cortex; state-dependent functional connectivity; sympathovagal balance; thalamic connectivity; ventral attention network.

Figure 1

Comparison of mean RR interval and heart rate variability (HRV) measures between epilepsy patients and healthy controls. The bottom and top of each box correspond to the 25th and 75th percentiles of the sample distribution, the line in the box corresponds to the median and the crosses indicate outliers, defined as values that are more than 1.5 times the interquartile range away from the edges of the box. The epilepsy patients showed significantly lower values of mean RR interval and HR-HRV than healthy controls. n.s., not significant.

Figure 2

Association of regional BOLD fluctuations with changes in heart rate at the group level. Statistical map of one-sample t-test considering patients and controls (n = 44), thresholded with a voxel-wise threshold of family-wise error (FWE) rate p < 0.05 corrected for multiple comparisons using the random-field theory (51). The unthresholded statistical map is available at https://neurovault.org/collections/9452/.

Figure 3

Mixed analysis of variance (ANOVA) for evaluating the effects of the group (epilepsy vs. controls; 1st column) and HRV state (low vs. high, patients and healthy controls; 2nd column) on whole-brain connectivity, and their interactions (3rd column). The lower triangular matrix corresponds to the f-test for pairs of parcels whereas the upper triangular matrix corresponds to the f-test for pairs of the eleven networks. The connectivity strength between pairs of networks that was used in the mixed ANOVA was defined as the mean correlation averaged across all pairs of parcels that belonged to the two corresponding networks. Pairs of networks with p < 0.0125 after FDR correction are indicated with an asterisk (*). Note that the lower triangular matrices are only shown to provide a qualitative description of the interactions of networks at the parcel level, and that their significance levels are not assessed. We observe that the connectivity strength of the thalamus with the ventral attention network and basal ganglia has strong interactions between the group and the LF-HRV state, which led us to examine thalamic connectivity with the ventral attention network and basal ganglia more carefully.

Figure 4

Functional connectivity (FC) for pairs of parcels with strong group (epilepsy vs. controls)-LF-HRV interactions. In the healthy controls (n = 16), the connectivity strength between the right caudal temporal thalamus and left dorsal granular insula [region of ventral attention (VA) network] was lower in times with high levels of LF-HRV (i.e., levels of LF-HRV in the highest quartile of a scan) compared to times with low levels of LF-HRV (i.e., levels of LF-HRV in the lowest quartile of a scan), whereas in people with epilepsy (n = 28) the connectivity strength was higher in times with high levels of LF-HRV. Similar results were observed for the connectivity strength between the left posterior parietal thalamus and the left ventromedial putamen [region of the basal ganglia (BG)].

Figure 5

Differences in seed-based FC between low and high LF-HRV state with seeds placed in the (top) right caudal temporal thalamus and (bottom) left posterior parietal thalamus. The first and second rows of each panel show the fisher-transformed correlations average across all healthy controls and epilepsy patients, respectively. Red color indicates higher (toward positive values) connectivity with the seed region in high LF-HRV state whereas blue color indicates lower connectivity. The last row of each panel shows the two-sample t-test thresholded at p < 0.001 (uncorrected). Even though no significant differences were found between the two groups after correcting for multiple comparison (FWE; p < 0.05), the differences observed with p < 0.001 (uncorrected) are consistent with the results obtained with the analysis in the atlas space (Figures 3, 4) where the spatial autocorrelation between voxels of the same parcel are implicitly taken into account. The unthresholded statistical maps are available at https://neurovault.org/collections/9452/.