Causal influence of brainstem response to transcutaneous vagus nerve stimulation on cardiovagal outflow

Fig. 1.. Experimental setup: partial brain coverage focused on the brainstem and consistent across participants was employed for fMRI. Electrocardiography (ECG) and respiration signals were acquired concurrently with imaging. The respiration signal was employed to gate left auricular stimulation, i.e., respiratory-gated auricular vagal afferent nerve stimulation (RAVANS). Stimulation was delivered through custom MRI-compatible electrodes placed in the cymba conchae of the ear.

Fig. 2.. –A brainstem mask defined in ICBM space was transformed to subjects’ individual fMRI space and used to retain brainstem voxels, mitigating physiological noise contamination from surface vessels and CSF that may have been introduced by partial volume effects or spatial smoothing. A high-resolution (0.2 mm isotropic voxels) ex vivo brainstem volume (courtesy of) was transformed to MNI space; inverse warps were then applied to the statistical fMRI maps to visualize clusters over this high-resolution anatomical volume to aid localization and comparison with brainstem atlases.

Fig. 3.. Conditioned causality (3-variable case, for explanatory purposes only). In this three-variable system, Y is only directly influencing X through path a. However, the dynamics of Y may also be relayed into X although the combined path B + C, which also involves variable Z. A correlation/GLM analysis will not be able to disentangle the first from the second path and would yield a single association measure influenced by both mechanisms. In contrast, conditioned GC (Y → X∣Z) will disentangle the direction and strength of the coupling due to path a only. Note that, typically, multiple paths and feedback loops coexist in the same system.

Fig. 4.. Outline of second level statistical analysis strategy. “Real” (i.e., within the same subject) SS-GC estimates are generated for each subject, hence giving rise to our data distribution to be tested against the null hypothesis that the median is zero. The null distribution against which to test this hypothesis is generated by calculating “fake” (i.e., combining different subjects in all possible permutations) SS-GC estimates which, by design, should be analytically zero. The median of the two distributions are then compared through a non parametric location test for all voxels within the brainstem mask.

Fig. 5. –. Group maps showing masked causal medullary responses to exhalatory taVNS (eRAVANS, top) and inhalatory taVNS (iRAVANS, bottom) overlaid on a high-resolution (0.2 mm) ex vivo brainstem. Group maps were derived by constructing the voxelwise null distribution for each specific hypothesis through within-subject permutation and comparing its medial to the median of the observed data distributions through statistical, nonparametric location testing cluster (p value < 0.001). The eRAVANS clusters are consistent with putative NTS and LC (ipsilateral to stimulation) and olive (counterlateral to stimulation), while the iRAVANS clusters are again consistent with NTS ipsilateral with stimulation (albeit to a smaller spatial extent and with dmPAG.

Fig. 6. –. Group maps showing masked medullary regions that were seen to exert a causal influence on the time-resolved HF-HRV signal during exhalatory taVNS (eRAVANS, top) and inhalatory taVNS (iRAVANS, bottom) overlaid on a high-resolution (0.2 mm) ex vivo brainstem. For Fig. 5, group maps were derived by constructing the voxelwise null distribution for each specific hypothesis through within-subject permutation and comparing its median to the median of the observed data distributions through statistical, nonparametric location testing cluster (p value < 0.001). The eRAVANS clusters are consistent with PBN (ipsilateral to stimulation) PN and RtP (counterlateral to stimulation), while the iRAVANS clusters are consistent with vlPAG and RtP (counterlateral to stimulation).