Sex-specific patterns of white matter microstructure are associated with emerging depression during adolescence

Abstract

Prior research has demonstrated associations between adolescent depression and alterations in the white matter microstructure of fiber tracts implicated in emotion regulation. Using diffusion tensor imaging, this study explored premorbid, sex-specific white matter microstructural features that related to future emergence major depressive disorder (MDD) during adolescence and young adulthood. Adolescents from the National Consortium on Alcohol and Neurodevelopment in Adolescence study, who were 12–21 years old at study entry and had not experienced major depression as of the baseline assessment, were selected for inclusion (N = 462, n = 223 female adolescents). Over five years of annual follow-up, 63 participants developed a diagnosis of MDD, as determined by the Computerized Semi-Structured Assessment for the Genetics of Alcoholism (n = 39 female adolescents). A whole-brain multivariate modeling approach was used to examine the relationship between fractional anisotropy (FA) at baseline and emergence into MDD, as a function of sex, controlling for age at baseline. Among female adolescents, those who developed MDD had significantly lower baseline FA in a portion of left precentral gyrus white matter, while male adolescents exhibited the opposite pattern. Such results may serve as indirect microstructural markers of risk and targets for the prevention of depression during adolescence.

1. Introduction

Adolescence represents a developmental period of particular vulnerability for the emergence of mood and anxiety disorders. Reports estimate that upwards of 13% of adolescents in the United States are diagnosed with at least one major depressive episode in a given year, and that this number has been steadily climbing for the last decade (SAMHSA, 2017). Importantly, emerging depressive symptoms in adolescence are a risk factor for subsequent clinical depression, recurring chronic depression later in life, as well as higher rates of suicidality (Fergusson et al., 2005; Klein et al., 2009). There is a notable rise in the prevalence of depressive symptoms after puberty (Cohen et al., 1993), particularly in girls (Hankin et al., 1998). The sex differences that emerge during this timeframe suggest that concomitant shifts in hormones, neurodevelopment, cognition, and psychosocial influences may contribute to an increased susceptibility for depression among female adolescents (Angold and Costello, 2006). Understanding the factors that lead to this differential risk for emerging depression during adolescence is crucial for informing decisions regarding both prevention and treatment.

To date, there have been a myriad of studies looking at the neuroimaging correlates of adolescent depression, the majority of which have implicated frontal-limbic circuitry. In general, sustained negative affect appears to be related to a diminished capacity for top-down emotional regulation (for review, see Gotlib and Joormann, 2010). The core symptoms of depression, including loss of interest or pleasure, depressed mood, difficulties concentrating, and fatigue, have been shown to be associated with smaller regional gray matter volumes in limbic and prefrontal regions (Redlich et al., 2018; Vulser et al., 2015), as well as altered functional connectivity between these areas (Cullen et al., 2014; Ho et al., 2015) and reduced structural cohesion of the white matter pathways that bridge them (Henderson et al., 2013; LeWinn et al., 2014; Zhu et al., 2011). However, it remains unclear whether these alterations associated with manifest depression might actually be present prior to symptom onset, thus serving as risk factors for symptom expression.

Several studies have provided evidence for premorbid neural risk phenotypes of depression. For example, smaller amygdalar volumes have been associated with emerging depressive symptoms (Chai et al., 2015) - though, depending on the age range examined, these relationships have been shown to vary by sex (Whittle et al., 2014). In terms of functional correlates, weaker functional connectivity of the subgenual cingulate cortex with other default-mode network regions has been associated with increases in depressive symptoms across middle and late adolescence (Strikwerda-Brown et al., 2015). Similarly, lower functional connectivity of the amygdala with prefrontal regions predicts escalations in symptom severity (Connolly et al., 2017; Scheuer et al., 2017). Taken together, these results seem to suggest that aberrant frontal-limbic connectivity likely plays a role in vulnerability for depression, but relatively few studies have examined the structural features of white matter pathways that may facilitate that functional phenotype. The most commonly reported metric of white matter microstructure is fractional anisotropy (FA), which is thought to reflect fiber coherence and myelination of axon bundles, though it can also be impacted by fiber diameter, fiber density, and membrane permeability (Alexander et al., 2007; Beaulieu, 2002; Jones et al., 2013). While some studies have demonstrated lower frontal-limbic FA in association with more depressive symptoms emerging over time (LeWinn et al., 2014; Xiao et al., 2015), others report positive associations between FA and depressive symptoms in overlapping brain regions (Aghajani et al., 2014).

Recently, in an attempt to bring about some consensus with a larger sample, researchers from the IMAGEN consortium demonstrated that lower FA in the right cingulum bundle and anterior corpus callosum mediated the transition from subthreshold depression at age 14 to clinical depression by age 16, with no dissociable sex differences (N = 96 adolescents with subthreshold depression, N = 336 healthy controls) (Vulser et al., 2018). However, this work was limited by focusing on a narrow outcome measure: a diagnosis of major depressive disorder (MDD) by age 16. Due to the fluctuating nature of depression, which may relapse and remit over time, we aimed here to identify premorbid white matter microstructural features related to the emergence of an MDD diagnosis at any point over the course of five years of follow-up in adolescents across a wider age range at baseline (12–21 years).

Importantly, there is ongoing maturation of white matter tracts during adolescence, and the trajectories of this maturation differ both regionally and by sex (Lebel et al., 2008; Simmonds et al., 2014). To account for these normative differences in development, we controlled for participant age at baseline, and we explicitly modeled potential sex differences. Based on previous literature focusing on white matter microstructure and depression risk (LeWinn et al., 2014; Vulser et al., 2018), we expected that lower FA in tracts that mediate emotion regulation, specifically the uncinate fasciculi and cingulum bundles, would be associated with future depression. By performing this analysis with a cohort of individuals who were healthy at baseline and had no history of depression, we aimed to dissociate microstructural brain features unique to risk for depression from those indicative of disease state.

2. Methods

2.1. Participants

Participants for the current analysis were selected retrospectively from the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA) study. The broader NCANDA project employs an accelerated longitudinal study design to examine neural correlates of adolescent risk for, and consequences of, alcohol use. However, the NCANDA study includes a large and demographically diverse sample that is well suited to the interrogation of other adolescent-emergent psychopathology, such as depression. The entire baseline sample was comprised of 831 youth between the ages of 12 and 21 at study entry, recruited from five sites across the United States: Duke University, Oregon Health & Science University (OHSU), University of Pittsburgh, SRI International, and University of California San Diego (UCSD). Further details regarding recruitment, demographics, and study procedures have been published previously (Brown et al., 2015). In short, participants were asked annually to complete a comprehensive battery of neuropsychological tests (Sullivan et al., 2016), assessments of behavior, psychiatric conditions and symptoms, and general functioning, reports on substance use, as well as a neuroimaging protocol with structural, resting-state, and diffusion tensor imaging (Pfefferbaum et al., 2016; Pohl et al., 2016). Adult participants provided voluntary informed consent, while minors provided their assent in addition to the informed consent of a parent or legal guardian. All study procedures were approved by the respective Institutional Review Board at each site. Neuroimaging data from the baseline visit, as well as behavioral data from the baseline visit and five subsequent follow-up visits, are analyzed in this report.

Exclusionary criteria for entry into the NCANDA study at baseline included current use of psychoactive medication, MRI contraindications (e.g. irremovable metal implants, pregnancy), current or persistent major Axis I psychiatric disorders that could interfere with protocol completion (e.g. psychosis), significant learning or developmental disorders, and serious medical problems or head injury (Brown et al., 2015). Although the majority of the NCANDA cohort had limited exposure to drugs or alcohol at enrollment, a small proportion were recruited who exceeded age-specific drinking thresholds for risky alcohol use according to the National Institute on Alcohol Abuse and Alcoholism guidelines (NIAAA, 2011). To rule out any effects of substance use on baseline neuroimaging metrics, only participants who did not exceed these thresholds were eligible for inclusion in the current analyses. In the interest of isolating microstructural features specific to depression risk, participants included in this report were also required to be depression-free at baseline (i.e. not meeting criteria for a diagnosis of MDD in their lifetime according to either parent- or youth-report on the Computerized Semi-Structured Assessment for the Genetics of Alcoholism – SSAGA). Summarily, of the 831 NCANDA participants, 148 were excluded for exceeding substance use thresholds at baseline, 175 were excluded for meeting diagnostic criteria for lifetime depression at baseline according to either parent- or youth-report, 17 were excluded for a missing diffusion-weighted imaging (DWI) scan at baseline, and 29 were excluded for poor quality diffusion imaging data at baseline, leaving a final sample of 462 participants. Of that final sample, 63 participants met criteria for major depression at one or more follow-up visits during the subsequent five years (MDD group), while the remaining 399 did not (control group).

2.2. Measures

A computerized version of the SSAGA, administered by trained research staff, was used to determine whether an individual met criteria for a diagnosis of MDD (as described in Brown et al., 2015). The SSAGA is a poly-diagnostic instrument designed to assess current (past two weeks) and lifetime endorsements of alcohol use and associated psychiatric disorders, including MDD (Bucholz et al., 1994; Hesselbrock et al., 1999). The MDD module specifically addresses how often and to what severity participants have experienced the core symptoms of MDD, including depressed mood, apathy or disinterest, feelings of worthlessness or guilt, difficulties concentrating, changes in weight and/or appetite, fatigue, and difficulties with sleep. Participants and their guardian/caregiver completed the SSAGA at baseline to determine lifetime endorsements of psychopathology, as well as at each annual follow-up visit to gauge current symptoms.

The Customary Drinking and Drug Use Record (CDDR) was administered to measure both current and lifetime substance use. The CDDR assesses frequency and quantity of alcohol and other drug use in adolescent populations (Brown et al., 1998). Like the SSAGA, this measure was administered at baseline and at each follow-up visit. Socioeconomic status at baseline was assessed using the highest level of education attained by either parent. Pubertal status at baseline was determined by responses on the Pubertal Development Scale (PDS), a sex-specific self-report measure of physical development (Petersen et al., 1988).

2.3. Neuroimaging procedures

2.3.1. Diffusion MRI acquisition

Diffusion-weighted images (DWI) were collected on either a 3T General Electric (GE) Discovery MR750 (SRI, Duke, and UCSD) or a 3T Siemens TIM TRIO (Pittsburgh and OHSU), as outlined previously (Pohl et al., 2016). At all sites, a 2D Axial Spin Echo, echo-planar imaging sequence was applied with gradient encoding pulses in 60 directions (b = 1,000 s/mm²), with two additional images collected at b = 0 s/mm² [repetition time = 8,000 ms; echo time = 89 ms (Pittsburgh and OHSU); echo time = 78 ms for participants < 160 lbs, 79 ms for participants ≥ 160 lbs (SRI, Duke, and UCSD); slices = 64; slice thickness = 2.5 mm; resolution = 2.5 × 2.5 × 2.5]. Further, a reverse-phase acquisition of this 2D Axial Spin Echo, echo-planar sequence with 6 directions (b = 500 s/mm²) and one additional image at b = 0 s/mm², were also collected, and this b = 0 s/mm² image was used to correct for susceptibility-induced distortions (see pre-processing steps below).

2.3.2. Image pre-processing

Raw DWI scans underwent strict visual inspection for motion and scanner-related artifacts (as described in Roalf et al., 2016). Previous studies suggest that inclusion of “poor” diffusion imaging data (> 20% volumes containing artifacts) significantly alters diffusion metrics, most notably in developmental samples (Roalf et al., 2016). Therefore, scans were excluded if 12 or more volumes contained artifacts. As such, 15 scans were excluded for excessive motion (e.g. 12 or more volumes containing motion-related artifacts, such as blurring, signal drop-out, or ghosting), 10 for scanner-related artifacts (including fat-saturation artifacts), and 4 for errors in image acquisition (e.g. incomplete whole-brain coverage). An additional 17 participants were excluded for a missing DWI scan at baseline.

Following visual inspection, DWI data were processed using FSL (v. 5.0.11) (Smith et al., 2004). First, using the reverse phase-encoded b = 0 s/mm² image from the 6-direction sequence and the first b = 0 mm/s² image from the 60-direction sequence (described above), susceptibility-induced distortions were estimated and corrected with FSL’s topup (Andersson et al., 2003). Then, FSL’s eddy was used to correct for eddy current distortion, intensity inhomogeneities, and head motion, while also adjusting the gradient table (Andersson and Sotiropoulos, 2016). Additionally, outlier replacement was carried out within FSL’s eddy, to detect and replace dropout slices that remained in the data following visual inspection (Andersson et al., 2016). Finally, FSL’s dtifit was used to calculate the diffusion tensors and eigenvalues for each voxel, and diffusion metrics (fractional anisotropy, FA; mean diffusivity, MD; axial diffusivity, AD; and radial diffusivity, RD) were calculated with FSL’s non-linear computational algorithm.

Next, Advanced Normalization Tools (ANTs) algorithms (Avants et al., 2011) were used to register participants to a study-specific template, and then to Montreal Neurological Institute (MNI) space, in a single-step interpolation, following procedures outlined previously (Schwarz et al., 2014). A Gaussian blur (sigma = 1 mm) was applied to all FA images. Then, to restrict whole-brain analyses to regions of white matter, a binary mask was created that included only voxels where mean FA across all subjects was greater than 0.3.

Finally, to minimize the effect of differences in scanner platform on FA, a voxel-wise correction factor determined by empirical Bayes methods was applied via the ComBat function in R (v.3.5.3). Age, sex, and group (MDD vs. control) were included as biological covariates. Although previous work has utilized a global scaling factor to address this issue (Pohl et al., 2016), it has since been demonstrated that ComBat outperforms such methods when applied to DWI (Fortin et al., 2017; Johnson et al., 2007). MD, AD, and RD maps were registered to standard space using the same transformations generated from registration of the FA maps, and adjusted via ComBat as well.

2.4. Statistical Analysis

All voxel-wise group-level comparisons on FA maps were run using AFNI’s 3dMVM (v. 17.1.03) with a voxel threshold of p < 0.01 (Cox, 1996). The residual estimates for the group-level model were used to estimate the spatial auto-correlation function (ACF) parameters via AFNI’s 3dFWHMx, which were then used with 3dClustSim to determine the appropriate cluster threshold, corrected for multiple comparisons via a non-parametric permutation analysis (default = 10,000 simulations, alpha = 0.05). Here, comparisons between whole-brain FA maps for participants who went on to develop depression (MDD) and those who did not (controls) were conducted using a two-way ANCOVA, including a group-by-sex interaction and baseline age as a covariate (FA ~ group + sex + group * sex + age; statistically significant cluster size requirement > 428 voxels as determined via AFNI’s 3dClustSim with family-wise error-rate correction). Due to the study design, it is possible that some participants experienced depression in between the follow-up visits, and began using SSRI’s or other antidepressants in the interim which then alleviated their symptoms. Such instances could lead to control participants being incorrectly classified. To address this issue, we re-ran our analyses a second time after removing any control participants who reported taking antidepressants (see list of antidepressants in Supplementary Materials) at any of the follow-up visits (n = 13 participants).

There are several other potential confounds that we addressed post-hoc. For these analyses, average FA values from significant clusters were extracted using AFNI’s 3dROIstats and entered into R (v. 3.5.3). Associations between FA and other relevant sociodemographic variables, including pubertal status at baseline, scanner platform, substance use (peak annual use reported during follow-up), and other psychopathology, were examined using a general linear model framework with the built-in stats package in R (R Core Team, 2018). Firstly, because MDD and substance use are often comorbid, we wanted to ensure that any observed variations in baseline white matter microstructure were associated with emerging depression above and beyond future alcohol and/or marijuana use. Just as depression can escalate and remit over time, an individual’s degree of substance use can also vary. For this reason, we examined the maximum use reported during the follow-up period (annual number of drinks of alcohol; annual number of days using marijuana) in relation to mean FA values from significant clusters post-hoc. Secondly, due to possible overlap in risk phenotypes for several forms of psychopathology (Jenkins et al., 2016), we also investigated the extent to which any identified differences in white matter microstructure at baseline related to the emergence of anxiety disorders, specifically generalized anxiety disorder (GAD) and obsessive compulsive disorder (OCD), using binary covariates. Finally, to determine whether any of the reported effects were related to subclinical variance in depression symptoms at baseline, we also examined any relationships between depression symptom counts at baseline and mean FA. Average MD, AD, and RD values were also extracted from significant clusters using AFNI’s 3dROIstats and examined in R.

3. Results

3.1. Participant Characteristics

A total of 462 participants were included in this study (n = 223 female adolescents). Demographic characteristics are provided in Table 1. There were statistically significant differences between individuals who went on to develop MDD and controls on several variables assessed at baseline, including sex (p = 0.020), data collection site (p < 0.001), scanner platform (p = 0.042), pubertal status at baseline (p = 0.018), and subclinical depression symptom counts at baseline (p < 0.001). There were also significant differences between these groups with respect to the emergence of several anxiety disorders over the course of follow-up (GAD, p < 0.001; OCD, p = 0.007). However, there were no significant differences between the groups in terms of either alcohol (p = 0.944) or marijuana use (p = 0.515) at baseline, or peak alcohol (p = 0.535) or marijuana use (p = 0.582) reported during follow-up.

Table 1.

Participant characteristics.

		Healthy Controls (N = 399)	Emerge into MDDa (N = 63)	t/Χ2	df	p
		n (%)	n (%)
Sex assigned at birth				5.432	1	0.020
	Female	184 (46.1%)	39 (61.9%)
	Male	215 (53.9%)	24 (38.1%)
Self-declared race/ethnicityb				3.128	5	0.680
	Caucasian/White	275 (68.9%)	47 (74.6%)
	African-American/Black	51 (12.8%)	4 (6.3%)
	Asian	31 (7.8%)	5 (7.9%)
	Native American/American Indian	3 (0.8%)	0 (0.0%)
	Pacific Islander	2 (0.5%)	0 (0.0%)
	Other	37 (9.3%)	7 (11.1%)
Scanner platform				4.154	1	0.042
	General Electric	285 (71.4%)	37 (58.7%)
	Siemens	114 (28.6%)	26 (41.3%)
Data collection site				19.137	4	<0.001
	University of Pittsburgh	52 (13.0%)	5 (7.9%)
	SRI International	89 (22.3%)	6 (9.5%)
	Duke University	95 (23.8%)	9 (14.3%)
	Oregon Health & Science University	62 (15.5%)	21 (33.3%)
	University of California – San Diego	101 (25.3%)	22 (34.9%)
Baseline
		Mean±SD	Mean±SD
Age		14.87±1.78	14.88±1.44	0.067	460	0.946
Subclinical depression symptoms (counts)		0.38±1.17	1.38±2.44	5.224	460	<0.001
Pubertal statusc (range 1–4)		2.94±0.73	3.17±0.65	2.381	456	0.018
Parental education (years)		17.0±2.46	16.4±2.84	1.737	460	0.083
Lifetime alcohol use (drinks)		0.56±2.56	0.59±2.64	0.071	460	0.944
Lifetime marijuana use (days)		0.47±2.50	0.70±2.97	0.065	460	0.515
Follow-Up
		n (%)	n (%)
Emerge into GAD during follow-upd		2 (0.5%)	8 (12.7%)	38.224	1	<0.001
Emerge into OCD during follow-upe		1 (0.3%)	2 (3.2%)	7.211	1	0.007
		Mean±SD	Mean±SD
Age at first MDD diagnosisa		N/A	16.8±1.59
Peak alcohol use across follow-up (annual drinks)f		163±265	139±188	0.621	323	0.535
Peak marijuana use across follow-up (annual days)g		70.5±111.0	80.4±118.0	0.551	251	0.582

^aMajor Depressive Disorder according to DSM-IV criteria; assessed using the Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA) (Bucholz et al., 1994; Hesselbrock et al., 1999).

^bCategories for self-declared race/ethnicity in this study were as follows: White (non-Hispanic), African American/Black, Hispanic/Latino, Asian, Bi-/Multi-Racial, Pacific Islander, and Native American/American Indian.

^cMissing/NA values for 4 controls; assessed using the self-report Pubertal Development Scale (PDS) (Petersen et al., 1988).

^dGeneralized Anxiety Disorder according to DSM-IV criteria; assessed using the SSAGA (Bucholz et al., 1994; Hesselbrock et al., 1999).

^eObsessive Compulsive Disorder according to DSM-IV criteria; assessed using the SSAGA (Bucholz et al., 1994; Hesselbrock et al., 1999).

^fMissing/NA values for 127 controls and 10 MDD participants; substance use measures derived from the Customary Drinking and Drug Use Record (CDDR) (Brown et al., 1998).

^gMissing/NA values for 194 controls and 15 MDD participants; substance use measures derived from the Customary Drinking and Drug Use Record (CDDR) (Brown et al., 1998).

3.2. ComBat Correction

The use of ComBat to address batch effects relating to scanner platform prior to voxel-wise analysis was effective in compensating for differences across this multi-site data (Supplemental Figure 1). Testing for effects of site and scanner post-hoc (when included either as linear covariates or using a random effects structure) confirmed that there were no significant effects of either scanner platform or data collection site on FA in either cluster (all p’s > 0.1).

3.3. Diffusion Tensor Imaging

When collapsing across sex and controlling for age at baseline, there was a significant difference in baseline FA between participants who later developed MDD and controls in a cluster comprised primarily of fibers from the anterior corona radiata (ACR), as well as a portion of cingulum white matter (492 voxels, η²_G = 0.0187; Figure 1). Mean diffusion indices extracted from this cluster (FA, MD, AD, and RD; Table 2) revealed that the MDD group exhibited significantly lower FA (p < 0.001), lower AD (p = 0.001), and higher RD (p = 0.004) in this region compared to controls (Figure 2; Table 2).

Figure 1. Relationship between fractional anisotropy (FA) at baseline and emergence into major depression (MDD) in the left anterior corona radiata (ACR).

There was a statistically significant main effect of group (MDD vs. controls) on baseline FA, while controlling for age at baseline and irrespective of sex. The identified cluster (492 voxels) is primarily comprised of fibers from the anterior corona radiata. Images are thresholded F-statistic maps overlaid on the mean FA map for the entire sample.

Table 2.

Diffusion indices extracted from significant clusters.

		Healthy Controls (N = 399)	Emerge into MDDa (N = 63)	t	df	p
		Mean±SD	Mean±SD
Left ACR – Main effect of group (control vs. MDD)
FA		0.495±0.033	0.476±0.027	4.497	460	<0.001
MD		0.791±0.030	0.793±0.025	0.462	460	0.644
AD		1.268±0.050	1.246±0.044	3.270	460	0.001
RD		0.553±0.036	0.567±0.030	2.917	460	0.004
Left SCR – Interaction effect between group (control vs. MDD) and sex (female vs. male)
Female Adolescents		N = 184	N = 39
	FA	0.442±0.026	0.428±0.025	3.010	221	0.003
	MD	0.732±0.022	0.739±0.024	1.659	221	0.099
	AD	1.105±0.039	1.101±0.034	0.700	221	0.485
	RD	0.546±0.025	0.558±0.027	2.766	221	0.006
Male Adolescents		N = 215	N = 24
	FA	0.438±0.026	0.458±0.026	3.549	237	<0.001
	MD	0.738±0.025	0.738±0.024	0.080	237	0.937
	AD	1.111±0.043	1.133±0.040	2.408	237	0.017
	RD	0.552±0.027	0.540±0.026	1.993	237	0.047

Anterior corona radiata (ACR); superior corona radiata (SCR); fractional anisotropy (FA); mean diffusivity (MD); axial diffusivity (AD); radial diffusivity (RD). MD, AD, and RD are in units of 10⁻³ mm²/s. FA is dimensionless.

^aMajor Depressive Disorder according to DSM-IV criteria; assessed using the Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA) (Bucholz et al., 1994; Hesselbrock et al., 1999).

Figure 2. Mean fractional anisotropy (FA) at baseline between MDD and control groups in left anterior corona radiata (ACR) cluster.

Boxplots of mean FA extracted from the cluster in the anterior corona radiata (ACR) where there was a significant main effect of group (MDD vs. controls). Participants who went on to develop MDD had lower mean FA in the identified cluster compared to controls. Median FA, interquartile range, and outliers are displayed.

Though not a primary focus of this study, there were significant main effects of both sex (Supplementary Figure 2) and age (Supplementary Figure 3) at baseline on FA in several large, diffuse regions across the brain. This is to be expected given the large age range at baseline (12–21 years) and known differences in developmental trajectories of white matter development by sex across adolescence (Lebel et al., 2008; Simmonds et al., 2014).

Additionally, there was a significant interaction between the effects of group (MDD vs. control) and sex for baseline FA in a cluster of precentral gyrus white matter bordering on the superior corona radiata (SCR; 446 voxels, η²_G = 0.0189 ; Figure 3). Mean diffusion indices extracted from this cluster (Table 2) showed that female adolescents who went on to develop MDD exhibited lower FA (p = 0.003), and higher RD (p = 0.006) in this region compared to controls, while male adolescents who later developed MDD exhibited higher FA (p < 0.001), higher AD (p = 0.017), and lower RD (p = 0.047) in this same region (Figure 4; Table 2).The clusters identified in these voxel-wise analyses remained unchanged when excluding the 13 control participants who reported using antidepressants during follow-up.

Figure 3. Sex-specific relationships between fractional anisotropy (FA) at baseline and emergence into major depression (MDD) in the left superior corona radiata (SCR).

There was a statistically significant interaction between the effects of group (MDD vs. controls) and sex on baseline FA, while controlling for age at baseline. The identified cluster (446 voxels) is comprised of precentral gyrus white matter, bordering on the superior corona radiata (SCR). Images are thresholded F-statistic maps overlaid on the mean FA map for the entire sample and boxplots of mean FA extracted from the cluster.

Figure 4. Mean fractional anisotropy (FA) at baseline between MDD and control groups, and between males and females, in left superior corona radiata (SCR) cluster.

Boxplots of mean FA 30 extracted from the cluster in the superior corona radiata (SCR) where there was a significant interaction effect between group (MDD vs. controls) and sex assigned at birth (male vs. female). Female adolescents who later developed MDD demonstrated significantly lower mean FA in the identified cluster compared to controls. Conversely, male adolescents who later developed MDD exhibited significantly higher FA in this same region compared to controls. Median FA, interquartile range, and outliers are displayed.

3.4. Post-Hoc Analyses

Despite a statistically significant difference between MDD and control groups with regard to pubertal status at baseline, this was not related to FA values extracted from either cluster, and when included as a covariate in the final model, did not change the significance or direction of any of the original results. None of the participants in this sample met criteria for a diagnosis of MDD, or an anxiety disorder (e.g. GAD, OCD), at the initial assessment. Although subclinical depression symptom counts at baseline were significantly different between MDD and control groups, there were no statistically significant relationships with mean FA values in either cluster (p’s > 0.05). Adding baseline depression symptoms as a covariate in the final model did not change the direction or significance of any of the reported results. Similarly, despite group differences in future development of anxiety disorders (GAD, OCD), there were no significant relationships with FA in either of the reported clusters (p’s > 0.05), and including these as covariates did not alter the results. It is also important to note that there were no statistically significant relationships between mean FA values in either cluster and future substance use variables (peak use during follow-up, all p’s > 0.05).

4. Discussion

In an effort to identify white matter microstructural markers of early risk for depression, here, we demonstrated differential, sex-specific patterns of white matter microstructure that relate to subsequent onset of depression in adolescents. While we showed that lower FA at baseline in a portion of the left ACR was generally associated with future development of major depressive disorder, irrespective of sex, we also identified a cluster of precentral gyrus white matter adjacent to the SCR, where male and female adolescents displayed opposing risk phenotypes. Female adolescents who later developed MDD showed lower FA in this region compared to healthy controls, whereas male adolescents who later developed MDD showed higher FA relative to controls in this same cluster. The white matter tracts implicated here do overlap with several previous reports demonstrating associations between FA and risk for emerging depression in similar age groups (Ganzola et al., 2018; Xiao et al., 2015). However, our findings were not specific to tracts directly involved in mood regulation, as we had initially hypothesized.

In this way, the evidence that we have provided for distinct sex-specific risk profiles extends the traditional conceptualization of depression beyond a simple model of frontal-limbic dysregulation and supports the notion of considering sex-specific markers of vulnerability in future work. Our findings may suggest that there are separate white matter microstructural precursors for depressive symptoms in female versus male adolescents. Alternatively, it is possible that variations in neurodevelopmental trajectories by sex underlie the emergence of these discrete profiles (Simmonds et al., 2014). In either case, it is imperative that we understand how even subtle alterations in white matter microstructure may confer susceptibility to emerging depression during this vulnerable developmental window.

Evidence of lower FA values relating to subsequent depression onset, specifically among female adolescents, mirrors the most commonly reported patterns in previous studies. Several cross-sectional analyses have provided evidence for lower coherence of white matter tracts in adolescents with manifest depressive disorders compared to healthy controls (Jenkins et al., 2016; LeWinn et al., 2014; Zhu et al., 2011); however, a majority of this work has focused a priori on particular tracts that mediate emotion regulation, such as the uncinate fasciculi and cingulum bundles. Our findings involve several other white matter pathways that may be important for characterizing the pathophysiology of depression risk. Notably, we have shown that lower FA values in the ACR are associated with future depression regardless of sex. Although the diffusion indices derived from DWI cannot resolve neurobiological changes (Jones et al., 2013), this pattern of lower FA coupled with lower AD (thought to be more sensitive to axonal properties) and higher RD (thought to be more sensitive to myelination), suggests altered or delayed maturation in this portion of the ACR among adolescents at risk for developing MDD. As a component of limbic-thalamo-cortical circuitry, the ACR is comprised of fan-shaped thalamic projection fibers that extend from the internal capsule to cortical regions (Catani et al., 2002; Wakana et al., 2004). This pathway has been implicated in executive control of attention and inhibition (Madsen et al., 2010; Niogi et al., 2010; Tamnes et al., 2012; Yin et al., 2013), as well as emotion regulation (Pavuluri et al., 2009; Sanjuan et al., 2013). Because the physical connections between brain regions ultimately underlie the neurophysiology of coordinated brain activity, diminished cohesion of white matter fibers may contribute to alterations in brain function, and by extension behavior. In this way, it’s possible that microstructural features of ACR fibers could underlie a propensity for the attentional symptoms of depression, such difficulties concentrating, internally or negatively focused attention, and rumination (Keller et al., 2019). This cluster also included a small portion of the cingulum, an associative white matter bundle which connects prefrontal and parahippocampal regions (Bubb et al., 2018). Lower FA in the cingulum has been implicated in the pathophysiology of depression risk, and specifically anhedonia (Huang et al., 2011; Keedwell et al., 2012). We have also characterized FA differences in precentral gyrus white matter bordering on the SCR that relate to subsequent depression onset in a sex-specific manner. Fibers in this region extend to the corticospinal tract and internal capsule (Mori et al., 2005; Wakana et al., 2004), and have been associated with psychomotor retardation in depression (Bracht et al., 2012; Bracht et al., 2018; Coloigner et al., 2019).

While it is possible that these integrative white matter pathways differentially contribute to the various affective, cognitive, and somatic domains of emerging depression, an alternative explanation could be that a general pattern of lower FA constitutes a non-specific risk phenotype for mood and anxiety disorders. At least one other group has suggested that a pattern of lower FA may be shared across multiple “emotional disorders,” including major depression, social anxiety, obsessive compulsive disorder, and bipolar disorder (Jenkins et al., 2016). However, in this analysis we did not observe any relationship between the variations in white matter microstructure at baseline that were associated with future depression and either future substance use (alcohol, marijuana) or future anxiety disorders (GAD, OCD). This finding may be a result of limited statistical power, given that only a few participants endorsed other anxiety disorders (n = 12); therefore, whether these risk phenotypes we have identified are specific to depression will warrant further study.

Although this work provides insight into white matter microstructural features that may relate to risk for depression in adolescence, several limitations should be considered. First, because our current analyses capture only a specific developmental window, it is possible that some participants who have not yet demonstrated increases in depressive symptoms will still go on to develop depression or other psychopathology in adulthood. The findings we have reported here will need to be replicated in larger samples, potentially using more complex modeling strategies that can take multiple longitudinal time points into account. Secondly, the SSAGA’s designation for “current” symptoms only includes those that have occurred in the past two weeks. Annual administration of this measure is unlikely to give us a comprehensive estimate of fluctuating depressive symptoms throughout the follow-up period, though it can provide intermittent snapshots of an individual’s mental health. Additionally, the sample bias in this study (predominantly Caucasian participants of high socioeconomic status) limits the generalizability of our findings. It should be noted that due to the exclusion of participants who exceeded age-specific drinking thresholds for risky alcohol use at baseline, this sample is also under-representative of substance use relative to the general population (SAMHSA, 2020). Though this also constrains generalizability, we maintain that for the purposes of examining premorbid risk phenotypes related to depression it was important to minimize the effects of substance use on baseline neuroimaging metrics by excluding these participants. Lastly, despite the fact that DWI is a useful tool for the non-invasive, indirect examination of in vivo white matter microstructure, it does not afford us the ability to make inferences about the specific neurobiological mechanisms (e.g. myelination, axonal density) underlying these observed differences in diffusion indices.

In sum, we have demonstrated differential associations between white matter microstructure and subsequent development of MDD in male and female adolescents. Female adolescents who later developed major depression exhibited lower baseline FA in a cluster of precentral gyrus white matter adjacent to the SCR compared to healthy controls, whereas male adolescents who went on to develop depression displayed higher FA values in this same region. These findings suggest that sex-specific variations in the coherence of major integrative white matter tracts, aside from those directly implicated in emotion regulation, may underlie vulnerability for adolescent-emergent depression. Future research should employ longitudinal analyses to tease apart the relationships between white matter microstructure maturation, symptom trajectories, and clinical outcomes to further inform treatment and prevention efforts.

Highlights.

• Patterns of white matter microstructure may indicate vulnerability for depression

• Adolescents who later develop depression mostly exhibit lower fractional anisotropy

• The relationship between diffusion indices and future depression is sex-specific