Role of the Locus Coeruleus in Response to Threat in Anxiety Disorders and Posttraumatic Stress Disorder: An Ultra-High-Field 7T Functional Magnetic Resonance Imaging Study

Abstract

Background

By harnessing the enhanced spatial resolution and signal power of high-field 7T magnetic resonance imaging (MRI), we assessed the functional involvement of the locus coeruleus (LC), together with the broader threat circuitry, in anxious arousal among individuals with anxiety disorders and posttraumatic stress disorder (PTSD).

Methods

Sixty-nine individuals with and without anxiety disorders or PTSD completed a modified no (N), predictable (P), and unpredictable (U) (NPU) threat task during a 7T functional MRI scan. Anxious arousal was measured using the Mood and Anxiety Symptom Questionnaire anxious arousal subscale. Individual data-driven LC segmentations were derived from ultra-high-resolution magnetization transfer contrast scans. We conducted LC functional activation and whole-brain data analyses during the NPU task using both transdiagnostic (Research Domain Criteria–based) and categorical (DSM-5–based) approaches.

Results

Greater LC activation during unpredictable threat was positively correlated with anxious arousal across all participants. In whole-brain analyses, the posterior cingulate cortex/precuneus and the subgenual anterior cingulate cortex were significantly activated during unpredictable threat, whereas the posterior insula was significantly activated during predictable threat. Greater activation within key structures of the threat circuitry, including the brainstem, the left hippocampus/amygdala, and the insula was positively correlated with anxious arousal across conditions and participants.

Conclusions

This translational and dimensional work advances our understanding of the role of the LC system and threat circuitry in pathological anxiety. Using 7T MRI, this study highlights the functional role of the LC in processing unpredictable threat in association with anxious arousal in individuals with anxiety disorders and PTSD.

Plain Language Summary

In this study, Boukezzi et al. utilized ultra-high field 7T MRI to investigate the role of the locus coeruleus (LC) during threat in individuals across a range of psychiatric conditions that present with pathological anxiety and hyperarousal, including anxiety disorders and posttraumatic stress disorder. The LC was preferentially activated in response to unpredictable threat in a manner that correlated with self-reported anxious arousal. A set of cortical and subcortical regions that associated with anxious arousal was also identified, which included the amygdala and insula. These findings increase our understanding of the role of the LC and other brain threat systems in pathological anxiety.

Plain Language Summary

In this study, Boukezzi et al. utilized ultra-high field 7T MRI to investigate the role of the locus coeruleus (LC) during threat in individuals across a range of psychiatric conditions that present with pathological anxiety and hyperarousal, including anxiety disorders and posttraumatic stress disorder. The LC was preferentially activated in response to unpredictable threat in a manner that correlated with self-reported anxious arousal. A set of cortical and subcortical regions that associated with anxious arousal was also identified, which included the amygdala and insula. These findings increase our understanding of the role of the LC and other brain threat systems in pathological anxiety.

Fear and anxiety are marked by shared and unique behavioral and neurobiological mechanisms. Fear, characterized by an immediate behavioral and physiological response to an acute and identifiable stressor, is essential for species survival (1). In contrast, nonpathological anxiety arises in response to an unidentifiable, unpredictable, or potential threat. While adaptive fear and nonpathological anxiety contribute to heightened arousal, behavioral activation, and adaptive functioning, their exaggeration can lead to pathological states. Common disorders of maladaptive fear and pathological anxiety include posttraumatic stress disorder (PTSD) and anxiety disorders, such as general anxiety disorder (GAD) and social anxiety disorder (SAD) (2). Anxiety disorders, as defined by DSM-5, are among the most prevalent psychiatric disorders (2), affecting 300 million individuals worldwide (3) and up to 33.7% of individuals during their lifetime (2). They are associated with significant health care costs and a substantial burden of disease. PTSD can develop after experiencing or witnessing one or several traumatic events, with a lifetime prevalence of about 6% in the United States (4).

Given the overlap of symptoms across different disorders as defined by the DSM-5, the Research Domain Criterion (RDoC) approach proposed by the National Institutes of Health offers a transdiagnostic framework for exploring disorders in terms of shared neurobehavioral domains, such as biology, behavior, and cognition (5). This approach presents potential advantages over the categorical approach (e.g., DSM-5 diagnosis), such as the integration of various levels of analysis and identification of biomarkers, which can lead to advancements in personalized medicine. However, the DSM may provide greater clinical utility, and a categorical approach is more easily implemented in clinical settings than a transdiagnostic approach (e.g., RDoC) (6). We suggest that these 2 approaches complement one another and can be used together to enhance our understanding of pathological anxiety.

The RDoC constructs most applicable to pathological fear and anxiety include negative valence and arousal systems. The no (N), predictable (P), and unpredictable (U) (NPU) threat task (7) provides a robust RDoC experimental paradigm to explore the negative valence and anxious arousal clinical dimension. This task has demonstrated distinct behavioral and neural responses to predictable and unpredictable threats in both healthy control participants (HCs) (7, 8, 9, 10) and individuals with psychiatric disorders (11,12). Its transdiagnostic application can be used to investigate the behavioral and neurobiological mechanisms underlying responses to fear and anxiety.

The neurobiological mechanisms that underlie fear and anxiety exhibit distinct but overlapping features (13,14), which have been debated in neurobiological models of emotions and preclinical models (15,16). Previous work suggests that in humans, fear responses are primarily associated with the involvement of subcortical structures such as the amygdala, while anxiety may recruit both subcortical and cortical structures including the insula and the posterior cingulate cortex (PCC) (13,16,17). With a key role in attention regulation and hyperarousal, the locus coeruleus (LC) has previously been studied in the context of anxiety disorders (18) and PTSD (19,20), both in animal models (21,22) and in humans (18, 19, 20). Situated in the pons, the LC projects norepinephrine (NE) afferents to various corticolimbic structures, including the brainstem, ventral tegmental area, hippocampus, amygdala, and broadly, the prefrontal cortex. Fear extinction deficits and alterations in the arousal system, mediated by the LC-NE system, are notable features of fear and anxiety disorders. In normal states, arousal involves different tonic and phasic activities of the LC, transitioning smoothly between sleep and anxiety (22). One hypothesis suggests that anxiety disorders involve hyperarousal due to excessive increased tonic firing of the LC-NE system (23). While the role of the LC has been explored in fear and anxiety disorders (19,20,24), the precise exploration of this structure has been challenging due to the limitations in spatial resolution and contrasts in standard MRI methods. Advancements in neuroimaging technology in terms of both acquisition (24) and data-driven processing pipelines, based on physiological principles related to the underlying tissue (24), have enabled increased study of small brainstem nuclei, including the LC. However, to our knowledge, there are no ultra-high-field 7T MRI studies characterizing the functional role of the LC in human pathological anxiety, in particular in transdiagnostic anxious arousal.

In this study, we aimed to delineate the functional role of the LC and key threat circuitry in pathological anxiety by examining neural activation during threat processing during ultra-high-field 7T MRI using both an RDoC-based transdiagnostic and a DSM-5-based categorical statistical approach. First, we investigated the relationship between LC activation and anxious arousal during the NPU task across all participants. We hypothesized that LC activation during anticipation of threat would be positively correlated with anxious arousal transdiagnostically. Then we compared LC activation in individuals with pathological anxiety and HCs. Next, using a whole-brain approach, we explored threat circuitry activation during the NPU task and associations with anxious arousal. Last, using a categorical approach, we analyzed group differences in brain activation across different task conditions.

Methods and Materials

Participants

Participants (male and female; ages 18–55) were recruited from the local community and enrolled at the Depression and Anxiety Center for Discovery and Treatment at the Icahn School of Medicine at Mount Sinai (ISMMS) in New York City (April 2019–November 2023). Clinical diagnoses were established by trained interviewers using the Structured Clinical Interview for DSM-5 Research Version (25). The anxiety (ANX) group met current DSM-5 diagnostic criteria for one of the following disorders for a duration of at least 12 months: GAD, SAD, or anxiety not otherwise specified (Anx-NOS). The PTSD group met the current DSM-5 criteria for PTSD for a duration of at least 12 months. The HC group did not meet the criteria for any current or past DSM-5 disorders (see Supplemental Methods). All study procedures were approved by the institutional review board at ISMMS. Participants provided written informed consent and received compensation.

Self-Report Questionnaires

Dimensions of general distress, anhedonia, and anxiety were measured with the Mood and Anxiety Symptom Questionnaire (MASQ). This is a validated questionnaire based on the tripartite model of affect, proposed to account for comorbidity between depression and anxiety disorders (26), and has 3 subscores: general distress, anhedonia, and anxious arousal. We focused our analyses on the MASQ anxious arousal (MASQ-AA) dimension as our symptom domain of interest. Relationships have been reported between LC activation and both MASQ anhedonia (MASQ-AD) and general distress (MASQ-GD) subscales.

NPU Task

In our 7T in-scanner adaptation of the NPU task (7), participants saw visual cues—red squares (P), blue triangles (U), and green circles (N) (Figure 1A). Instead of electric shocks, we used a compound aversive stimulus [unpleasant image from the International Affective Picture System picture database (27) and aversive sound]. This task included 3 runs of N, P, and U blocks counterbalanced across participants (e.g., PNUNUNP, UNPNPNU, PNUNUNP). Each block had 3 trials consisting of a cue, 1 intertrial interval (ITI), and a potential aversive stimulus. During each P block, a compound aversive stimulus was administered upon cue offset, while during each U block, the compound aversive stimulus was administered either during the ITI or the cue presentation. Two of 3 trials in P and U blocks included stimuli; N blocks had none. White noise (WN) startle probes (40 ms duration, 50 dB) were delivered via in-scanner headphones, calibrated to 80% of each participant’s maximum perceived volume. Probes were evenly distributed across the presentation of the cues and ITIs. In total, 12 WN startle probes per run were administered during the P and U conditions (6 during cues, 6 during ITIs), and 18 WN startle probes per run were administered during the N condition. Participants were told when to expect stimuli (predictable, unpredictable, or none). After each block, they rated anxiety levels (0–5) using an on-screen dial. Mean anxiety ratings were calculated for cue and ITI periods (Figure 1B, C; see Supplemental Methods).

Figure 1.

Illustration of the no (N), predictable (P), unpredictable (U) (NPU) threat experimental paradigm and behavioral responses. (A) NPU experimental paradigm. The in-scanner NPU task consisted of N, P, and U blocks, counterbalanced across participants in the following orders: PNUNUNP, UNPNPNU, and PNUNUNP. Each block included 3 trials, and each trial comprised a cue, an intertrial interval (ITI), and depending on the condition, an aversive stimulus (AS). For illustration purposes, only 2 of the 3 trials per block, followed by anxiety ratings, are shown. In P blocks, aversive stimuli were delivered at cue offset; in U blocks, they occurred either during the cue or the ITI. Two of the 3 trials in both P and U blocks were associated with an AS. No ASs were presented in N blocks. Each block also featured 6 white noise startle probes (40 ms, 50 dB), delivered through in-scanner headphones (depicted by headphone icons), calibrated to 80% of each participant’s maximum perceived volume. These white noise probes were evenly spaced across cue and ITI periods. At the end of each block, participants rated their anxiety using an on-screen Likert-style dial (0 = not anxious at all to 5 = extremely anxious). (B) Anxiety ratings during cue: Subjective anxiety ratings during cue-related stimuli for the N, P, and U conditions for healthy control participant (HC), anxiety disorder (ANX), and posttraumatic stress disorder (PTSD) groups. (C) Anxiety ratings during ITI: Subjective anxiety ratings during ITI for the N, P, and U conditions for HC, ANX, and PTSD groups. ∗∗∗p < .001.

MRI Acquisition and Preprocessing

Participants underwent 7T MRI scanning (Magnetom; Siemens) with a 32-channel transmit coil at the BioMedical Engineering and Imaging Institute, ISMMS, New York. A T1-weighted structural dual-inversion magnetization-prepared 2 rapid acquisition gradient-echo (MP2RAGE) scan was acquired with whole-brain coverage for segmentation and alignment using the following parameters: TA = 10 minutes 8 seconds, TR/TE = 6000/5.14 ms, inversion time (TI)1/TI2 = 1050/3000 ms, flip angle (FA) = 4°/5°, FOV = 224 × 224 mm, voxel resolution = 0.7 mm³, 224 slices, slice thickness = 0.70 mm, PF = 6/8, PAT = generalized autocalibrating partially parallel acquisitions (GRAPPA), acceleration factor = 3, bandwidth = 130 Hz/px. A magnetization transfer (MT) scan was acquired using a 3-dimensional segmented GRE readout (turbo-FLASH) with 20 MT pulses (190 V) and the following parameters: TA = 7 minutes 13 seconds, TR/TE = 1000/3.61 ms, FA = 8°, FOV = 192 × 192 mm, voxel resolution = 0.4 mm³, 56 slices, slice thickness = 0.50 mm, PF = 6/8, PAT = none, acceleration factor = 3, bandwidth = 140 Hz/px, MT frequency offset = 2000 Hz. Afterward, we acquired an identical slab without MT (non-MT) using the following parameters: TA = 5 minutes 41 seconds, TR/TE = 505/3.61 ms, FA = 8°, FOV = 192 × 192 mm, voxel resolution = 0.4 mm³, 56 slices, slice thickness = 0.50 mm, PF = 6/8, PAT = none, acceleration factor = 3, bandwidth = 140 Hz/px, MT frequency offset = 2000 Hz. This method allowed computing the relative signal enhancement based on neuromelanin content. Functional NPU task data were acquired using multiecho, multiband (MB) echo-planar imaging (EPI) with the following parameters: TA = 9 minutes 32 seconds, TR/TEs = 2100/14.0, 37.87, 61.74, FA = 60°, FOV = 1206 × 1206 mm, voxel resolution = 1.5 mm³, 69 slices, slice thickness = 1.50 mm, MB = 3, PF = 6/8, PAT = GRAPPA, iPAT acceleration factor = 3, pixel bandwidth = 1866 Hz/px. A subset of the data was acquired with our institute’s previous functional MRI (fMRI) sequence with the following characteristics: multiecho MB EPI pulse sequence with the parameters: TA = 9 minutes 25 seconds, TR/TEs = 1850/8.5, 23.17, 37.84, 52.51, FA = 70°, FOV = 640 × 640 mm, voxel resolution = 2.5 mm³, MB = 2, PAT = GRAPPA, iPAT acceleration factor = 3, 50 slices, slice thickness = 2.50 mm, pixel bandwidth = 1786 Hz/px. Fifty-four participants were scanned with the 1.5-mm³ sequence (19 HC, 11 GAD, 7 SAD, 2 Anx-NOS, and 15 PTSD), while 15 participants were imaged with the 2.5-mm³ sequence (8 HC, 6 GAD, and 1 PTSD).

Data analyses were conducted in AFNI, with images undergoing robust preprocessing and denoising of motion and physiological noise using ME-ICA version v3.2 beta1 (28,29). ME-ICA works by leveraging multi-echo fMRI data to separate the blood oxygen level–dependent (BOLD; neural activity-related) signal from non-BOLD noise (e.g., motion, physiology). It uses the different T2∗ decay rates captured at multiple echo times to identify components; BOLD components show consistent T2∗ scaling, while non-BOLD components do not. This method applied independent component analysis to decompose the data into independent components, retaining BOLD-related ones and discarding noise (28, 29, 30) (see Supplemental Methods).

Statistical Analyses

Demographic, behavioral, and clinical data were analyzed using JASP (v.0.18.3). Demographic variables were assessed using χ² tests. The Mann-Whitney U tests were applied for non-normally distributed data. Pearson or Spearman correlation analyses were conducted to examine associations between neuroimaging data and clinical scale scores, as appropriate. z Transformation tests were utilized to assess the relationship between coefficient correlations.

Two-way repeated-measures analyses of variance (ANOVA) were conducted to examine subjective anxiety ratings collected during cue and ITI presentations across 3 conditions (N, P, U) for the HC, ANX, and PTSD groups (see Supplemental Methods).

At the neuroimaging level, the first-level analyses were conducted using the 3dDeconvolve function, with regressors of interest modeling U-cue, P-cue, N-cue, aversive stimuli, and ITI using block functions. WN was modeled using a wave function. Normalized/blurred beta images corresponding to U-cue, P-cue, and N-cue were utilized in the subsequent second-level analyses, consistent with previous research (31).

The LC segmentation has been described in detail elsewhere (24). Relative signal enhancement maps were computed as the ratio of the MT map to the non-MT map. As a next step, FreeSurfer 6.2 was used on each participant’s T1-weighted image to segment their brainstem, which was coregistered to the native ultra-high-resolution space to exclude nonbrainstem tissue before the automated clustering procedure via Gaussian mixture modeling. For each voxel, the algorithm computes the probability of belonging to either the LC, cerebrospinal fluid, or white matter using expectation maximization. After the clustering, brainstem nuclei including the substantia nigra, ventral tegmentum, and LC were visible. For each participant, the LC tissue was isolated from the rest of the brainstem by applying a 4-mm³ dilated mask of the fourth ventricle from the FreeSurfer segmentation, which included the lateral portion of the brainstem where the LC is located. Lastly, the individualized masks were normalized to Montreal Neurological Institute (MNI) 2009 space and further constrained to remove voxels from cerebellar white matter and periaqueductal gray using masks from the Brainstem Navigator (29) and the cerebellar white matter atlas (30), respectively. The resulting LC masks were then used to compute the mean masks across 69 participants with 3dMean and binarize them with fslmaths (30% threshold). Parameter estimates were extracted from the mask and used in the second-level analyses. First, using a transdiagnostic approach, Spearman correlations were computed between MASQ-AA and LC parameters for each condition (N, P, U). Fisher z transformations were applied to the data. Differences between correlation coefficients were then tested (32). Second, using a categorical approach, LC parameter estimates were compared between groups (ANX, PTSD, and HC) using analysis of covariance (ANCOVA) including condition (U-cue, P-cue, N-cue) as an independent factor and with age and sex as covariates.

At the whole-brain level, the following second-level analyses were conducted: first, using a transdiagnostic approach, an ANCOVA was computed for cue-related stimuli using AFNI’s 3dMVM function, with condition (U-cue, P-cue, N-cue) as the within-subjects factor, MASQ-AA as a covariate of interest, and age and sex as additional covariates. Second, using a categorical approach, a separate ANCOVA was computed with condition (U-cue, P-cue, N-cue) as the within-subjects factor, group (HC, PTSD, ANX) as the between-subjects factor, and age and sex as additional covariates (see Supplemental Methods for description of post hoc analyses). The U-cue captured putative activation related to anxiety, whereas the P-cue captured putative activation in response to fear. The combination of U + P-cue captured activation in response to both predictable and unpredictable (e.g., general) threat. Data were corrected for multiple comparisons using 3dClustSim (33) (whole-brain cluster-corrected familywise error [FWE] p < .05, voxelwise p = .005, k > 78).

Results

Demographic and Clinical Characteristics

Groups did not differ significantly in terms of age, sex, education, race, ethnicity, employment, or relationship status. Participants in the ANX and PTSD groups scored higher on the MASQ than participants in the HC group (Table 1).

Table 1.

Demographic and Clinical Characteristics

	Groups			Between-Group Comparisons
	HC, n = 27	PTSD, n = 16	ANX, n = 26
Age at Enrollment, Years	35.00 (12.34)	35.70 (11.67)	31.60 (10.56)	F2,66 = 0.83, p = .44
Male	14 (51.85%)	4 (25.00%)	6 (23.08%)	χ2 = 5.71, p = .06
Race
American Indian or Alaskan Naïve	0	1 (6.25%)	0	χ2 = 14.00, p = .17
Asian	7 (25.93%)	2 (12.50%)	7 (43.75%)
Black or African American	8 (29.63%)	3 (18.75%)	1 (3.85%)
More than 1 race	1 (3.70%)	3 (18.75%)	3 (11.53%)
White or Caucasian	11 (40.74%)	6 (37.5%)	14 (53.85%)
Unknown or do not wish to disclose	0	1 (6.25%)	1 (3.85%)
Ethnicity
Hispanic/Latino	1 (3.70%)	3 (18.75%)	3 (11.54%)	χ2 = 4.36, p = .36
Not Hispanic/Latino	26 (96.29%)	12 (75%)	22 (84.62%)
Unknown or do not wish to disclose	0	1 (6.25%)	1 (3.85%)
Employment
Employed full time	11 (40.74%)	6 (37.50%)	12 (46.15%)	χ2 = 10.32, p = .24
Employed part time	4 (14.81%)	5 (31.25%)	5 (19.23%)
Not working due to disability	0	2 (12.50%)	0
Retired	1 (3.70%)	0	1 (3.85%)
Unemployed	11 (40.74%)	3 (18.75%)	8 (30.77%)
Education
Graduated/professional degree	9 (33.33%)	4 (25%)	5 (19.23%)	χ2 = 11.38, p = .33
Graduated 2-year college or trade school	0	2 (12.50%)	2 (7.69%)
Graduated 4-year college	11 (40.74%)	6 (37.50%)	11 (42.31%)
High school diploma/GED	0	3 (18.75%)	2 (7.69%)
Some college	3 (11.11%)	0	3 (11.54%)
Some graduate/professional school	4 (14.81%)	1 (6.25%)	3 (11.54%)
Relationship Status
Divorced	0	1 (6.25%)	1 (3.85%)	χ2 = 3.52, p = .48
Married	6 (22.22%)	1 (6.25%)	6 (23.07%)
Never married	20 (74.07%)	14 (87.50%)	19 (73.08%)
MASQ General Distress	13.11 (4.46)	28.00 (10.44)	24.58 (7.75)	F2,66 = 25.26, p < .001
MASQ Anhedonia	25.70 (5.72)	40.38 (9.92)	16.46 (6.21)	F2,66 = 22.94, p < .001
MASQ Anxious Arousal	10.56 (0.89)	20.19 (9.33)	16.46 (6.21)	F2,66 = 14.69, p < .001
Current Medicationa	0	2	4	–
Trauma Type
Death, actually witnessed happening to others in person	–	6	–	–
Sexual violence	–	6	–	–
Serious injury	–	4	–	–
Trauma Duration, Months	–	17.18 (8.76)	–	–

Values are presented as n, n (%), or mean (SD). The ANX group includes individuals with general anxiety disorder (n = 17), social anxiety disorder (n = 7), or anxiety not otherwise specified (n = 2). Race and ethnicity were reported by the study participants.

ANX, anxiety disorder; GED, General Educational Development; HC, healthy control participant; MASQ, Mood and Anxiety Symptom Questionnaire; PTSD, posttraumatic stress disorder; SSRI, selective serotonin reuptake inhibitor.

^aCurrent medication: ANX: SSRIs (n = 3), benzodiazepines (n = 1); PTSD: SSRIs (n = 1), benzodiazepines (n = 1).

Subjective Anxiety Ratings

Analysis of anxiety ratings during cue presentation showed a significant main effect of condition (F_2,13 = 88.30, p < .001), driven by higher ratings during P-cue compared with N-cue (p < .001) and during U-cue compared with N-cue (p < .001) but not between P-cue and U-cue (p = .55) conditions (Figure 1). Analysis of anxiety ratings during ITI presentation also showed a significant main effect of condition (F_2,13 = 86.38, p < .001), driven by higher ratings during P-ITI compared with N-ITI (p < .001) and higher ratings during U-ITI compared with N-ITI (p < .001) but not between P-ITI and U-ITI (p = .89) (see Supplemental Results).

fMRI Results: LC Data Analyses

Following a transdiagnostic approach, a positive correlation was observed between MASQ-AA and LC activation during the U-cue across groups (ρ = 0.37, p = .002). This association was attenuated when we controlled for diagnosis (r = 0.215, p = .078). There were no significant correlations between MASQ-AA and LC activation during the P-cue (ρ = −0.02, p = .90) or the N-cue (ρ = −0.01, p = .92) conditions (Figure 2A). Spearman correlations between LC activation and both MASQ-GD and MASQ-AD were not statistically significant (Table S1).

Figure 2.

Illustration of the locus coeruleus (LC) activation using both transdiagnostic and categorical approaches. (A) Illustration of the LC masks segmented and computed across 69 participants with 3dMean and binarized with fslmaths (30% threshold). (B) Research Domain Criteria–based transdiagnostic approach: Pearson’s correlations between beta weights extracted from the LC during unpredictable (U)-cue, predictable (P)-cue, no (N)-cue, and Mood and Anxiety Symptom Questionnaire anxious arousal (MASQ-AA). (C) DSM-5–based categorical approach: group comparisons performed on beta weights extracted from the LC during U-cue, P-cue, and N-cue. ANX, anxiety disorder; HC, healthy control participant; ns, nonsignificant; PTSD, posttraumatic stress disorder.

Fisher z transformations were applied to the data, and differences between correlation coefficients were analyzed. The correlation coefficients were statistically significantly different from each other (z = 2.32, p = .01).

Following a categorical approach, no group differences were found in LC activation during U-cue (F_2,64 = 1.43, p = .25), P-cue (F_2,64 = 1.66, p = .20), or N-cue (F_2,64 = 0.17, p = .84) conditions (Figure 2B).

fMRI Results: Whole-Brain Data Analyses

A main effect of condition was observed in structures of the visual and auditory cortices, posterior insula, precentral/postcentral gyrus, supplementary motor area, precuneus, and PCC (Figure 3A and Table S2). Post hoc analyses showed that the PCC/precuneus and subgenual anterior cingulate cortex (sgACC) showed higher activation during the U-cue compared with the P-cue and N-cue conditions. The posterior insula was mainly activated during P-cue. The PCC/precuneus, sgACC, and middle/superior frontal gyrus (Brodmann area 9/46) were mainly activated during U + P-cue (Figure 3B). The posterior insula, hippocampus, and mid-cingulate cortex (MCC) were activated during N-cue (see Supplemental Results).

Figure 3.

Main effect of the no (N), predictable (P), unpredictable (U) threat task on brain activation during cue-related stimuli. Whole-brain voxelwise statistical F maps on a canonical brain, showing a significant main effect of condition at p < .005, familywise error corrected, with a cluster size of ≥78 contiguous voxels. (A) Main effect of condition: This panel features transversal, sagittal, and coronal views arranged in a 2 × 2 layout to visualize activation of brain structures in response to cue-related stimuli at the defined statistical threshold. (B) Post hoc comparisons; brain activation based on the blood oxygen level–dependent (BOLD) signal during the U, P, U + P, and N conditions. Panel (B) features transversal, sagittal, and coronal views arranged in a 2 × 1 layout to visualize activation of brain structures in response to cue-related stimuli at the defined statistical threshold. Peak Montreal Neurological Institute coordinates for the displayed clusters are provided in Tables S2–S4. ACC, anterior cingulate cortex; BA, Brodmann area.

Following a transdiagnostic approach, greater activation in the right hippocampus/amygdala, left parahippocampal gyrus, right insula, brainstem, PCC/precuneus, and left precentral/postcentral gyrus was positively associated with anxious arousal (Figure 4A and Table S4). Moreover, significant MASQ-AA × condition interactions were observed in the substantia nigra (SN), left hippocampus/parahippocampal gyrus, MCC, brainstem (Figure 4B and Table S4), and amygdala, although the latter did not survive FWE correction (k = 35) (Figure 4B). These interactions were characterized by a positive correlation between activation during U-cue and anxious arousal in the SN (U: r² = 0.10, p < .01) and brainstem (U: r² = 0.11, p < .01); a positive correlation during U-cue and a negative correlation during P-cue with anxious arousal in the MCC ([P: r² = 0.06, p < .001]; [U: r² = 0.08, p < .05]); and a positive correlation between activation during P-cue and anxious arousal in the amygdala (P: r² = 0.12, p < .005) (see Supplemental Results).

Figure 4.

Main effect of no (N), predictable (P), unpredictable (U) threat task on brain activation during cue-related stimuli using a transdiagnostic approach. Whole-brain voxelwise statistical F maps on a canonical brain, showing a significant main effect of anxious arousal and anxious-arousal × condition interaction at p < .005, familywise error corrected, with a cluster size of ≥78 contiguous voxels. (A) Main effect of anxious arousal on brain activation based on the blood oxygen level–dependent (BOLD) signal across conditions. This panel features transversal, sagittal, and coronal views arranged in a 2 × 1 layout to visualize activation of brain structures in response to cue-related stimuli at the defined statistical threshold. (B) This panel shows transversal, sagittal, and coronal views of clusters of activation, together with beta weights extracted from the peak of activation defined by a 5-mm sphere plotted against Mood and Anxiety Symptom Questionnaire anxious arousal (MASQ-AA) scores for post hoc analyses. In panel (B) only, the right amygdala activation did not survive Bonferroni correction but is depicted for visualization purposes. Peak Montreal Neurological Institute coordinates for the displayed clusters are provided in Tables S2–S4.

Following a categorical approach, a main effect of group was observed in the right middle frontal gyrus/right precentral gyrus (F_2,66 = 9.56, p < .001) and was driven by higher activation in the HC group compared with the ANX (p < .005) and PTSD (p < .001) groups. There was no difference between the ANX and the PTSD groups (p = .55) (Figure 5A and Table S4). Moreover, a group × condition interaction was found in the right occipital gyrus (F_4,13 = 4.40, p < .005), which was driven by higher activation during U-cue in the ANX group compared with the HC group (p < .005), higher activation during P-cue than U-cue in the HC group (p < .05), and higher activation during N-cue than U-cue in the HC group (p < .005) (Figure 4B and Table S4).

Figure 5.

Main effect of the no (N), predictable (P), unpredictable (U) threat task on brain activation during cue-related stimuli using a categorical approach. Whole-brain voxelwise statistical F maps on a canonical brain, showing a significant main effect of group at p < .005, familywise error corrected, with a cluster size of ≥78 contiguous voxels. (A) Main effect of group on brain activation based on the blood oxygen level–dependent (BOLD) signal across conditions. (B) Group × condition interaction on brain activation based on the BOLD signal. These panels feature transversal, sagittal, and coronal views arranged in a 2 × 1 layout to visualize activation of brain structures in response to cue-related stimuli at the defined statistical threshold (left panel) together with beta weights extracted from the peak of activation defined by a 5-mm sphere plotted for each condition for post hoc analyses (right panel). Peak Montreal Neurological Institute coordinates for the displayed clusters are provided in Tables S2–S4. ∗∗∗p < .001. ANX, anxiety disorder; HC, healthy control participant; ns, nonsignificant; PTSD, posttraumatic stress disorder.

Discussion

This is the first 7T MRI study to investigate the role of the LC and broader threat circuitry during predictable and unpredictable threat processing in individuals with and without a range of anxiety or stress-related disorders. Behaviorally, increased anxiety levels were observed in response to both predictable and unpredictable threats, with trends for higher levels in the PTSD and ANX groups compared with the HC group. LC activation positively correlated with anxious arousal during unpredictable threat. There was higher posterior insula activation during predictable threat and higher sgACC, precuneus, and PCC activation during unpredictable threat. Other key threat circuitry structures including the hippocampus, brainstem, and amygdala showed greater activation associated with anxious arousal. Together, these findings highlight the roles of distinct nodes of the salience and arousal networks during the processing of specific threat conditions related to either more anxiety-like or fear-like dimensions.

Heightened response to unpredictable threats is a common feature in PTSD and anxiety disorders (34). Previous studies have shown that patients with PTSD exhibited elevated anxiety during unpredictable conditions compared with individuals with GAD and HCs (35). Patients with panic disorder were overly sensitive to unpredictable threat (11) while showing normal responses to predictable threat. At the neural level, the central nucleus of the amygdala is implicated in phasic response to predictable threats, while the bed nucleus of the stria terminalis mediates tonic responses to unpredictable threat (11,36). While previous research has implicated the LC as hyperactive in anxiety disorders and PTSD (18,24), our study is the first to demonstrate a specific activation during unpredictable threat, providing novel insights into the role of the LC in pathological anxiety. No group differences were found in LC activation, highlighting its potential as a transdiagnostic biomarker.

The insula plays a key role in the pathophysiology of anxiety disorders. Acting as a hub for integrating various sensory inputs, it evaluates the emotional and motivational significance of stimuli, serving as a bridge between external stimuli and internal motivational states (37). While the anterior insula is implicated in emotional processes (38), the mid-posterior insula is pivotal for interoceptive awareness, encompassing the perception of bodily sensations (37). Individuals with anxiety disorders exhibited hyperactivation of the anterior and middle insula in response to unpredictable threats (39,40); however, this is the first study to show a specific role of the posterior insula in the context of predictable threats. One possible interpretation is that participants mobilized attentional processes, preparing the body to respond and activating the somatosensory system, potentially initiated within the posterior insula.

Heightened precuneus and PCC activation during unpredictable threat is consistent with previous research, indicating that threat-induced anxiety affects parietal cortical regions (41). Heightened activation of the precuneus, a key default mode network structure dysfunctional in pathological anxiety (42) and PTSD (43), during emotional stimuli tasks suggests its prominent role in evaluating threat significance. As for the sgACC, existing research suggests its involvement in emotional and cognitive processing, particularly in emotion regulation and stress response modulation (44). Our results of sgACC activation during unpredictable and general threats are consistent with previous research (45).

Our findings also showed significant positive associations between anxious arousal and key structures of the threat circuitry. The amygdala is crucial for conditioned fear responses (46), while the hippocampus plays a key role in processing contextual information (47). In rodents, damage to the hippocampus has been associated with deficits in fear conditioning (48) that prevent freezing responses in response to the context while leaving cue-specific fear conditioning intact. This suggests that contextual anxiety involves both the amygdala and the hippocampus, while stimulus-induced fear conditioning only requires amygdala functioning (49).

Using a categorical approach, decreased activation in the right middle frontal gyrus was found in the ANX and PTSD groups compared with the HC group. Conversely, higher activation was observed in the occipital gyrus during exposure to unpredictable threat in the ANX and PTSD groups compared with the HC group. This suggests that individuals with pathological anxiety may experience dysregulation in top-down processing when handling salient stimuli as shown by previous research (50).

Our study has several limitations. First, the sample size was moderate, and the number of participants with PTSD was small, resulting in limited power to detect small effects. Our sample had a higher proportion of males in the HC group compared with the ANX and PTSD groups, which might have influenced the anxiety ratings. However, sex was included as a covariate of non-interest to account for this potential confounding effect in our data analyses. Second, our transdiagnostic sample did not include individuals with all types of anxiety disorders, which prevented subgroup analyses with participants who may be more susceptible to altered fear and threat mechanisms. Our primary approach was transdiagnostic, which allowed us to explore domains rather than diagnoses. Future studies should include larger transdiagnostic samples. Third, a few participants were taking psychotropic medications; however, including medication as a covariate was not feasible due to collinearity assumptions (51). This sample reflects the naturalistic population of individuals with anxiety disorders, many of whom use psychotropic medications. Finally, the inclusion of both 1.5- and 2.5-mm isotropic resolutions in our scanning sessions presents a limitation. The current primary voxel-level threshold of p < .005 was used to account for multiple comparisons, which could be more conservatively adjusted in future studies with larger sample sizes (52).

Conclusions

This study provides new insights into high-resolution fMRI research, demonstrating a positive correlation between LC activation and anxious arousal in response to unpredictable threat. Furthermore, it identifies distinct activation patterns within the threat circuitry during both predictable and unpredictable threat. By using both transdiagnostic and categorical approaches, we offer complementary insights into how different levels of anxiety affect critical brain regions involved in threat processing. Future research should aim to unravel the functional interplay between the LC and key structures within the threat circuitry, thereby paving the way for targeted interventions for anxiety-related conditions.