In anxiety disorders, initially adaptive defensive reactions are more sustained, out of proportion to the initial threat and lead to excessive avoidance behaviour impairing daily functioning. Reference Pittig, Treanor, LeBeau and Craske1 Cognitive–behavioural therapy, with exposure as a core component, is the most used treatment option for anxiety disorders. Reference Hofmann, Asnaani, Vonk, Sawyer and Fang2 However, although this therapy is effective for many, approximately 50% of people do not benefit, drop out or relapse in the long term, Reference Loerinc, Meuret, Twohig, Rosenfield, Bluett and Craske3 driving the need for treatment optimisation or alternative treatment options. Promising novel treatment options such as transcranial magnetic stimulation or functional magnetic resonance imaging (fMRI)-guided neurofeedback Reference Krause, Kogias, Krentz, Lührs, Goebel and Hermans4,Reference Krause, Linden and Hermans5 exist, but rely on our understanding of functional neuroanatomy in anxious compared with non-anxious individuals.
Classical fear conditioning as a model of anxiety
Our understanding of the functional neuroanatomy of maladaptive defensive reactions is mostly based on the use of classical fear-conditioning paradigms. Although it does not explain the complete phenomenology, the acquisition of conditioned fear responses is considered a model for the onset of clinical anxiety. Reference Beckers, Hermans, Lange, Luyten, Scheveneels and Vervliet6 In fear-conditioning paradigms, a conditioned threat stimulus (CS+) is an initially neutral stimulus that is repeatedly paired with an intrinsically salient, or unconditioned stimulus, such as an electrical shock. After sufficient pairings, this initially neutral stimulus elicits a conditioned response on its own; for example, enhanced activity of the sympathetic nervous system or self-reported fear. The neutral ‘safe’ stimulus (CS−) is, conversely, a stimulus that is never paired with the unconditioned stimulus, thus signalling safety. Patients with anxiety disorder are less able to differentiate between salient and neutral stimuli, and show increased fear toward this neutral ‘safe’ stimulus, compared with healthy controls. Reference Duits, Cath, Lissek, Hox, Hamm and Engelhard7
Functional neuroanatomy of classical fear conditioning
Research from past decades has examined individual brain regions activated during fear conditioning. Reference Fullana, Harrison, Soriano-Mas, Vervliet and Cardoner8–Reference Tovote, Fadok and Lüthi11 However, because sets of brain regions tend to systematically activate or deactivate together in large-scale intrinsic connectivity networks (ICNs), Reference Beckmann, DeLuca, Devlin and Smith12–Reference Seeley, Menon, Schatzberg, Keller, Glover and Kenna15 focusing on activation on isolated regions might be an oversimplification, as it disregards the fundamental network organisation of the brain. Exploring the temporal dynamics of ICNs during fear acquisition would enhance our understanding of its underlying mechanisms and pave the way for new brain-based interventions. Knowledge about the functional reorganisation of ICNs during fear processing is being accumulated. Reference Marstaller, Fynes‐Clinton, Burianová and Reutens16–Reference McMenamin, Langeslag, Sirbu, Padmala and Pessoa20 For example, decreased connectivity in the default mode network (DMN) was reported in response to CS+ stimuli. Reference Marstaller, Fynes‐Clinton, Burianová and Reutens16 However, much is still to be established, especially in relation to the temporal dynamics of the acquisition of fearful associations as a model for the onset of clinical anxiety. Moreover, it is unknown how ICNs change over time in the process of fear acquisition in young individuals who are at risk of developing anxiety disorders.
Aims
Although fear-conditioning paradigms are classically utilised to investigate the mechanistic underpinnings of anxiety- and trauma-related disorders, their relevance extends significantly to the broader affective disorder spectrum. Mood and anxiety disorders are highly comorbid, Reference Moffitt, Caspi, Taylor, Kokaua, Milne and Polanczyk21 and patients with primary affective disorders frequently exhibit disrupted fear and anxiety processing, such as altered fear acquisition and extinction. Reference Otto, Moshier, Kinner, Simon, Pollack and Orr22,Reference Dibbets, van den Broek and Evers23 This transdiagnostic overlap is particularly pronounced during youth – a critical developmental window characterised by heightened affective sensitivity, profound brain plasticity (e.g. in the fronto-amygdala circuitry; as reviewed by Towner et al) Reference Towner, Chierchia and Blakemore24 and the sharpest increase in the incidence of anxiety disorders. Reference Beesdo, Höfler, Leibenluft, Lieb, Bauer and Pfennig25–Reference Solmi, Radua, Olivola, Croce, Soardo and Salazar de Pablo27 The World Health Organization defines youth as the 15- to 24-year age group. 28
Building on this transdiagnostic framework, the current study investigates the dynamic functional reorganisation within and between three ICNs during fear processing: the salience network, executive control network (ECN) and DMN. Specifically, we examine these temporal network dynamics during fear acquisition in young individuals with and without mild-to-subclinical mixed affective complaints, aiming to gain fundamental insight into the neural onset of clinical anxiety and related affective pathologies.
Method
Participants
The data used are part of a larger randomised controlled trial that investigated the effects of a psychological intervention on psychopathology in emerging adulthood (Smartscan, Dutch Trial Register Number: NTR380). Participant recruitment is described in detail elsewhere. Reference Lange, Goossens, Bakker, Michielse, Marcelis and Wichers29 Briefly, three groups of young adults aged 16–25 years (mean age 20.79 (s.d. = 2.17) years, 118 females v. 22 males) were included. Group 1 comprised participants with a DSM-IV diagnosis of specific phobia of spiders (n = 46), assessed by the MINI-International Neuropsychiatric Interview (MINI), Reference Overbeek, Schruers and Griez30 who were not undergoing psychiatric treatment and did not have any current or past psychiatric diagnosis other than the specific phobia. Group 2 comprised participants with subclinical mixed psychopathology (n = 50), presenting with a positive subscale frequency score ≥10 and/or positive subscale distress score ≥2 on the Community Assessment of Psychic Experiences (CAPE) Reference Stefanis, Hanssen, Smirnis, Avramopoulos, Evdokimidis and Stefanis31,Reference Konings, Bak, Hanssen, Van Os and Krabbendam32 , and/or moderate but subthreshold depressive symptoms, defined as a score ≥10 on the Montgomery–Åsberg Depression Rating Scale (MADRS) Reference Montgomery and Åsberg33,Reference Müller, Himmerich, Kienzle and Szegedi34 . Group 3 comprised healthy controls. Exclusion criteria for the subclinical mixed psychopathology group were a current psychiatric treatment and/or a significant need for care. Finally, healthy controls (n = 44) were recruited based on a MADRS score <10, CAPE positive subscale frequency score <10 and CAPE positive subscale distress score <2, no current and/or lifetime diagnosis on the MINI and no current and/or lifetime psychological or psychiatric treatment. General exclusion criteria for all groups were left-handedness, a history of neurological disorder and magnetic resonance imaging (MRI) contraindications.
Previous publications of this dataset report on behavioural ratings and region-based fMRI analysis of the fear-conditioning task, in the healthy control and spider phobia groups. Reference Lange, Goossens, Bakker, Michielse, Marcelis and Wichers29
The Medical Ethics Committee of Maastricht University Medical Centre approved all study procedures (protocol number: NL41929.068.12/METC 12-2-072). All participants provided written informed consent, as did a parent when participants were <18 years of age.
Psychopathology measures
Spielberger‘s Stait-Trait Anxiety Inventory
Spielberger‘s Stait-Trait Anxiety Inventory (STAI) is a self-report questionnaire that measures trait anxiety. Reference Spielberger35 Each trait has 20 items, with total scores varying between 0 and 80.
Inventory of Depressive Symptomatology – Self-Report
The Inventory of Depressive Symptomatology – Self-Report (IDS-SR) is a self-report questionnaire that measures depressive symptoms over the past week. Reference Rush, Carmody and Reimitz36 It has 30 items, with total scores varying between 0 and 84.
Fear of Spiders Questionnaire
The Fear of Spiders Questionnaire (FSQ) is a self-report questionnaire measuring current symptoms of spider phobia. The questionnaire has 18 items, with total scores between 18 and 126. Reference Szymanski and O’Donohue37 Higher scores reflect more symptoms of phobia.
Experimental design
Task procedure
All participants underwent a 3-day fear conditioning and extinction paradigm in an MRI scanner. The current study focuses on the data of the first day, comprising the habituation and acquisition phase. Thus, here only details of the paradigm relevant to the current analysis are reported. Details of the remainder of the paradigm can be found in a previously published paper. Reference Lange, Goossens, Bakker, Michielse, Marcelis and Wichers29
The fear-conditioning paradigm was presented via the Presentation software package (Neurobehavioral Systems, Inc., Albany, California, USA; used on the Windows platform; https://www.neurobs.com). A circle or a rectangle was presented as the conditioned threat stimulus (CS+), and a triangle served as the neutral ‘safe’ stimulus (CS−). An electrical shock (200 ms) delivered to the left ankle (BioPac STM100C stimulator module with a STIMSOC isolation unit; BioPac Systems, Goleta, California, USA; used on the Windows platform; https://www.biopac.com) served as intrinsically salient unconditioned stimulus. Intensity of the unconditioned stimulus was individually calibrated at the start, to reach an intensity that was perceived as ‘highly annoying but not painful’.
In all phases, stimuli were presented for 4.4 s, with an inter-stimulus interval of either 2.2 or 4.4 s. Between trials, a fixation cross was presented. The order of stimuli was quasi-random to ensure that no more than two consecutive presentations of the same stimulus occurred.
In the habituation phase, all stimuli were presented 12 times without the unconditioned stimulus. Next to the CS+ and CS−, six rectangles/circles were used as generalisation stimuli; these stimuli were the same shape as the CS+, but parametrically differed in size. These were considered as conditions of no interest in the current analysis.
In the acquisition phase, the CS− was still presented without the unconditioned stimulus, whereas the CS+ was co-terminated with the unconditioned stimulus (following a 66% reinforcement schedule). Again, the CS+ and CS− were presented 12 times. The first four presentation trials of each stimulus were considered as early trials, the next four as middle trials and the last four trials were considered as late trials.
Behavioural ratings
Before and after the habituation and acquisition phase, participants rated the fear of each stimulus on a visual analogue scale ranging from 0 (not fearful) to 100 (very fearful). Additionally, arousal and valence ratings were given for each stimulus before and after the habituation and acquisition phase.
MRI acquisition and preprocessing
Details on the MRI acquisition and preprocessing can be found in the Supplementary Material available at https://doi.org/10.1192/bjp.2026.10718.
Data analysis
Potential differences in age, trait anxiety and depression levels between the three groups were assessed with a one-way analysis of variance (ANOVA). In case of significant F-tests, follow-up t-tests were performed to identify which specific groups differed.
MRI data analysis
Preprocessed data were analysed with the CONN toolbox (release 21.a; https://web.conn-toolbox.org), Reference Nieto-Castanon and Whitfield-Gabrieli38 within MATLAB (R2020b; The MathWorks, Inc., Natick, Massachusetts, USA; https://www.mathworks.com) running on the Windows platform. To analyse functional connectivity changes in relation to the task, we conducted whole-brain generalised psychophysiological interaction analysis (gPPI) Reference McLaren, Ries, Xu and Johnson39 , using a regressor for each task condition (CS+_Habituation, CS−_Habituation, CS+_Early, CS+_Mid, CS+_Late, CS−_Early, CS−_Mid, CS−_Late; as well as the conditions that are not of interest in the current study) and six motion regressors as covariates of no interest. Regions of interest (ROIs) were defined based on the most detailed version of the Shen et al Reference Shen, Tokoglu, Papademetris and Constable40 parcellation scheme (i.e. 278 subcortical and cortical grey matter ROIs). An average blood-oxygenation-level-dependent time-series was computed per ROI, and subsequently used as seed in the ROI-to-ROI gPPI analysis. The gPPI analysis yielded ROI×ROI matrices of the beta-weights reflecting the task-modulated ROI-to-ROI connectivity of each task condition for each subject. In subsequent analyses, only ROIs within the DMN, salience network and ECN were considered, following the organisation adopted by Markett et al. Reference Markett, Jawinski, Kirsch and Gerchen41 A detailed overview of the regions used in the current analysis can be found in Supplementary Table 1.
Mean connectivity
First, to assess how within-network connectivity differs between the networks between the groups and over the course of acquisition for the CS+ and the CS−, mean beta-weights of all ROI-to-ROI connectivity were computed within each network (DMN, salience network and ECN) and for each task condition. Before computing the mean beta-weights, the matrices were made symmetrical by taking the mean of A-to-B and B-to-A beta-weights, to refrain from interpreting directionality. Next, a network (three levels: DMN, salience network, ECN)×stimulus type (two levels: CS+, CS−)×task phase (four levels: habituation phase, early, mid, late)×group (three levels: healthy controls, spider phobia, subclinical mixed psychopathology) repeated-measures ANOVA was performed. In case of significant F-tests, follow-up t-tests were performed. p-Values of those t-tests were Bonferroni corrected.
ROI-to-ROI connectivity
Second, to assess specific changes in ROI-to-ROI connectivity within and between networks in more detail, we used general linear models Reference Nieto-Castanon42 to identify specific connections that show a stimulus type×task phase×group interaction. An exploratory analysis of the stimulus type×group interaction is reported in Supplementary Fig. 1.
Significance test
ROI-to-ROI gPPI analyses were conducted across 107 predefined ROIs organised into three large-scale networks (salience network, ECN, DMN). All possible ROI-to-ROI connections were included in the search space. Group-level gPPI analysis were considered significant with an uncorrected connection threshold of p < 0.01, and a false discovery rate-adjusted p-value (pFDR) of <0.05 cluster-size threshold. Reference Chumbley, Worsley, Flandin and Friston43 Thus, the analysed ROI-to-ROI connections were tested individually for significance, considering the pFDR as <0.05. Subsequently, cluster-level inference was performed in connection space. Only ROI-to-ROI connections that formed a cluster of logically adjacent connections were considered significant, meaning that only clusters of connections within one of the networks (salience network, ECN or DMN) or between a pair of networks (e.g. a cluster of multiple connections between the salience network and ECN) were considered as significant. Because of the exploratory nature of this analysis, we consider this somewhat liberal connection threshold with the follow-up correction of the cluster threshold to be valid for identifying effects. Reference McLaren, Ries, Xu and Johnson39
Results
Participants
Table 1 provides an overview of participant characteristics. The healthy control and spider phobia groups did not differ in age, trait anxiety and depression levels. The subclinical mixed psychopathology group did not differ from the healthy control and spider phobia groups in age; however, the anxiety and depression levels were higher in the subclinical mixed psychopathology group (compared with healthy controls: trait anxiety (STAI): t(94) = 9.575, p < 0.001; depression (IDS-SR): t(94) = 10.207, p < 0.001; compared with spider phobia: trait anxiety (STAI): t(94) = 9.683, p < 0.001; depression (IDS-SR): t(94) = 7.781, p < 0.001). The FSQ score differed between the three groups (spider phobia versus subclinical mixed psychopathology: t(94) = 10.68, p < 0.001; spider phobia versus healthy controls: t(88) = 22.98, p < 0.001; subclinical mixed psychopathology versus healthy controls: t(92) = 3.57, p < 0.001).
Overview of demographic and clinical variables for all participants, separately for the healthy control, spider phobia and subclinical mixed psychopathology groups
STAI, Stait-Trait Anxiety Inventory; IDS-SR, Inventory of Depressive Symptomatology – Self-Report; FSQ, Fear of Spiders Questionnaire.
A high educational level corresponds to Bachelor‘s level or higher; a mid-educational level to vocational education or higher.
Manipulation checks
Repeated-measures stimulus type (CS+ versus CS−)×task phase (habituation versus acquisition)×group (healthy controls, spider phobia and subclinical mixed psychopathology) ANOVAs were performed for the three behavioural ratings (fear, valence and arousal visual analogue scales), to check whether participants learned to differentiate between the CS+ and CS− stimuli after the fear acquisition phase and whether this differed per group (illustrated in Fig. 1). As expected, a stimulus type×task phase interaction was found for all behavioural measures (fear: F(1,131) = 455.57, p < 0.001; valence: F(1,131) = 347.04, p < 0.001; arousal: F(1,131) = 394.36, p < 0.001). This was the same in each group, as there were no significant stimulus type×task phase×group interactions (ps > 0.283). Follow-up t-tests confirmed that the CS+ and CS− ratings changed differently from the habituation to the acquisition phase (paired-samples t-test comparing CS+_Acquisition − CS+_Habituation versus CS−_Acquisition – CS−_Habituation: fear: t(133) = 21.38, p < 0.001; valence: t(133) = −18.71, p < 0.001; arousal: t(133) = 19.95, p < 0.001).
Behavioural ratings given after the habituation and fear acquisition phase. Mean fear, arousal and valence ratings are represented per group, and error bars represent the standard error of the mean. CS+, conditioned threat stimulus; CS−, neutral ‘safe’ stimulus; SMP, subclinical mixed psychopathology.
Mean connectivity
The network (three levels: DMN, salience network, ECN)×stimulus type (two levels: CS+, CS−)×task phase (four levels: habituation, early, mid, late)×group (three levels: healthy controls, spider phobia, subclinical mixed psychopathology repeated-measures ANOVA revealed a significant network×stimulus type×task phase three-way interaction (F(6, 792) = 4.170, p < 0.001), indicating that the difference in functional connectivity during CS+ and CS− presentations changed distinctively in the three networks over the course of acquisition (Fig. 2). There was no effect of group. Follow-up t-tests were performed to compare the difference in functional connectivity during CS+ and CS− (CS+ − CS−) between each task phase, separately for the networks. This analysis (with a Bonferroni-corrected threshold of p = 0.05/15 = 0.003) revealed that the difference between CS+ and CS− changed significantly over the course of acquisition only for the DMN, specifically when comparing the early with the late phase (t(139) = 3.4457, p < 0.001). In the early phase, the mean connectivity is stronger in response to the CS+ compared with the CS− (mean difference (CS+ − CS−) = 0.0484), whereas during the late phase, this is reversed (mean difference (CS+ − CS−) = −0.0622).
Mean connectivity strength per network, task phase and stimulus type. Mean connectivity strength is computed as the mean beta-weights from the first-level gPPI analysis, with the beta-weight representing the change in ROI-to-ROI synchronisation in response to a certain task condition. Error bars represent the standard error of the mean. CS+, conditioned threat stimulus; CS−, neutral ‘safe’ stimulus; DMN, default mode network; ECN, executive control network; ROI, region of interest. *p < 0.001.
Fig. 2 Long description
Panel A: A line graph depicts the mean connectivity strength of the default mode network (DMN) across different task phases: Habituation, Early, Mid, and Late. The y-axis represents mean connectivity strength, ranging from -0.14 to 0.04. The x-axis represents the task phase. Two lines are shown: one for the conditioned threat stimulus (CS+) in yellow and one for the neutral safe stimulus (CS-) in purple. The graph shows a notable decrease in connectivity strength for both stimuli during the Early phase, with a significant difference indicated by an asterisk. Panel B: A line graph depicts the mean connectivity strength of the salience network across the same task phases. The y-axis represents mean connectivity strength, ranging from -0.1 to 0.04. The x-axis represents the task phase. The lines for CS+ and CS- show minor fluctuations but generally remain close to zero. Panel C: A line graph depicts the mean connectivity strength of the executive control network (ECN) across the same task phases. The y-axis represents mean connectivity strength, ranging from -0.1 to 0.04. The x-axis represents the task phase. The lines for CS+ and CS- show a gradual increase in connectivity strength from Habituation to Late phase.
ROI-to-ROI connectivity
We found a cluster of connections within the DMN showing a significant stimulus type×task phase×group interaction effect (F(27,377.39) = 2.01, pFDR = 0.014) (Fig. 3 and Supplementary Fig. 2). This cluster comprised (sub)regions of the bilateral precuneus and cingulate, right cuneus and left cingulate. Follow-up t-tests showed, for the healthy control group, increased anti-correlation within the DMN in response to the CS− compared with the CS+ trials at the early phase of acquisition (t = 0.024, p = 0.022). In the late phase of acquisition, the opposite is found, with increased anti-correlation within the DMN in response to the CS+ compared with the CS− trials (t = −0.391, p < 0.001). In the spider phobia group, the same pattern can be observed at the early phase of acquisition, with increased anti-correlation within the DMN in response to CS− compared with CS+ trials (t = 0.011, p = 0.033). In the middle phase, this pattern reverses (t = −0.222, p = 0.042), similar to what is seen in the late phase in the healthy control group. In the late phase, we found no difference between the two stimuli. In contrast, the subclinical mixed psychopathology group showed no significant differences between the stimuli at any measurement point over the course of acquisition.
Significantly altered ROI-to-ROI functional connectivity by the interaction of stimulus type×task phase×group. All identified connections are within the default mode network. (a) Left view significant connections. (b) Right view. (c) Superior view. (d) Mean beta-weights of the significant cluster, for the different stimulus types per task phase in the three groups. Error bars represent the standard error of the mean. CS+, conditioned threat stimulus; CS−, neutral ‘safe’ stimulus; ROI, region of interest. *p <0.05.
The differential DMN connectivity at the end of the fear conditioning was not correlated with measures of trait anxiety nor with stimulus-related anxiety (details of this analysis can be found in the Supplementary Material). No connections within the ECN or salience network, or connections between the networks, showed a significant stimulus type×task phase×group interaction effect.
Discussion
To unravel the behaviour of large-scale brain networks in relation to the onset of clinical anxiety, the current study assessed functional connectivity changes within and between the salience network, ECN and DMN over the course of a fear-conditioning paradigm in healthy young adults and young adults with spider phobia and mild mixed psychopathology. We mainly identified differences within the DMN in the dynamic fear learning process. In the whole sample, mean functional connectivity within the DMN significantly changed in response to a stimulus that predicts a threat (CS+) compared with a safety outcome (CS−) over the course of fear conditioning. Specifically, we observed an initial decreased mean connectivity for both CS+ and CS−, which increased again over time for the CS−, but not the CS+. Moreover, when exploring specific connections within the DMN, we could highlight differences between the groups over the course of acquisition. By the end of the acquisition phase, healthy controls displayed significantly different DMN connectivity in response to the CS+ compared with the CS−, whereas this differentiation was absent in the group with mixed psychopathology. In the participants with a spider phobia, this differential DMN connectivity was seen already in the middle phase of acquisition.
The DMN presumably plays a role in the generation of internal thoughts. Reference Andrews-Hanna, Smallwood and Spreng44 A threat-predicting stimulus, such as the CS+, may disturb the generation of these internally focused thoughts, resulting in decreased connectivity. Several recent findings support this hypothesis. For example, the DMN has shown a negative generalisation gradient, Reference Berg, Ma, Rueter, Kaczkurkin, Burton and DeYoung45 with most deactivation in response to the CS+, and increasing activation when stimuli are more similar to the CS−. Similar to the results of Marstaller et al, Reference Marstaller, Fynes‐Clinton, Burianová and Reutens16 we found the same difference between threat and safety stimuli with more anti-correlation within the DMN for CS+ compared with the CS−. However, this difference only became apparent at the late phase of acquisition in the current analysis, when the associations with the CS− and CS+ are established. This effect of learned associations became apparent in another recent study as well, Reference Wen, Chen and Milad46 which found an increase in DMN connectivity for the CS+ toward the end of an extinction task. Taking these results together further supports the hypothesis that a stimulus predicting an aversive outcome disturbs DMN-related processes, since the predictive value of the CS+ increases over the course of acquisition and decreases over the course of extinction.
However, when looking more in detail to the functional connectivity of the DMN, young adults with subclinical mixed psychopathology, who reported the highest levels of anxiety show a different connectivity pattern over time. The altered DMN connectivity in this subclinical mixed psychopathology group, as identified in our results, lines up with a growing body of research indicating an important role of the DMN in affective disorders. Functional connectivity of the DMN was found to be altered in patients with social anxiety disorder, Reference Hahn, Stein, Windischberger, Weissenbacher, Spindelegger and Moser47 late-life anxious depression Reference Andrews-Hanna, Smallwood and Spreng44 and post-traumatic stress disorder. Reference Barredo, Aiken, van‘t Wout-Frank, Greenberg, Carpenter and Philip48 Patients with anxiety disorder have been shown to have more difficulty in learning that the CS− signals safety. Reference Duits, Cath, Lissek, Hox, Hamm and Engelhard7 In the current results, the connectivity in the DMN is indeed affected similarly when presenting safe and threatening stimuli in the subclinical mixed psychopathology group. Although the results in the healthy control group suggest that the participants learned that the CS− is safe, perhaps allowing for task-unrelated processes to take place in the brain, this change does not happen in the subclinical mixed psychopathology group. Enhancing DMN connectivity in response to the safe CS−, possibly by using transcranial magnetic stimulation or fMRI-guided neurofeedback, Reference Krause, Kogias, Krentz, Lührs, Goebel and Hermans4,Reference Krause, Linden and Hermans5 could offer a potential new intervention. Feasibility of neurofeedback-induced changes in DMN connectivity has been shown by several recent studies. Reference Zweerings, Hummel, Keller, Zvyagintsev, Schneider and Klasen49,Reference Zhang, Raya, Morfini, Urban, Pagliaccio and Yendiki50
The connectivity in the group of people with a spider phobia changed similarly over the course of acquisition as that in the healthy controls. This is not surprising, since specific phobia is characterised by a high level of fear toward a very specific stimulus (spiders in this case) and not by a generally elevated level of anxiety. This is illustrated in the present study by the low generalised anxiety scores (STAI) that did not differ between individuals with a spider phobia and healthy controls. Consequently, it could be hypothesised that the functional connectivity over the course of acquisition is related to the individual level of anxiety, as in the current sample, the subclinical mixed psychopathology group reported higher levels of anxiety compared with the spider phobia and healthy control groups. However, further exploration showed neither a correlation between the observed connectivity difference between the CS+ and CS− with proxies of the task-induced anxiety levels (as measured with stimulus-specific fear, valence and arousal ratings, described in the Supplementary Material), nor with trait anxiety. This mismatch between neural and self-reported behavioural measures in relation to fear and anxiety is consistent with the two-circuit view of threat processing, which posits that subjective experiences of fear often do not correlate well with underlying physiological and neural defensive responses. Reference LeDoux and Brown51 Similar dissociations have been reported in interventions successfully altering threat-related neural activity without conscious awareness. Reference Taschereau-Dumouchel, Cortese, Chiba, Knotts, Kawato and Lau52 Consequently, the observed DMN connectivity changes in our study may reflect implicit, non-conscious threat processing rather than the explicit generation of subjective fear. This suggests that other behavioural outcomes could potentially be linked to the observed connectivity results. For example, intolerance of uncertainty – a transdiagnostic vulnerability factor across anxiety disorders Reference Mertens, De Wolf, Bouwman and Engelhard53 – might better capture the behavioural relevance of these network dynamics. Future studies should elaborate on what other behavioural variables might be related to the lack of stimuli differentiation in DMN connectivity.
Limitations
A few considerations should be taken into account when interpreting the current results. First, the gPPI analysis is based on a regression analysis, with a directed ROI-to-ROI matrix as output. Although we were interested in changes in functional connectivity within and between the salience network, ECN and DMN throughout fear learning, which can be tested with our gPPI analysis, we refrained from analysing and interpreting this directionality. Second, the current paradigm contains a limited number of trials per condition, which is not ideal for functional connectivity analysis. Reference Masharipov, Knyazeva, Korotkov, Cherednichenko and Kireev54 Because of the rapid nature of the temporal dynamics of neural responses during fear conditioning, Reference Lonsdorf, Menz, Andreatta, Fullana, Golkar and Haaker55,Reference Wen, Raio, Pace-Schott, Lazar, LeDoux and Phelps56 we deliberately chose to split the acquisition in an early, mid and late phase so as to not wash out transient responses. However, this decision could limit the power of the current results. To validate our results, we performed a secondary analysis, addressing the same research question with an alternative method. Specifically, we supplemented our analysis with a gPPI analysis with deconvolution applied ( Reference Masharipov, Knyazeva, Korotkov, Cherednichenko and Kireev54 compared with using convolved data in our primary analysis), which yielded similar results (Supplementary Material). Future studies could address the current results in data-sets optimised for functional connectivity analysis; however, increasing the number of trials per task phase will remain challenging, if not impossible, to combine with the rapid acquiring of conditioned associations.
Clinical implications
Taken together, the observed group differences in DMN dynamics during fear acquisition suggest that vulnerability to maladaptive fear might be linked to altered intrinsic network functioning. Importantly, this is the first study to examine these dynamic network patterns in a young adult sample, a critical developmental period for the emergence of anxiety-related vulnerabilities. These findings point to an underappreciated aspect of fear learning; namely, dynamic changes in DMN configuration that are not captured by static or region-based approaches, and advance a shift from region-centric to network-dynamic models. Together with a growing body of research identifying altered DMN functioning in anxiety, the current results suggest that the DMN could be a promising research target for innovative neuromodulation interventions such as transcranial magnetic stimulation or neurofeedback.
Supplementary material
The supplementary material is available online at https://doi.org/10.1192/bjp.2026.10718
Data availability
Raw data were generated at Maastricht University. Derived data supporting the findings of this study and the analytic code are available from the corresponding author, J.H.d.V., on request.
Author contributions
J.H.d.V. contributed to study conceptualisation, methodology, formal analysis, and writing the original draft of the manuscript. S.B. and L.G. contributed to study conceptualisation, methodology and reviewing and editing the manuscript. F.K. contributed to study methodology and reviewing and editing the manuscript. I.L., J.B. and S.M. contributed to study conceptualisation, methodology, investigation and reviewing and editing the manuscript. N.K.L. and D.E.J.L. contributed to study supervision and reviewing and editing the manuscript. M.M., J.v.O. and T.v.A. contributed to reviewing and editing the manuscript. K.R.J.S. contributed to study conceptualisation, methodology, supervision and reviewing and editing the manuscript.
Funding
This study was funded by a research grant from Stichting De Weijerhorst awarded to J.v.O. Open access funding provided by Maastricht University.
Declaration of interest
None.
Open access
eLetters
No eLetters have been published for this article.