The sample consisted of 12 healthy women and 12 healthy men who were recruited from the local community via advertisements for two previously published neuroimaging studies 2833 and reanalyzed to test our hypotheses. One of the original studies was not initially designed to investigate sex differences in fear extinction, therefore women were in different phases of the menstrual cycle (4 follicular phase, 4 late luteal phase, 4 unknown). Table 1 shows that subjects are right-handed and primarily Caucasian with a relatively high education level (at least some college, on average). Men were older and had more years of education. We controlled for these differences when comparing males and females (see below).

Subjects were excluded if they had neurologic, endocrinologic, or other medical conditions affecting central nervous system function. Subjects were also screened for Axis-I psychiatric disorders, including substance-use disorders, using the Structured Clinical Interview for DSM-IV 34. There were no significant sex differences in anxiety measures (see Table 1). No participant was using psychoactive or other potentially confounding drugs or medications, and women had abstained from oral contraceptives or hormone replacement for at least three months. After a complete description of the procedures, written informed consent was obtained from all subjects in accordance with the requirements of the Partners Healthcare Human Research Committee.

The two-day fear conditioning and extinction procedures have been described elsewhere 35 (see Figure 1). Briefly, two digital photographs of rooms (an office and a library) were the visual contexts in which a lamp was switched from the off position (no color) to one of three colored lights (red, yellow, blue), constituting the conditioned stimuli (CSs). All images were displayed on a computer monitor located approximately two feet behind the subject and viewed on a mirror while the subject was in a 3 T MRI scanner. The unconditioned stimulus (US) was a 500 ms electric shock delivered through electrodes attached to the second and third fingers of the right hand. The subjects had previously selected a shock intensity they found “highly annoying but not painful” 3637. The electrodes were attached to the fingers during each phase of the study, but the US was presented only during conditioning. On day 1, subjects underwent the habituation phase in which the conditioning context and the extinction context were displayed four times while each CS was presented two to three times. In the conditioning phase, the (to later be) extinguished conditioned stimulus (CS + E) and the (to later not be extinguished) conditioned stimulus (unextinguished CS + or CS + U) were each presented eight times with 62.5% partial reinforcement (five shocks each), while the conditioned stimulus that was never followed by a shock (CS-) was intermingled and presented 16 times. All CSs were presented in the same context. The selection of CS + and CS- colors was pseudorandom and counterbalanced across subjects. After conditioning, subjects were briefly interviewed to ensure the CS-US pattern was observed. All subjects were aware of the CS-US contingency. The extinction phase immediately followed in which the CS + E and the CS- were each shown 16 times in a new, “safe” context. On day 2, the recall phase was in the extinction context and included the presentation of the CS + U along with the CS + E and CS-.

During each trial, the context images were presented for nine seconds: three seconds with the light off immediately followed by six seconds in combination with the CS. The mean inter-trial interval was 15 seconds. Skin conductance responses (SCR) were calculated by subtracting the maximum response during cue presentation from the average response of the two seconds immediately preceding context onset. The SCR values were then square-root transformed to reduce heteroscedasticity. Skin conductance levels 38 were measured during the five seconds preceding the onset of each habituation session trial and then averaged across eight trials to yield a baseline SCL.

To evaluate the amount of fear during the different phases of the experiment, SCR to the CS + and SCR to the CS- were compared as follows: during conditioning, SCR to the CS + was compared to CS-; during extinction, SCR to the CS + E was compared to CS-; and during extinction recall, SCR to the CS + E was compared to CS + U. All data are reported as means ± the standard error of the mean (S.E.). A repeated-measures analysis of variance 39 was used to analyze data across experimental phases. A Student’s t-test was used when appropriate.

Image acquisition parameters were identical to previous reports 284041. A Trio 3.0-Tesla whole-body, high-speed imaging device with a 12-channel gradient head coil was used (Siemens Medical Systems, Iselin, New Jersey). An automated scout image was obtained and shimming procedures were performed followed by high-resolution, three-dimensional magnetization prepared rapid gradient echo sequences (repetition time [TR]/echo time [TE]/flip angle = 7.25 ms/3 ms/7^o; 1 mm X 1 mm in plane X 1.3 mm), which were collected for spatial normalization and positioning the subsequent scans. Registration of individual functional scans was based on T1 (TR/TE/flip angle = 8 sec/39 msec/90^o) and T2 (TR/TE/flip angle = 10 sec/48 msec/120^o) sequences. fMRI images were acquired with gradient–echo T2*-weighted sequences (TR/TE/flip angle = 3 sec/30 msec/90^o). The T1, T2, and gradient-echo functional images were all collected in the same plane (45 coronal oblique slices parallel to the anterior-posterior commissure line, tilted 30^o anterior) with the same slice thickness (3 mm X 3 mm X 3 mm) except for the T1 (1 mm X 1 mm X 1 mm).

Each subject’s functional time series was first examined for global-signal artifacts (e.g., artifact caused by head movement) using the Artifact Detection Tool (ART) software package (http://web.mit.edu/swg/art/art.pdf) in order to control for this artifact during first-level statistical analyses. “Outlier” volumes were flagged if the average global-signal intensity of the image (i.e., average signal intensity across all voxels) was more than 3.0 standard deviations from the overall mean for all images (ART z-threshold = 3.0), the scan-to-scan translation movement was more than 0.6 mm or the scan-to-scan rotation movement was more than 0.004 radians. Once flagged, outlier volumes were modeled as regressors of no interest in the first-level general linear model (GLM) following standard fMRI pre-processing procedures using SPM8 software (http://www.fil.ion.ucl.ac.uk/spm/software/spm8).

For this pre-processing procedure, images from each functional run were first slice-timing corrected and realigned to the first image of the run. This procedure generated realignment parameters for each run that were also used as covariates of no interest in the first-level GLM, as well as a mean image for each run. These mean images and the MPRAGE were then co-registered to the mean image of the first imaging run to facilitate later transformation of the series into MNI space. Next, the MPRAGE was segmented and spatially normalized to the T1 MNI305 template included in SPM8 (Montreal Neurological Institute, MNI). The resulting spatial-transformation parameters were applied to the EPI time series to transform them to the common anatomical coordinate space (MNI305), and voxels were re-sliced to a dimension of 2 mm isotropic. Finally, to mitigate the effects of residual spatial-transformation noise, the normalized functional images were smoothed using an 8 mm full-width-at-half-maximum Gaussian kernel.

For the first level GLM, we used an epoch model and modeled context and CS as three-second and six-second events, respectively. During conditioning, we modeled the US as a 0.5-second event. Experimental regressors were convolved with the SPM canonical hemodynamic response function (HRF), but the regressors of no interest (i.e. outliers and motion parameters) were not convolved. The time series was subjected to a 128-second high-pass filter to correct for low-frequency signal drift.

First-level statistical parametric maps were calculated using the general linear model for the contrast of interest across the time window 23. The Stimulus Factor contrasted all 16 CS + trials vs. all 16 CS- trials in the conditioning phase (CS+ > CS-), the last 4 CS + E trials vs. the last 4 CS- trials in the extinction learning phase (CS + E > CS-), and the first 4 CS + E versus the first 4 CS + U trials in the extinction recall phase (CS + E > CS + U). The first four trials were used for extinction recall because we wanted to minimize any confound introduced by additional extinction learning during this phase, and electrophysiological data from rodents indicates that the vmPFC only signals extinction recall during the beginning of extinction recall 42.

First-level SPM contrasts were then grouped during second-level independent-groups t-tests that compared men and women using age and years of education as covariates. For this analysis, we used an uncorrected voxel-level statistical threshold of p < 0.005, and we report only peak-voxels from clusters of activation within our primary anatomical regions of interest (ROIs; vmPFC, insula, dACC, amygdala, and hippocampus). This statistical threshold was used given our focus on specific anatomical ROIs based on prior studies’ findings and a priori hypotheses about where sex differences in brain activation should be found. Following this analysis focusing on previously identified fear conditioning circuitry regions, we conducted a stringent functional-ROI analysis of sex differences in stress-response circuitry regions during fear conditioning.

Given our hypothesis that the brain’s stress-response circuitry and fear conditioning circuitry partly overlap, sex differences found in previous fMRI studies of the stress-response circuitry should also be apparent during fear conditioning. We identified a priori anatomical ROIs using Goldstein and colleagues findings 426 of functional sex differences in the following regions during an emotional arousal task that activates the stress-response circuitry: orbitofrontal cortex (OFC), anterior cingulate cortex (ACC), the periaqueductal gray brain stem area 43, hippocampus (HIPP), anterior hypothalamus (HYP), and amygdala (AMG). However, shared functional sex differences within a broad anatomical area (particularly OFC and ACC) do not necessarily suggest any similar function for these regions in both paradigms. Therefore, we further limited our search within these anatomical regions to functional ROIs (spheres) around Goldstein’s (2010) exact coordinates (Table 2). To do this, we created either 8 mm (cortical rois) or 4 mm (subcortical ROIs) spheres using the WFU Pickatlas tool for SPM8. We then examined sex differences in these spheres with an initial uncorrected voxel-level threshold of p < .05, which was corrected to a FWE p-value of p < .05 using small-volume correction. To characterize activation magnitude differences in both anatomical and functional ROIs, we calculated the average GLM beta value for each ROI by averaging across the beta value for all voxels within an ROI.

*Goldstein et al. (2010) did not report significant bi-lateral activation in these regions. Given the role of laterality in fear conditioning in these regions, we created ROls in the contralateral hemisphere of Goldstein and colleague’s significant findings

Men and women showed differential conditioning. An ANOVA conducted on the SCR data revealed a significant Stimulus main effect (F_(1,22) = 30.2, p < 0.001) with greater responses to the CS + than to CS- (first 4 trials of each type during conditioning) in women and men, indicating that both were able to learn the CS-US association. No significant main effects of Group (F_(1,22) = 2.9, p = 0.11) or Group X Stimulus interaction (F_(1,22) = 0.57, p = 0.81) were observed, indicating no significant sex differences in SCR during conditioning, even though males showed higher SCR than females but not significantly so (see group effect above, at p = .11) (see Figure 2A).

During extinction training on day 1, an ANOVA for the late extinction SCR data (last 4 CS + E vs. last 4 CS- trials) revealed no significant main effect of Stimulus (F_(1,21) = 0.21, p = 0.65) or Group (F_(1,21) = 2.23, p = 0.09) and no significant Group X Stimulus interaction (F_(1,21) = 0.14, p = 0.71), suggesting that comparable extinction learning had been achieved in both groups (data not shown). An ANOVA for the early extinction recall SCR data (first 4 CS + E vs. first 4 CS + U trials) also revealed no significant main effects of Stimulus (F_(1,22) = 3.19, p = 0.09), Group (F_(1,22) = 2.33, p = 0.14) or Group X Stimulus Interaction (F_(1,22) = 0.37, p = 0.55) (Figure 3A). Analyses of the extinction retention index, which controls for the level of fear acquired during the conditioning phase 23, revealed no significant group differences in extinction retention confirming that no significant differences between groups during extinction recall (data not shown). Collectively, we did not observe any statistically significant differences between men and women in our psychophysiological measures during any experimental phase. It is important to note that several statistical sex difference trends were observed with men tending to show higher SCR to the CSs throughout conditioning and extinction even though the differences were not statistically significant.

During fear conditioning, females responded to the conditioned stimulus with significantly greater activation, relative to males, in the following a priori anatomical regions of interest: the dACC, rACC, and the amygdala (see Table 3 for coordinates and statistical results and figure 2B). The average beta values for these ROIs (extracted from the between-group activation maps) show that these sex differences were present in dACC, mPFC and amygdala activation in females and deactivation in males (see Figure 2C).

ROIs are based on sexually dimorphic areas of the stress response circuitry identified by Goldstein and colleagues (2010).

We did not observe any significant sex differences in our anatomical ROIs during the extinction learning phase. During extinction recall, greater activation in the rostral region of the left rostral ACC (rACC) was observed in males relative to females. A trend in the same direction was also observed in the right rACC, which corresponds to a similar trend in right vmPFC [(6, 34, 0); z = 2.76, p = 0.006]. Additionally, greater insula activation was observed in females relative to males (see Figure 3B, Table 3). The average beta values from rACC showed that males were activating the rACC while females were deactivating this region. In contrast, females were activating insula, while males were deactivating this region (see Figure 3C).

We extended our ROI analyses based on the anatomy of the fear-conditioning circuitry by conducting an analysis based on previously reported 4 functional sex differences in the stress-response circuitry, such as the anterior hypothalamus. These data are summarized in Table 3. Although we found sex differences in the fear conditioning circuitry in our anatomical ROIs, no significant sex differences were observed in functional stress-response ROIs (data not shown). Conversely, Table 3 shows that during extinction learning (when no sex differences in fear circuitry were found), males showed significantly greater activation in right hypothalamus, and females showed greater activation in left hypothalamus, dACC, and mPFC. Lastly, extinction recall showed sex differences in fear and stress-response circuitry, as males exhibited significantly higher signal changes relative to females in our functional ROIs (Table 3).