Neural Correlates of Non-Specific Skin Conductance Responses During Resting State fMRI

Abstract

Skin conductance responses (SCRs) reliably occur in the absence of external stimulation. However, the neural correlates of these non-specific SCRs have been less explored than brain activity associated with stimulus-elicited SCRs. This study modeled spontaneous skin conductance responses observed during an unstructured resting state fMRI scan in 58 adolescents. A Finite Impulse Response (FIR) fMRI model was used to detect any type of hemodynamic response shape time-locked to non-specific SCRs; the shape of these responses was then carefully characterized. The strongest evidence for signal change was found in several sub-regions of sensorimotor cortex. There also was evidence for engagement of discrete areas within the lateral surfaces of the parietal lobe, cingulate cortex, fronto-insular operculum, and both visual and auditory primary processing areas. The hemodynamic profile measured by FIR modeling clearly resembled an event-related response. However, it was a complex response, best explained by two quickly successive, but opposing neuronal impulses across all brain regions – a brief positive response that begins several seconds prior to the SCR with a much longer negative neuronal impulse beginning shortly after the SCR onset. Post hoc exploratory analyses linked these two hemodynamic response phases to different emotion-related individual differences. In conclusion, this study shows the neural correlates of non-specific SCRs are a widespread, cortical network of brain regions engaged in a complex, seemingly biphasic fashion. This bimodal response profile should be considered in replication studies that attempt to directly link brain activity to possible homeostatic mechanisms or seek evidence for alternative mechanisms.

1. Introduction

Measurements of electrodermal activity provide a useful non-invasive approach to assess sympathetic nervous system activation during mental activities (Figner and Murphy, 2010). Skin conductance levels can be deconvolved into discrete skin-conductance-responses (SCRs) that reflect peripheral nerve impulses sent to eccrine sweat glands (Benedek and Kaernbach, 2010), usually after a stimulus or event transiently increases autonomic arousal. Deconvolution accounts for the delay between onset of sweat secretion and the observed rise in skin conductance, as well as accommodating overlapping sweat secretion responses (Boucsein, 2012). Because these SCRs appear to reflect the strength of sympathetic nervous response, they have proved useful in a wide variety of psychological research, including classical conditioning, fear conditioning, emotional reactivity, attention, and cognitive effort (Boucsein, 2012).

Although electrodermal activity and SCR-related brain function mostly has been studied within active, stimulus-evoked contexts, electrodermal activity increases also reliably occur in the absence of any external stimulation. These nonspecific SCRs typically are defined as spontaneous increases in skin conductance that surpass a threshold between 0.01–0.05 microsiemens (Braithwaite et al., 2015). They occur about 5 times per minute on average in most persons, and as often as 10 times per minute (Zimmer, 2000). In the applied experimental literature, stimulus-evoked SCRs have been utilized as evidence of task engagement and intensity (Boucsein, 2012) whereas these spontaneous SCRs typically are interpreted as a trait-like measure of electrodermal response lability vs. stability that has a measureable genetic influence (Crider, 2008). Electrodermal lability vs. stability has been proposed as an endophenotype (Gottesman and Hanson, 2005), that reliably correlates with individual differences in emotional expressiveness, agreeable behavior, and self-control (Crider, 2008). Most notably of these, electrodermal response stability predicts future antagonistic behavior and incarceration in adolescents (Raine and Venables, 1984), suggesting it is a marker for impaired emotional or behavioral self-control. Nonspecific SCRs also have been reported to correlate with habituation, orienting response, negative emotional arousal, subjective arousal, and anticipatory stress, but these findings are less clear and require further investigation (Boucsein, 2012).

Over the past 20 years, functional magnetic resonance imaging (fMRI) has helped to characterize how the central nervous system participates in prompting or reacting to SCRs. Because all but a few of these studies have examined event-locked SCRs, most of what we know comes from stimulus-evoked SCR studies. A meta-analysis of 19 fMRI studies identified many brain regions that are engaged during affective, somatosensory-motor, and cognitive stimulus-evoked electrodermal activity (Beissner et al., 2013). The cortical regions included dorsal, mid-, pregenual and subgenual cingulate regions, right anterior insula and left posterior insula, small regions of right dorso- and ventrolateral prefrontal cortices, primary somatosensory cortex, and the surpramarginal gyri/superior parietal lobules bilaterally in the parietal lobe. Subcortical regions were left amygdala, thalamus, the ventral tegmental area of the brainstem, and right cerebellum. This meta-analysis also found only a subset of these brain regions – posterior and mid-cingulate, left amygdala, right anterior insula, and left posterior insula – were engaged regardless of the different task processing demands. Moreover, some neural correlates of stimulus-elicited SCRs were context specific. Emotional tasks were more likely to engage primary somatosensory cortex and thalamus, but not ventromedial prefrontal cortex. Somatosensory tasks engaged both supplemental motor area cortex and brainstem not seen in other contexts. Cognitive tasks were more likely to engage the cerebellum.

Numerous electrodermal studies have identified several neural pathways that appear capable of producing SCRs in animals or humans (e.g., (Sequeira and Hamed, 1995; Tranel and Damasio, 1994; Wang, 1964); reviewed in (Boucsein, 2012)). Because multiple pathways can be involved, it is not clear whether the brain regions involved in the generation and representation of stimulus-evoked SCRs are the same as those involved in non-specific spontaneous SCRs. The human neural correlates of non-specific spontaneous fluctuations in electrodermal activity have garnered far less investigation than brain activity associated with stimulus-evoked SCRs. In the earliest of the few relevant studies, Critchley et al., (2000) examined the correspondence of electrodermal and fMRI-measured brain activity while 6 participants were engaged in a decision-making task involving immediate positive and negative reinforcement. While this was not a true stimulus-free experiment, it is relevant because their analyses statistically covaried reward task-related variance from the fMRI timeseries. Activity seen only immediately before SCR onsets was found in anterior cingulate, cerebellum, fusiform and lingual gyri, which was interpreted as implicating these regions in the generation of SCRs. In contrast, inferior medial prefrontal cortex was active after SCRs. Other brain activity was found in both phases, including posterior cingulate, orbitofrontal cortex, and separate locations in the previously mentioned regions. These regions were interpreted as being involved with the representation of peripheral autonomic states. In another small sample of 6 participants, Patterson et al., (2002) compared the relationship between electrodermal and fMRI BOLD signals during a gambling task, a “2-back” working memory task, and several 1-minute long rest blocks interleaved between the two tasks. This study found non-specific SCR and fMRI signals during the 1-minute rest blocks covaried in many of the same brain regions as observed in active tasks. Furthermore, Patterson linked these relationships to many of the same regions found in Critchley et al., (2000). Only one study has examined the correlation of nonspecific SCRs to fMRI-measured brain activity timeseries during several minutes of unstructured resting state fMRI (Fan et al., 2012). This study found hemodynamic activity covaried with SCR timeseries in many brain regions. Positive associations were found in bilateral dorsolateral and mid-lateral prefrontal cortex, anterior cingulate, bilateral insula, supplemental motor area, inferior parietal lobule, precuneus, calcarine gyri, caudate, and cerebellum. Interestingly, a large handful of regions had negative correlations with electrodermal measurements – brain regions whose relative activity decreased during SCRs. These regions included paracentral lobule, primary sensorimotor cortex, posterior cingulate, ventromedial prefrontal cortex, and several regions on the lateral surface of the temporal lobe.

Taken together, these few fMRI studies lay a foundation that suggests a broad cortical and limited subcortical network of brain regions is likely engaged during spontaneous fluctuations of electrodermal activity. These regions probably resemble those activated during stimulus-elicited SCRs, but the inconsistency of prior nonspecific SCR fMRI study results makes exact localization uncertain. Moreover, these relationships seem more complex than initially assumed. Different results are found depending on whether investigators focus on brain activity before or after non-specific SCR onset. These prior studies also are limited by small samples, the use of various methodologies, restriction of focus to only a few brain regions, and a reliance on correlation between hemodynamic and electrodermal signals. The latter is especially problematic because some brain regions might be under-represented due to subtle differences in shape and peak latency between hemodynamic response and skin conductance response models, while other findings might be spurious if the correlation is driven by long periods of simultaneous inactivity. Even when formal modeling of SCR-related brain activity was done, assumptions about the relative timing of events of interest were made. As such, there is a need to better characterize which brain regions are engaged during stimulus-free SCR responses so we can understand if they are elicited using the same core SCR network as stimulus-driven responses or if they engage different brain regions. There also is a need to learn if different brain regions show characteristically different types of responses. As suggested by Critchley et al., (2000), perhaps only some regions have tightly SCR-locked responses that precede SCR onset which could imply a role in neural generation. Other brain regions might only engage later, e.g., in response to ascending peripheral nervous system signals as they re-enter the brain.

The goal of this study was to address these questions by characterizing neural responses that coincide with nonspecific SCRs during task-free, resting state fMRI, in a moderate-sized sample. Our approach was to identify any brain region that showed evidence for hemodynamic response immediately before, during, or after nonspecific SCRs using an fMRI model with few assumptions about the shape or extent of hemodynamic activity within this time window. The modeling of non-specific SCRs during unstructured fMRI utilized in the present study is most similar to the approaches used in stimulus-evoked SCR studies, but remain capable of capturing regional activity time-locked to spontaneous phasic SCRs that had any hemodynamic onset, latency, or shape. A second, exploratory goal was to characterize the onset and activation timecourse of those regions’ hemodynamic responses by testing their fit to different types of conventional hemodynamic models. This analysis also sought to determine if different brain regions showed different response profiles. We expected brain activity linked to nonspecific SCRs would primarily reflect simple transient neuronal impulses that occur prior to the onset of electrodermal activity and would resolve several seconds later (i.e., resemble brief event-related hemodynamic responses). We hypothesized these nonspecific SCRs would engage a neural network that resembled those described in prior reports of task-elicited SCRs, particularly the dorsal cingulate, left amygdala, and anterior/posterior insular regions that are commonly engaged during task-elicited SCRs regardless of experimental context and may represent the core of the neural system underlying SCR-related processing.

2. Materials and Methods

2.1. Participants

The final sample after a handful of exclusions for data quality control comprised 58 healthy adolescents between ages 12–18 (32 females; mean age 15.62 ± SD 1.73) who participated in an NIMH-funded study (R01MH102854). The current effort analyzed available data collected to supplement the parent project’s primary aims. Participants were recruited via advertisements at the Olin Neuropsychiatry Research Center (Olin NRC) in Hartford, CT. Written informed consent was required using a protocol approved by Hartford Hospital’s Institutional Review Board. For legal minors, parents provided written permission and minors gave assent. All participants had IQ estimates at least in the average range or better.

2.2. Electrodermal Methodology

Electrodermal activity was recorded with an EDA100C-MRI amplifier. This recording system (www.biopac.com) was designed specifically for MRI acquisition use, incorporating signal processing circuitry to remove spurious MRI-induced artifacts in the electrical recording at the source. It used a 0.5 constant voltage signal across a pair of single-use dry electrodes. Biopac’s biocompatible, isotonic gel (GEL101) was applied to ensure maximum electrical conductivity. Electrodes were placed on the palmar surface of the left hand before the participant entered the MRI scanner. The MRI scanner room was kept at the same temperature (65⎕) and humidity (35%) for every participant. Skin conductance levels were allowed to stabilize during 45 minutes while brain structure and other project-related scans were collected. Electrodermal data acquisition was triggered by the start of the fMRI scan. Data was collected continuously at 1000 Hz during the entire fMRI resting state data collection run.

Using Biopac’s Acqknowledge software, the raw data were downsampled to 62.5 Hz. Median smoothing with a 63 sample window then was applied to remove signal contamination from fMRI gradient switching. A 1 Hz FIR low-pass filter was then applied and any differences between this resulting waveform and the original waveform that were >0.05 μS indicated additional signal artifacts (e.g., from hand motion) and were removed by linear interpolation in Matlab 2013 (Mathworks, Inc., Natick, MA) software using a digital inpainting function (D’Errico, 2006). The resulting file was saved in a format compatible with Ledalab software (Benedek and Kaernbach, 2010).

Only participants who showed a relatively large SCR (greater than 0.1 microsiemens) were examined. This threshold ensured adequate signal-to noise-ratio, which is especially low when observing innate physiological changes as opposed to laboratory generated responses. Two subjects whose data did not meet this threshold were excluded from further analysis. The focus on the strongest SCRs reduced concerns that overlapping phasic electrodermal responses from closely contiguous events might compromise the results, as such responses were relatively infrequent. Participants averaged approximately 47 ± 36 SCRs throughout the 7 minute resting state scan (i.e., approximately 6.7 per minute). Neither the number of SCRs (t=1.239, p=0.221), nor mean amplitude of SCRs (t=1.441, p=0.155) differed by sex. Participant age was likewise uncorrelated with these measures (SCR frequency r=0.097, p=.467 and SCR amplitude r=−.082, p=.543). Moreover, self-ratings of subjective anxiety prior to MRI using the Profile of Mood States-2 (Heurchert and McNair) was unrelated to SCR frequency (r=−.123, p=.371) or amplitude (r=−.203, p=.137). These factors were not further considered in hypothesis-testing.

2.3. fMRI Methodology

All participants were scanned at the Olin NRC using a Siemens 3T Skyra. Urine collected prior to MRI was tested for the presence of drug metabolites and pregnancy. MR sequences were chosen for compatibility with Human Connectome Project (HCP) pre-processing pipelines (Marcus et al., 2013), which provide highly accurate, structural image-guided brain atlas normalization for fMRI. Several types of MR sequences were obtained: fMRI 947 gradient echo planar images EPI (TR/TE 475/30 msec, flip 60 °, multi-band AF=8), 7:50 min each. Fieldmaps (TR/TE 731/4.92, flip 50°, AF=1, 1:45 min; and TR/TE 7220/73 msec, flip 90 °, AF=1, 0:15 min, run twice with reversed A>>P phase encoding) (EPI/fieldmap sequences had 3 mm isotropic voxels, 48 interleaved slices). T1-weighted (3D MPRAGE, TR/TE/TI=2400/2.07/1000 msec, flip 8°, 0.8 mm isotropic vox, 8:19 min) and T2-weighted (TR/TE=3200/387, 0.8 mm isotropic vox; 11:06 min) structural scans also were acquired. All structural images were radiologist-assessed to be free of macroscopic pathology. Daily MR stability and QA measurements ensured scans were of equal quality throughout the entire project.

Data preparation pipelines included scripts that operated on structural T1/T2 data (Pre-FreeSurfer, FreeSurfer, Post-FreeSurfer) or functional MRI data (Volume and Surface). Briefly, registered, distortion fieldmap-corrected, and finally MNI152 atlas-registered using FSL FLIRT+nonlinear FNIRT algorithms (Jenkinson et al., 2012). To denoise the data from known artifacts (e.g., cardiac and respiratory aliasing, global signal confounds, high-frequency noise, etc.) ICA-FIX with the “aggressive cleanup” algorithm utilized FSL MELODIC ICA output for regression of the full space of all noise components and head motion confounds (Griffanti et al., 2014; Salimi-Khorshidi et al., 2014). FreeSurfer-based registration, skull-stripping, and pial extraction on 1 mm-downsampled T1/T2 data were used to create structural volume/cortical ribbon files (Fischl et al., 1999). EPI data were registered to FreeSurfer cortical ribbon output, resampled to atlas space, intensity normalized, smoothed at 2 mm FWHM, and high-pass temporal filtered (2000 sec) before being written as a timeseries for analysis. Data from the dense timeseries were averaged within parcels identified in a recently-developed, multi-modal cortical atlas parcellation from the HCP group (Glasser et al., 2016a) prior to fMRI activation modeling. The use of this well-validated, detailed neuroanatomical atlas offer confidence in precise localization of any hemodynamic activation abnormalities.

All datasets met rigorous fMRI data quality control criteria for head motion. Framewise displacement (FD) (Power, 2012) values were estimated to quantify scan-to-scan motion using all six translation (i.e., X, Y, and Z axes) and rotation realignment parameters (i.e., pitch, yaw, and roll). Two participants who had acceptable skin conductance data, but whose fMRI timeseries showed a mean FD > .30 were excluded from analyses, consistent with the most rigorous available criteria (Power et al., 2012) used in prior fMRI studies of adolescents who can be more prone to fidget in the scanner. Mean FD was greater in males than females (t=2.10, p=.04) and greater in younger participants (r=−0.319, p=.015). Mean FD did not correlate with either number of non-specific SCRs (R²<0.001, p=0.819) or mean SCR amplitude (R²=0.019, p=0.29).

2.4. Hemodynamic Response Modeling

We used a Finite Impulse Response (FIR) model to localize brain activity because our goal was to identify any type of hemodynamic signal change within a reasonable window around each SCR (Henson and Friston, 2007). FIR models in Statistical Parametric Mapping software (SPM12; www.fil.ion.ucl.ac.uk/spm) assess whether each data point within an event-locked timeseries window differs from the overall mean, making the fewest possible assumptions about the actual shape of the hemodynamic response. This flexibility was well-suited to our data due to our lack of a priori knowledge exactly when SCR-linked brain activity might start or stop for these presumably endogenous physiological processes. After localization of SCR-linked brain activity, typical hemodynamic response curves could be depicted as activation estimate averages for each timepoint across the sample. These averages should be relatively robust to any overlapping phasic SCRs that remained after yoking events to SCRs that surpassed the 0.1 microsiemens threshold. Our 16.15 sec FIR window had intervals defined by the 0.475 sec TR. The window began 3.8 sec before each SCR because brain activity that might prompt a peripheral skin conductance responses (and the start of its corresponding delayed hemodynamic response) likely began several seconds before SCR peaks. The window was long enough to capture the typical peak of both SCR (2–4 sec; (Figner and Murphy, 2010)) and contemporaneous event-related hemodynamic responses (5–8 sec; (Henson and Friston, 2007)). This window also allowed us to determine if SCR-related brain function resolves in a characterizable event-related manner or might be sustained over time. Low frequency drifts in signal were removed in the FIR model using a standard high-pass filter with a 128 second cutoff. Serial correlation was estimated using an autoregressive AR(1) model.

An SPM12 factorial model assessed if the 34 parameter estimate maps (one for each window timepoint) differed from each other. Independence between levels was not assumed. Variance was assumed to be unequal. The F statistics generated for each voxel were corrected for multiple comparisons using False Discovery Rate (Benjamini and Hochberg, 1995). The critical value for significance corresponding to α = 0.05 for F_33,1938 = 1.694.

Human Connectome Project Workbench software was used to visualize FIR modeling results using a recently-developed multi-modal cortical atlas parcellation (Glasser et al., 2016b). This detailed neuroanatomical atlas offers confidence in precise localization of the brain regions whose activity levels changed within the FIR window. To focus reporting on only the most convincingly ‘active’ parcels, we selected only those that had an above average proportion of activated vertices/voxels, i.e., engagement of the majority of values within each discrete brain region (>1 SD above the sample mean, i.e., 41.6% of the parcel). This offers a more conservative approach than reporting any parcels for which there might be only a few values out of possibly hundreds within that region whose values met whole-brain significance thresholds. The mean proportion of active vertices/voxels across parcels was 18.7% ± 22.9%. All other parcel results are described in the Supplemental material (Table S1).

A supplemental re-analysis was done where each participants’ FIR model included an estimate of a proxy for global signal (i.e., average value of grey matter voxels across the brain). Although the effects of global signal are typically a concern only in the context of cross-correlation analyses for functional connectivity, these supplemental analyses were done here to ensure the validity of localized findings.

2.5. Characterization of Hemodynamic Responses

A consequence of FIR modeling is that the model will identify brain regions whose activity levels change in any way, not necessarily only signal changes that resemble characteristic hemodynamic responses to neuronal impulses (e.g., slow linear shifts, abrupt and difficult-to-interpret activity level shifts, etc.). To identify which brain regions showed typical brain activity response curves, and to fully characterize the onset, shape, and latency of hemodynamic responses for the brain regions identified as showing signal change within the FIR window, we extracted the mean of the brain surface vertex or voxel values within each parcel for each timepoint. These SCR-locked timecourses had 34 average beta-weight values for each parcel. Then, to learn if these hemodynamic profiles showed a similar or different shape across different brain regions, the diceR package in R software was used to test if these regions could be subdivided into classes across 12 commonly-used clustering algorithms (Chiu and Talhouk, 2018). Statistical significance for 2-, 3- and 4-cluster solutions was tested against the null hypothesis that the data were drawn from a single Gaussian distribution. These tests used a Cluster Index defined as the sum of within-class sums of squares about the mean divided by total sums of squares (Liu et al., 2008). We then averaged all regional SCR-locked averages into a grand mean for visualization.

We next conducted analyses to determine what sorts of neuronal activity impulses might have elicited the nonspecific SCRs. We compared the grand average of hemodynamic activity across all parcels that was detected within the FIR window to a) a series of single brief canonical positive-going response impulses modeled using SPM12’s canonical hemodynamic response function, each having an onset at every roughly half second interval (TR=0.475 sec) throughout the response window; b) a series of single brief canonical negative-going response impulses, again with activity onset roughly every half second; and c) a combination of two responses – one positive- and a second negative-going. The order of the positive and negative going impulses was not assumed. The latter not only tested all possible onsets at the TR intervals for both positive and negative impulse onsets, but also modeled different event durations to ensure the model considered all possible combinations of event durations and onsets within the FIR window. Both datasets were resampled to facilitate the hemodynamic convolution at the TR of 0.475 sec and to match a final sampling rate at 2 Hz to avoid possibly inflating Pearson r values from artificially high data dimensionality. Next, we iteratively tested all possible parameters of the various models until the best-fitting correlation was found to identity the ‘winning’ model. Pearson r correlation coefficients were used to quantify the correspondence between each hemodynamic model and the grand average of all regional SCR-locked averages for each model type. Finally, we subjected this result to a reliability analysis that randomly assigned study participants to 5 separate split-half sub-samples. These sub-samples were re-examined iteratively across all possible model parameters to a) confirm the best-fitting model type and b) seek convergence of specific final model parameters for each sub-sample winning model. Matlab 2013 code for this analysis will be provided upon request.

To determine if different phases of brain activity (e.g., before or after the SCR itself) might reflect individual differences in sympathetic nervous system reactivity, post hoc analyses displayed in Table 2 determined whether either a) the frequency of nonspecific SCRs within the 7-minute resting state run, or b) the average amplitude of the SCRs across all 58 participants was associated with any of the 34 average beta-weight values from their SCR-locked grand averages. Spearman’s rho was used because data screening indicated that all of the FIR estimates of brain activity were not normally distributed. We also examined the association of SCR-locked activity to data available from the parent R01MH102854 study of adolescent emotion regulation and emotion-cognition interaction. These analyses were done to explore the possible functional significance of the activity profiles. The specific measures are described in the Supplement. A visual aid for understanding the Materials and Methods is provided in the Supplement (Figure S3).

Table 2.

Correlations of FIR Analysis with Behavioral Measure

Scan Start Time (Sec)	SCR Number	SCR Average Amplitude	Emotional Reactivity Scale Total	ERQ Reappraisal	ERQ Suppression	EDP Attention Bias Hit Mean RT	Emotional Stroop Hit Mean RT	EATQ Activation	EATQ Attention	EATQ Inhibition	POMS Anxiety
−2.375	−0.06	−0.05	0.15	−0.11	0.171	0.291*	0.07	0.11	0.01	0.05	0.19
−1.9	−0.19	−0.17	0.16	−0.03	0.184	.276*	0.09	0.15	0.02	0.10	0.19
−1.425	−.411**	−.352**	0.16	−0.01	0.159	0.24	0.25	0.16	0.07	0.05	0.09
−0.95	−.362**	−.277*	0.17	0.03	0.095	0.21	0.23	0.16	0.08	−0.01	0.10
−0.475	−.385**	−.301*	0.11	0.08	0.200	0.24	0.26	0.17	0.07	0.01	0.09
0.000	−.282*	−.274*	0.08	−0.03	0.103	0.13	0.23	0.18	0.11	0.09	0.03
0.475	−0.22	−0.17	0.12	−0.05	0.046	0.10	0.15	0.21	0.14	0.11	−0.02
0.950	−0.10	−0.12	0.06	−0.06	0.073	0.13	0.10	0.25	0.10	0.10	−0.04
1.425	−0.11	−0.08	0.08	−0.08	0.032	0.16	0.15	0.19	0.08	0.07	−0.07
1.900	0.10	0.14	0.09	−0.20	0.073	0.13	0.04	0.09	−0.04	0.10	−0.09
2.375	0.15	0.17	0.10	−0.17	0.084	0.18	0.06	0.03	−0.11	0.12	−0.08
2.850	0.15	0.20	0.14	−0.18	0.088	0.08	−0.02	0.04	−0.06	0.11	−0.08
3.325	0.10	0.12	0.16	−0.13	0.037	0.06	0.02	0.07	−0.11	0.14	0.05
3.800	0.03	0.07	0.25	−.268*	−0.033	0.01	0.09	0.11	−0.09	0.04	0.07
4.275	0.09	0.10	.324*	−.303*	0.016	0.02	0.02	0.13	−0.13	−0.02	0.07
4.750	0.19	0.15	.351**	−.309*	−0.032	0.02	0.06	0.15	−0.10	−0.03	0.15
5.225	.259*	0.21	.356**	−.346**	−0.048	0.00	0.04	0.16	−0.09	−0.04	0.16
5.700	.319*	.285*	.347**	−.341**	0.018	0.01	0.04	0.10	−0.13	−0.02	0.20
6.175	.417**	.364**	.264*	−.357**	0.067	−0.02	0.00	−0.02	−0.12	0.07	0.11
6.650	.425**	.370**	0.24	−.376**	0.017	0.01	0.03	0.01	−0.07	0.04	0.09
7.125	.417**	.345**	0.19	−.371**	0.051	−0.01	0.06	0.00	−0.08	−0.01	0.03
7.600	.463**	.388**	0.21	−.448**	0.048	−0.01	0.06	0.06	−0.03	0.05	0.13
8.075	.400**	.338**	0.11	−.386**	0.084	−0.01	0.08	0.06	0.00	0.07	0.14
8.550	.443**	.367**	0.08	−.444**	0.051	0.03	0.10	0.06	0.04	0.09	0.12
9.025	.467**	.394**	0.00	−.465**	0.074	0.04	0.13	0.10	0.05	0.09	0.06
9.500	.468**	.382**	−0.05	−.485**	0.062	0.05	0.11	0.15	0.09	0.05	0.05
9.975	.482**	.438**	−0.05	−.495**	0.029	0.02	0.10	0.16	0.13	0.08	0.09
10.450	.461**	.366**	−0.02	−.467**	0.036	0.01	0.11	0.21	0.12	0.14	0.14
10.925	.328*	.273*	0.02	−.451**	0.035	0.04	0.11	0.25	0.10	0.03	0.21
11.400	.379**	.299*	0.01	−.385**	0.173	0.03	0.07	0.20	0.08	0.20	0.16
11.875	.319*	0.22	−0.01	−.344**	0.038	−0.02	0.07	0.20	0.05	0.07	0.16

Scan Start Time is in seconds referenced to skin conductance response onset time.

EATQ-Early Adolescent Temperament Questionnaire, ERQ-Emotional Regulation Questionnaire, EDP-Emotional Dot-Probe Task, RT-Reaction Time, POMS-Profile of Mood States. See supplement for summary of assessments and select psychometrics.

All values are Spearman rho correlations.

^*uncorrected p<0.05

^**uncorrected p<0.01

3. Results

3.1. Modeling SCRs During Resting State fMRI

Fifty-seven out of the 388 brain regions described by the Human Connectome Project atlas had vertices or voxels that met p<0.05 FDR significance thresholds and engaged a convincing proportion of each parcel (i.e., >1 SD above the mean coverage, i.e., >41.6% of the parcel). Table 1 lists these regions along with their atlas parcel name/label, the degree of coverage, and its peak F statistic. There were many parcels in what typically is labeled as sensorimotor cortex. There also was evidence for engagement of discrete areas within the lateral surfaces of the parietal lobe, cingulate cortex, fronto-insular operculum, and both visual and auditory primary processing areas. All other brain region results are listed in the Supplement (Table S1). Most of the latter had FIR model evidence for signal change in a convincingly low proportion of their vertices or voxels; 19 parcels had no evidence for signal change at all. Figure 1 displays all these regions, with the parcels reported in Table 1 outlined in blue. This surface representation shows the strongest FIR modeling evidence for signal change was found in sensorimotor, parietal, and fronto-insular operculum. Specifically, the most statistically significant findings included right posterior operculum/somatosensory area II (OP1), a right inferior parietal opercular region (PFop), left hemisphere primary motor regions (area 1), anterior (p32p) and posterior cingulate (POS2), and visual cortex (V1).

Table 1.

Parcels with percentage of significant voxels or vertices greater than 1 SD above the mean (41.6%)

	Area Name (Atlas Label)	Region Area	% Covered	Max F
SensoryMotor
	Right Area OP1/SII (R OP1 ROI)	Posterior Opercular Cortex	100.0%	4.69
	Left Area OP2–3/VS (L OP2–3 ROI)	Posterior Opercular Cortex	93.8%	2.68
	Left Area OP1/SII (L OP1 ROI)	Posterior Opercular Cortex	93.6%	3.85
	Right Area PFcm (R PFcm ROI)	Posterior Opercular Cortex	92.6%	4.82
	Left Area PFcm (L PFcm ROI)	Posterior Opercular Cortex	92.2%	3.91
	Left Area OP4/PV (L OP4 ROI)	Posterior Opercular Cortex	91.9%	3.84
	Right Area OP4/PV (R OP4 ROI)	Posterior Opercular Cortex	91.8%	4.03
	Right Area OP2–3/VS (R OP2–3 ROI)	Posterior Opercular Cortex	87.4%	3.76
	Right Area 43 (R 43 ROI)	Posterior Opercular Cortex	85.7%	3.04
	Right Dorsal area 6 (R 6d ROI)	Premotor Cortex	78.1%	3.28
	Left Area 43 (L 43 ROI)	Posterior Opercular Cortex	76.4%	3.03
	Left Dorsal area 6 (L 6d ROI)	Premotor Cortex	74.5%	3.16
	Right Frontal Opercular Area 1 (R FOP1 ROI)	Posterior Opercular Cortex	71.6%	2.42
	Right Area 2 (R 2 ROI)	Somatosensory and Motor Cortex	70.7%	3.65
	Left Area 1 (L 1 ROI)	Somatosensory and Motor Cortex	66.7%	3.86
	Left Frontal Opercular Area 1 (L FOP1 ROI)	Posterior Opercular Cortex	65.6%	2.94
	Right Ventral Area 6 (R 6v ROI)	Premotor Cortex	64.3%	2.57
	Left Area 2 (L 2 ROI)	Somatosensory and Motor Cortex	64.2%	3.51
	Right Area 1 (R 1 ROI)	Somatosensory and Motor Cortex	63.4%	3.68
	Left Primary Sensory Cortex (L 3b ROI)	Somatosensory and Motor Cortex	58.8%	2.95
	Right Supplementary and Cingulate Eye Field (R SCEF ROI)	Paracentral Lobular and Mid Cingulate Cortex	58.4%	2.85
	Right Primary Sensory Cortex (R 3b ROI)	Somatosensory and Motor Cortex	53.5%	3.52
	Left Supplementary and Cingulate Eye Field (L SCEF ROI)	Paracentral Lobular and Mid Cingulate Cortex	51.9%	2.68
	Left Area 3 a (L 3 a ROI)	Somatosensory and Motor Cortex	46.9%	2.67
	Left Ventral Area 6 (L 6v ROI)	Premotor Cortex	44.2%	2.59
Parietal
	Right Area PF Opercular (R Pfop ROI)	Inferior Parietal Cortex	96.7%	3.96
	Left Lateral Area 7A (R 7A (L ROI)	Superior Parietal Cortex	81.7%	3.39
	Right Area PFt (R PFt ROI)	Inferior Parietal Cortex	78.7%	3.50
	Right Area 7PC (R 7PC ROI)	Superior Parietal Cortex	76.0%	2.82
	Left Lateral Area 7A (L 7A (L ROI)	Superior Parietal Cortex	75.0%	3.11
	Right Area PF Complex (R PF ROI)	Inferior Parietal Cortex	64.7%	4.63
	Left Area PF Opercular (L PFop ROI)	Inferior Parietal Cortex	64.1%	3.81
	Left Area 7PC (L 7PC ROI)	Superior Parietal Cortex	55.6%	3.10
	Left Area PF Complex (L PF ROI)	Inferior Parietal Cortex	48.7%	4.17
Cingulate
	Right Parieto-Occipital Sulcus Area 2 (R POS2 ROI)	Posterior Cingulate Cortex	73.2%	2.91
	Left Parieto-Occipital Sulcus Area 2 (L POS2 ROI)	Posterior Cingulate Cortex	65.4%	3.49
	Left Prime Area p32 (L_p32p (R ROI)	Anterior Cingulate and Medial Prefrontal Cortex	53.9%	2.66
	Right Dorsal Transitional Visual Area (R DVT ROI)	Posterior Cingulate Cortex	53.7%	3.12
	Right Prime Area p32 (R_p32p (R ROI)	Anterior Cingulate and Medial Prefrontal Cortex	51.4%	2.68
	Right Area anterior 32 prime (R a32p (R ROI)	Anterior Cingulate and Medial Prefrontal Cortex	47.3%	2.29
	Left Dorsal Transitional Visual Area (L DVT ROI)	Posterior Cingulate Cortex	43.1%	2.52
Visual	Left Sixth Visual Area (L V6 ROI)	Dorsal Stream Visual Cortex	71.8%	2.69
	Right Sixth Visual Area (R V6 ROI)	Dorsal Stream Visual Cortex	67.5%	3.00
	Left Area V3A (L V3A ROI)	Dorsal Stream Visual Cortex	60.3%	2.57
	Left Primary Visual Cortex (L V1 ROI)	Primary Visual Cortex	55.5%	3.46
	Right Primary Visual Cortex (R V1 ROI)	Primary Visual Cortex	53.6%	3.25
	Right Area V3A (R V3A ROI)	Dorsal Stream Visual Cortex	50.0%	2.59
	Left Posterior InferoTemporal Complex (L PIT ROI)	Ventral Stream Visual Cortex	45.3%	3.02
	Left Medial Superior Temporal Area (L_MST_ROI)	MT+ Complex and Neighboring Visual Areas	44.0%	2.20
	Right Area Lateral Occipital (R_LO3_ROI)	MT+ Complex and Neighboring Visual Areas	43.8%	2.79
Fronto-InsularOperculum
	Right Insular Granular Complex (R Ig ROI)	Insular and Frontal Opercular Cortex	56.8%	3.10
	Right Frontal Opercular Area 3 (R FOP3 ROI)	Insular and Frontal Opercular Cortex	54.5%	2.84
	Right Frontal Opercular Area 2 (R FOP2 ROI)	Insular and Frontal Opercular Cortex	51.8%	2.59
	Left Frontal Opercular Area 2 (L FOP2 ROI)	Insular and Frontal Opercular Cortex	47.0%	2.52
Auditory and TPO Junction
	Right PeriSylvian Language Area (R PS (L ROI)	Temporo-Parieto-Occipital Junction	45.8%	2.88
	Right Lateral Belt Complex (R LBelt ROI)	Early Auditory Cortex	45.5%	2.28
	Left Primary Auditory Cortex (L A1 ROI)	Early Auditory Cortex	44.6%	2.50

Significant voxels and vertices passed a threshold of False Discover Rate corrected p>0.05.

The table is organized by “super regions” designated by Glasser et al., 2016.

Figure 1.

Significant signal change at FDR corrected 0.05 threshold. Significantly active regions listed in table 1 are outlined in blue. FIR model window began 3.8 seconds prior to SCR onset and lasted 16.15 seconds.

3.2. SCR Hemodynamic Response Shape Characterization

The post hoc analyses of event-locked SCR hemodynamic signal levels describe the shape of the signal change. Figure 2 shows the grand mean response across all brain regions reported in Table 1. The response profile measured by FIR modeling appeared to be event-related; there was a rise in hemodynamic signal that began a couple of seconds before the point at which descending peripheral nerve signals would begin signaling a skin conductance reaction. However, the average BOLD response profile did not exactly resemble a canonical hemodynamic response function. The characteristic rise seemed truncated, and the subsequent dip below baseline activity levels was both greater and much longer than expected. When we assessed the association between this SCR-locked activation waveform and various simple canonical hemodynamic response models, we found the best of the single onset, positive event model fit the average fMRI data with Pearson (r = 0.759). The best-fitting negative-going response model had relatively poorer fit (r = 0.677). The optimal hemodynamic model combined a brief positive brain response prior to the SCR with a subsequent prolonged negative neuronal impulse. This combination of two closely proximal responses had a near perfect correlation with brain function measured by FIR (r = 0.990). The split half reliability analysis across the five randomly-generated sub-samples narrowed down the range of positive-going neuronal activity onsets to −4 to −2.5 seconds, with durations from 1–6 seconds (Table S3). The average onset time across sub-samples for this initial positive-going response was −3.6 sec prior to the SCR peak, and lasted for an average of 3.4 sec across models. Parameters for the subsequent negative-going response across sub-samples had an average onset of 1.4 sec after the SCR peak (range −2 to 4.5 sec) and an average duration of 12 sec (range 9 to 15 sec). In summary, the exact parameters of the biphasic model varied somewhat across the sub-samples, but clearly indicated two distinct response phases with a fair consistency across sub-sample estimates (see Supplemental Table 3). The positive-going response reliably occurred before the SCR onset for a brief duration, while there was a much longer neuronal impulse that prompted the negative-going hemodynamic response beginning at or shortly after the peak of the phasic SCR.

Figure 2.

The grand mean response is the average hemodynamic signal from all brain regions deemed to be active across all subjects. The average is timelocked to the non-specific SCRs > 0.1 microsiemens, beginning 5 seconds prior to SCR onset and continuing to 15 seconds after. A series of positive hemodynamic models and negative hemodynamic models were compared and the best matches for positive and negative models are displayed, along with their Pearson correlation coefficient to the grand average.

All 57 brain regions appeared to have the same general hemodynamic response profile. When they were examined by cluster analysis, none of classification algorithms in the diceR package in R was able to produce a clustering solution that was statistically different from a null distribution. Consistent with this, when the 2- and 3- cluster solutions from a straightforward k means cluster approach were visualized (see Supplemental Figure 1), it was clear that enforced groupings had only minor differences in the amplitude of signal change to either the earlier or later phases of this biphasic hemodynamic response.

3.3. Post Hoc Analysis Results

Post hoc analyses found both early and late phases of the SCR time-locked response in how brain activity levels correlated with individuals’ number and amplitude of SCRs. Four contiguous timepoint values in the first phase were significant at uncorrected p<0.05, from 1.425 sec before SCR onset up to 0.95 sec SCR onset for both frequency and amplitude (correlation coefficient rho values ranged from −.274 to −.411). SCR frequency also was correlated with fifteen contiguous timepoints in the second phase, from 5.225 to 11.875 sec after SCR onset. SCR amplitude also correlated with 13 of those 15 timepoints in the second phase, from 5.7 sec to 11.4 sec. These Spearman rho values ranged from .285 to .357.

Post hoc analysis also revealed 3 of the 9 available cognitive control and emotion-related measures correlated with either the first or second phases of the SCR-related hemodynamic response. Mean reaction time during emotional attention bias E-Dot Probe trials positively correlated with the two initial time points of the first phase. In the second phase, the mean of Reappraisal scores on the Emotion Regulation Questionnaire correlated with eighteen time points between 3.8 and 11.875 seconds after SCR onset, and total score on the Emotional Reactivity Scale correlated with five time points between 4.275 and 6.175 seconds after SCR onset. An additional post hoc analysis used linear regression to statistically remove the influence of SCR number from brain activity estimates across the timepoints. In this way, we could reexamine all these associations controlling for a likely index of overall autonomic arousal. These results are tabulated in the Supplement (Table S2). There generally were no meaningful changes to the results above.

4. Discussion

The goals of this study were to localize and characterize the nature of hemodynamic brain activity that occurs during task-free non-specific skin conductance responses. As hypothesized, fMRI modeling found many brain regions had hemodynamic signal changes associated with nonspecific SCRs. The exact regions roughly correspond to a synthesis of findings from the few prior studies of nonspecific SCRs (Critchley et al., 2000; Fan et al., 2012; Patterson II et al., 2002), but are more widespread and specific than previously has been described. These regions also overlap with those found to be engaged during stimulus-elicited SCRs in prior studies (Beissner et al., 2013), but with some important differences. First, the neural correlates of nonspecific SCRs appear to be predominantly if not wholly cortical. Even at the least stringent statistical thresholds (see Supplement; Table S1), there is scant evidence for cerebellum, thalamus, amygdala, or brainstem engagement. The absence of amygdala engagement to non-specific SCRs is noteworthy because task-elicited SCRs consistently engage the amygdala in fMRI meta-analyses (Beissner et al., 2013) and are impaired by amygdala lesions in most functional contexts (Dallakyan, 1970). An amygdalectomy in a human patient did not impair SCRs to attention orienting responses (Tranel and Damasio, 1989) and the amygdala is not engaged during the perception and experience of pain (Fauchon et al., 2019; Seifert et al., 2013). This raises interesting new questions about whether or not the neural system underlying non-specific SCRs might be the same as in those functional contexts, or why limbic processing is not necessary for generating some SCRs. Second, non-specific SCRs do not appear to focally engage the lateral surfaces of the frontal lobes. Evidence for activity there was diffuse and sporadic, never convincingly converging within discrete homogenous brain parcels. This is notably different than task-elicited SCRs. Third, the strongest evidence for brain activity during nonspecific SCRs was found in secondary somatosensory cortex and regions in the posterior parietal operculum. Although these are not regions shown by fMRI to be consistently engaged for stimulus-elicited SCRs (Beissner et al., 2013), parietal cortex lesions are known to completely abolish SCRs to psychological or physical stimuli (Tranel and Damasio, 1994). The current results offer more anatomical precision than possible in those lesion studies, which can serve as a foundation for future studies seeking to assess the exact contribution of parietal cortex to SCRs. Finally, this study identified several other regions not previously implicated in task-elicited SCRs, including posterior cingulate, auditory cortex, and regions in the temporal-parietal and occipital cortex.

Our analysis of hemodynamic activity in these regions found they all had a similar, if somewhat complex response profile. While there was a clear event-related response locked to the SCR onset, it did not resemble a hemodynamic response to a single, discrete impulse. Instead, exploratory post hoc simulations suggested that it seemed to combine two proximal event-related responses. The hemodynamic model with the strongest correlation with fMRI data combined a brief positive-going response beginning several seconds before the SCR peak with a prolonged, negative-going response starting at or shortly after that electrodermal peak. Every brain region linked to nonspecific SCRs showed essentially this same biphasic profile with only minor variations in the height of the first or second response. An important implication of this observation is that fMRI hemodynamic models which presume a single, simple discrete hemodynamic response will fail to capture the brain’s actual nonspecific SCR-linked responses. The model also suggests the exact same cortical network is engaged twice during nonspecific SCRs, but in different ways. The observation that these two response phases correlated in characteristically different ways with participant individual differences also suggests each phase could serve characteristically different purposes. If these phases are in fact prompted by distinct neural impulses, the first response might reflect when the peripheral nerve response is instigated, as it begins approximately where one would expect after adjusting for the hemodynamic lag and brief additional peripheral nerve conduction time. Our split-half reliability analyses emphasize that the exact timing of the second response remains to be conclusively pinned down. Although it appears to begin coincident with the SCR peak or shortly thereafter, the exploratory model fit raised the possibility this second neural response could start slightly before the SCR actually peaks. If so, any interpretation would have to explain how the second phase begins sooner than nerve conduction time from the brain to hand and back. More likely however, the timings recovered from exploratory modeling suggest this second phase represents re-afferent connections back into the brain from the peripheral nervous system conveying information about bodily states. This is bolstered by the observation that the strongest signal change was found for secondary sensorimotor cortex which is known to represent a somatotopic map (Harding-Forrester and Feldman, 2018). The long duration of this protracted negative-going response also should prompt speculation it might represent a sustained period of neuronal inhibition, possibly a homeostatic mechanism. If this biphasic response represents some sort of homeostasis within this distributed cortical system, future studies should seek its cellular basis and determine how it operates. For instance, SCRs are produced by an organ system whose primary function is to maintain a body temperature (Hori, 1991), with dual GABAergic and glutamateric descending pathways for thermoregulation (Romanovsky, 2018). As such, it is possible these two neurotransmitter pathways might modulate these two fMRI-measured response phases, perhaps in different ways. Researchers might also look for clues about possible homeostatic functions in biological systems other than thermoregulation, such as pain processing or even emotional response. These systems engage brain regions that also are linked to SCRs for which models have been proposed involving homeostasis (e.g., (Craig, 2003). Not only do our results include brain regions outside of classically-defined thermoregulation pathways (e.g., locomotive and limbic-affective circuits, Wang 1964; reviewed in (Boucsein, 2012), our post hoc analyses linked different brain activity phases to various cognitive-emotional individual participant differences. Because emotional reactivity and emotion regulation reappraisal correlated to brain activity in inverse ways, future studies ultimately might link the interplay between these individual differences directly to the function of homeostatic mechanisms in different psychological contexts. Some examples for further study are emotional expressiveness (Cacioppo et al., 1992; Jones, 1950), and orientation and habituation (Eisenstein et al., 2001; Williams et al., 2000). The possibility that this represents some homeostatic process also would be bolstered if specific synaptic mechanisms can be identified that mediate both excitatory and inhibitory effects within the same neuronal assemblies (e.g., pre-synaptic lateral inhibition or similar mechanisms) that are expressed throughout all these cortical regions. Any such explanation for this apparently biphasic response also should satisfactorily explain the purpose of such a prolonged refractory period after the system’s initial engagement.

Alternative interpretations might consider recent findings that vascular processes explain BOLD signal response variance during unstructured states in ways that could have relevance to the autonomic processes examined in this study. For instance, Özbay and colleagues recently observed a similar-appearing bimodal hemodynamic response in a sleep study whose goal was to link photoplethysmography-measured vascular response characteristics to BOLD signal changes (Özbay et al., 2018). As more researchers look into these complex issues, links between vascular or other characteristics of the BOLD response might be found that also have direct links to the sympathetic nervous system processes and brain function examined in this study. For instance, studies that employ near-infrared or photoplethysmography methods could shed light on the exact nature of complex relationships between neuronal activity in these regions and SCR-linked BOLD signal response. These are likely to result in better-informed interpretations about any type of possible homeostatic mechanism that might explain the bimodal hemodynamic response before and during nonspecific SCRs.

A strength of this study is the Finite Impulse Response hemodynamic modeling method, which can detect and model activity of any response shape. Another is our conservative approach to identifying which regions are engaged. Our approach to summarizing this evidence considered not only whether change occurred, but also what it looked like and if it convincingly engaged brain regions known to be discrete cortical regions (Blumensath et al., 2013). The latter is important because the anatomical precision of the Human Connectome Project parcellation atlas for cortex does not similarly extend to our analysis of subcortical structures (i.e., it was necessary to more closely inspect subcortical structure activation patterns). Care was taken to remove many BOLD signal artifacts in the fMRI timeseries using ICA-FIX, which lends confidence that our findings were likely unrelated to commonly-encountered fMRI confounds (e.g., respiration, cardiac pulsality, etc. (Birn et al., 2006; Glover et al., 2000; Wise et al., 2004). ICA-FIX also at least partially addresses concerns over global signal effects mainly arising when resting state fMRI data are examined using functional connectivity cross-correlation methods (Glasser et al., 2018). Moreover, the Supplement presents re-analysis results that show the focal points of our findings are largely intact, albeit with smaller effect sizes, even when mean gray matter signal is regressed out of our data. But because ICA-FIX largely preserves signal that corresponds to well-described cortical intrinsic connectivity brain networks, it is not yet known if the relative absence of subcortical results in this study might be because those regions had more noise, as is typical when measurements use a multi-channel MRI headcoil. Future studies might attempt to replicate our findings either by directly controlling potential sources of signal artifacts, or perhaps by utilizing the Human Connectome Project’s temporal ICA denoising pipeline (Glasser et al., 2018). An additional limitation is the spatial resolution of fMRI typically limits the ability to detect signal change in small individual thalamic and brainstem nuclei. Further, the decision about which evidence was most convincing was arbitrary. Some of the regions described in the Supplement as having relatively few voxels with signal change ultimately may prove to be meaningfully linked to non-specific SCRs. For instance, there was some limited evidence for activity in caudate head prior to nonspecific SCRs, similar to what Fan et al., 2012 reported (Fan et al., 2012). Another study limitation is fMRI itself cannot directly measure neuronal activity that instigates SCR activity. We are limited to inferring the likely timing of neuronal events that might generate non-specific SCRs using the relatively coarse temporal resolution of fMRI. This limitation is mitigated somewhat by the fairly rapid TR (475 msec) MRI sequence that we used. But in general, this limitation prevents offering any firm conclusions about both neuronal firing or causality among these brain regions. Such observations ultimately could be useful in disentangling exactly how these brain regions instigate non-specific SCRs or support other processes linked to its possible refractory period. However, future studies will be able to build on our findings to better understand the relationships among non-specific SCR-related brain regions using fMRI techniques like Dynamic Causal Modeling (Friston et al., 2003) or by utilizing more invasive methods such as local field potential recording in animals(Scholvinck et al., 2010)(Scholvinck et al., 2010)(Scholvinck et al., 2010)(Scholvinck et al., 2010)(Scholvinck et al., 2010). Because we focused on relatively strong SCRs (i.e., > .1 microsiemens), future studies might find lesser SCRs have different neural correlates or response profiles. One final notable issue is this study was conducted in adolescents. Since their brains are still developing, it remains possible that age-related differences might explain our lack of significant cerebellum, caudate, and basal ganglia activity association to non-specific SCRs observed in at least one prior study (Fan et al., 2012). However, arguing against this was the lack of significant correlations between participant age and either non-specific SCR frequency, amplitude, or brain activity.

In conclusion, this study characterized the neural correlates of non-specific SCRs as a widespread, likely exclusively cortical network of regions engaged in a complex, biphasic fashion that implies some sort of local homeostatic process. This cortical response profile overlaps more with the neural correlates of task-induced SCRs from past studies than they differ, implying a common system underlying the generation and processing of electrodermal responses. Given that this study utilized less-frequently used fMRI methodology purposefully chosen to provide an inclusive, broad examination of all types of SCR-linked brain activity, these findings should be viewed as preliminary until replicated. The findings raise several questions that should be addressed in future neuroimaging research of SCRs. Foremost among these is whether a direct comparison of the shape of hemodynamic response to stimulus-elicited SCRs to that from nonspecific SCRs during resting state would reveal regions with similar biphasic response as found here. It would be equally interesting to learn if those stimulus-elicited responses were biphasic as seen here, or if there might be a mix of simple and complex responses in different brain regions. The brain activity profiles associated with non-specific SCRs found here provide a new target to seek in functional neuroimaging studies of SCRs elicited by external stimulation. Direct comparison of nonspecific and task-elicited SCR fMRI data also might help isolate which regions’ cortical activity is related to the causes of SCRs or their downstream effects. This study also raises new questions about the mechanisms, purpose, and targets of the homeostatic function implied by this biphasic nature of brain activity. Finally, the fact that brain activity changes during SCRs cannot be explained by a canonical hemodynamic response model also should prompt investigators to carefully reconsider their assumptions about how best to model SCR-linked brain function, both nonspecific SCRs as well as those elicited by different types of stimuli.