Longitudinal development of resting-state functional connectivity in adolescents with and without internalizing disorders

Abstract

Longitudinal studies using resting-state functional magnetic resonance imaging (rs-fMRI) focused on adolescent internalizing psychopathology are scarce and have mostly investigated standardized treatment effects on functional connectivity (FC) of the full amygdala. The role of amygdala subregions and large resting-state networks had yet to be elucidated, and treatment is in practice often personalized. Here, longitudinal FC development of amygdala subregions and whole-brain networks are investigated in a clinically representative sample.

Treatment-naïve adolescents with clinical depression and comorbid anxiety who started care-as-usual (n = 23; INT) and healthy controls (n = 24; HC) participated in rs-fMRI scans and questionnaires at baseline (before treatment) and after three months. Changes between and within groups over time in FC of the laterobasal amygdala (LBA), centromedial amygdala (CMA) and whole-brain networks derived from independent component analysis (ICA) were investigated.

Groups differed significantly in FC development of the right LBA to the postcentral gyrus and the left LBA to the frontal pole. Within INT, FC to the frontal pole and postcentral gyrus changed over time while changes in FC of the right LBA were also linked to symptom change. No significant interactions were observed when considering FC from CMA bilateral seeds or within ICA-derived networks.

Results in this cohort suggest divergent longitudinal development of FC from bilateral LBA subregions in adolescents with internalizing disorders compared to healthy peers, possibly reflecting nonspecific treatment effects. Moreover, associations were found with symptom change. These results highlight the importance of differentiation of amygdala subregions in neuroimaging research in adolescents.

1. Introduction

Internalizing disorders like anxiety and depression often start in adolescence and are highly prevalent in this stage of life. Sadly, recent studies imply that comorbid anxiety disorders are rather the rule than the exception in clinically depressed adolescents, as more than 50% of adolescents with a clinical depression have one or more comorbid anxiety disorders (Melton et al., 2016). These adolescents often display more severe symptoms, poorer response to treatment and increased risk of suicidal behaviors compared to clinically depressed adolescents without comorbid psychopathology (Melton et al., 2016). Models suggest that complex interactions between environment, genes, neurobiological characteristics, and the timing of developmental stages contribute to the development of these mental illnesses (Toenders et al., 2019; Swartz et al., 2014; Bas-et al., 2020). To attain insights in neurobiological underpinnings of adolescent internalizing disorders, investigation of function and structure of neural networks is vital.

Activity of neural networks at rest, also called functional connectivity (FC), is widely investigated by use of resting-state functional MRI (rs-fMRI) (Buckner et al., 2011; Yeo et al., 2011). Rs-fMRI focuses on spontaneous, low-frequency fluctuations (<0.1 Hz) in the blood oxygenation level-dependent (BOLD) signal that occur in the absence of a task or stimulus, which can be used to identify functional neural networks (Cole et al., 2010; Lee et al., 2013; Tomasi et al., 2011). Two common methods to identify these networks are seed-based correlation analysis (SCA), a hypothesis-driven approach to investigate the primary signal of a region-of-interest (ROI) to the whole brain (Cole et al., 2010), and independent component analysis (ICA), a fully data-driven, exploratory approach to extract resting-state networks using a multivariate exploratory analysis, which divides the brain into independent spatial regions according to their signal fluctuations (Beckmann, 2012). To our knowledge, fully data driven approaches are still scarce in current literature despite their advantages, like being unbiased to a priori selected regions. When comparing SCA and ICA, secondary signals expressed by the brain are missed in SCA (Cole et al., 2010), whereas distinct signals occurring in a particular region can go unnoticed due to the limited number of components that is being extracted in ICA (Beckmann, 2012). Therefore, SCA and ICA can be seen as complementary methods to investigate FC of neural networks and brain regions, such as the amygdala.

As one of the most extensively investigated brain regions in humans and animals, the function and structure of the amygdala is quite well known (Duvarci et al., 2014; Janak et al., 2015). It consists of several subregions, like the laterobasal amygdala (LBA), involved in regulation of fear and the perception hereof, and centromedial amygdala (CMA), mainly concerned with acute stress-reactions (Duvarci et al., 2014; Janak et al., 2015; LeDoux, 2007). Recently, it has been hypothesized that these subregions each contribute to different aspects of mental disorders (Janak et al., 2015; Michely et al., 2020), but their role in adolescent internalizing disorders has yet to be elucidated. Involvement of either or both subregions is likely, as meta-analyses of cross-sectional studies reported changes of full amygdala FC in depressed and anxious adolescents and adults when compared to healthy peers (Tang et al., 2018; Zugman et al., 2023), such as hypoconnectivity between the amygdala and the medial PFC (Zugman et al., 2023), and decreased amygdala FC within the executive control network, which has been found in regulating emotions and initiating goal-directed responses (Miller et al., 2001) although a recent meta-analysis did not replicate these findings in depressed adolescents (Tse et al., 2023). These alterations may underly the characteristic imbalance between affective-cognitive networks in adolescent internalizing disorders (Casey et al., 2008; Crone, 2009; Vink et al., 2014). To expand the understanding of the course of these alterations, development of amygdala FC and its subregions needs to be investigated. Unfortunately, longitudinal studies in this population are scarce and have reported inconsistent and isolated findings on full amygdala FC (Toenders et al., 2019; Baumel et al., 2022; Chattopadhyay et al., 2017; Straub et al., 2017; Villa et al., 2020; Klimes-Dougan et al., 2022; Cullen et al., 2016), like increased FC of the amygdala to the right medial and middle frontal gyrus and decreased FC of the amygdala and the right posterior cingulate cortex, which were associated with decreases in depressive symptoms (Chattopadhyay et al., 2017; Cullen et al., 2016). These studies suggest a normalizing or compensatory effect of treatment on amygdala FC (Toenders et al., 2019; Baumel et al., 2022; Chattopadhyay et al., 2017; Straub et al., 2017; Villa et al., 2020; Klimes-Dougan et al., 2022; Cullen et al., 2016), although this must be interpreted cautiously due to the scarcity of studies and their inconsistent, isolated results. Altogether, even though longitudinal studies are inconclusive, alterations in FC of amygdala subregions are highly suspect to be involved in adolescent internalizing disorders.

Current models suggest that several whole-brain resting-state networks are involved in internalizing disorders as well (for a comprehensive overview, see Table S2 and Fig. 1 in (Bas-et al., 2022)). Cross-sectional alterations in FC within and between these networks have been reported in depressed adolescents and anxious adults when compared to healthy peers, such inter- and intranetwork alterations in the limbic, default mode, central executive, and salience networks (Tang et al., 2018; Tse et al., 2023; Xu et al., 2019). It is therefore likely that, in addition to the changes in amygdala FC, complex and widely spread changes in FC of other networks are involved in the development and course of adolescent internalizing disorders.

Fig. 1.

Different development of LBA FC in adolescents with internalizing disorders compared to healthy peers.

(a) FC of networks derived from data-driven ICA analysis. (b) Positive (red) and negative (blue) FC of networks derived from SCA using amygdala subregions as seeds (green). (c) Significant group × time interaction in left LBA FC to the frontal pole (INT (T2 > T1) > HC (T2 > T1)), coordinates of displayed slices (MNI, x, y, z): 0, 50, −6. (d) Significant group × time interaction in right LBA FC to the postcentral gyrus (INT (T2 > T1) > HC (T2 > T1)), coordinates of displayed slices (MNI, x, y, z): 60, −12, 18. (e) Distribution of mean z-scores for INT and HC on baseline and after three months in the significant cluster in the frontal pole. Dots represent individual z-scores. (f) Distribution of mean z-scores for INT and HC on baseline and after three months in the significant cluster in the postcentral gyrus. Dots represent individual z-scores. Images are displayed according to radiological convention: right in image is left in the brain. LBA: laterobasal amygdala; INT: internalizing group; HC: healthy controls; MNI: Montreal Neurological Institute. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Besides the need to explore the longitudinal trajectory of FC of these regions and networks in this population, outcomes of prior longitudinal studies on FC development in adolescents with internalizing disorders may not reflect practical applicability in clinical settings, as they predominantly investigated treatment effects using standardized protocols (Chattopadhyay et al., 2017; Straub et al., 2017; Villa et al., 2020; Klimes-Dougan et al., 2022; Cullen et al., 2016). In clinical practice, however, especially adolescents typically receive more personalized treatments tailored to their individual needs, adjusted to their level of emotional and cognitive development and in collaboration with their family (Singh et al., 2018; Walter et al., 2020, 2023).

To gain more insight in the development of neural networks, we have conducted the Emotional Pathways’ Imaging Study in Clinical Adolescents (EPISCA) study (van den Bulk et al., 2013). In EPISCA, we used a study design in which participants received care-as-usual and were followed over the course of six months without study interventions, thus reflecting real world development of neural networks in a clinically heterogeneous group as close as possible. Using this longitudinal dataset, we have previously demonstrated that baseline measures of white matter microstructure are associated with changes in depressive symptoms over time in adolescents with internalizing disorders (Roelofs et al., 2022).

In the present study, FC in adolescents with internalizing disorders and healthy controls over a period of three months was investigated to obtain more insight in the development of FC of amygdala subregions and whole brain networks between and within these groups, possibly reflecting treatment effects. Based on previous findings, we expected FC differences between groups over time in regions involved in emotion processing, such as limbic structures, and emotion regulation, like frontal regions.

2. Methods

2.1. Participants

Data for this study were derived from the EPISCA study (van den Bulk et al., 2013). Briefly, two groups were investigated over time and in both groups, the majority of participants identified themselves as being female. The first group consisted of treatment-naïve adolescents with clinical depression and comorbid anxiety (internalizing disorders group, ‘INT’, n = 30 at baseline). They were included in the study before start of care-as-usual and were diagnosed with clinical depression and at least one clinical anxiety disorder as assessed by categorical measures of DSM-IV depressive or anxiety disorders. The second group consisted of healthy control peers (HC, n = 32 at baseline). All participants were investigated three times in a six-month timespan, during which the INT group received personalized variants of cognitive behavioral therapy (CBT) and family-based interventions. Extensive information about recruitment of participants, in- and exclusion criteria and demographic data of the full sample can be found in previous publications (van den Bulk et al., 2013; Roelofs et al., 2022; Pannekoek et al., 2014a; Pannekoek et al., 2014b; Aghajani et al., 2014; van den Bulk et al., 2016; van Hoof et al., 2017; van den Bulk et al., 2014; van Hoof et al., 2019; Riem et al., 2019; Rinne et al., 2016; Aghajani et al., 2016; van Hoof et al., 2015; Rinne et al., 2020; Rinne et al., 2017; van et al., 2020; de et al., 2019; Wang et al., 2021; Dennis et al., 2019; Han et al., 2021; Zhu et al., 2023; van et al., 2022; Sun et al., 2022a; Sun et al., 2022b) and in the supplemental information. Detailed information about personalized treatments, MRI data acquisition, preprocessing of the rs-fMRI data and statistical analyses of demographical data, including multiple comparison corrections, is also included in supplemental information.

In the present study, rs-fMRI data obtained at baseline (before treatment; T1) and after three months (T2) were analyzed, as loss-to-follow-up was substantial after six months (loss-to-follow-up after three months: 16%, loss-to-follow-up after six months: 24%). At baseline, two participants (n = 1 INT, n = 1 HC) were excluded due to an anomaly on the structural T1-scan. Furthermore, one HC participant was excluded as criteria for clinical psychopathology were met. Two HC participants were excluded due to technical issues and six participants (n = 4 INT, n = 2 HC) did not follow-up after the first visit, leaving data from 25 INT and 26 HC participants for initial analysis. In addition, data from several questionnaires was included if available on both timepoints. To assess severity of depressive and anxious symptoms, total scores on the Children's Depression Inventory (CDI) (Kovacs, 1985) and on the anxiety subscale of the Revised Child Anxiety and Depression Scale (RCADS) (Chorpita et al., 2000) were included. Pubertal stage was taken into account by using baseline scores on the self-report Pubertal Development Scale (PDS) (Petersen et al., 1988), according to the following categories: 1) prepubertal, 2) early pubertal, 3) midpubertal, 4) late pubertal, and 5) postpubertal.

2.2. General preprocessing of resting-state data

Image pre-processing and analyses were performed using the Oxford Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library (FSL), version 6.0.3 (RRID:SCR_002823). Data underwent several preprocessing steps following the procedures described in (Pruim et al., 2015a, 2015b; Parkes et al., 2018). First, non-brain removal was performed on T1 and high-resolution EPI data using the Brain Extraction Tool (BET) (Smith, 2002). Then, BOLD data were preprocessed using FEAT (FMRI Expert Analysis Tool) (Woolrich et al., 2001), including motion correction using MCFLIRT (Jenkinson et al., 2002), spatial smoothing using a Gaussian kernel of full-width half-maximum (FWHM) 6.0 mm and grand-mean intensity normalization of the entire 4D dataset by a single scaling factor to enable higher-level analyses and registration. Resting-state fMRI scans were first registered to high-resolution EPI images, which were registered to T1 images, which in turn were registered to the Montreal Neurological Institute (MNI) T1-template brain (resolution 2 mm) using FNIRT nonlinear registration (Jenkinson et al., 2001, 2002; Andersson et al., 2007; Jenkinson et al., 2001). These transformation matrices were then combined to obtain a native-to-MNI-space transformation matrix per participant.

Next, ICA-AROMA (ICA-based Automatic Removal Of Motion Artifacts) was used to remove motion-related artifacts (Pruim et al., 2015a, 2015b). Afterwards, native T1-weigthed images were segmented into white matter (WM), central spinal fluid (CSF) and grey matter (GM) probability maps using FSL FAST (Zhang et al., 2001). Without extensive erosion of WM and CSF masks, these signals can be correlated with the GM signal (Power et al., 2017). Therefore, we applied 2 erosion cycles for the CSF mask and 4 erosion cycles for the WM mask to prevent partial voluming. If a mask contained fewer than 5 voxels after a given erosion cycle, the previous cycle was selected. Then, these masks were linearly aligned into native resting state space using FSL's FLIRT (Jenkinson et al., 2001, 2002; Andersson et al., 2007; Jenkinson et al., 2001). BOLD data were then submitted to FEAT to perform high-pass temporal filtering (Gaussian-weighted least-squares straight line fitting, cutoff 0.01 Hz) and subsequently nuisance regression with WM and CSF masks was performed. The processed resting-state fMRI data of each participant were then warped to MNI space using the combined native-to-MNI transformation matrix.

Data quality was assessed by inspecting each participant's processed EPI data as a time series (i.e., “carpet plot”) alongside framewise displacement (FD), as calculated by FSL_motion_outliers. Participants were excluded if mean FD over time was >0.2 mm, or >20% suprathreshold FDs, or any FD > 5.0 mm (Parkes et al., 2018).

2.3. Identification of resting-state networks

Rs-fMRI data were investigated twofold using the Oxford Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library, version 6.0.3 (FSL; RRID:SCR_002823). First, SCA was used to investigate FC of two bilateral amygdala subregions, creating seeds derived from the Juelich histological atlas (Aghajani et al., 2016). In short, voxelwise bilateral regions-of-interest (ROIs) of the laterobasal amygdala (LBA) and centromedial amygdala (CMA) were created by including voxels only if the probability of the voxel belonging to either subregion was higher than 40%. Each voxel was exclusively assigned to one region, overlapping voxels were assigned to the region with the greatest probability. Using these ROIs as seeds, FC of each subregion to the whole brain was investigated. First, we explored FC of each subregion for every individual participant on each timepoint. These results, containing data from both timepoints and all participants, were subsequently used as input to generate group-level FC maps of each subregion to the whole brain, using a cluster-forming threshold of Z > 3.1 and a cluster-corrected extent threshold p < 0.05 (Fig. 1B and Table S2). The activated networks were successively binarized and used as a mask for the statistical analyses below, ensuring that analyses were confined within the respective networks.

Second, to determine group-level data-driven resting-state networks, Probabilistic Independent Component Analysis (PICA) (Beckmann et al., 2004) and dual regression analysis (Nickerson et al., 2017) were used. First, PICA was applied to separate 4D functional data into spatial maps with an associated time course for each map, starting with temporally concatenating all preprocessed subjects’ data for both timepoints in Montreal Neurological Institute (MNI)-standard space into one dataset. Subsequently, this dataset was decomposed in 20 independent components, consisting of independent vectors describing signal variation across the spatial (maps) and temporal (time-course) domain. This number of components results in a representative set of functional connectivity networks (Smith et al., 2009a) in line with previous work in our group (Bas-et al., 2021). Afterwards, dual regression was used to generate subject-level maps for each of the 20 group components. This resulted in a set of maps (one for each original group-level ICA component) that describe the network structure based on the data from each participant. The 20 spatial maps at group-level were visually inspected. Components with spatial similarity to the networks of interest were selected based on descriptions of functional networks in previous work: default mode, dorsal attention, frontoparietal (separated in a left and right component), executive control, salience and affective network (Fig. 1A and Table S2) (Smith et al., 2009a; Barkhof et al., 2014; Witt et al., 2021; Pan et al., 2017).

2.4. Statistical analysis of resting-state networks

After investigating FC of the LBA and CMA seeds (Fig. 1B) and confirming the seven networks as defined by ICA (Fig. 1A), further statistical analyses were conducted. First, differences in head motion between groups were investigated by comparing individual mean relative and absolute motion parameters created by FSL's MCFLIRT using the Mann-Whitney U test in R (Table 1; R Project for Statistical Computing (RRID:SCR_001905)) (Jenkinson et al., 2002; R Core Team R, 2023).

Table 1.

Motion parameters.

	INT (n = 23)	HC (n = 24)
Head motion (median ± SD)
mean absolute head motion T1	0.18 ± 0.10	0.16 ± 0.08
mean relative head motion T1	0.07 ± 0.02	0.08 ± 0.02
mean absolute head motion T2	0.22 ± 0.12	0.17 ± 0.19
mean relative head motion T2	0.08 ± 0.02	0.08 ± 0.02

Then, several analyses were conducted. First, SCA was applied to explore group × time interactions in connectivity of two bilateral amygdala subregions, namely the LBA and CMA. To underline the importance of investigating subregions, we conducted additional analyses using bilateral whole amygdala seeds. Second, exploratory group × time interactions were investigated in seven resting-state networks derived from ICA: default mode, frontoparietal (separated in a left and right component), executive control, dorsal attention, salience and affective network (Smith et al., 2009a; Bas-et al., 2021; Barkhof et al., 2014; Witt et al., 2021; Pan et al., 2017; Zhang, 2017). Specifically, multi-level block permutation analyses were conducted in FSL's PALM (Winkler et al., 2014, 2015) (RRID:SCR_017029) to examine voxel-wise changes in FC in the networks of interest (four networks related to bilateral amygdala subregions; seven whole-brain networks) over time between INT and HC, using 5000 permutations and threshold-free cluster enhancement (TFCE) (Smith et al., 2009b). Two contrasts were investigated in each time direction: i) group × time interaction HC (T2 > T1) > INT (T2 > T1) and ii) group × time interaction INT (T2 > T1) > HC (T2 > T1). In other words, the first interaction explores for each voxel within the functional networks whether HC had a greater change in FC after three months than INT, while the second interaction investigates whether the change in FC after three months was greater in INT compared to HC (design is provided in Table S3). Age at the time of scanning was accounted for by subject-specific regressors. To assure these assumptions, additional analyses were conducted to control the effect of age on group × time interactions by adding age at baseline as a fixed, non-time varying covariate. Two extra contrasts were added to the main design to test positive and negative effects of age on changes in FC. Results are compared to the main design in the Supplemental Text (Table S7 and Fig. S1). Correction for multiple testing considered the fact that we investigated both directions for interaction effects (Alberton et al., 2020). Analyses across seeds and hemispheres were corrected for multiple comparisons at a false discovery rate (FDR) of 5% (Benjamini et al., 1995). In case of a significant interaction, directionality of the interaction within each group was explored (Table S3) and a separate design was used to investigate whether changes in FC were associated with symptom change within INT (Table S4). Multi-level block permutation analyses were conducted in PALM with 5000 permutations and TFCE to investigate associations with changes in anxiety symptoms, as measured by the RCADS, and changes in depressive symptoms, as measured by the CDI. An F-test was added to assess their overall effects on FC change, which reflects significance without providing directionality or a joint score of both symptom scores. INT-participants were included in these association analyses if the questionnaire (RCADS and/or CDI) was filled out on baseline (before treatment) and after three months.

3. Results

3.1. Exclusion of participants

After motion control, four participants (n = 2 INT, n = 2 HC) were excluded, leaving 23 INT participants and 24 HC participants for further analysis. Motion parameters were not significantly different between INT and HC (Table 1).

3.2. Demographics

Characteristics of the final sample are summarized in Table 2. Groups did not differ in demographic measurements. As expected, patients experienced significantly higher levels of depression and anxiety on both timepoints compared to healthy controls (p < 0.000). Within INT, depression and anxiety severity significantly decreased over time (p = 0.017, p < 0.001, resp.). There were no significant symptom changes within HC.

Table 2.

Demographic characteristics of participants with and without clinical depression and comorbid anxiety.

	INT (n = 23)	HC (n = 24)
Demographics at baseline
Age in years (mean ± SD; range)	15.86 ± 1.47 (13–18)	15.12 ± 1.74 (12–20)
Male/Female (n)	2/21	3/21
IQ (mean ± SD)	105.30 ± 8.35	107.08 ± 7.92
Puberty stage (median ± SD; range) a	4.00 ± 0.73 (2–5)	4.00 ± 0.73 (3–5)
Days between visits (median ± SD)	100.00 ± 11.76	98.00 ± 12.59
Ethnicity (% white)	91.30	95.83
Primary diagnosesb(n)
MDD + GAD	15
MDD + SAD	13
MDD + specific phobia	8
MDD + panic disorder	4
MDD + separation anxiety	7
Secondary (externalizing) diagnosesb(n)
ADHD; inattention	1
ADHD; hyperactive	1
Oppositional Defiant Disorder	1
Behavioral Disorder	1
Self-report measures (median ± SD)
CDI on baseline c	20.00 ± 9.20	4.00 ± 3.62
CDI after three months	11.50 ± 9.40	2.00 ± 3.31
RCADS on baseline	33.0 ± 14.70	10.00 ± 11.14
RCADS after three months d	25.00 ± 15.56	10.00 ± 8.60

MDD: Major Depressive Disorder; GAD: Generalized Anxiety Disorder; SAD: Social Anxiety Disorder; ADHD: Attention Deficit Hyperactivity Disorder; CDI: Children's Depression Inventory; RCADS: Revised Child Anxiety and Depression Scale; SD: Standard Deviation.

^aDiagnoses are not mutually exclusive.

^bThree healthy controls and two patients did not complete the questionnaire.

^cOne patient did not complete the questionnaire.

^dTwo patients did not complete the questionnaire.

3.3. Voxelwise changes in FC over time between patients and controls

Significant group × time interactions were found in FC of the left and right LBA (Table 3; Fig. 1C/1D). Specifically, development of FC between the left LBA and frontal pole (Fig. 1C; Table 3) was significantly different between INT and HC. Furthermore, significant differences were found in development of FC between the right LBA and right postcentral gyrus between INT and HC (Fig. 1D; Table 3). However, results did not stand when corrections for multiple testing were applied. No significant interactions were observed within the ICA-defined networks or in FC of the CMA or whole amygdala bilateral seeds.

Table 3.

Coordinates of significant clusters in group × time interaction.

	Voxels	p-value	MNI coordinates			Regions involved
			X	Y	Z
Left LBA seed	581	0.003	0	50	−6	Frontal Pole; Frontal Medial Cortex; Paracingulate Gyrus; Anterior Cingulate Gyrus
Right LBA seed	74	0.021	60	−12	18	Postcentral Gyrus; Central Opercular Cortex

LBA: laterobasal amygdala. Threshold-free cluster enhancement (TFCE) and family-wise error (FWE) corrected at p-values <0.05.

3.4. Voxelwise changes in FC over time within patients and controls

Additional analyses concerning the left and right LBA further revealed effects of time within INT (Table S6). Fig. 1E/1F illustrates the change in z-score between and within the groups, indicating an increase of FC in the INT group and a decrease in the HC group of the bilateral LBA seeds over time. Results were not significant at the FDR-corrected level. At the uncorrected level, FC of the left LBA significantly increased to the frontal pole (Fig. 1E) and decreased to the postcentral gyrus, but FC of the right LBA did not significantly change. FC did not significantly change over time within HC.

3.5. Associations with symptom changes in significant clusters

Within INT, associations between changes in symptom scores and changes in FC of each LBA seed were further explored. There were no significant associations at the corrected significance level, but an association was found between changes of FC between the right LBA and the postcentral gyrus and the overall effect of changes in anxious and depressive symptoms at the uncorrected level (Table S6) (p = 0.032, 30 voxels; p = 0.037, 13 voxels; p = 0.049, 3 voxels. TFCE and FWE corrected at p-values <0.05). Fig. 2 illustrates the associations between changes in symptoms and changes in FC of the right LBA, indicating an association between a decrease of symptoms and an increase of FC of right LBA to the postcentral gyrus. This association was only significant when analyzing the joint effect of anxious and depressed symptoms. Correlation analyses between changes in FC of LBA seeds and changes in anxious or depressive symptoms separately did not reveal any significant associations.

Fig. 2.

Associations between longitudinal changes in anxious and depressive symptoms and changes in functional connectivity between the right LBA and the postcentral gyrus within adolescents with depression and comorbid anxiety.

(a) Association between changes in depressive symptoms (Δ CDI) and changes in functional connectivity (Δ z-score). Changes are calculated as score after three months minus score on baseline. Dots represent individual associations between changes in z-score and changes in depressive symptoms. Individual z-scores are derived from mean timeseries extracted from significant clusters. (b) Association between changes in anxious symptoms (Δ RCADS) and changes in functional connectivity (Δ z-score). Changes are calculated as score after three months minus score on baseline. Dots represent individual associations between changes in z-score and changes in anxious symptoms. Individual z-scores are derived from mean timeseries extracted from significant clusters.

CDI: Children's Depression Inventory. RCADS: Revised Child Anxiety and Depression Scale.

4. Discussion

In this study, we investigated longitudinal development of FC in a cohort of adolescents with internalizing disorders (e.g. with clinical depression and at least one comorbid anxiety disorder) who started care-as-usual and their healthy peers over the course of three months. Specifically, FC of amygdala subregions, the whole amygdala, and large resting-state networks derived from data-driven analysis was examined. Initial evidence was found for different FC development of the left and right laterobasal amygdala (LBA) in adolescents with internalizing disorders compared to healthy peers. Moreover, change in FC of the right LBA was associated with the overall effect of changes in anxious and depressive symptoms in INT.

To start, development of FC between the left LBA and frontal pole was significantly different between patients and healthy peers (Fig. 1C). Further analyses revealed that over time, FC increased in patients but did not change in healthy peers (Fig. 1E). There were no significant associations between changes in FC of the left LBA and changes in anxiety symptoms (measured with the RCADS), depressive symptoms (measured with the CDI) or an overall effect of these symptom changes in the INT group. This difference in FC development of the left LBA to the frontal pole between INT and HC fits nicely with current models of corticolimbic network dysregulation in affective disorders, in which abnormal processing within and between regions involved in cognitive control (i.e. prefrontal cortices) and emotions (i.e. limbic structures) are thought to be part of neurobiological mechanisms underlying depression and anxiety disorders (Swartz et al., 2014; Bas-et al., 2020). The frontal pole, a distinct segment within the frontal cortex, maintains both functional and structural connections with the LBA, thereby positing its likely involvement in the regulation of affective and social-cognitive abilities (Bludau et al., 2014). Indeed, changes in function and structure of the LBA and the frontal pole have frequently been reported in cross-sectional studies on adolescent and adult internalizing disorders and thus support models concerning corticolimbic dysregulation (Toenders et al., 2019; Schmaal et al., 2017; Bludau et al., 2016). This finding is also generally in line with previous longitudinal studies, which have reported increases in FC between the entire amygdala and frontal regions over time in depressed adolescents. For example, three studies demonstrated an increase in FC of the amygdala to the bilateral prefrontal cortex (PFC) (Straub et al., 2017), left medial PFC (Villa et al., 2020) and the dorsal anterior cingulate cortex (Klimes-Dougan et al., 2022) after several weeks of psychotherapy (either five weeks or at least six sessions of CBT or 16 weeks of interpersonal psychotherapy, resp.). Interestingly, in two studies, some of these findings correlated with improvement of depression severity (Straub et al., 2017; Klimes-Dougan et al., 2022), but this result was not replicated in the study by Villa et al. (2020) nor in the present study. Another study in depressed adolescents reported an increase in amygdala FC to the right medial and middle frontal gyri after eight weeks of selective serotonin reuptake inhibitor (SSRI) treatment (Cullen et al., 2016), which was again associated with a decrease of depressive symptoms. Considering these results, we cautiously propose that the increase in FC to the frontal pole in patients could be due to a nonspecific effect of treatment, regardless of which treatment is offered. However, the control group were healthy adolescents rather than a non-treated patient group, due to ethical concerns of withholding treatment. Therefore, we cannot conclude whether this increase was due to treatment effects or non-specific effects of improvement in symptoms related to time. Nonetheless, our hypothesis is supported by the fact that previous studies showed a similar pattern of increased FC between the amygdala and frontal regions related to treatment, while all investigated different treatment methods. The lower functional coupling to regulatory regions in the frontal cortex, such as the frontal pole, at baseline could reflect aberrant psychophysiology, while the increase in FC to the frontal pole over time possibly mirrors new skills learned in treatment, although we did not find an association between this FC change and symptom levels as measured with the RCADS and CDI. We cautiously hypothesize that the changes in FC might reflect adaptive changes or nonspecific effects of treatments. Furthermore, internalizing disorders have a wide range of symptoms, which are only partly captured by questionnaires like the RCADS and CDI. Perhaps other symptoms, which are not measured by these questionnaires, would be associated with these FC changes.

Second, using the right LBA as a seed region, a significant group × time interaction between groups was found in FC to the right postcentral gyrus (Fig. 1D). Further analyses revealed no effect of time within each group (Fig. 1F). However, changes in connectivity to the postcentral gyrus were associated with the overall effect of changes in anxious and depressive symptoms within INT. Other associations with only anxious or depressive symptoms were not significant. Structural and functional studies have confirmed connectivity between the amygdala and sensorimotor regions, including the postcentral gyrus (Rizzo et al., 2018). Functionally, the amygdala-sensorimotor pathway is involved in multiple steps of emotion processing, among others emotion recognition and emotional regulation (Tomasi et al., 2011; Kropf et al., 2019; Roy et al., 2009). For example, high levels of activation in this pathway have been reported when attention is focused on emotional state and in response to unconscious perceived threats (Kropf et al., 2019). Previous findings concerning this pathway in internalizing disorders are inconsistent and the relation to clinical symptoms in adolescents is unknown. However, one study demonstrated increased FC of the amygdala as a whole to the postcentral gyrus in adolescents with depression (Mao et al., 2020) and increased grey matter volume of the postcentral gyrus has been reported in adolescents with anxious depression, but not in healthy controls or depressed adolescents (Wehry et al., 2015). Because the overall effect of symptom changes was associated with FC change in the right LBA-postcentral gyrus pathway, we posit that the amygdala-sensorimotor pathway might be especially involved in internalizing psychopathology with anxiety and depressive components, although the specific role of the LBA remains to be investigated.

Taken together, these results indicate that FC development of amygdala subregions is different in adolescents with internalizing disorders while they receive case-as-usual compared to healthy peers. Results were specifically located in FC of the LBA, a subregion connected to, among others, frontal regions, while analyses on FC of the CMA, mainly connected to the brainstem nuclei that generate behavioral and visceral correlates of acute stress-reactions (Duvarci et al., 2014; Janak et al., 2015; LeDoux, 2007), and analyses on FC of the whole amygdala did not reveal any significant interactions. Therefore, these results provide more insight in involvement of amygdala subregions in adolescent internalizing disorders and highlight the importance of differentiation between these subregions.

These findings are generally in line with previous work on longitudinal data of the EPISCA study, where we have demonstrated greater baseline cross-sectional connectivity of the uncinate fasciculus (UF), a white matter tract connecting the amygdala to the frontal cortex (Aghajani et al., 2014) and greater full amygdala FC to the dorsolateral prefrontal cortex (dlPFC) in adolescents with internalizing disorders compared to healthy peers (Pannekoek et al., 2014a), although longitudinal analyses on white matter revealed no changes in the UF (Roelofs et al., 2022).

Interestingly, we did not find any changes in FC of large resting-state networks derived from ICA. This could be due to several reasons, such as our sample size, which might be too small relative to the complexity of our data, or the time frame, which might not be long enough to detect any differences between groups over time. Our findings are in line with a large recent population-based study in adolescents with internalizing symptoms, which reported small changes in FC from ICA derived networks over four years that did not survive multiple testing correction (Dall'Aglio et al., 2023).

4.1. Strengths, limitations & future research

This study is unique in the way that it investigated FC of amygdala subregions and data-driven whole-brain networks in a heterogeneous cohort of adolescents with internalizing psychopathology. This approach offers possibilities to expand our knowledge of FC development in adolescents with internalizing disorders compared to the healthy population, while simultaneously reflecting the clinical practice in which adolescents typically receive personalized treatment, focused on their individual needs. However, a few limitations of the present study need to be mentioned. First, this study design did not use a standardized treatment protocol but followed patients over three months’ time, while they received care-as-usual; thus, it does not allow to evaluate treatment efficacy and any precise hypotheses about underlying specific reasons why FC and disease severity change in this population are relatively hard to generate. However, this study design provides valuable information about how FC and symptoms change over time, regardless of specific interventions, and is representative of clinical practice. Second, the study groups predominantly consisted of female participants, but the sex distribution was not significantly different between the clinical and healthy control group. Yet, as girls are more likely to develop internalizing psychopathology and the female sex has been associated with an increased risk of comorbid anxiety and depression, we feel our study is representative of the clinical population although the sex ratio might be more unbalanced compared to ratios reported in current literature (Costello et al., 2003; Martel, 2013). Third, due to the imbalance in sex within groups, correction for sex was not possible, even though a mediating role in FC development has been reported (Alarcon et al., 2015). Fourth, sample size and study duration were modest. In this heterogeneous group, adolescents were diagnosed with at least one out of five anxiety disorders as noted in Table 1 while receiving variants of personalized treatment as noted in Supplemental Table S1, and the sample reflected the adolescence in ages ranging from 12 to 20. In the light of the small sample size, this heterogeneous group could make interpretation of results difficult. With regards to the study duration, previous longitudinal studies in adolescents with internalizing disorders have assessed longitudinal effects of cognitive behavioral therapy (CBT) over three months and reported functional brain changes in amygdalar – prefrontal connections, which in some studies were associated with improvement of clinical symptoms (Villa et al., 2020; Connolly et al., 2017; Cyr et al., 2020; La et al., 2021). Taken together, we feel our sample represents clinical practice as close as possible and is thus indicative of the development of neural networks across adolescence in the real world, and we expected three months to be sufficient to detect any changes. Fifth, interpretation of results is limited as we did not include a clinical control group due to ethical reasons. Sixth, results did not stand after FDR correction for multiple testing across seeds and hemispheres. Lastly, information about social-economic status or illness duration was not collected and therefore could not be accounted for.

Future research could focus on several aspects. First, previous studies used different interventions and timespans, complicating comparison and generalization of the results. Therefore, a project within a large consortium like the ENIGMA-Anxiety Working Group might be interesting to pool imaging data, increase sample size and improve reliability of results within a mega- or meta-analysis (Zugman et al., 2023; Bas-et al., 2022; Zugman et al., 2022). Second, future studies should consider longitudinal investigation of amygdala subregions to further explore their unique role in adolescent internalizing disorders. Third, this study provides new insights in amygdala FC development in this population, which could lead to identifying possible new targets for innovative treatments such as fMRI-based neurofeedback (Quevedo et al., 2020; Zhang et al., 2023; Zich et al., 2020). Fourth, future studies could consider more frequent assessment in a large timeframe to detect small FC changes in ICA derived networks.

4.2. Conclusion

Concluding, this study provides initial evidence that FC development over the course of three months is different in adolescents with internalizing disorders who receive care-as-usual compared to their healthy peers, possibly reflecting nonspecific treatment effects, and highlights the importance of differentiation of amygdala subregions.

Disclosure

The authors report no biomedical financial interests or potential conflicts of interest.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Handling Editor: A. Meyer-Lindenberg

Appendix A. Supplementary data

The following are the Supplementary data to this article.