Ventromedial prefrontal cortex connectivity during and after psychological stress in women

Abstract

The ventromedial prefrontal cortex (vmPFC) integrates sensory, affective, memory-related, and social information from diverse brain systems to coordinate behavioral and peripheral physiological responses according to contextual demands that are appraised as stressful. However, the functionality of the vmPFC during stressful experiences is not fully understood. Among 40 female participants, the present study evaluated (a) functional connectivity of the vmPFC during exposure to and recovery following an acute psychological stressor, (b) associations among vmPFC functional connectivity, heart rate, and subjective reports of stress across individuals, and (c) whether patterns of vmPFC functional connectivity were associated with distributed brain networks. Results showed that psychological stress increased vmPFC functional connectivity with individual brain areas implicated in stressor processing (e.g., insula, amygdala, anterior cingulate cortex) and decreased connectivity with the posterior cingulate cortex and thalamus. There were no statistical differences in vmPFC connectivity to individual brain areas during recovery, as compared with baseline. Spatial similarity analyses revealed stressor-evoked increased connectivity of the vmPFC with the so-called dorsal attention, ventral attention, and frontoparietal networks, as well as decreased connectivity with the default mode network. During recovery, vmPFC connectivity increased with the frontoparietal network. Finally, individual differences in heart rate and perceived stress were associated with vmPFC connectivity to the ventral attention, frontoparietal, and default mode networks. Psychological stress appears to alter network-level functional connectivity of the vmPFC in a manner that further relates to individual differences in stressor-evoked cardiovascular and affective reactivity.

1. INTRODUCTION

The ventromedial prefrontal cortex (vmPFC) is implicated in physiological control processes, particularly involving the autonomic and cardiovascular systems (Gianaros & Wager, 2015; Roy, Shohamy, & Wager, 2012; Sinha, Lacadie, Constable, & Sao, 2016; Tobia, Hayashi, Ballard, Gotlib, & Waug, 2017; Wager, van Ast et al., 2009; Wager, Waugh et al., 2009; Winecoff et al., 2013; Yang et al., 2018). The vmPFC has also been implicated in ascribing meaning to or evaluating the personal significance of emotionally evocative contexts (e.g., threatening stimuli) and stressors (Levy & Glimcher, 2012; Montague & Berns, 2002). In this regard, it has been proposed that the vmPFC may act as a hub brain system that integrates sensory, affective, memory-related, and social information from diverse brain systems to coordinate behavioral and peripheral physiological responses according to contextual demands, particularly those demands that are appraised as psychologically stressful (Roy et al., 2012).

In support of the latter proposal, human neuroimaging studies demonstrate that psychological stressors reliably engage the vmPFC and alter its activation (Roy et al., 2012; van der Werff, van den Berg, Pannekoek, Elizinga, & van der Wee, 2013; Wager, Waugh et al., 2009; Wheelock et al., 2016). Moreover, vmPFC activity changes that are evoked by psychological stressors have been associated with individual differences in self-reported emotional intensity during and after stressor exposure (Orem et al., 2019; Tobia et al., 2017; van der Werff et al., 2013; Wheelock et al., 2016; Yang et al., 2018). Recent analyses examining 270 participants across 18 studies demonstrated that patterns of vmPFC activity are reliably associated with negative affective responding (Kragel et al., 2018).

In addition to features of negative emotionality, the vmPFC appears to play a role in regulating (influencing and representing) peripheral physiology through visceromotor (efferent) and viscerosensory (afferent) pathways (Chang et al., 2013; Eisenbarth, Chang, & Wager, 2016; Gianaros & Wager, 2015, Ginty, Kraynak, Fisher, & Gianaros, 2017; Roy et al., 2012; Thayer, Ahs, Fredrickson, Solers, & Wager, 2012; Young et al., 2017). Indeed, animal evidence has long implicated the vmPFC as a visceromotor brain region that controls peripheral physiology (Frysztak & Neafsey, 1994; Resstel & Correa 2006; Ulrich-Lai & Herman, 2009). Such a role is supported by extensive vmPFC interconnections with cortical and subcortical regions whose circuit-level dynamics modify autonomic and endocrine outflow to peripheral organs, as well as relay peripheral afferent information to rostral brain systems through viscerosensory (i.e., body-to-brain) projections (Gabbott, Warner, Jays, Salway, & Busby, 2005; Ongur & Price, 2000; Saper, 2002). Hence, several imaging studies have demonstrated that vmPFC activity associates with diverse parameters of autonomic and cardiovascular physiology across a range of behavioral states (for reviews, see Critchley, Nagai, Gray, & Mathais, 2011; Gianaros & Wager, 2015; Roy et al., 2012). In the context of stress and autonomic control, functional neuroimaging studies have revealed increased activity within some portions of the vmPFC (albeit not exclusively) in association with decreased heart rate, possibly suggesting proparasympathetic control during some aversive behavioral states (Eisenbarth et al., 2016; Gianaros & Wager, 2015; Wager, Waugh et al., 2009).

Despite lines of evidence suggesting a role for vmPFC in stress processing and peripheral physiological regulation, there are several open questions about the extent to which vmPFC circuit function relates to particular aspects of stress reactivity and recovery, especially within an anticipatory stress paradigm. To elaborate, anticipatory stress paradigms, such as speech preparation tasks commonly used in psychophysiological and behavioral medicine studies, are known to increase heart rate across individuals (e.g., Brindle, Ginty, Phillips, & Carroll, 2014). A more recent line of brain imaging work also indicates that such paradigms alter activity in the vmPFC as well as activity in cortical and subcortical brain regions in association with parallel (concurrent) autonomic changes (Wager, van Ast et al., 2009; Wager, Waugh et al., 2009). To extend this prior work, we first tested whether a canonical psychological stressor involving social evaluative threat and anticipation alters the functional connectivity between the vmPFC and visceral control regions that are known to exhibit functional anatomical connections with the vmPFC in a sample of female participants. We then tested whether vmPFC connectivity changes following a stressor, namely, during recovery. Lastly, we tested whether circuit-level vmPFC neural correlates are similar for cardiovascular (i.e., heart rate) reactivity and self-reported stress. Evidence has long suggested weak or at best unreliable correlations of small effect size between stressor-evoked self-reported negative emotions and different parameters of cardiovascular reactivity, which may suggest a lack of convergence at the level of the brain (Brindle, Whittaker, Bibbey, Carroll, & Ginty, 2017; Feldman et al., 1999; Schwertfeger, 2004). Put differently, the neural correlates of self-reports may be spatially different or dissociable from those for cardiovascular stress reactivity. But, the latter possibility has not been formally tested to our knowledge, particularly with respect to the vmPFC.

On an exploratory basis, we examined whether the patterns of stress-related or recovery-related vmPFC connectivity associates with known intrinsic or canonical cortical networks (Yeo et al., 2011). The motivation for this exploratory analysis draws upon growing appreciation that psychological processes, such as stress, appear to engage neural networks comprising spatially distinct, yet neuroanatomically connected, brain areas (Kragel, Koban, Barrett, & Wager, 2018). Hence, our exploratory analyses aimed to supplement our primary research questions to characterize, at the network level of analysis, the main effects of stress and recovery as well as neural correlates of individual differences in self-reports and cardiovascular physiology.

2. METHOD

2.1. Participants

Participants were 40 female undergraduate students who were recruited through an already existing cardiovascular stress laboratory study (Ginty, Brindle, & Carroll, 2015). The original study took place between September 2012 and April 2013. Forty participants were randomly selected from this pre-existing study to take part in a separate functional neuroimaging testing session. The original study was predominately female (81%). Only female participants were selected for the current small-scale study due to the skew in gender in the original sample. See Table 1 for participant demographic information for the present study. All participants selected completed the session between July and August 2014. Exclusion criteria included non-MRI compatible (ferromagnetic) implants, claustrophobia, and history of cardiovascular disease. All participants gave informed consent. Participants received £20 for study participation and were compensated for travel to and from the university. The study was approved by the University Ethics Committee.

TABLE 1.

Demographic information and characteristics

	Mean (SD)
Age (years)	19.05 (0.22)
Ethnicity
White British	60%
Mixed/other	12.5%
Black/Black British	10%
Asian/Asian British	10%
Chinese	7.5%
Parental occupational status
Nonmanual	77.5%
Manual	22.5%
HADS depression	4.68 (3.58)
Body mass index (kg/m2)	22.98 (4.78)

N = 40 women.

Note. Abbreviation: HADS, Hospital Anxiety and Depression Scale.

2.2. Procedure

Participants completed testing at the Birmingham University Imaging Centre between 11 am and 7 pm. Participants were asked to refrain from alcohol or vigorous exercise (12 hr), caffeine (2 hr), and food and drinks other than water (1 hr) before testing. Upon arrival to the Centre, participants were familiarized with the MRI equipment and given the opportunity to ask questions. Participants completed the depression subscale of the Hospital Anxiety and Depression Scale (Zigmond & Snaith, 1983). Participants provided information about the occupational status of the parent who was the primary household provider, with options ranging from professional (e.g., doctor/lawyer) to unskilled (e.g., day laborer). Occupations were categorized into manual and nonmanual (Ginty, Phillips, Der, Deary, & Carroll, 2011). Then, participants completed the neuroimaging portion of the study, which consisted of a structural scan, a 6-min resting baseline period, two 2-min speech preparation tasks, and a 6-min recovery period. Neuroimaging data were obtained using a Philips 3.0T Achieva system with 8-channel head coil. Structural scans were acquired using the T1 TFE (turbo field echo) technique (repetition time, TR = 8.4 ms, echo time, TE = 3.8 ms, FoV = 232 mm, flip angle = 8° 288 × 288 matrix, 175 slices, voxel size 1 × 1 × 2 mm). All functional scans were acquired similarly using BOLD contrast weighted echoplanar imaging (EPI; TR = 2,000 ms, TE = 34 ms, FoV = 240 × 240 mm, 32 slices, no gap, 3.0 × 3.0 × 4 mm isotropic voxels).

2.3. fMRI stress task protocol

The fMRI stress task paradigm consisted of a 6-min resting state scan (baseline period) with eyes open, two 2-min speech preparation periods, and a recovery period (Wager, van Ast et al., 2009; Wager, Waugh et al., 2009). After the resting state scan (baseline period), participants were informed that they would be given a speech topic and 2 min to prepare a speech (henceforth, Speech A). They were also told that this would be followed by a second speech topic and a second 2-min preparation period (henceforth, Speech B). Finally, they were instructed that, following the two speech preparation periods, they would be randomly assigned to one of three conditions: give speech from the first preparation period, give speech from the second preparation period, or give no speech at all. The two speech topics were presented in a counterbalanced order. The speech topics were (a) defend themselves against being falsely accused of shoplifting a belt, and (b) reveal and describe their three best and three worst personal characteristics (Bosch et al., 2009). Moreover, participants were told that law professors would view and rate the quality and persuasiveness of their shoplifting defense speech, and peers would view and rate the quality of their personal characteristics speech. Finally, participants were told that anxiety experts would view the videotapes and rate their anxiety levels. In actuality, all participants were selected to give no speech at all (Tobia et al., 2017; Wager, van Ast et al., 2009; Wager, Waugh et al., 2009), and therefore none of these reviews occurred. There was a video camera in the scanner to communicate with the control room for safety purposes; however, the camera did not save recordings. After the second period of speech preparation, all participants were told that they had been selected not to give speeches. Following this disclosure, participants were asked to keep their eyes open and remained in the scanner for a 6-min recovery period. The speech preparation task is assumed to evoke psychological threat appraisal or an appraisal of a threat to self (Wager, van Ast et al., 2009; Wager, Waugh et al., 2009). Thus, it would be expected that the speech preparation task would engage the vmPFC and alter its interactions (connectivity) with other areas in a way that influences reports of psychological distress and autonomic outflow to the heart. In support of this, the speech preparation task has been shown to induce autonomic, neural, and self-report psychological changes (e.g., Tobia et al., 2017; Wager, van Ast et al., 2009; Wager, Waugh et al., 2009). Additionally, when compared to tasks such as mental arithmetic, it has been suggested that the speech preparation task elicits negative affect states that are more similar to those used in nonimaging (e.g., laboratory) settings (e.g., Fredrickson, Mancuso, Branigan, & Tugade, 2000; Wager, Waugh et al., 2009).

2.4. Measures

2.4.1. Psychological stress levels

Participants were asked to rate on a Likert-type scale from 0–6 their level of perceived stress, with anchor 0 representing not at all stressed and anchor 6 representing stressed. Participants provided ratings after the structural scan and after the resting baseline scan, which were averaged to create a baseline perceived stress level. After each preparation task, participants again rated on the same Likert-type scale how stressed they felt during the preparation period; these were averaged to create a stress level. Participants also provided a rating immediately prior to leaving the scan on the same Likert-type scale asking them how stressed they felt during the 6-min period after they were told they did not have to give the speech; this was used as the recovery level.

2.4.2. Cardiovascular activity

A continuous pulse wave form for computing heart rate was measured noninvasively throughout the imaging protocol via MRI-compatible pulse oximetry (InVivo 4500 MRI; Invivo Research Corp., Orlando, FL). A custom algorithm in Spike2 (v. 6.06; Cambridge Electronic Design Limited, Cambridge, UK) was used for automated R-peak detection. R-peak selection was manually inspected, and artifacts were removed where necessary. Average heart rate was then calculated for each of the following periods: baseline, Speech A, Speech B, and recovery. Heart rate was available for 28 of the 40 participants. Data were lost for 12 participants due to unusable raw pulse oximeter data. Heart rate recorded during the resting/baseline scan was averaged to calculate a baseline value. Heart rate during Speech A and Speech B were calculated separately. The two speech periods were highly correlated (r = .613, p = .001). Following prior cardiovascular reactivity studies and psychometric principles for the study of individual differences, the two periods were averaged to create a stress heart rate to improve reliability of cardiovascular reactivity measures (Kamarck et al., 1992; Kamarck, Jennings, Stewart, & Eddy, 1993). Averaging across similar stress sessions has also been shown to improve reliability in concurrent patterns of stressor-evoked neural and cardiovascular responses (Sheu, Jennings, & Gianaros, 2012).

2.5. Data analyses of psychological and cardiovascular activity

Repeated measures analyses of variance (ANOVAs), using values from the resting baseline, speech preparation, and recovery conditions, were applied to test for condition-related differences in heart rate and perceived stress. To examine our hypotheses regarding stress-related and recovery-related changes in psychological stress and heart rate reactivity, we tested a priori planned contrasts of changes in reactivity for (a) stress-related reactivity, the difference in speech preparation minus resting baseline, and (b) recovery-related reactivity, the difference in recovery minus resting baseline. Similarly, we computed change scores in subjective stress and mean heart rate following these planned contrasts. Specifically, stress-related reactivity was calculated as the difference between the average measurement taken during speech preparation minus the resting baseline period, and recovery-related reactivity was calculated as the difference between measurements taken during the recovery period minus the resting baseline period. These change scores were used to examine individual differences in psychological, heart rate, and vmPFC connectivity.

2.6. Data analysis assessing temporal stability of cardiovascular activity

As noted above, participants in this fMRI study were drawn from a larger study that included a laboratory stress reactivity protocol (Brindle et al., 2016; Ginty et al., 2015). The laboratory stress reactivity protocol used the 10-min paced auditory serial addition task (PASAT) as the main stress task. Briefly, the PASAT presents a series of single-digit numbers, and participants have to add the consecutive numbers together while remembering the previous number in order to add it to the next number. Participants present their answers verbally and are seated throughout the procedure. Additionally, the task includes elements of social evaluative threat (i.e., participants are videotaped and told that body language experts would review the tape to assess their anxiety levels; participants have to look in a mirror throughout the duration of the task; participants are told that they will hear a buzzer if they answer incorrectly or hesitate). The PASAT has been shown to have good test-retest reliability (Ginty, Gianaros, Derbyshire, Phillips, & Carroll, 2013) and to reliably perturb cardiovascular activity (Ring, Burns, & Carroll, 2002). The laboratory portion also used continuous electrocardiography to record heart rate using electrodes in a three-lead configuration (Brindle et al., 2016; Ginty et al., 2015). For a more detailed description of the laboratory testing session see Trotman et al., 2019. Bivariate correlation analyses were run to assess stability of reactivity and recovery between the laboratory and neuroimaging portions of the study.

2.7. fMRI data preprocessing and analysis

2.7.1. fMRI data preprocessing

Preprocessing of BOLD images was conducted in Statistical Parametric Mapping software (SPM8; Wellcome Trust Centre for the Study of Cognitive Neurology, www.fil.ion.ucl.ac.uk/spm). First, to realign BOLD images across all task periods (i.e., resting baseline, speech preparation, recovery) for a given subject, the first images from each period were aligned to the first image of the first period, and then images within each period were realigned to the first image of their respective period with the unwarp procedure to adjust for geometric distortion. Six realignment parameters computed by rigid body spatial transformation were generated for each period. A mean image of all functional periods was computed. In the within-subject (co)registration step, gray matter image from the New Segmentation procedure in SPM8 was coregistered to the mean functional image. In normalization, realigned and unwarped functional images were normalized by a 12-parameter nonlinear and affine transformation to the International Consortium for Brain Mapping 152 template (Montreal Neurological Institute, MNI). Normalized images were resliced to 2 × 2 × 2 mm voxels and smoothed with a 6-mm FWHM isotropic Gaussian kernel. Functional connectivity was computed using the CONN toolbox version 17f (Whitfield-Gabrieli & Nieto-Castanon, 2012). The time series in each voxel was denoised and adjusted for potential confounding effects due to movement and physiological noise by linear detrending at the single subject level. Physiological noise regressors corresponded to the first five principal components of white matter signal and first five principal components of cerebral spinal fluid signal (Behzadi, Restom, Liau, & Liu, 2007). Movement regressors were the six parameters estimated from realignment and their first temporal derivative. In addition, effects of each condition (i.e., rest, speech, recovery) were also entered into the single-subject design matrix along with nuisance regressors. Following nuisance regression, BOLD signal time series were filtered with a 0.005 Hz high-pass filter.

2.7.2. Neuroimaging analyses

A vmPFC mask (see Figure 1) to be used as a seed region was generated from Neurosynth (http://neurosynth.org; Yarkoni, Poldrack, Nichols, Van Essen, & Wagner, 2011) using the reverse inference command (now labeled association test in Neurosynth) for the term VMPFC and thresholded at p < .05 uncorrected. The generated mask was smoothed with a 6-mm FWHM kernel, and left and right hemisphere voxels were mirrored, creating a symmetrical vmPFC mask. This vmPFC mask is similar to the vmPFC mask in Roy et al., 2012, with a volume of 47,824 mm³, with a center of mass at MNI x = 0 mm, y = 43.6 mm, z = −3.4 mm (Figure 1). The vmPFC mask used in these analyses is available for download on NeuroVault (https://neurovault.org/collections/SFVWGXCY/; Gorgolewski et al., 2015).

FIGURE 1.

Ventromedial prefrontal cortex (vmPFC) mask used as the seed for analyses. The volume for the vmPFC mask was 47,824 mm³, with a center of mass at MNI x = 0 mm, y = 43.6 mm, z = −3.4 mm

At the single-subject level, seed-to-voxel functional connectivity analyses were computed using the CONN toolbox, employing generalized psychophysiological interaction analyses (gPPI; McLaren, Ries, Xu, & Johnson, 2012). Using this approach, we extracted the average BOLD time series from the vmPFC seed region mask and computed the following three variables to be used as PPI regressors:

1. The product of the vmPFC time series with a task regressor corresponding to the resting baseline condition convolved with hemodynamic response function (HRF);

2. The product of the vmPFC time series with a task regressor corresponding to the speech preparation condition convolved with HRF; and,

3. The product of the vmPFC time series with a task regressor corresponding to the recovery condition convolved with HRF.

These three condition-specific connectivity regressors were then submitted to a gPPI model to conduct task-modulated seed-to-voxel connectivity analyses (McLaren et al., 2012). Each seed-to-voxel gPPI map reflecting the regressors above were constructed for each subject. These seed-to-voxel gPPI maps were used to test for the main effects of between-condition contrasts at the group level, along our a priori hypotheses, testing for stress-related connectivity changes as differences between speech preparation and resting baseline and testing for recovery-related connectivity changes as differences between recovery period and resting baseline.

To test for the neural correlates of reactivity and recovery levels of heart rate and perceived stress at the group level, a priori planned gPPI contrast maps were constructed mirroring the main effects contrasts described above (i.e., speech preparation—resting baseline and recovery—resting baseline). Using the CONN toolbox, stress-related and recovery-related change scores in heart rate and perceived stress as described above were entered into separate group-level design matrices with their respective gPPI contrast maps as dependent measures.

For statistical significance testing in voxelwise analyses, we applied a false discovery rate (FDR) threshold of .05 (Benjamini & Hochberg, 1995) and contiguous cluster size of k ≥ 50 voxels. Unthresholded voxelwise maps are available on NeuroVault (https://neurovault.org/collections/SFVWGXCY/; Gorgolewsky et al., 2015). In ancillary sensitivity tests, we explored the contributions of individual differences in fMRI signal motion as well as condition-related differences in fMRI signal motion to observed effects and associations. To this end, we computed mean framewise displacement values for each subject and each condition (Power, Schlagger, & Petersen, 2015). Condition-related differences in motion were examined using repeated measures ANOVA and associations between individual differences in motion with the aforementioned planned change scores in perceived stress and heart rate.

2.7.3. Large-scale spatial pattern similarity analysis

To investigate whether the distributed patterns of stress-related or recovery-related vmPFC connectivity relate to known brain networks, we conducted spatial pattern similarity analyses. These analyses follow the methods of previously published work (Eisenbarth et al., 2016; Kragel, Koban et al., 2018; Kraynak, Marsland, Wager, & Gianaros, 2018) and were conducted using custom MATLAB routines (accessible at https://canlab.github.io). Specifically, we tested the spatial pattern similarity between each participant’s distributed patterns of stress-related (i.e., speech preparation − resting baseline) and recovery-related (i.e., recovery − baseline) vmPFC connectivity maps with an established set of seven intrinsic networks derived from a large sample of 1,000 men and women (Yeo et al., 2011). For every participant, we computed cosine similarity, on a voxelwise basis, between their unthresholded stress-related and recovery-related vmPFC PPI connectivity maps with maps reflecting each of the seven intrinsic networks, resulting in seven stress-related and seven recovery-related spatial pattern estimates for each participant, respectively. These spatial pattern estimates (cosine similarity metrics) reflect the extent to which a participant’s task-modulated vmPFC connectivity map associates with a brain network across all voxels. Main effects of task condition on spatial pattern estimates were examined using one-sample t tests. Associations between spatial pattern estimates and individual differences in heart rate and subjective stress reactivity were explored using bivariate correlations.

3. RESULTS

3.1. Psychological and cardiovascular responses and recovery from stress task

The speech preparation task significantly increased subjective reports of psychological stress, F(2, 78) = 111.64, p < .001, η² = .741. Post hoc Bonferroni-corrected analyses indicated that participants reported more stress during speech preparation (M perceived stress = 3.42, SD = 1.19; p < .001) compared to both baseline (M perceived stress = 1.43, SD = 1.13) and recovery (M perceived stress = 1.55, SD = 1.01; p < .001). The stress testing paradigm also increased heart rate, F(2, 54) = 53.05, p < .001, η² = .663. Post hoc and Bonferroni-corrected analyses demonstrated that participants exhibited faster heart rates during speech preparation (M = 84.29, SD = 13.91) compared to baseline (M = 71.34, SD = 8.51; p <. 001) and recovery (M = 74.34, SD = 10.40; p < .001). Heart rate was also higher during recovery compared to baseline (p = .007). In addition, individual differences in heart rate reactivity were not statistically correlated with self-reported perceived stress reactivity, r(26) = .25 p = .203. Similarly, individual differences in heart rate recovery were not statistically correlated with self-report perceived stress recovery, r(26) = −.07, p = .716.

3.2. Temporal stability of cardiovascular activity

Correlation analyses demonstrated that heart rate reactivity during the neuroimaging protocol correlated positively with heart rate reactivity from the laboratory portion of the study, r(26)_{ΔHRlab•ΔHRfMRI} = .55, p = .002. Likewise, heart rate recovery during the neuroimaging protocol correlated positively with heart rate recovery from the laboratory portion of the study, r(26)_{recoveryHRlab•recoveryHRfMRI} = .65, p < .001. These findings replicate previous studies indicating temporal stability in individual differences in cardiovascular responding across laboratory and MRI testing contexts (Gianaros, Jennings, Sheu, Derbyshire, & Matthews, 2007; Ginty et al., 2013).

3.3. Stressor-evoked changes in vmPFC connectivity.

At the voxelwise level of analysis, the contrast of speech preparation versus resting baseline revealed increased connectivity of the vmPFC during preparation with the bilateral insular cortex, encompassing its anterior, middle, and posterior territories. The specific cluster of voxels encompassing insular cortical areas contiguously extended, as well, into the inferior frontal gyrus, precentral gyrus, orbitofrontal cortex, midcingulate cortex, amygdala, and putamen within the basal ganglia. Speech preparation also increased vmPFC connectivity with the occipital cortex, extending into the parietal lobule and precuneus. Other areas exhibiting increased stressor-evoked connectivity with the vmPFC, such as the dorsolateral prefrontal cortex, inferior and middle temporal gyrus, and cerebellum, are illustrated and summarized in Figure 2 and online supporting information, Table S1. By contrast, speech preparation evoked a decrease in vmPFC functional connectivity with the dorsomedial prefrontal cortex, posterior cingulate cortex (extending into the precuneus), thalamus, inferior temporal cortex, and medial portion of the cerebellum (see Figure 2 and supporting information, Table S1).

FIGURE 2.

Color-scaled parametric t-statistic maps of brain areas exhibiting significant changes in vmPFC connectivity (speech vs. rest) are shown for the sagittal (a), axial (b), coronal (c) views. (d) and (e) demonstrate the left and right view and (f) represents the right limbic region. K = 10 with a false discovery rate (FDR) of 0.05 to control for multiple statistical testing

At the network level of analysis, spatial similarity tests indicated that the pattern of increased vmPFC connectivity during speech preparation associated with the dorsal attention network, ventral attention network, and frontoparietal network (Ts > 3.12, FDR-corrected ps < .012). Conversely, the pattern of decreased vmPFC connectivity during speech preparation associated with the default mode network (T = −6.68, FDR-corrected p < .001). These results are in Table S2 and depicted as similarity plots in Figure 3.

FIGURE 3.

Polar plots demonstrating spatial pattern similarity between patterns of stress-related (red) and recovery-related (blue) vmPFC connectivity maps to seven known networks derived from a large sample (N = 1,000; Yeo et al., 2011). Red solid (stress-related) and blue solid (recovery-related) lines depict the mean similarity measure (cosine similarity) with each network. Solid gray lines around colored lines depict standard error. See Table S2a for details on these statistics

3.4. Main effects of speech recovery versus resting baseline

At the voxelwise level of analysis, we observed no clusters of statistical differences in vmPFC connectivity between the resting baseline and speech recovery periods. At the network level of analysis, spatial similarity tests indicated that the pattern of increased vmPFC connectivity during recovery associated with the frontoparietal network (T = −2.53, FDR-corrected p = .044).

3.5. Heart rate reactivity and self-reports

At the voxelwise level of analyses and using FDR thresholding, we observed no clusters showing statistical associations between stress-related changes in vmPFC connectivity and stress-related changes in either heart rate or perceived stress. Nor did we observe clusters showing associations during recovery using voxelwise FDR thresholds. However, at the network level of analyses, there was a negative association between stress-related vmPFC–ventral attention network coupling with stress-related changes in perceived stress (r = −.56, p < .001). There were no corresponding associations with stress-related changes in heart rate at the network level. Moreover, comparing individual differences in recovery-related vmPFC connectivity changes at the network level, there was a positive association between vmPFC–default mode network coupling with changes in heart rate recovery (r = .41 p = .032), as well as a negative association between vmPFC–frontoparietal network coupling with changes in perceived stress (r = −.33, p = .040). These results are depicted as scatter plots in Figure 4. In sensitivity analyses removing multivariate outliers exceeding four times the mean, identified using Cook’s distance, vmPFC–ventral attention network coupling with stress-related changes in perceived stress (r = −.59, p < .001) and vmPFC–default mode network coupling with changes in heart rate recovery (r = .40, p = .042) were maintained. However, vmPFC–frontoparietal network coupling with changes in perceived stress did not meet a conventional threshold for statistical significance after outlier removal (r = −.25, p = .081).

FIGURE 4.

Scatter plots demonstrating significant associations between (a) stress-related changes in perceived stress and vmPFC–ventral attention network coupling, (b) recovery-related heart rate changes and vmPFC–default mode network coupling, and (c) recovery-related changes in perceived stress and vmPFC–frontoparietal network coupling

3.6. Sensitivity tests of motion

Stressor-evoked functional connectivity changes could have been influenced by some other variable, such as participant motion in the scanner (Satterthwaite et al., 2019). However, main effects of task condition on mean framewise displacement, a measure of participant movement in the scanner, did not reach statistical significance after sphericity correction for repeated measures (Jennings, 1987; Jennings & Wood, 1976). Similarly, across subjects, individual differences in condition-related changes in motion did not associate with their respective change scores in heart rate or perceived stress (all ps > .21).

4. DISCUSSION

The aims of the present study were to (a) examine whether a stress paradigm altered the circuit level function of the vmPFC during and after exposure to an acute psychological stressor, (b) examine interindividual covariation among vmPFC connectivity, heart rate reactivity, and subjective reports of stress during and after an acute psychological stressor, and (c) test whether the patterns of stress-related or recovery-related vmPFC functional connectivity were associated with known intrinsic cortical networks in a sample of otherwise healthy female subjects. Results demonstrated that an anticipatory psychological stressor including social evaluative threat increased vmPFC functional connectivity with areas previously implicated in stressor processing (e.g., insula, amygdala, midcingulate cortex; for a list, see Table S1) and decreased vmPFC functional connectivity with the posterior cingulate, thalamus, dorsomedial prefrontal cortex, inferior temporal cortex, and medial portion of the cerebellum. Spatial similarity analyses indicated that the anticipatory stress condition resulted in an increase in vmPFC connectivity with components of the dorsal attention, ventral attention, and frontoparietal networks and decreased vmPFC connectivity with the default mode network. There were no observed voxelwise differences between the resting baseline and stress recovery periods in vmPFC connectivity after correction for multiple testing; however, at the network level, results demonstrated increased vmPFC connectivity with components of the frontoparietal network. Although the stress task increased both HR and perceived stress, there were no observed covariations at the voxelwise level between stressor-evoked vmPFC connectivity changes and corresponding individual differences in HR or self-reported stress; however, we observed associations at the network level of analysis between heart rate and perceived stress and vmPFC connectivity with the ventral attention, frontoparietal, and default mode networks.

The present study demonstrated stressor-evoked increased functional connectivity between the vmPFC and cortical and subcortical regions with which the vmPFC has extensive anatomical interconnections (Gabbott et al., 2005; Ongur & Price, 2000; Saper, 2002). There was increased stressor-evoked functional connectivity between the vmPFC and areas associated with visceral control, such as the insular cortex, orbitofrontal cortex, and amygdala (Saper, 2002). vmPFC, amygdala, and insula activation have been related to autonomic changes across multiple types of stressors (Chang et al., 2013; Gianaros & Sheu, 2009; Gianaros & Wager, 2015; Gianaros et al., 2017; Ginty et al., 2017; Muscatell & Eisenberger, 2012). Lesion studies and animal work suggest that these regions are important for cardiovascular and autonomic control (Oppenheimer & Cechetto, 2016; Shoemaker & Goswami, 2015; Shoemaker, Norton, Baker, & Luchyshyn, 2015). There was decreased functional connectivity during stress between the vmPFC and the thalamus, which is somewhat surprising given the importance of the thalamus in visceral control (Bandler, Keay, Floyd, & Price, 2000; Ongur & Price, 2000; Saper, 2002). However, the thalamus expresses neuroanatomical connections to virtually all cortical brain regions (Jones, 2001), and connections between specific thalamic nuclei and distinct cortical brain regions comprise “corticothalamic loops” (Sherman & Guillery, 2014). Corticothalamic processing along these loops are thought to support sustained cortical processing (Hwang, Bertolero, Liu, & D’Esposito, 2017). The vmPFC in particular expresses connections to the mediodorsal nucleus of the thalamus (Giguere & Goldman-Rakic, 1988). Interestingly, the thalamus cluster demonstrating decreased stress-related vmPFC connectivity appears to specifically comprise the mediodorsal nucleus. It is possible that decreased connectivity during stress could reflect an “under-recruitment” of this mediodorsal thalamus–vmPFC loop and/or a rerouting from corticothalamic loop processing toward interactions with the other higher-order cortical brain regions.

Previous functional connectivity research using seed-based approaches demonstrated increases in functional connectivity between the vmPFC with the dorsolateral prefrontal cortex, anterior prefrontal cortex, and inferior parietal lobe while passively viewing unpleasant pictures compared to a resting baseline (Sinha et al., 2016). Insofar as viewing unpleasant pictures evokes an aversive behavioral state approximating psychological stress, the present study similarly showed increased functional connectivity between the vmPFC and the dorsolateral prefrontal cortex and parietal lobe but not the anterior prefrontal cortex. Applying thresholding and correction, there were additional changes in stressor-evoked functional connectivity between the vmPFC and multiple areas (see Table S1, Figure 2). The methodological differences between the two studies could potentially explain the differences in study findings, insofar as passively viewing pictures is unlikely to evoke similar stressor appraisals as in the current paradigm. Hence, in the current study, the stressor was an effortful and active task that required participants to prepare a speech (cf. Wager, van Ast et al., 2009; Wager, Waugh et al., 2009), whereas the study conducted by Sinha and colleagues (2016) used a passive task in which the ability to evoke psychological stress appraisals and physiological changes considered as canonical stress responses are unclear. Notably and consistent with the latter view, a recent study using the same active stress paradigm as our study has demonstrated networkwide functional connectivity changes with the vmPFC during speech compared to baseline (Tobia et al., 2017).

The current study aimed to extend previous research by examining whether circuit level neural correlates are similar for cardiovascular reactivity and self-reported stress (Sinha et al., 2016; Tobia et al., 2017). The vmPFC has been associated with individual differences in heart rate changes in response to stress (Chang et al., 2013; Eisenbarth et al., 2016; Thayer et al., 2012; Young et al; 2017). Wager et al. (2009) demonstrated that patterns of stressor-evoked activation in areas including the medial prefrontal cortex were associated with heart rate reactions to a speech task. However, no study has examined the relationship between stressor-evoked functional vmPFC changes using a seed-based approach and stressor-evoked physiological changes. vmPFC activation has also been associated with self-reported task-related stress and anxiety (van der Werff et al., 2013; Wager, Waugh et al., 2009; Wheelock et al., 2016; Yang et al., 2018) and higher trait-level coping (Sinha et al., 2016). A recent study found large-scale brain networks involving the vmPFC to be associated with individual differences in stress-induced emotions (Tobia et al., 2017). Despite the speech task altering cardiovascular system activity and increasing subjective feelings of stress, there were no significant covariations between stressor-evoked vmPFC changes and corresponding individual differences in heart rate or individual differences in self-reported stress at the voxelwise level using a mass univariate testing approach with multiple testing thresholding in this sample. The failure to detect any correlations may be attributable to low statistical power for such mass univariate approaches. Recent work suggests that voxelwise associations between stressor-evoked neural patterns and individual differences in cardiovascular reactivity may be relatively small, accounting for approximately 10% or less of the variance across individuals (Eisenbarth et al., 2016; Gianaros et al., 2017). Although the current sample size may have been adequately powered to detect group level changes corresponding to the main effects of our task, it may have not been adequate power to detect interindividual covariation among stress-related parameters on a voxelwise basis. Further research in a larger sample examining the association between stressor-evoked vmPFC functional connectivity changes and stressor-evoked cardiovascular and subjective stress reactions is warranted.

In contrast to voxelwise analyses, it may be fruitful for future research to employ network-level analyses, as provisional and exploratory findings here suggest some covariation between vmPFC connectivity with known brain networks during stress and recovery. In particular, stress appeared to increase connectivity between the vmPFC and ventral attention network, the latter comprising the anterior insula and dorsal anterior cingulate (Yeo et al., 2011). This network is also referred to as the salience network (Seeley et al., 2007) and is implicated in autonomic regulation (Jennings et al., 2016; Sturm et al., 2018; Young et al., 2017). Invasive studies in humans, moreover, complement functional neuroimaging studies to suggest roles for the insula and dorsal anterior cingulate in autonomic cardiovascular control (Kim et al., 2019; Oppenheimer & Cechetto, 2016). In addition, individual differences in perceived stress correlated negatively with vmPFC–ventral attention network coupling during stress (Figure 4a), suggesting that coupling between the vmPFC and components of the ventral attention network may be associated with encoding or appraising situations and contexts as stressful to the individual. Put differently, participants who exhibited lesser stress-related elevations in connectivity within this circuit also reported higher levels of perceived stress. Across the sample, coupling between the vmPFC and the frontoparietal network appeared to remain increased during recovery as compared to rest, the latter network comprising regions such as the lateral PFC and posterior parietal cortex. We observed a negative association between vmPFC–frontoparietal coupling and perceived stress during recovery, suggesting that individuals who did not sustain this coupling during recovery also reported sustained levels of perceived stress (Figure 4c). The frontoparietal network is implicated in cognitive control (Zanto & Gazzaley, 2013), and some evidence suggests it may be involved in regulatory aspects of mood and mental health (Cole, Repovš, & Anticevic, 2014). One possibility for future testing is that there may be different regulatory roles for vmPFC–ventral attention and vmPFC–frontoparietal coupling for stress responsivity and stress recovery, respectively. Finally, although we observed covariation across individuals in heart rate recovery and vmPFC connectivity within the default mode network (DMN; Figure 4b), we view this observation with caution. First, there was no main effect during recovery on vmPFC–DMN connectivity. Second, we did not exclude spatial territory encompassed by the vMPFC seed in the DMN similarity analyses. Accordingly, this observation awaits replication in future work.

It has been proposed that the vmPFC acts as a central network hub that communicates with other areas of the brain to construct affective meaning from contexts to coordinate and drive physiological and behavioral responses (Roy et al., 2012). A recent interpretive framework proposes that the vmPFC may engage in predictive processing, specifically orchestrating changes that are anticipatory or preparatory of metabolic needs generated by a context (Gianaros & Jennings, 2018; Ginty et al., 2017). This framework is supported in the current study by demonstrating that the vmPFC has functional connectivity during anticipatory stress with regions associated with both emotional and physiological responding (Hermans et al., 2011; Sinha et al., 2016; Tobia et al., 2017; Wager, Waugh et al., 2009). This framework warrants specific testing by examining whether stressor-evoked vmPFC functional connectivity changes are associated with stressor-evoked cardiovascular reactions in excess of metabolic needs (e.g., additional heart rate; cf. Carroll, Phillips, & Balanos, 2009).

This report is the first to our knowledge to demonstrate that psychological stress modulates vmPFC connectivity with known networks (e.g., increased connectivity with dorsal attention, ventral attention, frontoparietal, and decreased connectivity with the default mode network). A recent article reported that negative emotions were associated with increased connectivity in the dorsal attentional and frontoparietal networks and decreased connectivity in the default mode network (Kragel, Kano et al., 2018). Another recent study identified linear and quadratic relationships between the ventral attention, default mode, and executive control networks in the context of stressor-evoked heart rate reactivity but did not focus on the vmPFC (Young et al., 2017). Similarly, a study in 335 healthy participants demonstrated a relationship between individual cortisol responses and reductions in DMN connectivity in response to a mental and physical stress task (Zhang et al., 2019). A study focusing on whole brain network efficiency reported a relationship between increased levels of cortisol responses to stress and decreased network efficiency during stress (Wheelock et al., 2018).

The current study is not without limitations. First, the study consisted of only female participants. Some review articles have suggested the possibility of sex differences in neural activation in response to emotionally salient cues (Filkowski, Olsen, Duda, Wanger, & Sabatinelli, 2017; Hamann, 2005; Stevens & Hamann, 2012), while others have found no sex differences in the processing of negative events (Garcia-Garcia et al., 2016). Kogler, Gur, and Derntl (2015) examined sex differences in stressor-evoked, neural subjective stress, skin conductance, and cortisol responses in 43 healthy adults. Results indicated women exhibited greater neural and levels of perceived stress compared to men during the stress task, but there were no sex differences in galvanic skin or cortisol responses (Kogler et al., 2015). Some conceptual formulations (Taylor et al. 2000) have also suggested the possibility of sex differences in peripheral physiological responding to stress. However, a meta-analysis on cardiovascular stress reactivity demonstrated men exhibit significantly higher blood pressure and sympathetic nervous system reactions to stress but no differences in heart rate responses to stress (Brindle et al., 2014). Given the mixed findings in previous research examining sex differences in neural and peripheral physiological responses to stress, future studies should include both male and female participants in their studies with sufficient power to test for interactions or effect modification. This will allow for a more complete examination of sex differences and a fuller account of generalizability. Second, the sample size is relatively small and consists of all undergraduate students. Future studies should include a larger sample and a range of ages to increase statistical power and improve the generalizability of the results. Third, the self-report stress ratings were taken immediately after each speech preparation period. It would have been beneficial to include a continuous measure of stress throughout the entire speech task to better capture self-reported stress (Wager, Waugh et al., 2009). Fourth, there was no measurement of hormonal contraceptive use or phase of the menstrual cycle at the time of fMRI testing. In the parent study from which the present sample was drawn, contraceptive use was not related to heart rate reactivity or perceived stress—either in the larger sample or in the subsample selected for this study. However, future research should measure these variables. Fifth, it is currently unclear how the results observed in our anticipatory stress paradigm generalize to other stressful experiences, particularly those that do not invoke an anticipatory context. Prior studies show that nonanticipatory stress paradigms (e.g., mental arithmetic, Stroop interference) similarly modulate vmPFC activity (Gianaros et al., 2017; Sheu et al., 2012), yet it is unclear whether they similarly modulate vmPFC connectivity. Some work has suggested that the fMRI context itself represents an anticipatory stressor that alters physiological responding (e.g., Gossett et al., 2018); however, we note that cardiovascular stress responses in the current sample of women were strongly correlated across fMRI and laboratory testing sessions—suggesting relatively stable individual differences in reactivity across contexts. In these regards, examining vmPFC connectivity in these and other stress paradigms will increase our understanding of these processes. Lastly, the vmPFC is a large and heterogeneous region of the brain, which encompasses many subregions that may have different functional characteristics during periods of psychological stress. The vmPFC anatomical mask used in the present study likely encompassed many functionally distinct areas (Hiser & Koenigs, 2018). A study using a smaller vmPFC seed, specifically targeting the intersection of the default mode network and salience network, demonstrated vmPFC connectivity with the pgACC (anterior cigulate cortex), dorsal ACC, mid-ACC, and left superior temporal gyrus positively correlated with heart rate variability (Jennings et al., 2016). vmPFC subareas exhibit extensive interconnections with one another (Ongur & Price, 2000), and the present study was not designed to explicate whether there is a segregation or integration of self-report and autonomic stress-related neural correlates across vmPFC subareas. This issue could be addressed in future work using multiple vmPFC seeds. For example, different vmPFC areas might differentially or interactively influence autonomic outflow. To elaborate, it has been speculated that more dorsal territories of the vmPFC are associated with prothreat appraisals and prosympathetic functions, whereas more ventral territories and subareas are associated with prosafety appraisals or proparasympathetic functions (Critchley, 2004; Dum, Levinthal, & Strick, 2016; Myers-Schulz & Koenigs, 2012; Roy et al., 2012; Wager, Waugh et al., 2009). However, this dorsal-ventral distinction is not absolute, as there is likely to be colocalization of appraisal-related processes, as well as sympathetic and parasympathetic central influences across cortical and subcortical brain systems (see Gianaros & Wager, 2015; Kim et al., 2019; Opennheimer & Cechetto , 2016).

In summary, acute psychological stress appears to alter functional connectivity between the vmPFC and regions associated with autonomic control and emotional processing. In the present study, connectivity of the vmPFC also appeared to return to resting baseline levels after psychological stress. Further research using larger and more diverse samples is needed to fully determine whether stressor-evoked changes in vmPFC relate to individual differences in cardiovascular reactivity, cardiovascular recovery, and self-reported stress in the context of mental and physical health.