Our study has introduced a combined HRV-fMRI method which derives the central neural correlates of a continuous and causal estimate of efferent cardiovagal activity. The brain regions implicated are likely components of the network modulating central command of parasympathetic outflow to the heart. Our data analysis demonstrated that fMRI activity in several important CAN brain regions including the parabrachial nucleus/locus ceruleus, cerebellum, periaqueductal gray, hypothalamus, amygdala, and insular and dorsomedial prefrontal cortices demonstrated significant correlation with cardiovagal activity assessed by HF power. While these regions have been implicated by invasive animal studies, our non-invasive approach also implicated other brain regions which may be unique to the human: dorsolateral prefrontal cortex, mediodorsal thalamus, hippocampus, caudate, septal n. and middle temporal gyrus.
Our results should be interpreted from the context of the known relationship between HF power and cardiovagal physiology. Under physiological conditions and without a significant change in respiratory signal power within the HF power band (confirmed in our data), HF power of the HR signal is an expedient marker of parasympathetic modulation (Malik et al. 1996
). The SA node responds very quickly to vagal (~150msec latency, steady state at 1–2sec), in contrast to sympathetic (~1–2sec latency, steady state at 30–60sec), influence (Spear et al. 1979
; Levy 1997
). Hence, continuous HF power is an excellent metric to use as a regressor within the fMRI GLM framework, given the temporal resolution of fMRI.
Some of the brain regions where fMRI signal was correlated with HF were components of the CAN as implicated by more invasive animal studies, serving as corroborative evidence for our methodology. These regions included several brainstem and cerebellar regions. Sensitivity in these regions was enhanced by cardiac-gated fMRI and brainstem-specific coregistration techniques. The PBN / LC in the right pons were found to be negatively correlated to the HF regressor. The PBN is an important viscerosensory nucleus, relaying information from the NTS to a wide array of target regions including the PAG, hypothalamus, amygdala, and other forebrain nuclei (Gauriau et al. 2002
). The LC is the source of noradrenergic input in the brain and is associated with arousal and the hypothalamic-pituitary-axis (HPA) stress response, relaying afferent signaling to the hypothalamus and amygdala (Benarroch 1993
; Ziegler et al. 1999
). Thus, a negative correlation to HF in the LC would be consistent with decreased stress response during increased cardiovagal modulation. The right cerebellum was also negatively correlated with HF. The cerebellum has been implicated in autonomic modulation arising from motor (e.g. grip task) and vestibular activity (Balaban 1999
) via direct connections with the hypothalamus (Dietrichs et al. 1994
). The PAG, which was also negatively correlated with HF, has been found by intracranial recording to respond to exercise with spectral power increase across multiple frequency bands (Green et al. 2007
). It is directly connected with both sensory (NTS) and premotor (DMNX, NAmb) cardiovagal nuclei (Farkas et al. 1997
), as well as sympathetic premotor nuclei.
FMRI signal in the hypothalamus was found to be positively correlated with HF. This brain structure is considered the principal controller of autonomic and neuroendocrine homeostasis in the body and is the organizational center for many behavioral autonomic response patterns, including exercise-induced ANS outflow (Loewy et al. 1990
). FMRI signal in the amygdala was also found to be positively correlated with HF. Behavioral-associated (e.g. emotion) input from the amygdala to the hypothalamus is transmitted via the ventral amygdalofugal pathway (Parent 1996
) and translates affect into sympathetic and/or parasympathetic response (Thayer et al. 2006
). In addition, fMRI signal in several cortical areas was also either positively or negatively correlated with HF power. In monkeys, autonomic modulation by the DMPFC may result from interconnections with both the PAG (An et al. 1998
) and hypothalamus (Ongur et al. 1998
). The posterior insula is a known ANS sensorimotor region (Saper 2002
). This region is interconnected with the hypothalamus and its stimulation in the rat can induce either a pressor or depressor response, depending on the particular site of stimulation (Yasui et al. 1991
Due to phylogenetic divergence, higher cortical regions implicated by our method in humans are the most likely to be different from CAN regions evaluated in lower mammalian models. The DLPFC, positively correlated with HF, has not been strongly implicated in studies using animal models and may be unique to the more developed human neocortex. Furthermore, fMRI signal in the mediodorsal thalamus, hippocampus, caudate, septal n. and middle temporal gyrus correlated negatively with HF power. These regions have also not received extensive attention from animal studies of the CAN, but are strongly associated with limbic and/or higher cortical processing and may play a greater role for humans compared to lower mammalian models. For instance, the DLPFC is an important brain region governing human cognition (see Duncan et al. 2000
for review) and connects mono- and multi-synaptically with paralimbic CAN regions (Petrides 2005
). Hence, the DLPFC may influence outflow commensurate to perceived
grip task effort in a similar manner as was found for the ACC in a SPECT study which used hypnotism to suggest a perceived grip task (Williamson et al. 2002
Several other studies have correlated or covaried cardiac metrics with neuroimaging data. Gianaros et al. covaried HF with H215
O PET during graded memory tasks (Gianaros et al. 2004
) and found that regional cerebral blood flow (rCBF) covaried positively with HF in the right VMPFC, left insula, and left amygdala-hippocampal complex, while covarying negatively in the right cerebellum. Lane et al. similarly covaried HF with PET neuroimaging during an emotion-laden task (Lane et al. 2001
). Their results showed positive covariation in the left DMPFC, and left anterior insula/orbitofrontal cortex. A study by Critchley et al. used a bandpass filtered RR-series as a regressor to an fMRI GLM during both cognitive and isometric grip tasks (Critchley et al. 2003
). The results found positive correlation in the cerebellum and several cortical regions including the supplementary motor area, cingulate and DMPFC. Our methodology benefited from this last approach, but incorporated important methodological advances such as improved brainstem sensitivity for fMRI and, perhaps most importantly, an accepted measure of continuous cardiovagal modulation (Malik et al. 1996
), as opposed to forming regressors by bandpass filtering the RR time-series in the LF or HF band – an approach which has not been cross-validated or correlated with cardiovagal activity. These methodological differences may have resulted in unique HF-correlated activity in the hypothalamus, PAG by our method, while the cingulate, implicated by Critchley et al. and Matthews et al. (Matthews et al. 2004
) and anatomically connected with the insula and amygdala (implicated by our approach), could not be assessed by our methodology due to partial brain coverage. However, when the methodology reported by Critchley et al. was applied to our own data, no brain regions passed an identical post-processing threshold cutoff used for our methodology. Other studies have attempted to correlate HR or MAP with fMRI (Kimmerly et al. 2005
) or PET (Critchley et al. 2000
). However, measures such as mean heart rate (or mean RR interval) do not proportionally quantify either sympathetic or parasympathetic modulation (Malik et al. 1996
), and are thus less likely to relate to activity in discrete brain regions with known sympathetic or parasympathetic connections.
The most direct interpretation of brain regions derived by our method is that they are involved with central command of cardiovagal modulation – either premotor or via multi-synaptic higher cognitive control. However, other indirect possibilities exist and should be discussed. As our method is correlational, if afferent vagal signaling or efferent sympathetic signaling is correlated with efferent cardiovagal modulation, then the brain regions implicated by our method may also reflect viscerosensory processing. However, sympathetic outflow is not strictly antagonistic to vagal outflow (Janig 2006
), while the afferent/efferent cardiovagal feedback loop is even less consistent and remains to be more fully described (Fallen 2005
). Thus, the more likely interpretation of our results is that the regions implicated are indeed related to central command of what HF is a measure of – efferent parasympathetic modulation to the heart.
Several limitations in our study design should be discussed. Firstly, as we used cardiac-gated fMRI, our field of view could not cover the entire brain. As our hypotheses were focused on brainstem and subcortical autonomic regions, we chose to exclude the most anterior and posterior portions of the brain. We preferred to compromise brain coverage for improved fidelity in the brain regions we did image. Secondly, we made an assumption that brain regions correlating with HF power were somehow involved with central
cardiovagal activity. We feel this assumption was justified given that humoral factors (renin-angiotensin, adrenomedullary catecholamines, muscle metaboreflex etc.) mainly influence lower frequencies, while the influence of respiratory-induced mechanical stretching of the SA node should be minimal in our subject population (Berntson et al. 1991
). In addition, artifactual “activity” was found in the lateral ventricles. While our methods were focused on mitigating sources of physiological artifact in the brainstem, other sources, such as CSF pulsatility, may play a greater role in regions next to the lateral ventricles, diminishing our confidence for results in these regions (septal nucleus, mdThalamus). Finally, our derived brain correlates may be specific to cardiovagal modulation by exercise and may not be generalizable to other autonomic challenges (e.g. cognitive tasks) nor to autonomic outflow to other effector systems (e.g. sudomotor, pupillary, etc.).
In conclusion, we have described a combined HRV-fMRI approach to derive the neural correlates of exercise-induced cardiovagal outflow in humans. Our methodology incorporated a continuous estimate of HF power that better matched the temporal resolution of fMRI leading to a more accurate assessment of central command for cardiovagal outflow. Furthermore, cardiac gating and improved brainstem coregistration were used to improve the fidelity of measurements in brainstem and hypothalamic regions. This approach should be optimized and applied to study the human brain correlates of ANS modulation for various stimuli in physiological and pathological states.