Familial risk of major mood disorders and brain functional connectivity in the default mode, cognitive executive, and salience networks

Abstract

Background:

Mood disorders, including depressive and bipolar disorders, begin in late adolescence to early adulthood, tend to run in families, and present early with subthreshold symptoms. They have been associated with differential connectivity in 3 core networks: the default mode network (DMN), cognitive executive network (CEN), and salience network (SN), but it remains unclear whether differences in connectivity in the DMN, CEN, and SN are associated with familial risk for mood disorders.

Methods:

We recruited youth aged 9–19 years, including offspring of parents with major depressive or bipolar disorders (familial high risk [FHR]) and offspring of parents with no mood disorder (controls) for a resting-state functional magnetic resonance imaging study. We tested associations between family history of major mood disorders and connectivity within and between the DMN, CEN, and SN.

Results:

We included 205 youth: 126 at FHR with a mean age of 13.38 (standard deviation [SD] 2.91) years and 79 controls with a mean age of 13.17 (SD 2.67) years. Mean connectivity in the DMN (β = 0.003, 95% confidence interval [CI] −0.023 to 0.029), CEN (β = −0.009, 95% CI, −0.070 to 0.089), and SN (β = −0.010, 95% CI −0.071 to 0.051) in the FHR group was similar to that of controls. Moreover, DMN, CEN, and SN connectivity was not significantly associated with depressive symptoms.

Limitations:

Given that brain connectivity changes over the developmental period, longitudinal studies would improve understanding of how this change occurs in familial risk groups to identify critical time periods for intervention or prevention of mood disorders.

Conclusion:

Connectivity within and between the DMN, CEN, and SN is not a neural indicator of familial risk for major mood disorders.

Introduction

Major mood disorders, which include major depressive disorder and bipolar disorder, tend to run in families and typically begin in late adolescence to early adulthood.^1,2 These disorders have been linked with unusual patterns of brain functional connectivity,³ which suggests that changes in the communication between brain regions may be a key mechanism underlying mood disorders. It is not known whether these atypical connectivity patterns develop before the onset of the mood disorder or whether they are associated with major risk factors such as family history of major mood disorders or early subthreshold depressive symptoms.

People living with major mood disorders may have atypical functional connectivity patterns across multiple networks.^4,5 However, 3 core networks have been suggested as crucial for understanding mood disorder psychopathology, namely the default mode network (DMN), the cognitive executive network (CEN), and the salience network (SN).⁶ Collectively, these networks constitute the triple-network model of self-reference, saliency, and cognitive function.⁶

The DMN’s importance lies in its involvement with internal thought processes regarding the self, including rumination, a core feature of mood disorders.^7,8 The key regions of the DMN include the medial prefrontal cortex and the posterior cingulate cortex.⁹

The CEN is highly activated during executive function or cognitive tasks.¹ It underlies cognitive functions such as goal-directed behaviour, decision-making, and working memory, ¹⁰ which are related to familial and personal risk of mood disorders.^11,12 The CEN is also known as the cognitive control network, and its activity is negatively correlated with that of the DMN. The key regions of the CEN are the dorsolateral prefrontal cortex and the posterior parietal cortex.¹⁰

The SN is involved in detecting relevant information and directing attention to important aspects of the environment; dysregulation in this network can lead to impaired assignment of attention between the external environment and internal emotions.¹³ For this reason, the SN is proposed to mediate the engagement of the CEN and the disengagement of the DMN.⁶ The key regions of the SN are the anterior insula, dorsal anterior cingulate cortex, amygdala, and ventral tegmental area.⁶

Given the interconnected roles of the DMN, CEN, and SN, studies have investigated potential disruptions in these networks in relation to mood disorders.⁶ Although several associations between mood disorders and connectivity of these 3 networks have been reported, the findings remain mixed. Some case–control studies have reported decreased connectivity within the DMN in mood disorders,^14–16 while others found an increase in connectivity.^17–19 Similarly, several studies reported mixed findings regarding CEN^14,15,19 and SN connectivity in mood disorders.^20–22 A recent study showed that greater connectivity between the DMN and SN is implicated in depression,²³ but connectivity between these networks in mood disorders remains unclear. This is because previous studies have primarily focused on changes within specific brain networks; when they have assessed connectivity between networks as a secondary analysis, the findings have been inconsistent (Appendix 1, Table S1, available at www.jpn.ca/lookup/doi/10.1503/jpn.250002/tab-related-content).^15,16,19 Given the inconsistency in the findings, it is crucial to further investigate connectivity changes between and within DMN, CEN, and SN and determine whether any differences precede or arise as a consequence of these mood disorders.

Familial high-risk studies are well suited for addressing this gap. This study design allows for the investigation of brain connectivity alterations using family history as an early and stable indicator of risk, before typical onset of mood disorders. By examining brain connectivity in people at high familial risk, we can gain evidence on whether atypical changes in brain connectivity precede or result from the emergence of mood disorders. If alterations in brain connectivity are observed in high-risk individuals before disorder onset, this would suggest that these changes precede and potentially contribute to the development of major mood disorders.

Previous research on DMN, CEN, and SN connectivity in unaffected relatives of people with major mood disorders has yielded contradictory results. Some researchers reported increased DMN connectivity,²⁴ while others reported no differences in connectivity.^25,26 Although 1 study reported decreased DMN to SN connectivity in a familial risk group,²⁷ other studies have reported mixed findings. Specifically, there are inconsistent results regarding connectivity within the CEN and SN, as well as between network connectivity in the 3 networks.^{14,24–26,28,29} These discrepancies may stem from differences among previous studies in the selection of brain regions to define the 3 networks, the age range of participants, the inclusion or exclusion of participants who had already developed a mood disorder, and the severity of symptoms at the time of the brain scan (Appendix 1, Table S2).^30–33 Therefore, an investigation applying standard definitions of the DMN, CEN, and SN at developmental stages before typical mood disorder onset is needed to determine whether atypical connectivity in these networks predates onset of major mood disorders.

In the current study, we sought to examine the connectivity of the DMN, CEN, and SN in a cohort including people at high familial risk of mood disorder while accounting for sources of heterogeneity that might have affected previous studies. Although we expected that family history would be associated with connectivity within the DMN, CEN, and SN, we did not hypothesize about the directionality of this connectivity given that both increased and decreased connectivity has been reported in familial risk studies.^25,26 Based on previous findings that between-network connectivity may vary in familial risk groups, we also explored within-network connectivity in our cohort. To understand how connectivity might be related to psychopathology, we examined the association between depressive symptoms and within and between network connectivity of the DMN, CEN, and SN.

Methods

Participants

Participants were children and adolescents, including offspring of biological parents with a lifetime diagnosis of a major depressive disorder or bipolar disorder and participants from control families whose biological parents did not have a major mood disorder. We recruited participants as part of the Families Overcoming Risk and Building Opportunities for Well-Being (FORBOW) study, a longitudinal study enriched for children of parents with mental illness.³⁴ The familial high-risk (FHR) families were recruited by referral from adult mental health services or clinicians who were treating parents, while controls were recruited from schools and communities in Nova Scotia that matched the socioeconomic background of FHR participants.

We scheduled study participants for magnetic resonance imaging (MRI) at baseline and at 1- and 2-year follow-up. Additionally, we invited one-quarter of participants for a reliability scan 2 weeks after their yearly scan.

Inclusion and exclusion criteria

To examine brain connectivity in a high-risk sample before onset of mood disorder,³⁵ we included participants between the ages of 9 and 19 years, a period associated with rapid brain development. We included FHR participants if they had a biological parent with a lifetime diagnosis of a major mood disorder (major depressive disorder or bipolar disorders 1 and 2) as defined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5).³⁶ We included control participants if they did not have a family history of major mood or psychotic disorders.

We excluded participants who had a diagnosis of a neurologic disorder, a history of head trauma with loss of consciousness, a family history of schizophrenia, or contraindications to MRI.

Measures

We completed diagnostic assessments and collected demographic information at baseline and annual follow-ups.

Trained assessors interviewed parents using the Structured Clinical Interview for DSM-5 Disorders (SCID-5).³⁷ Youth assessors blind to parent psychopathology assessed the offspring. For offspring younger than 18 years, they used the Kiddie Schedule for Affective Disorders and Schizophrenia, Present and Lifetime Version (K-SADS-PL), while for those aged 18 years or older, they used the SCID-5.^38,39

Mood disorder diagnoses were confirmed in consensus meetings with adult and child psychiatrists, blind to information on other family members.

Depressive symptoms were measured using the 5-item subscale of the Youth Experience Tracker Instrument (YETI), a 26-item self-report questionnaire that tracks symptoms relevant to mood disorders in youth.⁴⁰ The YETI has a high degree of validity, and its depressive symptom items correlate strongly with established measures of depression used in clinical practice.^40,41 Participants completed the YETI before the MRI scan. Since antecedent depressive symptoms are a common predictor of major depressive disorder and bipolar disorder, we used the depressive symptom score (range 0–15) as the antecedent of interest.^42,43 Additionally, we used anxiety symptom items from the YETI in a sensitivity analysis, as anxiety is another common precursor of depression and bipolar disorder.⁴³ More details on YETI can be found at https://www.youthhealthmeasures.com/.

We collected information on sex, age, race and ethnicity, and socioeconomic indicators. We scored socioeconomic status (range 0–5) based on maternal and paternal levels of education, family household annual income, ownership of primary residence, and ratio of bedrooms to residents in the household.⁴⁴ We measured intelligent quotient (IQ) scores with the Wechsler Abbreviated Scale of Intelligence.⁴⁵ We measured height and weight using calibrated scales.

Imaging

Participants were scheduled for a scanning session at the Biomedical Translational Imaging Centre at the Queen Elizabeth II Health Sciences Centre in Halifax, Nova Scotia. Before the scan, an MRI technologist screened participants for MRI contraindications.

The scan session lasted around 35 minutes, during which anatomic and resting-state images were acquired using a 3 T General Electric Discovery MR750 scanner equipped with a 32-channel head coil.

We acquired T₁-weighted anatomic images (168 slices, 224 × 224 voxels, 1 mm³ isotropic resolution). We also acquired T₂-weighted images at the same resolution and voxel size (repetition time 5100 ms). After the anatomic scans, we also conducted 8-minute resting-state functional echoplanar imaging (51 axial–oblique slices, 3 mm³ voxels, slice thickness 3 mm with no gap, repetition time 950 ms, echo time 3 ms, flip angle 60°, multiband factor 3) while participants lay in the scanner with their eyes closed.

Preprocessing

We used a modified version of the Human Connectome Project pipeline⁴⁶ to preprocess the structural and functional MRI data. This analysis pipeline is available on Github (https://github.com/forbow-lab/neuro-structural-pipeline). Specifically, we used the fMRIPrep BIDS app (v.20.2.1),⁴⁷ a Nipype-based tool,⁴⁸ to preprocess the data following standard procedures. The steps involved are described in detail in Appendix 1.

Connectivity analysis

We conducted connectivity analysis with a custom Python script using the Nilearn libraries.⁴⁹ We followed the Friston-24 model to minimize noise signals resulting from head movement.^50,51 The steps for this analysis included regression of 6 parameters obtained by head motion correction and regression of the white matter and cerebrospinal fluid signals, averaged from white matter and cerebrospinal fluid brain regions. We also incorporated more complex measures into our analysis. Specifically, we included the first-order derivatives (R′) — which are measures of how the white matter and cerebrospinal fluid signals change — and the squares (R²) — which help to quantify the intensity of these signals — as regressors. These preprocessing steps effectively reduce variance that is unlikely to reflect neural activity.⁵² Following nuisance regression, we demeaned and detrended the data and, finally, applied bandpass filtering (0.009–0.08 s).⁵⁰

Since head motion is a major concern in functional connectivity, specifically for a pediatric cohort, we also denoised the individual data using the scrubbing method, which censors time points where the framewise displacement (FD) is greater than 0.3 mm⁵⁰ or the standardized DVARS (the root mean square of the temporal change of the functional MRI signal) is greater than 3.⁵³ We excluded participants for whom more than 20% (i.e., 100) of their time points were scrubbed. Following this step, we calculated the mean FD of all participants across the entire resting-state scan period (i.e., 8 min). We excluded participants with excessive motion (mean FD > 0.4 mm) from further analysis.

Definition of regions of interest

According to our review of the literature, the most standard approach to defining these networks is by using the hub nodes — critical regions that coordinate network activity.⁵⁴ Therefore, we used each network’s hub nodes for our analysis. We characterized the DMN using its hub nodes in the dorsal medial prefrontal cortex (dmPFC) and posterior cingulate cortex (PCC). For the SN, we focused on the anterior insula (AI) and dorsal anterior cingulate cortex (dACC) nodes, while for the CEN, we used the dorsolateral prefrontal cortex (dlPFC) and posterior parietal cortex (PPC) nodes.^10,13

For each participant, we defined each node using 6-mm radius spheres centred on the coordinates of the respective region of interest (ROI) within the DMN, CEN, and SN (Figure 1). These coordinates, from J. D. Power’s seminal paper that divided the whole brain into functional networks,⁵⁵ were also used in a recent paper examining functional connectivity between the frontoparietal and default mode networks.³³ We extracted these networks separately for the left and right hemispheres. This process yielded 8 measures of connectivity within networks (right hemisphere DMN, left hemisphere DMN, right hemisphere CEN, left hemisphere CEN, right to left hemisphere CEN, right hemisphere SN, left hemisphere SN, right to left hemisphere SN) and 9 measures of connectivity between networks (right hemisphere DMN to CEN, left hemisphere DMN to CEN, right to left hemisphere DMN to CEN, right hemisphere SN to DMN, left hemisphere SN to DMN, right to left hemisphere SN to DMN, right hemisphere SN to CEN, left hemisphere SN to CEN, right to left hemisphere SN to CEN).

Figure 1.

Brain plots of key regions of the (A) default mode network, (B) cognitive executive network, and (C) salience network. dACC = dorsal anterior cingulate cortex; dlPFC = dorsolateral prefrontal cortex; dmPFC = dorsomedial prefrontal cortex; PCC = posterior cingulate cortex; PPC = posterior parietal cortex.

Extracting estimates of functional connectivity

We measured functional connectivity during rest as the temporal correlation of the blood oxygen level–dependent (BOLD) signal activities within the DMN, CEN, and SN. We calculated Pearson correlation coefficients between mean time series for all voxels in each ROI for each participant. We used these correlations in the linear mixed-effects models to examine group differences.

Statistical analyses

In a post hoc power analysis for each of the 17 networks, we determined that we had a power of 0.77 to detect a medium effect size (Cohen d = 0.5) at the false discovery rate–corrected α of 0.007.

We probed test–retest reliability of the DMN, CEN, and SN as the intraclass correlation (ICC) of connectivity between yearly scans and corresponding reliability scans’ ROIs. The ICC was defined by the formula below, where λ is the object of measurement (i.e., connectivity), π is specified source of error (i.e., repeated runs), and ɛ is unspecified source of error.⁵⁶

$$(ICC)=\frac{{\sigma }_{\lambda }^2}{{\sigma }_{\lambda }^2+{\sigma }_{\pi }^2+{\sigma }_{\varepsilon }^2}$$

To investigate the association of DMN, CEN, and SN connectivity with familial risk, we ran mixed-effects linear regressions. We then ran mixed-effects linear regressions to explore the effects of familial risk on between-connectivity of the DMN, CEN, and SN. For all tests, we corrected for age. We included the corrected age and sex as fixed effects since these have been shown to influence functional connectivity.²⁵ Additionally, we included family and participant identifiers as random effects to account for the nonindependence of repeat scans of the same participants and siblings.⁴²

As a follow-up to the main analysis, we performed sensitivity analyses examining socioeconomic status, IQ, scrubbed time points, depressive symptoms, anxiety diagnosis, other disorders (e.g., substance use, attention-deficit/hyperactivity disorder), and medication status as additional covariates in mixed-effects linear regressions to rule out confounding by these variables.^42,57 Since the primary analysis included people with a mood disorder diagnosis, we also included this diagnosis as a covariate in the sensitivity analysis.

To investigate the moderation effects of sex, we ran linear mixed models and included the interaction term of sex and connectivity as a predictor.

To study the association of within- and between-connectivity of the DMN, CEN, and SN and depressive symptoms, we ran a repeated-measures correlation between age-corrected connectivity measures and YETI depression scores.

To examine potentially specific or differential effects of familial risks for depression versus familial risk for bipolar disorder on connectivity, we ran separate linear mixed model analysis on participants at familial risk for depression and those at familial risk for bipolar disorder.

Results

We completed brain scans with 205 participants, including 79 controls and 126 FHR participants (Table 1). The control and FHR groups had an equal distribution of males and females, and depressive symptoms did not vary across the 2 groups. Of the 205 participants, 26 had a major depressive disorder diagnosis at baseline and 5 additional participants received a diagnosis on follow-up. Among the 31 participants with a major depressive disorder diagnosis, 20 were female, 16 of whom were in the FHR group.

Table 1.

Descriptive characteristics of participants by family group

Characteristic	No. (%) of participants*		Test statistic
	Controls n = 79	FHR n = 126
Sex, female	39 (49.4)	61 (48.4)	χ2448 = 0.02, p = 0.9, Cramer V = 0.01
Age, yr, mean ± SD	13.17 ± 2.67	13.38 ± 2.91	t446 = 1.05, p = 0.3, d = −0.07 (95% CI 0.07 to −0.21]
Depression score, mean ± SD	1.4 ± 3.1	1.56 ± 2.67	t446 = 0.60, p = 0.6, d = −0.06 (95% CI 0.02 to −0.14]
Mood disorder diagnosis	6 (7.6)	25 (19.8)	χ2448 = 0.02, p = 0.9, Cramer V = 0.16
Scrubbed time points†, mean ± SD	44.72 ± 51.50	39.23 ± 48.78	t446 = 1.11, p = 0.3, d = 0.11 (95% CI −2.49 to 2.72)
Framewise displacement, mean ± SD	0.16 ± 0.08	0.15 ± 0.07	t446 = 1.13, p = 0.3, d = 0.11 (95% CI 0.11 to 0.12)
No. of scans
Initial	79	126	–
Reliability	19	60	–
Follow-up	56	108	–
Parent diagnosis (mother, father)
Depression	0 (0.0), 0 (0.0)	67 (53.2), 38 (30.2)	–
Bipolar	0 (0.0), 0 (0.0)	24 (19.0), 14 (11.1)	–
Anxiety	11 (13.9), 1 (1.3)	1 (0.8), 4 (3.2)	–
Substance use	0 (0.0), 5 (6.3)	0 (0.0), 4 (3.2)	–

CI = confidence interval; FHR = familial high risk; SD = standard deviation.

^*Unless indicated otherwise.

^†Scrubbed time points represent the mean value of repetition times that were removed in a functional connectivity run because of excessive motions.

We excluded 34 participants (47 scans) from the primary analysis (association of connectivity of the DMN, CEN, and SN with familial risk) because of excessive motion; we excluded an additional 2 participants (3 scans) from the secondary analysis (associations of depressive symptoms and connectivity of the DMN, CEN, and SN) because of missing depression scores. A comparison of the included and excluded participants showed that these groups differed in their depression and motion (FD) scores (Table 2).

Table 2.

Descriptive characteristics of participants included in the study versus those excluded because of motion

Characteristic	No. (%) of participants*		Test statistic
	Included n = 205	Excluded n = 34
Sex, female	100 (53.7)	10 (29.4)	χ2239 = 3.66, p = 0.06
Age, yr, mean ± SD	13.28 ± 2.79	12.05 ± 2.45	t46 = 0.11, p = 0.6
Depression score, mean ± SD	2.96 ± 2.89	0.45 ± 2.18	t46 = 0.35, p < 0.05
Mood disorder diagnosis	31 (15.1)	3 (8.8)	χ2237 = 0.35, p = 0.6
Framewise displacement, mean ± SD	0.16 ± 0.08	1.47 ± 6.24	t33 = −1.16, p < 0.05
No. of scans
Initial	205	34	–
Reliability	79	4	–
Follow-up	164	7	–

SD = standard deviation.

^*Unless indicated otherwise.

Test–retest reliability

We calculated test–retest reliability for within-network connections. We found a marked difference between the DMN, (ICC 0.6, reliable) versus the CEN and SN (ICC < 0.4, unreliable). Between-network connectivity and global values across all network connections were unreliable (ICC < 0.4; Table 3).

Table 3.

Test–retest reliability results for within- and between-network connections

Connection	ICC (95% CI)
Within DMN
Right DMN	0.350 (0.068 to 0.580)
Left DMN	0.605 (0.383 to 0.761)
Within CEN
Right CEN	0.425 (0.156 to 0.636)
Left CEN	0.436 (0.169 to 0.644)
Right-to-left CEN	0.284 (−0.005 to 0.530)
Within SN
Right SN	0.108 (−0.187 to 0.385)
Left SN	0.271 (−0.019 to 0.520)
Right-to-left SN	0.469 (0.209 to 0.668)
Between DMN and CEN
Right DMN to CEN	0.332 (0.048 to 0.567)
Left DMN to CEN	0.270 (−0.020 to 0.519)
Right-to-left DMN to CEN	0.189 (−0.105 to 0.454)
Between DMN and SN
Right DMN to SN	0.130 (−0.164 to 0.405)
Left DMN to SN	0.141 (−0.154 to 0.414)
Right-to-left DMN to SN	0.279 (−0.011 to 0.526)
Between SN and CEN
Right SN to CEN	−0.159 (−0.428 to 0.137)
Left SN to CEN	0.038 (−0.253 to 0.325)
Right-to-left SN to CEN	−0.055 (−0.338 to 0.238)
Global*	0.219 (−0.067 to 0.472)

CEN = cognitive executive network; CI = confidence interval; DMN = default mode network; ICC = intraclass correlation; SN = salience network.

^*Global values across all network connections.

Association between familial risk and within-network connectivity

We tested whether FHR participants differed from controls in DMN, CEN, and SN connectivity. We found no differences between the 2 groups in the connectivity within these networks in the right or the left hemispheres (Figure 2, Figure 3, and Appendix 1, Table S3). To separate correlates of familial risk from consequences of a mood disorder, we examined participants with a diagnosis as a separate group; connectivity in the mood disorder group did not differ from that of controls and FHR groups for any ROI (Figure 2 and Figure 3). These results remained consistent across sensitivity analyses performed to rule out the effects of potential confounders (Appendix 1, Table S4).

Figure 2.

Within-network connectivity in the (A, B) right and (C, D) left hemisphere of the default mode network (DMN), and in the (E, F) right and (G, H) left hemisphere of the cognitive executive network (CEN). The left column shows participants’ family history group, irrespective of whether they have a mood disorder diagnosis; the right column shows participants with family history without a mood disorder, those with a mood disorder, and controls without a mood disorder. In all networks, connectivity did not differ between the groups. FHR = familial high risk.

Figure 3.

Within-network connectivity in the (A, B) right, (C, D) left, and (E, F) right-to-left hemispheres of the salience network (SN). The left column shows participants’ family history group, irrespective of whether they have a mood disorder diagnosis; the right column shows participants with family history without a mood disorder, those with a mood disorder, and controls without a mood disorder. In all networks, connectivity did not differ between the groups. FHR = familial high risk.

There was no significant difference in the mean head motion between the control and FHR groups (t₄₄₆ = −1.13, p = 0.3).

Association between familial risk and between-network connectivity

In terms of between-network connectivity, FHR participants did not differ from controls in the DMN, CEN, and SN (Figure 4, Figure 5, and Appendix 1, Table S3). Likewise, the presence of a mood disorder diagnosis did not have an effect on between-network connectivity (Figure 4 and Figure 5). These results remained the same after we performed a sensitivity analysis controlling for potential confounders (Appendix 1, Table S4).

Figure 4.

Between-network connectivity in the (A, B) right, (C, D) left, and (E, F) right-to-left hemisphere of the default mode network (DMN) and the cognitive executive network (CEN), and in the (G, H) right and (I, J) left hemisphere of the DMN and the salience network (SN). The left column shows participants’ family history group, irrespective of whether they have a mood disorder diagnosis; the right column shows participants with family history without a mood disorder, those with a mood disorder, and controls without a mood disorder. In all networks, connectivity did not differ between the groups. FHR = familial high risk.

Figure 5.

Between-network connectivity in the (A, B) right-to-left hemisphere of the default mode network (DMN) and the salience network (SN), and in the (C, D) right, (E, F) left, and (G, H) right-to-left hemisphere of the SN and the cognitive executive network (CEN). The left column shows participants’ family history group, irrespective of whether they have a mood disorder diagnosis; the right column shows participants with family history without a mood disorder, those with a mood disorder, and controls without a mood disorder. In all networks, connectivity did not differ between the groups. FHR = familial high risk.

The sensitivity analysis revealed that motion was associated with all between-network connectivity measures (Table 4). These motion effects persisted after correction for multiple testing (p < 0.05). We reran between-network analyses with 177 low-motion participants (FD < 0.2 mm) (Table 5). Motion was associated with between-network connectivity for the left DMN to CEN and right-to-left DMN to SN (Table 6).

Table 4.

Association between motion and between-network connectivity from linear mixed-effects models

Model	β (95% CI)
DMN to CEN
Right DMN to CEN	1.025 (0.766 to 1.285)
Left DMN to CEN	0.860 (0.581 to 1.139)
Right-to-left DMN to CEN	0.854 (0.586 to 1.122)
DMN to SN
Right DMN to SN	0.913 (0.633 to 1.192)
Left DMN to SN	1.081 (0.790 to 1.372)
Right-to-left DMN to SN	1.029 (0.741 to 1.317)
SN to CEN
Right SN to CEN	0.349 (0.096 to 0.603)

CEN = central executive network; CI = confidence interval; DMN = default mode network; SN = salience network.

Table 5.

Descriptive characteristics of low-motion participants (framewise displacement < 0.2 mm)

Characteristic	No. (%) of participants*		Test statistic
	Controls n = 64	FHR n = 113
Sex, female	33 (51.6)	56 (49.6)	χ2177 = 0.01, p = 0.9, Cramer V = 0.01
Age, yr, mean ± SD	13.43 ± 2.77	13.64 ± 2.95	t175 = −0.48, p = 0.6, d = −0.08 (95% CI 0.03 to −0.18)
Depression score, mean ± SD	1.42 ± 2.77	1.74 ± 2.64	t175 = −0.77, p = 0.4, d = −0.12 (95% CI 0.03 to −0.27)
Mood disorder diagnosis	4 (6.2)	22 (19.5)	χ2177 = 4.69, p = 0.03, Cramer V = 0.16
Framewise displacement, mean ± SD	0.12 ± 0.04	0.12 ± 0.04	t175 = −0.24, p = 0.8, d = −0.04 (95% CI −0.036 to −0.037)
No. of scans
Initial	64	113	–
Reliability	17	40	–
Follow-up	35	80	–

CI = confidence interval; FHR = familial high risk; SD = standard deviation.

^*Unless indicated otherwise.

Table 6.

Association between motion and between-network connectivity among low-motion participants, from linear mixed-effects models

Model	β (95% CI)
DMN to CEN
Right DMN to CEN	1.355 (0.302 to 2.410)
Left DMN to CEN	1.851 (0.697 to 3.005)
Right-to-left DMN to CEN	1.362 (0.277 to 2.446)
DMN to SN
Right DMN to SN	0.489 (−0.644 to 1.622)
Left DMN to SN	1.407 (0.258 to 2.557)
Right-to-left DMN to SN	1.051 (0.462 to 1.640)
SN to CEN
Right SN to CEN	−1.380 (−2.417 to 0.342)

CEN = cognitive executive network; CI = confidence interval; DMN = default mode network; SN = salience network.

Association between network connectivity and depressive symptoms

Depressive symptoms did not have a significant effect on connectivity (Appendix 1, Table S5). Repeated-measures correlation showed that there was no significant association between connectivity and depressive symptoms in the control or FHR groups (Appendix 1, Table S6). All results were nonsignificant before and after performing a Benjamini–Hochberg false discovery rate correction for multiple testing (p > 0.05).

Separate analyses of familial risk for bipolar disorder and familial risk for depression groups identified no significant effect of familial risk on connectivity (Appendix 1, Tables S7 and S8). Finally, there were no significant moderation effects of sex on connectivity (Appendix 1, Table S9).

Discussion

We investigated between- and within-network connectivity in the DMN, CEN, and SN and the association between connectivity with familial risk and depressive symptoms. We found that functional connectivity within and between the 3 major brain networks was unrelated to family history of major mood disorders or to depressive symptoms among children and adolescents.

Our test–retest reliability results were similar to that of mean global connectivity measures reported in a meta-analysis of 25 studies of adult participants (ICC 0.29, 95% confidence interval 0.23 to 0.36), including the Human Connectome Project from which our preprocessing pipeline is derived. ⁵⁶ In agreement with previous studies, we found that the DMN connectivity measure was more reliable than CEN and SN connectivity measures; this difference in ICC can be partially attributed to the positioning of the ROIs within the 3 networks.⁵⁶ Given that increasing the scan length improves reliability, future studies should aim to use a longer scan duration and increase the number of acquired volumes.⁵⁸

Our findings of no differences in connectivity between youth at familial risk for major mood disorders and controls are consistent with 1 previous report,²⁶ but disagree with other studies that found altered connectivity among FHR participants.^14,24 By including participants with a mood disorder diagnosis, we were able to examine whether connectivity was associated with familial risk irrespective of illness; a sensitivity analysis accounting for diagnosis ruled out any differences that could be specific to participants with or without a diagnosis of a mood disorder themselves. Additionally, we were able to categorize participants at familial risk into 2 subgroups: those at risk for depression and those at risk for bipolar disorder. The results showed that familial risk for either disorder was unrelated to connectivity. These null results — and the mixed findings from previous studies showing either increased or decreased connectivity in familial risk groups — suggest caution for propositions that DMN, CEN, and SN connectivity is associated with familial risk for mood disorders.

Motion had a significant influence on between-network connectivity of the DMN to CEN, the DMN to SN, and the right SN to CEN. This is likely because pediatric cohorts generally move more in the scanner than adults.⁵⁹ We used strategies to reduce and correct for the effects of motion, including using motion and the number of scrubbed time points as covariates in the mixed-effects models and running the between-network analysis with low-motion participants. As there were no differences in motion or usable volumes between control and FHR groups, the association between motion and connectivity is unlikely to have confounded the primary results.

We did not find significant associations between connectivity and depressive symptoms in the control or FHR groups. These results agree with a previous longitudinal study by Lopez and colleagues,⁶⁰ but contradict previous research that showed a positive correlation between connectivity and depressive symptoms in an FHR group.²⁵ Some key differences may have influenced the likelihood of replicating previous positive findings, including differences in sample size (44 in Chai and colleagues²⁵ v. 205 in the present study), age range of the participants (8–14 yr in Chai and colleagues, ²⁵ 11–60 yr in Posner and colleagues,²⁴ 9–19 yr in the present study), and methods used for analysis (independent component analysis in Posner and colleagues²⁴ and in Heinze and colleagues²⁶ v. seed-based analysis in the current study). The varied sample sizes in previous studies, along with inconsistent results, indicate the need for larger samples.⁶¹ The different age ranges raise the possibility of effects limited to adolescence, a time when the brain is rapidly developing and the incidence of disorder onset increases. The use of data-driven versus seed-based methods complicates cross-study comparison and indicates the need for harmonized analytical methods. Overall, our findings did not confirm that DMN, CEN, and SN functional connectivity is associated with depressive symptoms in familial risk groups.

The present study had several notable methodological strengths. We had sufficient power and included participants at an age range just before the peak risk of mood disorder onsets. We performed a supplemental analysis to investigate familial risk for depression and bipolar disorder separately, given the heterogeneity of these disorders. We also ran a rigorous analysis and accounted for multiple testing. Therefore, our negative results highlight the need for further exploration of the previously reported associations between familial risk and connectivity. The lack of replication highlights the need for standardization of methodology and use of sufficiently powered samples in neuroimaging research.⁶¹

Limitations

Since this was a cross-sectional study, we could not determine whether the results observed persist in a longitudinal context. Brain connectivity is known to change over the developmental period, and an understanding of how this change occurs in familial risk groups might be useful in identifying critical time periods for intervention or prevention of mood disorders.⁶² Future longitudinal studies should aim to validate the present findings and assess the potential of DMN, CEN, and SN connectivity in identifying youth who are at highest risk of progressing to a mood disorder diagnosis. We ran a separate analysis for participants with depression and bipolar disorder. Although this allowed for a comparison to determine whether there were any differences between these 2 groups versus controls, the small sample size of participants with a mood disorder diagnosis limited the interpretability of our results. Future studies with equal sample groups would enable a thorough investigation of the differences between familial risk and mood disorder groups.

Conclusion

We found that brain network connectivity in youth was unrelated to familial risk of major depressive disorder or bipolar disorder, or to depressive symptoms. We suggest that connectivity in these networks might not be a marker of risk for mood disorders.