Structural covariance network in the striatum is modulated by non-suicidal self-injury in adolescents with major depressive disorder

Abstract

Background

Adolescents with major depressive disorder (MDD) significantly affect the development and well-being of this group. Approximately one-third of these adolescents engage in non-suicidal self-injury (NSSI) behavior, which notably increases their risk of suicide. Striatal abnormalities, particularly in the corticostriatal circuit, have been implicated in depression. The striatum is a crucial part of the reward system, and its abnormal connectivity with other brain regions within the system has been linked to repetitive NSSI behavior in adolescents. This study aims to investigate structural covariance (SC) connectivity patterns within the corticostriatal circuitry in adolescents with MDD and explore how NSSI modulates these patterns.

Methods

This study employed SC analysis to examine T1-weighted structural images from 92 adolescents with MDD and 81 healthy controls (HCs). Using voxel-based morphological analysis to assess gray matter volume, we assessed aberrant SC in the striatum of these patients and explored the influence of NSSI severity on this pattern.

Results

Compared with HCs, adolescents with MDD exhibited increased SC in the corticostriatal circuit. In addition, the NSSI total scores and its sub-dimensions significantly modulated SC in the striatum, medial orbitofrontal cortex, superior frontal gyrus, cingulate cortex, and thalamus. Increased SC was observed within the caudate nucleus and other components of the reward system, whereas a decrease was observed in the globus pallidus, putamen, and thalamus.

Conclusions

The corticostriatal circuit, vital for emotional processing, appears pivotal in the early stages of adolescents with MDD. Our findings underscore the critical effects of NSSI on neurophysiological alterations in these patients.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-025-07662-x.

Introduction

Adolescents with major depressive disorder (MDD) pose significant health risks, with elevated rates of morbidity and mortality [1]. Research indicates that up to 20% of adolescents will experience MDD before adulthood [2]. Compared with adults, these young individuals are associated with an increased risk of recurrent depressive episodes, impaired social and occupational functioning, and reduced patient quality of life [3, 4]. Additionally, childhood neuropsychiatric disorders are globally recognized as a leading cause of disease burden among young populations [5], emphasizing the importance of early identification and intervention in adolescents with MDD for enhancing public health outcomes.

Non-suicidal self-injury (NSSI) represents an important risk factor for MDD [6]. NSSI refers to deliberate acts of self-harm without suicidal intention [7, 8], including behaviors ranging from forms without obvious tissue injury (e.g., hitting, choking, and scratching) to those that cause obvious tissue injury (e.g., cutting and burning) [9]. Typically recurrent and socially unacceptable, these behaviors are common among adolescents and young adults and are associated with mood disorders and depressive symptoms [10], with a lifetime prevalence of 13–17% [11, 12]. In addition, NSSI behavior can lead to physical harm, impaired social functioning, and reduced productivity at work or school [13]. Despite its high incidence and association with high risk of adolescent suicide, the underlying mechanisms of NSSI behavior remain poorly understood [14], highlighting the need for further investigation into its neural mechanisms.

Structural covariance (SC) connectivity analysis offers a valuable tool for exploring neural developmental coordination across various cerebral cortex regions and assesses the anatomical connectivity between different brain regions [15]. An increase in SC indicates hyper-synchronous developmental coordination or maturation between brain regions of the same individual, whereas decreased SC suggests a mismatch in neural maturation rates [16, 17]. Unlike functional magnetic resonance imaging (fMRI), which is susceptible to state-dependent signal fluctuations and artifacts [18]. SC analysis provides insights into long-term connectivity patterns reflecting developmental maturity or specific characteristics [15], which are crucial for understanding the neuropathological mechanisms of depression. Previous research has demonstrated that the development of depression is intimately linked to abnormal neurodevelopmental trajectories and the connections among various brain regions [19]. However, research on SC changes in adolescents with MDD remains relatively limited.

Previous studies have linked diminished positive emotions in patients with MDD to striatal dysfunction within the reward system [20, 21]. Rodent research has established a connection between striatal function and depressive behaviors [22]. However, the etiology of depression is complex and likely involves interactions between specific brain circuits rather than dysfunction in a single brain region [23]. Recent studies have highlighted abnormalities in corticostriatal circuits in the diagnosis and severity of MDD [24, 25]. Gabbay et al. identified abnormalities in resting-state functional connectivity of the corticostriatum in patients with MDD, which was associated with impaired emotion regulation function [26]. Nonetheless, whether corticostriatal SC is abnormal in adolescents with MDD remains unclear, warranting further investigation.

The striatum, a crucial hub in the brain’s reward system, plays an important role in reward-punishment learning, goal orientation, and habitual behavior [27]. In addition to the striatum, the reward system includes the anterior cingulate cortex (ACC), orbitofrontal cortex (OFC), dorsal prefrontal cortex (DLPFC), ventral striatum, hippocampus, and thalamus, all involved in various aspects of reward processing [28] NSSI behavior shares similarities with behavioral addiction, characterized by compulsive and uncontrollable self-injurious behaviors [29]. Given the reward system’s role in addiction, it likely contributes to the neural mechanisms underlying NSSI behavior [30]. Therefore, crucial knowledge gaps remain. It is currently unknown whether adolescents with MDD exhibit aberrant SC within the corticostriatal circuit, and how the clinically critical behavior of NSSI modulates these long-term structural networks. Elucidating this relationship is key to understanding the neurodevelopmental basis of MDD and its association with NSSI behaviors.

Based on the established role of the corticostriatal circuitry in reward processing, motivation, and habitual behavior, we hypothesized that: (1) Adolescents with MDD would exhibit increased SC within the corticostriatal circuit compared to HCs, reflecting a potential compensatory mechanism or early disease-related neurodevelopmental synchronization. (2) The severity of NSSI would significantly modulate these SC patterns, with positive modulation expected in circuits subserving reward and emotion, and negative modulation in circuits underlying habitual behavior and executive control.

Methods

Participants

This study utilized baseline data from a registered clinical trial (registration number: ChiCTR2100042346, registration date: 2021-01-19), involving 92 adolescents diagnosed with MDD and 81 HCs. The study design adhered to the Helsinki Declaration (revised in 2008). Ethics approval was obtained from the Ethics Committee of The Affiliated Brain Hospital, Guangzhou Medical University, with written informed consent obtained from all participants and their parents. Participants in both groups were required to self-assess the severity of their depressive symptoms using the 17-item Hamilton Depression Rating Scale (HAMD-17), a standard clinical instrument for depression evaluation. Diagnoses of MDD were made by a psychiatrist trained in the criteria outlined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5).

Participants with MDD eligible for the study met the following inclusion criteria: (i) age 12–18 years, (ii) meeting DSM-5 criteria for depression, (iii) no psychiatric medications for 4 weeks before enrollment, and (iv) HAMD-17 score > 7. Exclusion criteria were (i) diagnoses other than depression, (ii) organic brain diseases or history of brain trauma/surgery, (iii) contraindications for MRI or claustrophobia, and (iv) significant cardiac, liver, or renal diseases.

HCs were adolescents aged 12 to 18 years, recruited from nearby communities through poster advertisements. To ensure the absence of psychiatric conditions, all HCs underwent the same structured clinical interview by a trained psychiatrist as the MDD group. Individuals who met the DSM-5 criteria for any major psychiatric disorder (including but not limited to depressive disorders, anxiety disorders, psychotic disorders, and substance use disorders) or had a history of such disorders were excluded. Those with serious medical or neurological illnesses, or contraindications for MRI were also excluded.

Clinical assessment

The Adolescent Non-suicidal Self-injury Assessment Questionnaire (ANSAQ) was compiled by Wan et al. [31] and further refined in 2018 [32]. The questionnaire evaluates self-injury behaviors over the past year through 12 items, divided into two dimensions: NSSI without obvious tissue injury (pinching or hitting oneself with a fist or object; items 1–7) and NSSI with obvious tissue injury (stabbing, cutting, or burning, potentially leading to bleeding, scratches, or other injuries; items 8–12). Based on the Likert 5-point scale, each item offers five options: “never”, “rarely”, “sometimes”, “often”, and “always”, scored from 0 to 4, respectively. Participants are classified as having NSSI behavior if they report engaging in NSSI at least once. The total score ranges from 0 to 48, with higher scores indicating more severe self-harm behavior. ANSAQ has satisfactory reliability and validity. In this study, Cronbach’s α of the questionnaire was 0.82 [32].

MRI data acquisition

MRI was conducted at the Magnetic Resonance Center of the Brain Hospital of Guangzhou Medical University using a Siemens Magnetom Prisma 3.0T scanner equipped with a 64-channel head coil. High-resolution T1-weighted anatomical images were acquired using a magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence in the sagittal orientation. The specific parameters were as follows: repetition time (TR) = 2000 ms; echo time (TE) = 2.32 ms; inversion time (TI) = 900 ms; flip angle = 8°; field of view (FOV) = 256 × 256 mm²; matrix size = 256 × 256; slice number = 208; and voxel size = 0.9 × 0.9 × 0.9 mm³. This parameter set was chosen to optimize the contrast between gray and white matter and to achieve high spatial resolution, which is critical for precise voxel-based morphometry analysis.

Voxel-based morphometry analysis

All structural images were processed and analyzed using the Computational Anatomy Toolbox (CAT12, http://dbm.neuro.uni-jena.de/cat12/) within the SPM12 framework, running in MATLAB R2019b. The preprocessing pipeline followed the standard CAT12 protocol, which included the following steps:

1. Spatial Registration: Images were initially aligned to a standard orientation.

2. Bias Field Correction: Inhomogeneities in the magnetic field were corrected to ensure uniform intensity across the image.

3. Tissue Segmentation: Images were segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using an adaptive maximum a posteriori technique and a hidden Markov Random Field model.

4. Spatial Normalization: The segmented GM images were non-linearly normalized to the Montreal Neurological Institute (MNI) space using the DARTEL algorithm.

5. Modulation: The normalized GM maps were modulated by the Jacobian determinants derived from the spatial normalization to preserve the total amount of GM volume from the original images.

6. Smoothing: The modulated GM maps were smoothed with an 8 mm full-width at half-maximum (FWHM) Gaussian kernel. This kernel size was chosen as it is a standard and widely used value in VBM studies, providing an optimal balance between increasing the signal-to-noise ratio and retaining the spatial specificity of the findings.

Total intracranial volume (TIV) was automatically estimated by CAT12 during the segmentation process as the sum of GM, WM, and CSF volumes. TIV was included as a covariate in all statistical models to control for individual differences in overall brain size.

Quality Control: All preprocessed GM maps were rigorously checked for segmentation and normalization accuracy using the built-in CAT12 quality control tools. No datasets exhibited excessive motion artifacts or poor registration quality that warranted exclusion.

Group-level differences in striatal SC

Based on previous studies implicating the striatum in reward processing and depression [20, 21, 27], we selected it as our region of interest (ROI). The striatum was segmented into six bilateral ROIs: the caudate nucleus, globus pallidus, and putamen. The specific MNI coordinates and the 6 mm spherical radius for these ROIs were defined based on prior work that established their functional and structural relevance in corticostriatal circuits [33, 34], thereby providing a strong a priori justification for our analysis.

SC analysis was conducted using the Brain Covariance Connection Toolkit (BCCT_V2.1; https://github.com/JLhos-fmri/BrainCovarianceConnectToolkitV2.1) in MATLAB (R2019b). This technique employs Pearson’s correlation analysis on the local GM volume (GMV) of selected areas and GMV of brain-wide regions at the voxel level to explore broad structural network patterns.

This study examined differences in group-level SC across groups using the standard general linear model (GLM) for testing, with the following results.

GLM examined variations in GMV correlations between the striatum and other brain areas in adolescents with MDD compared to HCs. The specific items of GLM are as follows: β₀ is the intercept term, and β₁, β_2, and β₃ are considered covariates, representing age, sex, and total intracranial volume (TIV). respectively. The relationships between V_i and V _Seed are captured by β₄. β₅ modeled the relationship between the group term and V_i, and β₆ represented the interaction term (group by V_seed) with V_i. Ultimately, the results were adjusted to account for multiple comparisons. A Gaussian random field (GRF) correction was used to account for multiple comparisons, with a voxel-wise p < 0.001 and a cluster-level p < 0.05 correction.

Modulation of striatal SC in adolescents with MDD

We investigated how the NSSI total score and sub-dimensional scores of self-injury behavior (with or without obvious tissue injury) modulated striatal SC. To further control for potential confounding variables affecting the interaction effects of the NSSI dimension, we included HAMD-17 depression scores as a covariate.

In the model similar to GLM previously discussed, β0 is the intercept term, β1, β2, and β3 represent covariates for age, gender, and TIV, respectively. β4 is a covariate of the HAMD depression score. β5 indicates the relationship between Vi and V_seed. β₆ modeled the association between V_i and the overall NSSI score and its two sub-dimensional scores, and β₇ represents the relationship between the interaction term (V_seed by score) and V_i. A two-tailed one-sample t-test was performed on β7 in the adolescents with MDD group. A positive interaction indicates that individuals with higher scores have higher SC and vice versa. GRF correction was used to account for multiple comparisons, with a voxel-wise p < 0.001 and a cluster-level p < 0.05 correction.

Statistical analysis

All statistical analyses were performed using the GLM framework within the BCCT_V2.1 toolkit. The GLM was selected as it provides a flexible and robust approach for testing hypotheses about the relationships between brain structure, clinical group, and behavioral measures, while allowing for the inclusion of continuous and categorical covariates.

For all models, age, sex, and TIV were included as covariates of no interest to control for their well-established effects on gray matter volume.

Critically, in the modulation analysis (section “Modulation of striatal SC in adolescents with MDD”), the HAMD-17 depression score was included as an additional covariate. This was done to statistically isolate the specific relationship between NSSI severity and striatal SC, ensuring that any observed modulation was not merely a proxy for the overall severity of depressive symptoms. Our goal was to test whether NSSI explains unique variance in brain structure beyond what is accounted for by the core symptoms of depression.

To account for multiple comparisons across the whole brain, we employed GRF theory for cluster-level inference. GRF correction was chosen because it controls the family-wise error (FWE) rate in spatially continuous data like structural images, providing a balance between sensitivity and specificity. The primary significance threshold was set at a voxel-level p < 0.001 in combination with a cluster-level p < 0.05 (FWE-corrected) [35, 36].

For all analyses, clusters surviving the GRF correction are reported. To quantify the magnitude and location of effects, we report the peak T-value and the cluster size (in voxels) for each significant cluster [37, 38]. The T-value serves as a standardized measure of effect strength at the peak voxel within a cluster. The degrees of freedom for all models are inherent in the calculation of the T-value and are determined by the sample size and number of regressors in the GLM.

Validation analysis of striatal SC

To address potential confounding factors and confirm the robustness of the observed striatal structural covariance (SC) patterns, we performed validation analyses in stratified subgroups. Specifically, the main analyses of group differences and NSSI modulation were repeated separately within the female subgroup, the male subgroup, and a medication-naïve MDD subgroup. Considering the potential reduction in statistical power within these smaller subgroups, we employed the following statistical thresholds: For the sex-stratified group comparisons, a Gaussian random field (GRF) correction was applied with a voxel-wise p < 0.05 and a cluster-level p < 0.05. For all other analyses, including the group comparison in the medication-naïve subgroup and all modulation analyses (within the full sample, sex subgroups, and medication-naïve subgroup), we maintained the more stringent threshold of voxel-level p < 0.001 and cluster-level p < 0.05.

Results

Demographic and clinical characteristics

Table 1 summarizes the demographic and clinical profiles of the adolescents with MDD and HCs. The two groups were well-matched in terms of age (t = -1.176, p = 0.241) and education years (t = -0.799, p = 0.425). However, as is often observed in adolescent MDD cohorts, the groups differed significantly in sex distribution (χ² = 33.956, p < 0.001), with a higher proportion of females in the MDD group (81.52%) compared to the HC group (38.27%).

Table 1.

Demographic characteristics and clinical data of patients with MDD and HCs

Characteristics	MDD (n = 92)	HCs (n = 81)	t/c²	p value
Age (years)	14.53 (1.66)	14.85 (1.91)	-1.176	0.241a
Sex			33.956	< 0.001 b
Female	75 (81.52%)	31 (38.27%)
Male	17 (18.48%)	50 (61.73%)
Education years	8.75 (1.73)	8.98 (1.97)	-0.799	0.425a
Duration of illness (months)	13.27 (1.81)
Lifetime medication history	24 (26.09%)
Current medication (at scan)
Yes	9 (9.78%)
No	83 (90.22%)
Antidepressant medication (SSRIs)
Yes	5 (5.43%)
No	87 (94.57%)
Atypical antipsychotic medication (Quetiapine)
Yes	1 (1.09%)
No	91 (98.91%)
Mood stabilizer medication (Sodium valproate)
Yes	1 (1.09%)
No	91 (98.91%)
Benzodiazepines (Alprazolam)
Yes	2 (2.17%)
No	90 (97.83%)
HAMD-17 score	22.10 (6.17)	1.32 (2.27)	30.081	< 0.001 a
NSSI total score	9.79 (9.63)	0.17 (0.54)	9.566	< 0.001 a
Without obvious tissue injury score	7.12 (6.98)	0.15 (0.50)	9.549	< 0.001 a
With obvious tissue injury score	2.67 (3.03)	0.02 (0.16)	8.370	< 0.001 a

Data are presented as the mean (standard deviation) or frequencies (proportion)

Abbreviations: MDD, major depressive disorder; HCs, healthy controls; SSRIs, Selective Serotonin Reuptake Inhibitors; HAMD-17, 17-item Hamilton Depression Rating Scale; NSSI, Non-Suicidal Self-Injury

^a Two-sample t-test

^b Chi-square test

Clinically, as expected, the MDD group reported a substantial duration of illness (13.27 ± 1.81 months) and exhibited significantly higher scores on the HAMD-17 (t = 30.081, p < 0.001) and all NSSI measures (all p < 0.001) compared to HCs.

Regarding medication exposure, 24 patients (26.09%) had a lifetime history of psychotropic medication use. Crucially, only 9 patients (9.78%) were actively receiving medication at the time of the MRI scan. This medicated subgroup was small, comprising individuals on selective serotonin reuptake inhibitors (SSRIs, n = 5), benzodiazepines (n = 2), an atypical antipsychotic (n = 1), and a mood stabilizer (n = 1).

Group-level differences in striatal SC

Our analysis revealed aberrant SC primarily within the corticostriatal circuit among adolescents with MDD (n = 92) compared to HCs (n = 81). Specifically, increased SC was noted between the left caudate nucleus and both the ipsilateral superior frontal gyrus and the middle temporal gyrus. Conversely, decreased SC was observed between the right middle temporal gyrus and right caudate nucleus. In addition, increased SC was found (1) between the bilateral globus pallidus and left middle occipital gyrus, right inferior occipital gyrus, and right cerebellum, (2) between the right putamen and right rectus, and (3) between the left putamen and left middle occipital gyrus, right inferior occipital gyrus, and right cerebellum. Overall, most SC connectivity within the corticostriatal circuit showed an increase. For detailed information, see Fig. 1; Table 2.

Fig. 1.

Altered structural covariance network of the bilateral striatum in MDD compared with HCs

Table 2.

Altered structural covariance network of the bilateral striatum in MDD compared with HCs

ROI	Clusters	Cluster sizes	Including regions	Peak T-value	Peak MNI coordinate (mm)
					X	Y	Z
Left caudate nucleus	1	631	Left caudate nucleus	-4.54	-15	26	-6
	2	313	Left middle temporal gyrus	3.94	-39	-50	17
	3	417	Left superior frontal gyrus	3.75	-17	54	36
Left globus pallidus	1	369	Left middle occipital gyrus	4.20	-21	-105	-3
	2	1317	Right inferior occipital gyrus	4.18	30	-87	-15
	3	623	Right cerebellum	4.52	54	-53	-45
Left putamen	1	320	Left middle occipital gyrus	3.86	-21	-105	-3
	2	1626	Right inferior occipital gyrus	4.13	29	-87	-15
	3	429	Right cerebellum	4.23	57	-53	-41
Right caudate nucleus	1	234	Right middle temporal gyrus	-4.16	60	-54	11
Right globus pallidus	1	399	Left middle occipital gyrus	4.35	-20	-105	-3
	2	1377	Right inferior occipital gyrus	4.08	20	-99	-2
	3	348	Right cerebellum	4.14	54	-53	-45
Right putamen	1	302	Right rectus	3.98	3	14	-20

Abbreviations: ROI: region of interest

Modulation of striatal SC in adolescents with MDD

Within the MDD group (n = 92), we further examined how NSSI total scores and its sub-dimensions modulated changes in striatal SC. Significant effects were observed on striatal SC for the NSSI total score and its sub-dimensions. Specifically, the NSSI total score and self-injury without obvious tissue injury positively modulated SC in the medial right superior frontal gyrus, left parahippocampal gyrus, and bilateral caudate nucleus. NSSI scores indicating obvious tissue injury positively modulated SC in the right medial superior frontal gyrus, left inferior frontal gyrus triangle, left medial orbitofrontal cortex, and bilateral caudate nucleus. Additionally, the intensity of SC between the left caudate nucleus and right cingulate gyrus was positively modulated by scores indicating obvious tissue injury. For detailed information, see Fig. 2; Table 3.

Fig. 2.

Modulation pattern of the NSSI total and item scores on structural covariance network of the bilateral caudate nucleus in MDD patients

Table 3.

Modulation pattern of the NSSI total and item scores on structural covariance network of the bilateral striatum in MDD patients

ROI/scores	Clusters	Cluster sizes	Including regions	Peak T-value	Peak MNI coordinate (mm)
					X	Y	Z
Left caudate nucleus
NSSI total	1	983	Right medial superior frontal gyrus	4.55	8	62	12
Without obvious tissue injury	1	317	Left parahippocampal gyrus	3.93	-12	-29	-17
	2	725	Right medial superior frontal gyrus	4.48	8	62	12
With obvious tissue injury	1	998	Right cingulate gyrus	4.18	6	23	35
	2	1409	Right medial superior frontal gyrus	4.28	9	53	-2
	3	520	Left inferior frontal gyrus triangle	3.85	-45	32	17
	4	363	Left medial orbitofrontal cortex	4.55	-14	44	-6
Right caudate nucleus
NSSI total	1	890	Right medial superior frontal gyrus	4.35	8	62	12
	2	357	Left parahippocampal gyrus	4.02	-15	-18	-24
Without obvious tissue injury	1	686	Right medial superior frontal gyrus	4.31	8	62	12
With obvious tissue injury	1	449	Right cingulate gyrus	3.70	8	24	38
	2	1296	Right medial superior frontal gyrus	4.08	9	53	-2
	3	745	Left inferior frontal gyrus triangle	4.10	-54	29	21
	4	359	Left medial orbitofrontal cortex	4.41	-14	44	-6
Left globus pallidus
NSSI total	1	769	Left thalamus	-4.53	0	-9	-3
Without obvious tissue injury	1	665	Right thalamus	-4.42	0	-9	-2
Left putamen
NSSI total	1	668	Left thalamus	-4.40	0	-8	-3
Without obvious tissue injury	1	609	Left thalamus	-4.32	0	-6	-2
Right putamen
Without obvious tissue injury	1	358	Right thalamus	-3.89	0	-8	-3

Abbreviations: ROI: region of interest; NSSI, Non-Suicidal Self-Injury

In contrast, the NSSI total score and self-injury without obvious tissue injury negatively regulated the SC of the globus pallidus and putamen. Specifically, as the NSSI total score and self-injury without obvious tissue injury score increased, SC decreased between the left globus pallidus and bilateral thalamus and between the left putamen and left thalamus. Additionally, NSSI scores indicating self-injury without obvious tissue injury negatively modulated SC between the right putamen and right thalamus. For detailed brain mapping, see Figs. 3 and 4.

Fig. 3.

Modulation pattern of the NSSI total and item scores on structural covariance network of the bilateral globus pallidus in MDD patients

Fig. 4.

Modulation pattern of the NSSI total and item scores on structural covariance network of the bilateral putamen in MDD patients

In summary, our findings suggest that in adolescents with MDD, alterations in SC predominantly involve connections between the striatum and other key regions within the reward system.

Validation analysis results

Sex-stratified group-level differences in striatal SC

In the female subgroup (MDD: n = 75, HC: n = 31), the between-group differences in SC did not survive the stringent threshold identical to the primary analysis (GRF correction, voxel-level p < 0.001, cluster-level p < 0.05). Acknowledging the potential reduction in statistical power due to the smaller sample size, we subsequently performed an exploratory analysis using a more lenient threshold (GRF correction, voxel-level p < 0.05, cluster-level p < 0.05). Under this condition, the overall pattern of increased SC within the corticostriatal circuit observed in the primary analysis was largely replicated (Fig. S1). Specifically, SC was increased across most primary corticostriatal connections, with the exception of the right putamen. In the male subgroup (MDD: n = 17, HC: n = 50), similar trends concordant with the primary findings were observed in the bilateral caudate nucleus. However, no significant group differences were detected in the bilateral globus pallidus or putamen (Fig. S2).

Sex-stratified modulation of striatal SC by NSSI

The modulatory effects of NSSI on striatal SC were further examined within the female MDD subgroup (n = 75). Consistent with the primary analysis, the NSSI total score and its sub-dimension scores significantly modulated the SC between the bilateral caudate nucleus and key regions of the reward system, including the medial superior frontal gyrus, parahippocampal gyrus, and orbitofrontal cortex (Fig. S3). Concurrently, the negative modulation of SC between the globus pallidus and thalamus by NSSI scores was also preserved (Fig. S4). In the male MDD subgroup (n = 17), no significant modulation effects were observed, likely attributable to the limited sample size.

Validation in medication-naïve subgroup

To definitively address potential confounding effects of psychotropic medication, we repeated our primary analyses in a subgroup of medication-naïve MDD patients (n = 68). When comparing these patients to all HCs (n = 81) under the original stringent threshold, the pattern of increased structural covariance within the corticostriatal circuit was successfully replicated (Fig. S5). Furthermore, within this medication-naïve subgroup, the modulatory effects of NSSI on striatal SC were also preserved. The NSSI total score and its sub-dimensions significantly and positively modulated the SC between the bilateral caudate nucleus and key reward regions (Fig. S6). Concurrently, the negative modulation of SC linking the globus pallidus and putamen to the thalamus by NSSI scores remained robust and significant (Figs. S7 and S8).

Discussion

This study employed SC analysis to delineate SC network abnormalities in the corticostriatal circuit among adolescents with MDD. Specifically, we observed significant alterations in SC between the striatum and multiple cortical regions, including the superior frontal gyrus, middle temporal gyrus, and middle occipital gyrus. Furthermore, regardless of the NSSI total score and severity of NSSI with or without obvious tissue injury, striatal SC abnormalities were significantly modulated, mainly reflected in changes within the reward system. These findings highlight the pivotal role of the corticostriatal circuit, which is implicated in emotional and behavioral processing, and the reward system in the early neurodevelopment of MDD in adolescents.

Previous research on SC networks has successfully advanced our understanding of mental and neurological disorders, including Alzheimer’s disease and schizophrenia [39, 40]. Despite these developments, studies specifically targeting depression, particularly among adolescents, remain relatively limited. Recent retrospective analyses of resting-state functional connectivity indicate that adolescents with MDD who engage in NSSI exhibit abnormal functional connectivity in limbic system regions like the amygdala and cingulate cortex [14]. However, these studies primarily focus on short-term brain activity and fail to reveal the developmental coordination changes of brain regions over longer time scales. Furthermore, while existing studies have examined the link between depression and the reward system, they predominantly focus on specific areas of the striatum such as the nucleus accumbens [41]. Our research uses three principal parts of the striatum as seed points to better understand the changes in structural covariance within the corticostriatal circuit in adolescent with MDD.

Our study revealed significant changes in the SC of the corticostriatal circuit in adolescents with MDD compared with HCs. Notably, there was a general increase in SC strength between the striatum and most cortical regions(Fig. 1). The cortex serves as a hub for complex emotional and cognitive regulation, whereas the striatum, which receives excitatory inputs from both the cortical and subcortical regions, plays a pivotal role in sustaining positive emotions [42–44]. The observed surge in corticostriatal circuit SC could reflect an early sign of the disease or a compensatory mechanism in response to emotional imbalances during the development of adolescents with MDD [26]. These findings hold significance for the early identification of patients with MDD and individuals at high future risk of developing the disorder.

In addition, our data highlight the important role of NSSI in modulating SC between the striatum and other brain regions within the reward system. We observed an increase in SC intensity between the bilateral caudate nucleus and key brain regions of the reward system (e.g., DLPFC, OFC, and ACC) as the total score of NSSI and severity of NSSI with or without obvious tissue injury increased (Fig. 2). This pattern of heightened SC may reflect aberrant neurodevelopmental synchronization in circuits underlying motivated behavior, possibly related to the compulsive characteristics of NSSI [16]. To interpret these findings, we propose a testable model centered on dopaminergic dysfunction. Given the established role of dopamine in reward signaling and reinforcement learning [45, 46], the observed SC increases in reward-related pathways might indicate a substrate for altered dopaminergic function in adolescents engaging in NSSI. One plausible hypothesis is that these structural patterns correlate with a dysregulated dopamine system, wherein NSSI behaviors may initially trigger transient dopamine release—potentially serving as a maladaptive coping mechanism [47]—but ultimately contribute to long-term alterations in dopaminergic homeostasis [48]. This framework offers a compelling neurobiological account for the addictive qualities often observed in NSSI [49, 50], though future studies directly measuring dopamine function are needed to validate this model.

The thalamus is considered the core integration hub of the brain, transferring information between the cortex and basal ganglia [51, 52]. The putamen and globus pallidus primarily receive inputs from sensorimotor areas and the thalamus, which play key roles in the development of habitual behaviors [53]. Disruptions in this circuit are considered the main cause of executive control dysfunction [54]. We noted a reduction in SC between the globus pallidum, putamen, and the reward system (e.g., thalamus) in NSSI cases (Figs. 3 and 4), suggesting a disruption in developmental coordination among these brain regions, which likely impairs behavioral control [55] and patients’ capacity to properly inhibit or manage NSSI behavior, thereby facilitating the recurrence of such behaviors [56].

Our validation analyses bolster confidence in the core findings. While the group-level SC differences in the female subgroup required a more lenient threshold, potentially due to reduced power, the key modulation effects of NSSI remained significant. Crucially, all primary results were robustly replicated in the medication-naïve subgroup under the strict criteria. This consistency indicates that the observed SC alterations are intrinsic features of adolescent MDD and its interaction with NSSI, rather than artifacts of medication or residual confounding.

This work provides valuable neuroimaging evidence with potential clinical implications. The distinct SC patterns associated with MDD and NSSI could, upon further validation, contribute to objective markers for identifying adolescents at high risk for self-injury. Furthermore, delineating the specific circuits modulated by NSSI (e.g., caudate-reward vs. pallidum-thalamus connections) could eventually inform more personalized intervention strategies, such as targeting reward-based interventions for some patients and cognitive control training for others.

Future studies should prioritize longitudinal designs to elucidate the causal and developmental trajectories of these SC changes. Combining SC with fMRI in a multimodal framework would help clarify the relationship between long-term structural maturation and short-term brain function. Finally, replicating these findings in larger, prospectively recruited medication-naïve samples will be crucial to solidify the present conclusions and further dissect the neurobiology of adolescent MDD from confounding treatment effects.

Conclusion

This study identified enhanced SC within the corticostriatal circuit in adolescents with MDD, indicative of early disease progression or compensatory mechanisms. In addition, NSSI modulates SC between the caudate nucleus and other brain regions of the reward system, potentially leading to excessive synchronous development of brain regions, disruption of the normal reward mechanisms, and NSSI addiction and dependence. Conversely, decreased SC between the globus pallidus, putamen, and thalamus suggests neural maturation gaps in these brain regions, contributing to behavioral control disorders and promoting repetitive NSSI behaviors. Taken together, these findings shed light on striatal SC abnormalities in adolescents with MDD and offer neuroimaging insights for identifying such patients with a history of NSSI.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MDD

Major depressive disorder

NSSI

Non-suicidal self-injury

SC

Structural covariance

ACC

Anterior cingulate cortex

OFC

Orbitofrontal cortex

DLPFC

Dorsal prefrontal cortex

ROI

Region of interest

HCs

Healthy controls

HAMD-17

17-item Hamilton Depression Rating Scale

DSM-5

Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition

ANSAQ

Adolescent Non-suicidal Self-injury Assessment Questionnaire

GM

Gray matter

FWHM

Full width at half maximum

GMV

Gray matter volume

GLM

General linear model

TIV

Total intracranial volume

GRF

Gaussian random field

Author contributions

LZJ made substantial contributions to the conception and design of the work; CYY was responsible for the acquisition, analysis, and interpretation of data. XYB performed the additional statistical analyses for the revision, assisted in drafting the response to reviewers, and critically revised the manuscript. QXW, LWC, HZB, LHN, WCY, LXF, MSM, LGX, ZF, ZYX, CXY, YZR, LYM, and CYF collected data and critically revised the manuscript. YL and YP also critically revised the manuscript. All authors have read and approved the final manuscript. LZJ and CYY contributed equally to this work.

Funding

This work was supported by the National Natural Science Foundation of China (grant number 82471546, 82322024), Guangzhou Health Science and Technology Project (grant number 20251A010033), Innovative Clinical Technique of Guangzhou (2024–2026), Guangzhou Key Clinical Specialty (Clinical Medical Research Institute).

Data availability

The data supporting the findings of this study are available from the corresponding authors upon reasonable request.

Declarations

Ethics approval and consent to participate

All subjects provided written informed consent. The study was approved by the Ethics Committee of The Affiliated Brain Hospital, Guangzhou Medical University. This study is registered in the Chinese Clinical Trial Registry (Clinical Trial Registration Number: ChiCTR2100042346, Registration Date: 2021-01-19) and was conducted in accordance with the revised Declaration of Helsinki (2008).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.