Revisiting sex differences in the acquisition and extinction of threat conditioning in humans

Abstract

Findings pertaining to sex differences in the acquisition and extinction of threat conditioning, a paradigm widely used to study emotional homeostasis, remain inconsistent, particularly in humans. This inconsistency is likely due to multiple factors, one of which is sample size. Here, we pooled functional magnetic resonance imaging (fMRI) and skin conductance response (SCR) data from multiple studies in healthy humans to examine sex differences during threat conditioning, extinction learning, and extinction memory recall. We observed increased functional activation in males, relative to females, in multiple parietal and frontal (medial and lateral) cortical regions during acquisition of threat conditioning and extinction learning. Females mainly exhibited higher amygdala activation during extinction memory recall to the extinguished conditioned stimulus and also while responding to the unconditioned stimulus (presentation of the shock) during threat conditioning. Whole-brain functional connectivity analyses revealed that females showed increased connectivity across multiple networks including visual, ventral attention, and somatomotor networks during late extinction learning. At the psychophysiological level, a sex difference was only observed during shock delivery, with males exhibiting higher unconditioned responses relative to females. Our findings point to minimal to no sex differences in the expression of conditioned responses during acquisition and extinction of such responses. Functional MRI findings, however, show some distinct functional activations and connectivities between the sexes. These data suggest that males and females might use different neural mechanisms, mainly related to cognitive processing, to achieve comparable levels of acquired conditioned responses to threating cues.

Sex differences in forms of learning and memory have been widely observed (Andreano and Cahill 2009). The influences of sex on emotion learning and memory are of special interest. From a basic neuroscience perspective, it is intriguing to understand whether and how males and females might differ in facing fundamental neurobiological phenomena such as threat. From a clinical perspective, examining sex differences in emotion learning and memory may facilitate our understanding of the observed prevalence disparity in fear- and anxiety-related disorders between males and females (Tolin and Foa 2008; McLean et al. 2011; Steel et al. 2014; Jalnapurkar et al. 2018). Despite the recent significantly increased interest in sex differences and the molecular, cellular, neural, and behavioral mechanisms underlying such differences, much remains to be understood, especially within the context of emotion regulation.

One widely used laboratory model to study emotional learning and memory is the Pavlovian threat conditioning and extinction model. In this model, a neutral conditioned stimulus (CS) is repeatedly paired with an aversive unconditioned stimulus (US), so that the presentation of the CS alone elicits a conditioned threat response. This threat association can then be extinguished via repeated presentations of CS in the absence of the US. Following extinction learning, subjects often demonstrate reduced conditioned threat responses to the presentation of the now extinguished CS, indicating the retention of extinction memory. In line with the call in the field to refine the use of the terms “fear” and “threat” to distinguish between conscious feelings of fear and nonconscious processes of threat detection and responding (LeDoux 2012; LeDoux and Pine 2016; but see Fanselow and Pennington 2017), we adopt this distinction in the present study. We note, however, that there is some broader debate about the utility of such distinction. Threat conditioning and extinction are known to engage several key nodes of the “threat network” (Shin and Liberzon 2010; Milad and Quirk 2012; Alexandra Kredlow et al. 2021), including the ventromedial prefrontal cortex (vmPFC), dorsal anterior cingulate cortex (dACC), insular cortex, hippocampus, and amygdala. Many of these regions are sexually dimorphic (Goldstein et al. 2001; Cosgrove et al. 2007; Sacher et al. 2013), which might associate with sex differences during emotional learning and memory.

Using threat conditioning paradigms, sex differences of neural activations within the “threat network,” as well as psychophysiological and behavioral responses, have been reported in both rodent (Milad et al. 2009a; Baran et al. 2010; Fenton et al. 2014; Gruene et al. 2015) and human (Lebron-Milad et al. 2012; Hwang et al. 2015; Lonsdorf et al. 2015) studies. Generally, these studies suggest comparable threat conditioning between males and females, but deficient extinction learning and extinction memory recall in females. Whereas some studies observed sex differences during threat conditioning and extinction, some did not observe any sex differences, and others observed sex differences in the opposite direction (for reviews, see Merz et al. 2018; Velasco et al. 2019). One possible reason for this inconsistency is that studies often have insufficient statistical power to robustly detect subtle sex differences. Most studies to date contain relatively small sample sizes (<50 for each group).

In this study, we examined sex differences in the acquisition and extinction of threat conditioning in a relatively large sample of healthy human participants. To increase detection power, we aggregated functional magnetic resonance imaging (fMRI) data (N = 395–435, depending on the experimental phase) and skin conductance response (SCR) data (N = 152–187, depending on the phase) from healthy individuals across multiple studies (Marin et al. 2016; Sevinc et al. 2019; Seo et al. 2021; Wen et al. 2021a) to examine sex differences in the acquisition of threat conditioning and extinction. All participants underwent a validated Pavlovian threat conditioning and extinction paradigm (Milad et al. 2007b, 2009b). During threat conditioning, two cues were paired with a shock (CS+) and one cue was not paired with a shock (CS−). During extinction learning, a CS+ and a CS− were presented without shock. The next day, extinction memory was tested by presenting all three cues: extinguished CS+ (CS+E), unextinguished CS+ (CS+U), and CS−. We examined sex differences in brain activations of several regions of interest (ROIs) of the “threat network” that are thought to be critical in threat processing. We then extended the analysis to whole-brain activation beyond the “threat network.” Furthermore, we conducted whole-brain functional connectivity analysis to examine potential sex differences in broadly distributed brain networks that are engaged during emotional learning and memory. SCR data from different experimental stages were also examined.

Results

Considering that prior human and rodent studies showed distinct neural activations across trials within each experimental phase (Quirk et al. 1995; Milad et al. 2007b, 2009b; Burgos-Robles et al. 2009; Milad and Quirk 2012), we divided each phase into multiple trial blocks. As in previous studies, we focused our analyses on the early (the first four trials CS+ vs. CS−) and late (the last four trials CS+ vs. CS−) stages during threat conditioning and extinction learning. For the extinction memory recall, we focused on the early CS+E versus CS+U tests (the first four trials) to minimize the confounds introduced by additional extinction learning during late stage. We examined the effect of sex on neural activations, functional connectivity, and SCR data within each trial block. Note that the available data for each phase are different. This was due to the following factors: (1) a subset of participants underwent threat conditioning outside the fMRI scanner, (2) there were sometimes difficulties in collecting SCR data in the fMRI scanner, and (3) some participants dropped out on day 2. In our analyses, all available data for each single phase were included.

Threat conditioning

SCR data from 73 males and 116 females were analyzed for this phase. We first examined the effect of CS type (CS+ vs. CS−) across all participants, which revealed that CS+ evoked stronger SCR than CS− during both early (t₍₃₇₆₎ = 12.72, P < 0.001) and late (t₍₃₇₆₎ = 6.80, P < 0.001) threat conditioning, indicating that participants exhibited significant differential acquisition. We then examined sex differences in differential SCR responses (CS+ minus CS−). We did not observe significant sex differences during either early (t₍₁₈₇₎ = 1.42, P = 0.16) or late (t₍₁₈₇₎ = 0.75, P = 0.45) threat conditioning, suggesting that both males and females exhibited comparable levels of differential threat conditioning. A two-way mixed (CS type [CS+ vs. CS−] × sex) analysis of variance (ANOVA) also suggested that males showed higher overall CS response relative to females (F_{(1, 187)} = 5.93, P = 0.016) during late threat conditioning (Supplemental Fig. S1A; see the Supplemental Material for detailed ANOVA results on SCR data).

Neuroimaging data from 138 males and 257 females were analyzed for this phase. During early conditioning, we did not observe significant sex differences in neural activation within the predefined ROIs or across the brain that survived correction for multiple comparisons. During late conditioning, however, males showed greater activation than females in multiple brain regions (Fig. 1), including the dorsal anterior cingulate cortex (dACC; MNI_xyz = [−6, −2, 36]), precuneus (MNI_xyz = [−8, −62, 42]), inferior frontal gyrus (IFG; MNI_xyz = [−48, 44, 8]), and postcentral gyrus (PCG; MNI_xyz = [−50, −22, 34]).

Figure 1.

Sex differences during threat conditioning. (A) Males and females exhibited similar SCR measures (CS+ – CS−) during early and late threat conditioning. (B) Males exhibited higher neural activations in multiple brain regions compared with females during late but not early threat conditioning.

We subsequently examined whether males and females differed in their unconditioned response—the response to the electric shock delivery. Analyses of the SCR data revealed that males showed higher unconditioned responses relative to females (t₍₁₈₇₎ = 2.75, P = 0.006). For neural activations (Fig. 2), females exhibited higher activation (compared with males) in the amygdala (small volume correction, MNI_xyz = [26, −6, −20], cluster size = 69, P_FWE = 0.011) and somatomotor area (whole-brain analysis, MNI_xyz = [−4, 8, 70]).

Figure 2.

Sex differences in unconditioned response. (A) Males showed higher unconditioned response to shocks. (B) Females showed higher activations in the amygdala and somatomotor area relative to males during shock delivery.

Extinction learning

SCR data from 70 males and 112 females were analyzed for this phase. Sex differences in differential SCR responses (CS+ minus CS−) were not significant during either early (t₍₁₈₀₎ = 0.84, P = 0.40) or late (t₍₁₈₀₎ = 1.43, P = 0.15) extinction learning. A two-way mixed (CS type [CS+ vs. CS−] × sex) ANOVA suggested that males showed marginally higher overall CS response relative to females (F_{(1, 180)} = 3.67, P = 0.056) during early extinction learning (Supplemental Fig. S1B; see the Supplemental Material).

Neuroimaging data from 138 males and 297 females were analyzed. During early extinction learning, there was no evidence of sex differences in neural activation either within the predefined ROIs or within our whole-brain analyses. During late extinction learning, we did not observe significant effects of sex on predefined ROIs. Whole-brain analyses, however, revealed that males exhibited higher activations (compared with females) in multiple brain regions (Fig. 3), including the dorsal lateral prefrontal cortex (dlPFC; MNI_xyz = [42, 22, 44]), superior frontal gyrus (SFG; MNI_xyz = [20, 26, 54]), rostral anterior cingulate cortex (rACC; MNI_xyz = [4, 58, 12]), and precuneus (MNI_xyz = [10, −60, 28]).

Figure 3.

Sex differences during extinction learning. (A) Males and females exhibited similar SCR measures (CS+ – CS−) during early and late extinction learning. (B) Males exhibited higher neural activations in multiple brain regions relative to females during late but not early extinction learning.

Extinction memory recall

SCR data from 62 males and 90 females were analyzed. As shown in Figure 4A, sex differences in SCR data (CS+E minus CS+U) were not significant during early extinction memory recall (t₍₁₅₀₎ = −0.13, P = 0.90).

Figure 4.

Sex differences during early extinction memory recall. (A) Males and females exhibited similar SCR measures (CS+E – CS+U) during early extinction memory recall. (B) Females exhibited higher neural activations in the amygdala and inferior parietal gyrus during early extinction memory recall.

Neuroimaging data from 133 males and 285 females were analyzed. During early extinction memory recall, females, relative to males, exhibited higher activation in the right amygdala (small volume correction; MNI_xyz = [−48, 44, 8], k = 29, P_FWE = 0.056) and inferior parietal gyrus (whole-brain analysis; MNI_xyz = [26, −52, 54]).

Connectivity analyses

Since brain regions across distributed neural systems are likely to interact with each other during threat conditioning and extinction (Fullana et al. 2018; Berg et al. 2020; Wen et al. 2021a), we conducted whole-brain functional connectivity analyses to examine potential sex differences in task-based connectivity. Based on previous studies (Schaefer et al. 2018; Tian et al. 2020), we divided the whole brain into 432 functionally homogeneous parcels, including 400 cortical and 32 subcortical regions. We used jackknife-based beta series correlation (Richter et al. 2015; Thompson et al. 2018; Wen et al. 2021a, 2022) to estimate connections between every two of the 432 functionally homogeneous regions. The Network-Based Statistic (NBS) procedure (Zalesky et al. 2010)—a validated method for controlling family-wise error rate (FWER)—revealed a significant network component (permutation test, P_FWE < 0.05, 215 connections/edges) showing higher connectivity in females, relative to males, during late extinction learning (Fig. 5A). To further examine the distribution of the identified connections, we divided the 432 regions into eight subnetworks according to previous resting-state fMRI studies (Thomas Yeo et al. 2011; Schaefer et al. 2018; Tian et al. 2020). As shown in Figure 5B, the component with greater connectivity in females involved mostly connections from the visual, ventral/dorsal attention, and somatomotor networks. There was no evidence of sex differences in functional connectivity during any of the other experimental phases.

Figure 5.

Sex differences in functional connectivity during late extinction learning. (A) Females showed higher functional connectivity compared with males across multiple neural systems. Each sphere represents a brain region; the color of the sphere represents its network assignment, and the size of the sphere represents the weighted number of connections that showed group differences. (B) The proportion of significantly different connections within or between the eight subnetworks. A darkly shaded cell indicates that the connections of that network pair (indexed from the X-axis and Y-axis) were showing massive sex differences.

Discussion

We evaluated sex differences in neural and physiological signatures during threat conditioning, extinction learning, and extinction memory recall. We combined data from >400 participants from multiple studies, which allowed us the power to detect potential subtle differences in neural and physiological measures between males and females. We observed significant sex differences in neural activations during late threat conditioning, late extinction learning, early extinction memory recall, and responses to unconditioned (shock) stimuli. Sex differences were also observed in functional connectivity during late extinction learning. In contrast to the sex differences observed in neuroimaging data, we did not observe any significant difference between males and females in differential learning and memory SCR measures. However, compared with females, males exhibited higher reactivity to shock and higher overall response to CS presentations in late conditioning and early extinction learning, regardless of CS types. The observed sex differences in the neural correlates of the acquisition and extinction of threat conditioning suggest distinct neural mechanisms used by females and males to achieve comparable levels of learning across experimental phases.

There were minimal sex differences in the differential activation to the conditioned stimuli (CS+ vs. CS−) within the nodes of the threat detection network. We observed significant increased BOLD signal within the amygdala in females in response to the delivery of the shock and during extinction memory retrieval. These results suggest that females engage the amygdala to a much higher level compared with males in order to maintain expression of conditioned and unconditioned threat responding. Interestingly, the increased amygdala activation in females in response to the shock coincided with lower—not higher—unconditioned responding as measured by SCR. The increased amygdala activation in females during extinction recall could possibly be related to the maintenance of fear/threat homeostasis during extinction recall, as opposed to increased expression of conditioned threat responding. These interpretations, although speculative, are consistent with accumulating evidence suggesting that increased amygdala activation is not synonymous with heightened fear/threat reactivity, but rather can also be related to signaling safety (as in threat extinction learning) (LeDoux 2007; Herry et al. 2008; Tovote et al. 2015). These results further support the complex and dynamic functioning of the amygdala not just as a brain region important for threat detection, but also as a region important for maintaining emotional homeostasis (Herry et al. 2008; Pare and Duvarci 2012; Tovote et al. 2015; Gothard 2020).

The only other node within the threat detection network that exhibited sex differences was the dACC; higher activation in males was observed during the acquisition of threat conditioning. Within the context of threat conditioning, this brain region is often associated with the expression of heightened monitoring of error detection and processing as well as threat conditioning (Botvinick et al. 2004; Milad et al. 2007a; Alexander and Brown 2019). There were no differences between the sexes in the expression of conditioned responding, as indexed by SCR, during threat conditioning. Thus, this increased activation in the dACC might reflect heightened error monitoring/detection in males. Of note, in addition to heightened reactivity to the shock, males also expressed generally higher skin conductance response to all stimuli during late conditioning and early extinction learning (see the Supplemental Material), likely indicating general, nonspecific elevation of arousal/reactivity to emotionally salient stimuli in men, and not a reflection of sex differences associated with learning and memory. We also note that the direction of sex difference we report in this study is opposite to our previous study (Lebron-Milad et al. 2012). Similarly, although previous studies reported sex differences in other regions of the “threat network” (Lebron-Milad et al. 2012; Hwang et al. 2015; Velasco et al. 2019) such as the vmPFC and insular cortex, we did not observe such differences in threat conditioning in the current large pooled sample using the same paradigm. We note two plausible explanations. One is that sex hormone variance within the current sample and prior samples might have contributed to these apparent discrepancies. The other, more parsimonious explanation is that prior studies (including our own) used smaller sample sizes. With the substantially larger sample size in the current sample, the results obtained here might be more robust for testing group-level differences.

We observed sex differences in brain activations beyond the “threat network.” While the contribution of the threat network is critical for the acquisition and extinction of conditioned threat responses, recent conceptualizations and empirical results show that threat conditioning and extinction engage interactions between distributed neural systems involved in attention control, conscious awareness, and sensory motor functioning (Fullana et al. 2016; LeDoux and Pine 2016; Taschereau-Dumouchel et al. 2019; Wen et al. 2021a, 2022; Zhou et al. 2021). Moreover, differences between males and females in structural volumes, structural connectivity, and functional connectivity across the brain have been widely reported (Gong et al. 2011; Ruigrok et al. 2014; Satterthwaite et al. 2015). Consistent with these studies, our results show that males and females exhibited different activations in high-order association cortices such as the IFC and dlPFC. Furthermore, we observed significant sex differences in task-based functional connectivity during late extinction learning, mainly involving the visual, ventral/dorsal attention, and somatomotor networks. These results suggest that the effects of sex on emotional learning might be distributed across multiple neural systems implicated in attention, motor, and sensory function. Given the lack of any measurable sex differences at the psychophysiological levels, at least as observed with SCR, these data suggest that males and females might engage higher cognitive structures differently in order to achieve the same behavioral/psychophysiological outcomes.

The sex differences observed in our present study were relatively minimal and do not necessarily align with clinical observations that females are more prone to psychopathology. Possibly, larger sex differences could be observed if sex hormones, menstrual cycle phase, and use of contraceptives were to be included in the analytic strategy. Indeed, in our prior studies conducted in rodents and humans, sex differences were initially absent and became apparent only when considering hormonal status in females (Milad et al. 2009a, 2010). Previous studies have reported impact of sex hormones on brain activations and functional connectivity (Lebron-Milad and Milad 2012; Andreano et al. 2018; Pritschet et al. 2020). Notably, estradiol significantly modulates the recall of an extinguished threat response, with higher circulating estradiol levels predicting successful extinction recall and exogenous estradiol administration improving recall in rats and humans (Graham and Milad 2013; Hwang et al. 2015; Maeng et al. 2017; Wen et al. 2021b; for a recent review, see Hammoud et al. 2020). In one of our previous studies, we showed that sex differences in fear extinction are influenced by the phase of the estrous cycle in female rats and of the menstrual cycle in women (Zeidan et al. 2011), an effect also seen with other psychophysiological indices including fear-potentiated startle (Glover et al. 2012). Larger sex differences in threat conditioning and extinction were observed when females were divided into low- and high-estradiol groups (Milad et al. 2009a, 2010). The use of contraceptives can also impact estradiol level as well as learning and memory (Montoya and Bos 2017). Therefore, factors such as endocrine status and use of oral contraceptives are important considerations when examining sex differences. Future studies considering these factors could facilitate our understanding of sex differences in threat learning and memory.

Findings in this study may have clinical implications given that females have a significantly higher prevalence of fear- and anxiety-related disorders than males, and that sex differences in trauma exposure do not fully account for this differential prevalence (Gater et al. 1998; Tolin and Foa 2008). These vulnerability differences might be mediated by brain-based differences in the processing of emotional stimuli in females compared with males. Threat conditioning and extinction paradigms have been used to investigate the psychopathology of emotional disorders. Abnormal psychophysiological responses and neural activations in individuals with anxiety disorders and posttraumatic stress disorder (PTSD) during threat conditioning and extinction have been documented in previous studies (Norrholm et al. 2011; Milad and Quirk 2012; Duits et al. 2015; Shalev et al. 2017; Norrholm and Jovanovic 2018; Ressler 2020; Alexandra Kredlow et al. 2021). Sex differences among individuals with fear- and anxiety-related disorders have emerged in threat conditioning and extinction paradigms (Glover et al. 2012; Shvil et al. 2014; Sartin-Tarm et al. 2020). For example, psychophysiological and neural activation patterns associated with deficits of extinction memory recall were observed in males but not females with PTSD (Shvil et al. 2014). Further investigation of sex differences during threat learning and memory might facilitate our understanding the mechanisms for the higher risk of developing fear- and anxiety-related disorders in females than males.

We note a few limitations of the present study that should be considered. First, we did not collect subjective reports of fear, which prevented us from linking the observed sex differences in neural and physiological signatures with the conscious feelings of fear. Second, because of difficulties in collecting SCR data in the fMRI scanner, we only collected SCR measures from a subset of the participants, which might have decreased the statistical power available to detect sex differences in SCR measures. Third, it is possible that the sex differences in psychophysiological and neural patterns observed in our study might differ from other studies given different experimental settings such as the stimulus types and reinforcement rates across threat conditioning paradigms. Fourth, our peripheral measure of threat reactivity was SCR. It is possible that other psychophysiological indices, like startle, might be more sensitive in detecting sex differences during threat conditioning. Therefore, we cannot rule out the possibility that the increased amygdala activation in females during extinction recall might be related to increased fear/threat responding that was not detected by SCR. There is evidence supporting sex differences in threat conditioning and extinction when using other complementary psychophysiological measures, such as fear-potentiated startle and stimulus expectancy ratings across species (Merz et al. 2018; Velasco et al. 2019). For example, female rats exhibited less extinction retention than male rats when measured by fear-potentiated startle (Voulo and Parsons 2017). In human participants, sex differences in fear-potentiated startle, fear ratings, and unconditioned stimulus expectancy were observed (Gamwell et al. 2015; Lonsdorf et al. 2015). Thus, it is possible that other data modalities (other than those used in our present study) might reveal sex differences during threat conditioning and extinction.

In summary, the present study demonstrates that males and females likely exhibit different patterns of brain activation and functional connectivity during threat conditioning and extinction learning and its memory recall. These differences, however, appear to be minimal within the threat detection network itself but are distributed across multiple neural systems engaged in cognition, attention, and memory processes. Our results also support the idea that females and males appear to use different neural mechanisms to achieve comparable behavioral outputs. It is also worth noting that when analyzing larger sample sizes, sex differences appear to be relatively small and can easily be undetected. It is therefore important to consider sex as a factor when investigating neural circuits during threat conditioning and extinction, especially when studying psychopathology.

Materials and Methods

Participants

Data from a total of 435 healthy participants (297 female and 138 male) across multiple studies were included in the analyses. Among the 435 participants, 135 (82 females and 53 males) were trauma-exposed individuals without diagnosis of psychiatric disorders. Results from different subsets of the participants included in the analyses had been previously published with different scientific questions and different analytic strategies (Marin et al. 2016; Sevinc et al. 2019; Seo et al. 2021; Wen et al. 2021a). All participants were recruited at Massachusetts General Hospital. All procedures were approved by the Partners HealthCare Institutional Review Board of Massachusetts General Hospital. All participants provided written informed consent before they participated in the study.

Experimental design

Participants underwent a 2-d threat conditioning and extinction paradigm (Milad et al. 2007b, 2009b; Marin et al. 2017, 2020) during fMRI scanning. On day 1, participants first completed the threat conditioning phase. They were presented with three cues (conditioned stimuli [CSs]), two of which were partially reinforced (reinforcement rate: 62.5%) with a mild electric shock (CS+, eight trials for each) while the other was not reinforced (CS−, 16 trials). Following conditioning, participants underwent the extinction learning phase, during which the CS− (16 trials) and one of the CS+ (termed CS+ extinguished [CS+E], 16 trials) were repeatedly presented without shock. Twenty-four hours later (day 2), participants underwent the extinction recall test in the fMRI scanner. Participants were presented with all three cues: the CS+E (eight trials), the CS+ that was not extinguished on day 1 (termed CS+U, eight trials), and the CS− (16 trials) without shock.

Image acquisition and preprocessing

Neuroimaging data were acquired using three different MRI settings. Data from 64 participants were acquired in a Trio 3.0 Tesla whole-body MRI scanner (Siemens Medical Systems) using an eight-channel head coil. Functional data were acquired using a T2*-weighted echo-planar imaging (EPI) pulse sequence (TR: 3.0 sec, TE: 30 msec, slice number: 45, voxel size: 3 × 3 × 3 mm). Data from 180 participants were acquired in the same Trio 3.0 Tesla MRI scanner using a 32-channel head coil. Functional data were acquired using a T2*-weighted EPI pulse sequence (TR: 2.56 sec, TE: 30 msec, slice number: 48, voxel size: 3 × 3 × 3 mm). Imaging data from 97 participants were acquired on a Siemen's Prisma 3.0T scanner equipped with a 32-channel head coil. Functional data were acquired using a T2*-weighted EPI pulse sequence (TR: 3.0 sec, TE: 30 msec, slice number: 48, voxel size: 2.5 × 2.5 × 2.5 mm). High-resolution anatomical images were acquired for image registration.

Preprocessing was performed using the default pipeline of fMRIPrep 20.0.2—a standard toolbox for automatic fMRI data preprocessing (Esteban et al. 2019). Functional images were corrected for slice timing, realigned, coregistered with the structural image, normalized into the Montreal Neurological Institute (MNI) space, and smoothed with a 6-mm full-width half-maximum Gaussian kernel.

Imaging data analyses

Task-based activation

For threat conditioning and extinction learning, we created contrast maps for the early (first four trials CS+ vs. CS−) and late (last four trials CS+ vs. CS−) phases. For the extinction memory test, we focused on the early CS+E versus CS+U (the first four trials) contrast map. The rationale for selecting these specific trials for each phase was based on prior fMRI and animal studies showing distinct neural activations for these trials within each examined phase (Quirk et al. 1995; Milad et al. 2007b, 2009b; Burgos-Robles et al. 2009; Milad and Quirk 2012). Individuals’ contrast maps were used as the dependent variable for the second-level group analyses and sex was used as an independent variable. The impact of age was controlled by including the age of participants as a covariate. Study site information was included as covariates in our analyses to control for site effects.

Based on prior studies on threat conditioning and extinction, we were specifically interested in the following regions of interest (ROIs): the amygdala, hippocampus, insular cortex, dorsal anterior cingulate cortex (dACC), and ventromedial prefrontal cortex (vmPFC). The masks of the amygdala, hippocampus, and insular cortex were created based on the Harvard–Oxford structural atlas (probability threshold: 50%). For dACC and vmPFC ROIs, we searched the keyword “conditioning” in the Neurosynth database (https://neurosynth.org) and identified two activation peaks (dACC: [0, 14, 28], vmPFC: [−2, 46, −10]). We created 8-mm spheres around each coordinate to generate ROI masks. Activations detected in these ROIs with an uncorrected voxel-level P < 0.005 were tested with small volume corrections (family-wise error rate corrected P_FWE < 0.05) using the corresponding masks. We also examined brain regions beyond the ROIs. In this analysis, group-level activation maps were thresholded with a voxel-level P < 0.005 and a cluster-level P_FWE < 0.05.

Task-based connectivity

We examined task-based functional connectivity among brain regions. We divided the whole brain into 432 functionally homogeneous regions, including 400 cortical regions (Schaefer et al. 2018) and 32 subcortical regions (Tian et al. 2020). Each region was assigned to one of eight canonical subnetworks according to previous studies (Thomas Yeo et al. 2011), including the frontoparietal control network (CON), default mode network (DMN), dorsal attention network (DAN), limbic network (LIM), ventral attention network (VAN), somatomotor network (SMN), visual network (VIS), and subcortical network (SUB). For every two regions, we combined the beta series correlations method (Rissman et al. 2004) and the jackknife correlation method to estimate task-based connectivity (Richter et al. 2015; Thompson et al. 2018; Wen et al. 2021a). We focused on functional connectivity during early and late threat conditioning/extinction learning and early extinction memory recall. We constructed a symmetrized 432 × 432 FC matrix for each condition, with each value representing an edge (connection) between paired regions. We used the Combat harmonization approach to correct site effects (Fortin et al. 2017; Yu et al. 2018). We used the Network-Based Statistic (NBS) procedure (Zalesky et al. 2010)—a well-validated method for controlling family-wise error rate (FWER)—to identify edges that showed significant sex difference. In the NBS procedure, the edge-level threshold was set to P < 0.001 and the cluster-level threshold was set to P_FWE < 0.05.

Physiological data analyses

Skin conductance response (SCR) data were analyzed as described in our prior studies (Milad et al. 2005, 2007b, 2009b). SCR to each stimulus was computed by subtracting the mean skin conductance level observed during the last 2 sec of context preceding CS onset from the maximal skin conductance level reached during CS presentation. Analyses were performed on square root transformed data (if negative values were obtained, square root transformation was applied to the absolute value and a negative sign was then assigned to the square root transformed value), with age and site effects regressed out. To align with the fMRI data analysis strategy, we used two-sample t-test to examine the effects of sex in differential SCRs (CS+ minus CS−) during early/late threat conditioning and extinction learning, and the difference between CS+E and CS+U during early extinction memory recall. For completeness, we also conducted two-way mixed analyses of variance (ANOVAs) on the SCR data, with sex (male vs. female) as a between-subjects factor and CS type (CS+ vs. CS− for conditioning and extinction learning, or CS+E vs. CS+U for extinction memory recall) as a within-subjects factor. The results are presented in the Supplemental Material.

Competing interest statement

Praxis Precision Medicines, Inc., provides partial salary support to E.F.P.-S. The other authors declare no competing interests.