Skin Conductance Responses and Neural Activations During Fear Conditioning and Extinction Recall Across Anxiety Disorders

Key Points

Question  Do individuals with anxiety disorders exhibit dysregulated psychophysiological and neuroimaging profiles during fear conditioning and extinction recall?

Findings  In this cross-sectional, case-control study, 61 individuals with anxiety disorders activated the ventromedial prefrontal cortex less during fear conditioning and extinction recall compared with 21 healthy controls. This hypoactivation was more pronounced in those diagnosed as having multiple anxiety disorders than in those having only one anxiety disorder.

Meaning  Gaining a better understanding of the structures of the fear circuitry that are dysregulated in anxious individuals might help in guiding better treatments targeting the neurobiological features of the disorder.

Importance  The fear conditioning and extinction neurocircuitry has been extensively studied in healthy and clinical populations, with a particular focus on posttraumatic stress disorder. Despite significant overlap of symptoms between posttraumatic stress disorder and anxiety disorders, the latter has received less attention. Given that dysregulated fear levels characterize anxiety disorders, examining the neural correlates of fear and extinction learning may shed light on the pathogenesis of underlying anxiety disorders.

Objectives  To investigate the psychophysiological and neural correlates of fear conditioning and extinction recall in anxiety disorders and to document how these features differ as a function of multiple diagnoses or anxiety severity.

Design, Setting, and Participants  This investigation was a cross-sectional, case-control, functional magnetic resonance imaging study at an academic medical center. Participants were healthy controls and individuals with at least 1 of the following anxiety disorders: generalized anxiety disorder, social anxiety disorder, specific phobia, and panic disorder. The study dates were between March 2013 and May 2015.

Exposures  Two-day fear conditioning and extinction paradigm.

Main Outcomes and Measures  Skin conductance responses, blood oxygenation level–dependent responses, trait anxiety scores from the State Trait Anxiety Inventory–Trait Form, and functional connectivity.

Results  This study included 21 healthy controls (10 women) and 61 individuals with anxiety disorders (36 women). P values reported for the neuroimaging results are all familywise error corrected. Skin conductance responses during extinction recall did not differ between individuals with anxiety disorders and healthy controls (ηp² = 0.001, P = .79), where ηp² is partial eta squared. The anxiety group had lower activation of the ventromedial prefrontal cortex (vmPFC) during extinction recall (ηp² = 0.178, P = .02). A similar hypoactive pattern was found during early conditioning (ηp² = 0.106, P = .009). The vmPFC hypoactivation was associated with anxiety symptom severity (r = −0.420, P = .01 for conditioning and r = −0.464, P = .004 for extinction recall) and the number of co-occuring anxiety disorders diagnosed (ηp² = 0.137, P = .009 for conditioning and ηp² = 0.227, P = .004 for extinction recall). Psychophysiological interaction analyses revealed that the fear network connectivity differed between healthy controls and the anxiety group during fear learning (ηp² range between 0.088 and 0.176 and P range between 0.02 and 0.003) and extinction recall (ηp² range between 0.111 and 0.235 and P range between 0.02 and 0.002).

Conclusions and Relevance  Despite no skin conductance response group differences during extinction recall, brain activation patterns between anxious and healthy individuals differed. These findings encourage future studies to examine the conditions longitudinally and in the context of treatment trials to improve and guide therapeutics via advanced neurobiological understanding of each disorder.

Fear conditioning and extinction paradigms are relevant for studying anxiety disorders. It has been proposed that pathological anxiety could emerge from dysregulated patterns of fear learning and that maintenance of anxiety-related symptoms could be explained by extinction deficits.^1,2 Until now, this paradigm has been mostly tested in populations with posttraumatic stress disorder (PTSD).^1-8 At the psychophysiological level, they exhibit generally normal conditioning and extinction learning but impaired extinction recall.^9-13 During extinction recall, individuals suffering from PTSD exhibit lower activations in brain regions promoting safety signal processing, such as the ventromedial prefrontal cortex (vmPFC) and hippocampus, and they exhibit greater activations in regions promoting fear signal detection, such as the amygdala and dorsal anterior cingulate cortex (dACC).^10,12-14

Until the DSM-5 release, PTSD was considered an anxiety disorder; it is now classified as a trauma and stress–related disorder. However, it remains unclear whether the physiological deficiencies in PTSD are also observed in conditions currently categorized as anxiety disorders. Although PTSD and anxiety disorders present overlapping features—notably, dysregulated fear levels—DSM-5 anxiety disorders have been less studied in the context of conditioning and extinction paradigms. Results of a meta-analysis¹⁵ suggested that anxious individuals exhibit higher fear in response to safety cues during conditioning and higher fear in response to danger cues during extinction. The meta-analysis included study samples with PTSD and dealt with complex clinical portraits (ie, comorbidity from the same or different diagnostic categories). Finally, it remains to be studied whether the magnitude of the pathophysiological deficit differs based on the number of anxiety disorders, without the confounds of other comorbidities.

This study aimed to elucidate some of these issues in individuals having anxiety disorders without other comorbidities. Using psychophysiological and neuroimaging tools, this study investigated how anxiety disorders influence the circuitry of fear conditioning and extinction recall. We then examined whether the presence of multiple anxiety disorders influences the circuitry relative to a single disorder. We hypothesized that, relative to healthy controls, individuals with anxiety disorders (1) would have lower differential fear conditioning and deficient extinction recall in terms of skin conductance response (SCR) and (2) would exhibit dysregulated activation patterns in the fear circuitry nodes during fear learning and extinction recall. We also hypothesized that there would be more pronounced dysregulations in those with multiple anxiety disorders. We conclude with a mechanistic focus investigating how activations during fear memory encoding relate to activations during recall.

We recruited 61 individuals meeting criteria for at least 1 of the following anxiety disorders: generalized anxiety disorder, social anxiety disorder, specific phobia, and panic disorder (45 had 1 disorder and 16 had ≥2 disorders), with no other current comorbidities. We included previous data from 21 healthy controls¹⁶ who underwent identical experimental procedures with use of the same scanner. For exclusion criteria and a description of the study sample, see the eAppendix in the Supplement.

Participants provided written informed consent in accord with the requirements of the Partners Healthcare Institutional Review Board, who approved the study. Participants underwent a Structured Clinical Interview for DSM-IV with one of us (N.B.L.) to determine the diagnoses and eligibility. They filled out the State Trait Anxiety Inventory–Trait Form (STAI-T¹⁷), examining self-reported anxiety levels. Participants underwent a 2-day fear conditioning and extinction paradigm^{9,10,13,14,16,18,19} in a functional magnetic resonance imaging scanner that included conditioned stimuli (CS) (eAppendix in the Supplement). On day 1, fear conditioning occurred, during which 2 cues (CS+) were reinforced and 1 cue (CS−) was not. This conditioning was followed by extinction learning, where 1 CS+ and the CS− were presented. The next day, extinction recall was tested, where the extinguished CS+ (CS + E), the nonextinguished CS+ (CS + NE), and the CS− were presented (details are provided in the eAppendix in the Supplement).

Skin conductance response and imaging data were computed using the previously used methods.^{9,10,13,16,18,19} Details are provided in the eAppendix in the Supplement.

For conditioning, equal numbers of trials are used for CS+ and CS−. However, 2 CS+s and only one CS− are used, suggesting that there might be more habituation to CS− (16 trials of 1 cue) relative to CS+ (16 trials based on 2 cues). We performed analyses to assess habituation effects between groups (eAppendix in the Supplement).

For conditioning, a stimulus (CS+ vs CS−) × time (early vs late) × group (healthy vs anxiety) analysis of covariance (ANCOVA) was performed on SCR. For the imaging analysis, between-group differences were investigated for early, late, and all conditioning. Additional analyses examined whether both groups had similar fear extinction levels (eAppendix in the Supplement). For extinction recall, a stimulus (CS + E vs CS + NE) × group (healthy vs anxiety) ANCOVA was performed on SCR. Similar analyses were performed for the imaging data.

To investigate differences between single vs multiple anxiety disorders, analyses were repeated for the anxiety cohort alone. For these analyses, group (single vs multiple) was the between-group factor.

To assess associations between anxiety severity and the fear network within the anxiety group, a voxelwise analysis was performed with STAI-T scores as a regressor for early conditioning (CS+ vs CS−), late conditioning (CS+ vs CS−), and extinction recall (CS + E vs CS + NE). Beta weights were extracted from the peak voxel to generate a correlation coefficient.

Psychophysiological interaction (PPI) analyses were performed during early conditioning and extinction recall. For these analyses, vmPFC was used as the seed.

Imaging analyses were performed with an initial threshold of P < .005 and 10 contiguous voxels. Activations detected with that threshold within the fear circuitry (amygdala, hippocampus, insular cortex, ACC, and vmPFC) were then tested for small-volume correction.

This study included 21 healthy controls, with a mean (SD) age of 25.8 (4.8) years, 47.6% (10 of 21) of whom were female, and 61 individuals with anxiety disorders, with a mean (SD) age of 30.4 (11.5) years, 59.0% (36 of 61) of whom were female. Healthy controls were younger (t₇₇ = 2.559, P = .01) and more educated (t₈₀ = 1.922, P = .06) than the anxiety group. Both groups had similar shock levels (t₇₃ = 0.346, P = .18) and sex distributions (χ₁ = 0.824, P = .36). Analyses comparing these groups were run with (main text) and without (eAppendix in the Supplement) age and educational level as covariates. Analyses comparing the single disorder group with the multiple disorders group did not include covariates because the groups did not statistically differ on any demographics.

During conditioning (Figure 1A), the SCR ANCOVA yielded a marginal effect of stimulus (F_1,77 = 3.009, P = .09, ηp² = 0.038), an effect of group (F_1,77 = 11.126, P = .001, ηp² = 0.126), and a time × group interaction (F_1,77 = 5.110, P = .03, ηp² = 0.062), where ηp² is partial eta squared. Compared with the anxiety group, healthy controls exhibited greater vmPFC activation during early conditioning (Montreal Neurological Institute x, y, z coordinates [hereafter MNI] −8, 50, −28; cluster size of 12; t₇₂ = 2.99; P = .009 familywise error; ηp² = 0.106) and greater hippocampal activation during late conditioning (MNI 32, −30, −6; cluster size of 10; t₇₂ = 2.99; P = .007 familywise error; ηp² = 0.113). Both groups underwent successful extinction learning (eAppendix in the Supplement).

For extinction recall, the SCR ANCOVA did not yield any significant results, (F_1,67≤1.109 for all, P ≥ .30 for all, ηp²≤0.016 for all). Relative to healthy controls, the anxiety group exhibited less activation in the rostral ACC (rACC) (MNI −12, 44, 8; cluster size of 26; t₆₃ = 3.21; P = .007 familywise error; ηp² = 0.117), the vmPFC (MNI −14, 46, −18; cluster size of 169; t₆₃ = 3.41; P = .02 familywise error; ηp² = 0.178), and the insular cortex (MNI −36, 10, −12; cluster size of 18; t₆₃ = 3.44; P = .003 familywise error; ηp² = 0.136).

During conditioning (Figure 2A), the SCR ANOVA yielded a main effect of stimulus (F_1,58 = 46.005, P < .001, ηp² = 0.442), a main effect of time (F_1,58 = 36.010, P < .001, ηp² = 0.383), a main effect of group (F_1,58 = 5.671, P = .02, ηp² = 0.089), a marginal time × group interaction (F_1,58 = 3.580, P = .06, ηp² = 0.204), and a time × stimulus interaction (F_1,58 = 14.876, P < .001, ηp² = 0.058). During early conditioning, the multiple disorders group had greater insular cortex activation (MNI −36, 2, 4; cluster size of 67; t₅₆ = 4.21; P = .001 familywise error; ηp² = 0.179) but less vmPFC activation (MNI 8, 48, −26; cluster size of 33; t₅₆ = 3.19; P = .009 familywise error; ηp² = 0.137) relative to the single disorder group. For late conditioning, the multiple disorders group activated the insular cortex (MNI −34, 10, 4; cluster size of 42; t₅₆ = 3.28; P = .008 familywise error; ηp² = 0.162) more than the single disorder group. When examining conditioning across all trials, the multiple disorders group had greater activation in the amygdala (MNI 14, −2, −18; cluster size of 86; t₅₆ = 3.99; P = .002 familywise error; ηp² = 0.231), the insular cortex (MNI −36, 0, 4; cluster size of 219; t₅₆ = 3.91; P = .006 familywise error; ηp² = 0.228), and the dACC (MNI 0, 28,16; cluster size of 59; t₅₆ = 3.39; P = .008 familywise error; ηp² = 0.151).

During extinction recall, the SCR ANOVA yielded no significant results (F_1,48≤0.281 for all, P ≥ .59 for all; ηp²≤0.006 for all). The functional magnetic resonance imaging analysis revealed less vmPFC (MNI 0, 38, −30; cluster size of 32; t₄₇ = 3.48; P = .004 familywise error; ηp² = 0.227) and less subgenual ACC (sgACC) (MNI 2, 18, −14; cluster size of 23; t₄₇ = 3.11; P = .009 familywise error; ηp² = 0.206) activations in the multiple disorders group compared with the single disorder group.

Given the different weights of the specific phobics in the single disorder group (n = 17) compared with the multiple disorders group (only present as comorbidity and not as the main disorder), we conducted analyses to rule out the possibility that the effects obtained were driven only by the specific phobics. Details are provided in the eAppendix in the Supplement.

For early conditioning (Figure 3A), STAI-T scores were positively correlated with sgACC activation (MNI −10, 16, −14; cluster size of 25; t₄₇ = 3.21; P = .008 familywise error; r = 0.424) and negatively correlated with insular cortex (MNI −34, −28, 18; cluster size of 31; t₄₇ = 3.16; P = .01 familywise error; r = −0.414) and vmPFC (MNI 14, 54, −18; cluster size of 38; t₄₇ = 3.17; P = .01 familywise error; r = −0.420) activations. During late conditioning, STAI-T scores were positively associated with rACC (MNI −16, 40, 16; cluster size of 18; t₄₇ = 3.05; P = .01 familywise error; r = 0.407) and sgACC (MNI 16, 32, −20; cluster size of 231; t₄₇ = 4.25; P = .001 familywise error; r = 0.562) activations. During extinction recall (Figure 3B), STAI-T scores were negatively associated with sgACC (MNI 16, 34, −18; cluster size of 34; t₄₁ = 3.73; P = .003 familywise error; r = −0.503) and vmPFC (MNI −10, 60, −20; cluster size of 14; t₄₁ = 3.35; P = .004 familywise error; r = −0.464) activations but were positively associated with dACC (MNI 2, 22, 20; cluster size of 23; t₄₁ = 3.15; P = .009 familywise error; r = 0.442) and insular cortex (MNI 44, 12, 4; cluster size of 45; t₄₁ = 2.95; P = .03 familywise error; r = 0.419) activations.

During early conditioning, the vmPFC seed had greater functional connectivity in healthy controls with the sgACC (MNI −14, 34, −12; cluster size of 10; t₇₁ = 3.40; P = .003 familywise error; ηp² = 0.095) but greater connectivity in the anxiety group with the hippocampus (MNI −26, −24, −12; cluster size of 17; t₇₁ = 2.81; P = .02 familywise error; ηp² = 0.101), the amygdala (MNI 26, −4, −26; cluster size of 384; t₇₁ = 3.71; P = .02 familywise error; ηp² = 0.176), and the insular cortex (MNI −38, 6, −12; cluster size of 13; t₇₁ = 2.80; P = .01 familywise error; ηp² = 0.088). These results are shown in Figure 4A.

During extinction recall, the vmPFC seed showed greater functional connectivity in healthy controls with the sgACC (MNI −8, 16, −24; cluster size of 10; t₆₃ = 2.95; P = .007 familywise error; ηp² = 0.111) but greater connectivity in the anxiety group with the amygdala (MNI −24, −10, −24; cluster size of 305; t₆₃ = 4.26; P = .002 familywise error; ηp² = 0.235), the insular cortex (MNI 40, 10, −22; cluster size of 169; t₆₃ = 3.39; P = .02 familywise error; ηp² = 0.158), and the vmPFC (MNI 6, 38, −26; cluster size of 37; t₆₃ = 3.37; P = .006 familywise error; ηp² = 0.143). These results are shown in Figure 4B.

As an exploratory analysis, we examined whether dysregulated activation patterns observed during conditioning could account for or be related to the activation pattern observed in extinction recall. Hippocampus activation during conditioning was not associated with vmPFC activation during recall (r = 0.098, P = .44). There was a suggestion of an association between vmPFC activation during conditioning and vmPFC activation during recall (r = 0.232, P = .06). We examined the pattern within each group and found that it was only present in healthy controls (r = 0.643, P = .004) and not in anxious individuals (r = −0.151, P = .31) (Figure 5A). We tested if similar patterns of associations were present for the SCR data. An exploratory analysis examined correlations between SCR during recall (SCR to the first 4 trials of CS + E minus SCR to the first 4 trials of CS + NE) and the connectivity values between the vmPFC and the following regions: sgACC, amygdala, insular cortex, and vmPFC. The connectivity value between vmPFC and sgACC was negatively associated with SCR (r = −0.281, P = .03), whereas the connectivity value between vmPFC and amygdala was positively associated with SCR (r = 0.271, P = .04) (Figure 5B). The vmPFC-vmPFC (r = 0.101, P = .44) and vmPFC-insula (r = −0.001, P = .99) connectivity values were not associated with SCR.

We recruited individuals diagnosed as having 1 or more anxiety disorders without other comorbid disorders. This sample allowed us to investigate the association between anxiety disorders and fear circuitry across different phases of a conditioning and extinction procedure and to examine whether the number of anxiety disorders differentially influences the circuitry.

During conditioning, SCR was blunted in the anxiety group relative to healthy controls. However, both groups differentiated between the CS+ and the CS−. This blunted pattern seems to be driven by individuals with a single disorder. The literature has suggested larger responses to the CS− in anxious individuals, which could result in lower differential acquisition.¹⁵ Based on our results, the number of diagnoses is an important factor to consider that could be used as an index of clinical severity.

In terms of imaging, individuals with multiple disorders activated more fear encoding and expression regions (amygdala, insular cortex, and dACC) during conditioning. This finding is consistent with studies^20-28 that have shown hyperactivation of fear-promoting regions during emotional tasks. During early conditioning, the vmPFC was less activated in the anxiety group compared with healthy controls. The number of diagnoses modulated that vmPFC hypoactivation such that individuals having multiple anxiety disorders showed reduced vmPFC activation compared with those having a single disorder. Moreover, vmPFC activation during early conditioning showed a negative correlation with STAI-T scores. These results suggest vmPFC hypoactivation in anxious individuals, an effect that is more pronounced in more severe cases (either greater symptoms or more disorders).

Previous studies^29-33 performed in social anxiety disorder and generalized anxiety disorder have also reported lower activation in the medial PFC during emotional tasks. Our PPI analyses revealed that this region was more functionally coupled with the sgACC in healthy controls but showed more functional coherence with the hippocampus, amygdala, and insular cortex in the anxiety group. This finding suggests that, for the anxious group, the vmPFC region that showed lower activation during early conditioning was also more coupled with regions known to support fear encoding and processing, which are typically more activated in anxious individuals.

Contrary to our hypothesis, during extinction recall, no SCR deficits were found between the healthy controls and the anxiety group. This result is in contrast with various psychopathological conditions, such as PTSD, obsessive-compulsive disorder, and schizophrenia, in which deficits were noted at the SCR level.^{9,10,12,13,34-36} This finding is an important psychophysiological distinction between PTSD and the anxiety disorders tested in our study, and it is also contrary to our hypothesis. Despite no group differences for SCR during extinction recall, brain activation patterns differed between the groups. In fact, the healthy controls activated the vmPFC, rACC, and insular cortex more relative to the anxious individuals. This result is consistent with investigations that have shown dysregulated rACC and vmPFC activation patterns in anxious individuals using various emotion regulation tasks.³⁷ Focusing on the anxiety group, results showed that vmPFC activation was reduced in those with multiple disorders. The regression analyses also revealed a negative correlation between the vmPFC activation and the trait anxiety levels. When looking at PPI analyses with the vmPFC as the seed, we again observed higher functional coherence with the sgACC in healthy controls. On the other hand, the same seed showed more connectivity with the amygdala and insular cortex in the anxiety group, as was the case in the PPI analyses conducted during conditioning, as well as with a vmPFC area. Similar to conditioning, the hypoactive vmPFC region in the anxiety group showed more functional coherence during extinction recall in anxious individuals with fear-promoting regions, which tend to be hyperactive in that same sample.

Similar patterns emerged during early conditioning and extinction recall with regard to vmPFC activation and its modulation by the number of diagnoses and trait anxiety levels, as well as with its functional coherence with the rest of the network. Activation of the vmPFC during early conditioning was positively associated with vmPFC activation during extinction recall but only in healthy controls. These exploratory analyses emphasize the importance of assessing how fear is initially encoded, which seems to influence how the safety memory will be retrieved later. In fact, deficits that have been reported in terms of activation patterns during extinction recall in different disorders might potentially be traced back to dysregulated activation patterns during the initial fear memory formation. In support of this hypothesis, Livneh and Paz³⁸ showed that the synchronization of amygdala and dACC activity during fear encoding predicts higher resistance to extinction.

As an exploratory analysis, we next examined whether brain activation patterns were associated with fear expression during extinction recall. Although both groups had similar SCR during recall, this measure carries great variability in individuals. The analysis revealed that greater connectivity between the vmPFC and the sgACC, which was more coupled in healthy controls, was associated with better extinction recall. In contrast, the connectivity value between the vmPFC and the amygdala, which was higher in the anxiety group, was associated with worse extinction recall. This finding is in line with animal investigations showing that specific patterns of medial PFC–amygdala correlate with fear expression.³⁹ These exploratory models highlight the need for studies to further examine such questions with cross-validation techniques in larger sample sizes.

Some limitations of this study should be highlighted. First, the anxiety group was older and less educated than the healthy controls. We have covaried for these variables throughout our analyses. We have also rerun all analyses without covariates, and most of our findings remained unchanged. Furthermore, the covariates were not significantly associated with any of our main outcomes (eAppendix in the Supplement). Second, the specific phobics were more represented in the single disorder group, which could suggest that the comparisons made for the number of diagnoses reflect a difference between specific phobia and the other disorders. Our supplemental analyses ruled out this effect by showing that the single disorder group with specific phobia was comparable to the single disorder group without specific phobia and that the differences between the single disorder group and the multiple disorders group remained when excluding individuals with only specific phobia. Third, there are sex differences pertaining to the prevalence of anxiety disorders,^40-42 and sex hormones modulate extinction learning.^43-46 We did not assay gonadal hormones, making it impossible to measure their influence. Fourth, we draw some parallels between our findings and those from other clinical study samples, notably PTSD. These inferences are based on patterns emerging from the data, and no direct comparisons were made between these 2 groups statistically.

Our results reveal no SCR deficits for differential acquisition and extinction recall. However, the imaging data suggest that the fear circuitry is dysregulated in individuals with anxiety disorders and that some differences are modulated by the number of disorders or the self-reported anxiety symptoms. The PPI analyses highlighted the importance of investigating the whole fear network and the association between its main nodes because an imbalance in the activation of fear-promoting regions and extinction-promoting regions at different stages throughout the paradigm may synergistically act in conveying a greater vulnerability to anxiety disorders. This study allowed identification of patterns applicable to the DSM-5 category of anxiety disorders that excludes trauma-related and stress-related conditions. Although we used a categorical approach, it would be informative to test similar questions using a dimensional approach. The correlations between the anxiety symptoms and brain activation patterns highlight aspects of the research domain criteria method and the importance of examining more extensively these questions from this approach.⁴⁷ From a clinical standpoint, our results provide a rationale for future work in further classifying each anxiety disorder because not all disorders may be equivalent. Understanding the similarities and differences between anxiety disorders may enable neurobiologically driven treatment development and selection tailored to a patient’s diagnosis, comorbidities, and level of anxiety severity.

Accepted for Publication: February 11, 2017.

Corresponding Author: Mohammed R. Milad, PhD, Department of Psychiatry, Massachusetts General Hospital, 149 13th St, Office 2.508, Charlestown, MA 02129 (milad@nmr.mgh.harvard.edu).

Published Online: April 12, 2017. doi:10.1001/jamapsychiatry.2017.0329

Author Contributions: Drs Marin and Milad had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Zsido, Killgore, Rauch, Simon, Milad.

Acquisition, analysis, or interpretation of data: Marin, Zsido, Song, Lasko, Killgore, Simon, Milad.

Drafting of the manuscript: Marin, Zsido, Milad.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Marin, Zsido, Song, Milad.

Obtained funding: Milad.

Administrative, technical, or material support: Marin, Song, Lasko, Killgore, Simon, Milad.

Study supervision: Marin, Killgore, Simon, Milad.

Conflict of Interest Disclosures: None reported.

Funding/Support: This project was supported by grant R01 MH097964 from the National Institute of Mental Health to Dr Milad and by a Banting Postdoctoral Fellowship to Dr Marin.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank the following research assistants who have been involved in the recruitment or data collection for the present project: Allison Campbell, BA, Laura E. Fischer, MSc, Aaron J. Landau, BA, Blake L. Rosenbaum, BA, and Peter Rosencrans, BA (all at Massachusetts General Hospital), as well as Lily A. Preer, BA (at McLean Hospital). No compensation was received.