Gray matter volume alterations in early-adulthood major depressive disorder patients with non-suicidal self-injury and suicide attempts

Abstract

Background

Major depressive disorder (MDD) is frequently accompanied by self-harm behaviors, including non-suicidal self-injury (NSSI) and suicide attempts (SA), which are important risk factors for suicide and poor clinical outcomes. Although neuroimaging studies have identified structural brain abnormalities related to suicidal behaviors in MDD, the neurobiological distinctions between NSSI and SA remain poorly understood. In particular, few studies have simultaneously compared MDD patients with NSSI, MDD patients with SA, and MDD patients without self-harm behaviors. Therefore, this study aimed to investigate gray matter volume (GMV) alterations and their associations with clinical symptoms in early adulthood (aged 18–30 years) MDD patients using voxel-based morphometry (VBM).

Methods

A total of 54 MDD patients with NSSI (MDD/NSSI), 68 MDD patients without NSSI (sMDD), 50 MDD patients with SA (MDD/SA), and 66 healthy controls (HCs) were included. VBM was used to examine GMV differences using high-resolution T1-weighted MRI scans. Age, sex, education, and intracranial volume were included as covariates. One-way ANOVA with Gaussian random field (GRF) correction was performed, followed by post-hoc t-tests. Correlations between GMV and clinical measures of depression (HAMD), anxiety (HAMA), and suicide risk (NGASR) were assessed.

Results

Significant GMV differences were observed in the superior/middle frontal gyrus, superior frontal gyrus, bilateral caudate, superior/middle temporal gyrus, and middle cingulate gyrus (voxel-level p < 0.001, GRF corrected p < 0.05). Post-hoc analyses showed reduced GMV in the MDD/SA group compared with the MDD/NSSI group in several regions. However, supplementary GLM analyses indicated that NSSI-related characteristics showed more consistent independent associations with regional GMV, whereas SA history was not independently associated with GMV in any examined region. Correlations between GMV and clinical measures did not survive FDR correction.

Conclusions

MDD patients with NSSI and SA may exhibit distinct patterns of gray matter alterations; however, these differences likely reflect clinical heterogeneity across self-harm subgroups rather than SA-specific structural abnormalities. Supplementary analyses further suggest that NSSI-related characteristics show stronger independent associations with GMV than SA history. These findings highlight the importance of distinguishing self-harm subtypes in the neurobiological characterization of MDD. All results should be interpreted cautiously given the cross-sectional design and the lack of robust associations after multiple-comparison correction.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12888-026-08333-1.

Introduction

Major depressive disorder (MDD) is a common and debilitating mental disorder characterized by persistent low mood, anhedonia, and cognitive impairment. With a lifetime prevalence of approximately 16%, MDD represents one of the leading contributors to global disease burden and years lived with disability worldwide [1–3]. Beyond its core affective and cognitive symptoms, MDD substantially impairs social functioning and quality of life and is strongly associated with a range of high-risk behaviors, particularly self-harm and suicide, making it a major global public health concern [4–6]. Suicide is one of the leading causes of premature mortality worldwide. Epidemiological data indicate that approximately 800,000 individuals die by suicide each year globally, accounting for nearly 1.5% of all deaths, and this number is projected to exceed one million by 2040 [7]. MDD has consistently been identified as the most important psychiatric risk factor for suicidal behavior, with a substantial proportion of individuals who attempt or die by suicide having experienced depressive episodes prior to the event [5, 6]. In addition to suicidal behavior, non-suicidal self-injury (NSSI) is also a frequent and clinically significant phenomenon among individuals with MDD [8].

From a psychological and neurocognitive perspective, individuals with MDD may be particularly vulnerable to engaging in self-harm behaviors due to impairments in emotion regulation, reduced inhibitory control, and altered value-based decision-making. These processes have been associated with fronto–limbic–striatal circuits involved in affective and cognitive regulation, which are consistently implicated in the pathophysiology of depression. Dysfunction within these circuits may compromise the ability to cope with intense negative affect, thereby increasing the likelihood of maladaptive behavioral responses such as self-harm [9–11]. Self-harm behaviors are commonly categorized into NSSI and suicide attempt (SA). NSSI refers to the deliberate, repetitive destruction of one’s own body tissue without suicidal intent and without socially sanctioned purposes [12], whereas SA involves potentially lethal self-injurious behavior carried out with at least some intent to die, but which does not result in death due to rescue or other factors [13]. Although these two behaviors differ fundamentally in terms of suicidal intent and lethality, they frequently co-occur in clinical and community samples. Studies involving adolescents and young adults in non-clinical populations with NSSI have found that 69.2–83.3% reported having engaged in SA behaviors. In another study, 24.4% reported having attempted suicide, suggesting a close clinical relationship between NSSI and suicidal behavior [14, 15]. Although NSSI and SA share certain behavioral and clinical characteristics, increasing evidence suggests that they may represent partially overlapping but distinguishable forms of self-harm with different motivational and neurobiological correlates [16]. However, accumulating evidence indicates that, despite behavioral similarities, NSSI and SA differ in their underlying motivations, functional roles, and developmental trajectories [11, 17]. Suicidal behavior has been consistently associated with overwhelming psychological distress, hopelessness, and elevated suicide risk [17], whereas NSSI is more commonly conceptualized as a maladaptive emotion regulation strategy, serving functions such as alleviating negative affect, reducing emotional numbness, self-punishment, or eliciting interpersonal support [18]. These distinctions suggest that NSSI represents a clinically meaningful construct that is not merely a less severe form of suicidal behavior [11, 17, 18].

Neuroimaging research has provided important insights into the neural mechanisms underlying MDD and self-harm behaviors. Structural and functional MRI studies consistently demonstrate that MDD is associated with widespread abnormalities in fronto–limbic–striatal networks, which play central roles in emotion regulation, impulse control, and reward-based decision-making. These networks encompass prefrontal cortices, anterior cingulate cortex, and subcortical structures critical for regulating affective responses and cognitive control. Voxel-based morphometry (VBM) meta-analyses have revealed robust reductions in gray matter volume (GMV) in prefrontal and limbic regions in patients with MDD compared with healthy controls [19–21]. Importantly, studies in medication-free MDD samples suggest that at least part of these structural alterations may not be solely attributable to treatment exposure [21]. Structural MRI studies have reported distinguishable structural patterns associated with suicidal behaviors in MDD. Large-scale ENIGMA consortium studies and meta-analytic evidence indicate that suicidal ideation and suicide attempts in MDD are associated with additional gray matter volume reductions in frontal, cingulate, insular, and parietal regions, beyond the core neuroanatomical alterations observed in major depressive disorder more generally [22–24]. A systematic review focusing on suicide attempts across different age groups identified widespread and age-dependent GMV alterations in medial, dorsolateral, and orbitofrontal cortices, highlighting shared and developmental aspects of suicidality-related brain changes in MDD [25]. More specifically, MDD patients with a history of suicide attempts show reduced GMV in dorsolateral and medial prefrontal cortices, angular gyrus, and parietal regions, which have been associated with executive dysfunction, altered decision-making, and emotional dysregulation in individuals with elevated suicide risk [25, 26]. Recent machine-learning studies further demonstrate that structural MRI features can distinguish MDD patients with suicide ideation or attempts from those without suicidal behaviors, supporting the presence of potentially distinct neuroanatomical patterns related to suicidality [27]. Structural MRI studies in adolescents with NSSI have reported significant gray matter volume reductions as well as surface-based cortical morphology alterations in fronto-limbic regions, including the insula, anterior cingulate cortex, putamen, and frontal gyri, compared with healthy controls. These alterations encompass changes in gray matter volume, cortical thickness, and cortical complexity across prefrontal, insular, and sensorimotor regions, providing convergent evidence for disrupted neural substrates underlying affect regulation and self-referential processing in NSSI [28]. Together, these findings suggest both overlapping and distinct neural substrates for MDD, SA and NSSI. Interpretation of structural MRI findings in self-harm populations remains challenging due to clinical heterogeneity and the potentially state-dependent nature of self-harm behaviors. Moreover, few investigations have included MDD patients without self-harm behaviors as a clinical comparison group. Studies simultaneously comparing MDD with NSSI, MDD with SA, MDD without self-harm, and healthy controls remain scarce. Consequently, the extent to which NSSI and SA are associated with distinguishable patterns of gray matter alterations within MDD remains unclear. To address these gaps, the present study employed VBM to compare gray matter volume among four groups: MDD with NSSI (MDD/NSSI), MDD with SA (MDD/SA), MDD without self-harm behaviors (sMDD), and healthy controls (HCs). Given the exploratory nature of this cross-sectional study, the present study aimed to characterize group-related structural differences rather than infer causal mechanisms.

Methods

Participants

Between July 2017 and December 2019, a total of 263 Han Chinese patients with MDD, aged 18–30 years, were recruited from Nanjing Brain Hospital. The cohort comprised patients with MDD categorized into three behavioral subgroups: MDD/NSSI, MDD/SA, sMDD. Additionally, 66 HCs, matched for age and sex, were recruited from the local community.

All psychiatric diagnoses were independently evaluated by two experienced psychiatrists according to DSM-5 criteria through clinical interviews. To reduce the possibility of undiagnosed bipolar disorder, manic symptoms were additionally assessed using the Young Mania Rating Scale (YMRS), and participants with clinically significant manic symptoms were excluded. Individuals with psychotic disorders, substance use disorders, or other major psychiatric disorders were also excluded. The inclusion criteria for MDD patients were as follows: (1) Met DSM-5 criteria for MDD; (2) Had a 17-item Hamilton Depression Rating Scale (HAMD-17) [29] score > 17 on the day of MRI scanning; (3) Had a score < 14 on the 32-item Hypomania Checklist and < 10 on the YMRS [30]. The inclusion criteria for HCs were as follows: (1) no major physical illnesses, psychiatric disorders, or neurological diseases; (2) no history of psychotropic medication use. Group-specific inclusion criteria: MDD/NSSI: (1) met DSM-5 proposed criteria for NSSI; (2) Had a Clinician-rated severity of non-suicidal self-injury (CRSNSSI) (DSM-5–based clinician-rated severity specifier) score ≥ 2; (3) No suicidal intent during self-injury; (4) Suicide attempts during the current episode were excluded. MDD/SA: (1) A suicide attempt within the past month; (2) MRI scanning performed ≥ 7 days after overdose attempts; (3) Explicit intent to die and had suicidal ideation on the scanning day; (4) No lifetime history of NSSI. sMDD: (1) No lifetime history of suicide attempts or NSSI (CRSNSSI = 0). Participants in the MDD/NSSI group were allowed to have a lifetime history of suicide attempts, provided that no suicide attempt occurred during the current depressive episode. To address the potential clinical heterogeneity introduced by this criterion, an additional sensitivity analysis excluding these participants was subsequently performed. General exclusion criteria for all participants: (1) Any psychiatric disorder other than borderline personality disorder; (2) Substance abuse or dependence within the past year; (3) History of neurological disorders, systemic medical illnesses, traumatic brain injury, or any other organic condition that may cause psychiatric symptoms; (4) Pregnancy or breastfeeding; (5) Contraindications to MRI scanning (e.g., metallic implants, claustrophobia).

This study was approved by the Research Ethics Review Committee of Nanjing Brain Hospital, affiliated with Nanjing Medical University. After providing participants with a detailed explanation of the study, written informed consent was obtained from all participants.

Research methods

Clinical assessments

Clinical data were collected using a structured questionnaire developed for this study to gather general information about the patients, including age, gender, years of education, age of onset of depression, age of onset of NSSI, frequency of depressive episodes, presence of borderline personality disorder, family history, and childhood abuse history. The severity of depressive symptoms was assessed using the HAMD-17, while anxiety symptoms were assessed using the Hamilton Anxiety Rating Scale (HAMA) [31]. The severity of NSSI was evaluated using the CRSNSSI scale, and suicide risk was assessed using the Nurses’ Global Assessment of Suicide Risk Scale (NGASR) [32]. Additionally, the Ottawa Self-Injury Inventory was used to assess NSSI behaviors. The number of self-injury incidents in the past week, month, six months, and twelve months was recorded. The presence of suicidal ideation during this episode and any history of suicide attempts during previous episodes (where the suicide attempt occurred more than one year before enrollment and there were no suicide attempts during the current episode) were also documented. To further characterize treatment exposure, a medication load index was calculated according to a previously published method that quantifies psychotropic medication exposure based on medication dosage and treatment regimen [33]. Higher scores reflected greater overall medication exposure.

MRI data acquisition

All MRI data were collected using a Siemens 3.0T MRI scanner (Germany) to acquire high-resolution three-dimensional T1-weighted structural images. Cotton earplugs were provided to the patients during the scan to facilitate participant comfort, to improve compliance with scanning procedures, and to help them adapt to the scanning environment more quickly. Participants were instructed to remain relaxed, keep their eyes closed, and minimize head movement during scanning. The scanning parameters were as follows: Repetition time (TR) = 1900 ms, Echo time (TE) = 2.48 ms, Field of view (FOV) = 250 mm × 250 mm, Matrix = 256 × 256, Slice thickness = 1 mm, Scan time = 4 min 18 s.

Data processing

Preprocessing

MRI data preprocessing and analysis were conducted on the MATLAB 2013 platform using SPM8 (Statistical Parametric Mapping) software (https://www.fil.ion.ucl.ac.uk/spm/). Data with imaging abnormalities, anatomical results inconsistencies, or artifacts were excluded from further analysis.

GMV analysis

Structural data for each participant were segmented using the New Segment method in VBM to obtain gray matter, white matter, and cerebrospinal fluid components [34]. The segmented gray matter images were iteratively processed using the DARTEL algorithm to create an average template. The original gray matter images were modulated, spatially normalized, and resampled into a 1.5 × 1.5 × 1.5 mm³ gray matter image. Finally, Gaussian smoothing was applied with an 8 mm full-width at half-maximum (FWHM) kernel, producing GMV values suitable for statistical analysis.

Statistical analysis

Demographic and clinical data were analyzed using SPSS version 19.0. Continuous variables were compared using one-way analysis of variance (ANOVA), followed by Bonferroni-corrected post hoc analyses when appropriate. Illness duration, NGASR scores, CRSNSSI scores, and medication load index were compared among patient groups only. Categorical variables were analyzed using chi-square tests or Fisher’s exact tests. Partial η² and Cramér’s V were calculated as measures of effect size for continuous and categorical variables, respectively.

Voxel-wise GMV analyses were performed using DPABI/SPM8 within the framework of a general linear model (GLM). Group was entered as the main factor, with age, sex, years of education, and total intracranial volume (TIV) included as covariates. Whole-brain group comparisons were corrected for multiple comparisons using Gaussian Random Field (GRF) correction, with a voxel-level threshold of p < 0.001 and a cluster-level threshold of p < 0.05 (GRF corrected). For regions showing significant group effects, Bonferroni-adjusted pairwise comparisons were subsequently performed.

To further examine the independent contributions of MDD diagnosis, NSSI history, and SA history, an additional GLM analysis was conducted. MDD diagnosis, lifetime history of NSSI, and lifetime history of SA were entered as fixed factors, while age, sex, years of education, and TIV were included as covariates. Detailed results are presented in Supplementary Table 2.

To further evaluate the robustness of the findings, two sensitivity analyses were conducted based on GMV values extracted from regions showing significant group effects in the primary whole-brain analysis. First, participants in the MDD/NSSI group with a lifetime history of suicide attempts were excluded. Second, an alternative GLM model was performed by replacing years of education with illness duration as a covariate and additionally adjusting for HAMD-17 scores and medication load index. Detailed results of these sensitivity analyses are provided in Supplementary Tables 3 and 4.

Pearson correlation analyses were performed between extracted GMV values and clinical variables. Multiple comparisons were controlled using the false discovery rate (FDR) correction, with a corrected significance threshold of q < 0.05.

Results

Demographic and clinical data

Of the 263 initially recruited patients with MDD, 62 were excluded due to incomplete clinical assessments, and 29 were excluded because MRI scanning was not completed. The final sample consisted of 54 patients with MDD/NSSI, 68 patients with sMDD, 50 patients with MDD/SA, and 66 HCs. Detailed demographic and clinical characteristics are presented in Table 1. Significant differences among the four groups were observed in age, sex, borderline personality disorder, family history, childhood abuse, history of suicide attempts, self-harm ideation, HAMD-17 scores, and HAMA scores (p < 0.05). Significant differences among patient groups were also observed in NGASR scores (p < 0.001). No significant differences were found in years of education, illness duration, or medication load index among patient groups (p > 0.05). Compared with the MDD/NSSI group, the MDD/SA group showed significantly higher age, HAMD-17 scores, and NGASR scores (p < 0.05), whereas no significant differences were observed in years of education, illness duration, HAMA scores, or medication load index (all p > 0.05).

Table 1.

Demographic and clinical data of the four groups of participants

Variable	MDD/SA(n = 50)	MDD/NSSI(n = 54)	sMDD(n = 68)	HCs (n = 66)	Overall p-value	Effect Size	MDD/SA vs. MDD/NSSI p-value
Age (years)	24.3 ± 4.5	20.8 ± 3.6	20.3 ± 1.9	21.1 ± 2.0	< 0.001*	0.19	< 0.001*
Gender(Male/Female)	20:30	6:48	12:56	8:58	< 0.001*	0.39	< 0.001*
Years of Education(years)	13.6 ± 2.7	13.0 ± 2.2	13.0 ± 1.8	13.6 ± 1.5	0.166	0.02	0.605
Total Duration of Illness(years)	4.4 ± 2.9	3.5 ± 3.0	3.7 ± 2.8	NA	0.278	0.01	0.401
Borderline Personality(yes/no)	5:45	9:45	0:68	0:66	< 0.001*	0.29	0.320
Family History(yes/no)	19:31	17:37	24:44	0:66	< 0.001*	0.36	0.540
Childhood Abuse(yes/no)	7:43	16:38	7:61	0:66	< 0.001*	0.31	0.055
History of Suicide Attempts(yes/no)	50:0	9:45	0:68	0:66	< 0.001*	0.91	< 0.001*
Self-harm Ideation(yes/no)	50:0	41:13	33:35	0:66	< 0.001*	0.74	< 0.001*
HAMD17 Total Score	26.5 ± 6.4	22.8 ± 4.3	21.5 ± 4.6	2.6 ± 1.1	< 0.001*	0.82	< 0.001*
HAMA Total Score	15.6 ± 5.7	13.2 ± 5.0	14.5 ± 5.9	2.6 ± 1.2	< 0.001*	0.56	0.065
NGASR Total Score	12.1 ± 1.6	9.9 ± 3.0	6.2 ± 2.5	NA	< 0.001*	0.51	< 0.001*
CRSNSSI Total Score	0 ± 0	3.2 ± 0.8	0 ± 0	NA	NA	NA	NA
Medication Load Index	1.3 ± 1.6	0.9 ± 1.4	1.0 ± 1.3	NA	0.306	0.01	0.396

Values are presented as mean ± standard deviation (SD)

Overall p values were calculated using one-way analysis of variance (ANOVA) or Kruskal–Wallis tests for continuous variables and chi-square or Fisher’s exact tests for categorical variables. Pairwise comparisons between the MDD/SA and MDD/NSSI groups were predefined based on the study hypothesis and are presented in the last column

Effect size was expressed as η² for continuous variables and Cramér’s V for categorical variables. CRSNSSI scores were assessed only in participants with NSSI behavior

A p value < 0.05 was considered statistically significant

“NA” indicates not applicable

Abbreviations: MDD/SA, major depressive disorder with suicide attempts; MDD/NSSI, major depressive disorder with non-suicidal self-injury; sMDD, major depressive disorder without self-harm behaviors; HCs, healthy controls; HAMD-17, 17-item Hamilton Depression Rating Scale; HAMA, Hamilton Anxiety Rating Scale; NGASR, Nurses’ Global Assessment of Suicide Risk; CRSNSSI, Clinician-Rated Severity of Non-Suicidal Self-Injury

GMV analysis results

Whole-brain voxel-wise analyses, controlling for age, sex, education, and TIV, revealed significant group effects in the right superior frontal gyrus (SFG)/rectus gyrus, bilateral caudate, bilateral SFG, right superior temporal gyrus (STG)/middle temporal gyrus (MTG), and bilateral middle cingulate gyrus (MCG) (voxel-level p < 0.001, cluster-level GRF-corrected p < 0.05), as shown in Fig. 1a. Post hoc pairwise comparisons demonstrated that, compared with the MDD/NSSI group, the MDD/SA group exhibited reduced GMV in the right rectus gyrus, bilateral caudate, right MTG, bilateral MCG, and right STG (Fig. 1b; Table 2). Compared with the sMDD group, the MDD/SA group showed reduced GMV in the right SFG (Fig. 1c). Compared with healthy controls, the MDD/SA group showed reduced GMV in the right STG/amygdala, right MTG, and right caudate (Fig. 1d). No other pairwise comparisons survived correction for multiple comparisons. Detailed results are summarized in Supplementary Table 1.

Fig. 1.

Gray matter volume differences between early-onset depression subgroups and healthy controls. (a) Results of one-way analysis of variance (ANOVA) showing gray matter volume differences across the four groups. Significant clusters were observed in the right superior frontal gyrus/Rectus gyrus, left superior frontal gyrus, bilateral caudate, right superior temporal gyrus/middle temporal gyrus, and bilateral middle cingulate gyrus. The color bar indicates F values. (b) Two-sample t-test results comparing the MDD/SA and MDD/NSSI groups, showing reduced gray matter volume in the right Rectus gyrus, bilateral caudate, right middle temporal, right superior temporal gyrus, and bilateral middle cingulate gyrus in the MDD/SA group. (c) Two-sample t-test results comparing the MDD/SA sMDD groups, revealing reduced gray matter volume in the right superior frontal gyrus in the MDD/SA group. (d) Two-sample t-test results comparing the MDD/SA and HCs groups, showing reduced gray matter volume in the right superior temporal gyrus/amygdala, right middle temporal gyrus, and right caudate in the MDD/SA group. All results are displayed at a voxel-wise threshold of p < 0.001 with cluster-level Gaussian random field–corrected p < 0.05. Warm and cool colors indicate relative increases and decreases in gray matter volume, respectively

Table 2.

Gray matter volume differences between the NSSI and SA groups

Brain Region(AAL)	Peak MNI Coordinates(x, y, z)	NSSI adjusted mean ± SE	SA adjusted mean ± SE	p	Partial η²
Right Rectus Gyrus	5 39 -24	0.683 ± 0.009	0.617 ± 0.009	< 0.001*	0.197
Right Caudate	12 -3 21	0.274 ± 0.005	0.248 ± 0.005	< 0.001*	0.108
Right Middle Temporal Gyrus	66 -30 -2	0.630 ± 0.010	0.581 ± 0.010	0.002	0.097
Left Middle Cingulate Gyrus	-2 45 50	0.495 ± 0.008	0.461 ± 0.008	0.005	0.076
Left Caudate	-14 8 0	0.404 ± 0.005	0.379 ± 0.006	0.004	0.084
Right Superior Temporal Gyrus	63 -14 12	0.563 ± 0.009	0.514 ± 0.009	< 0.001*	0.117
Right Middle Cingulate Gyrus	5 -33 38	0.848 ± 0.013	0.775 ± 0.013	< 0.001*	0.118

Brain regions are defined according to the Automated Anatomical Labeling (AAL) atlas. Peak MNI coordinates (x, y, z) indicate the location of the maximum statistical difference within each region. The regions shown in this table were identified from the whole-brain ANCOVA analysis using Gaussian random field (GRF) correction (voxel-level p < 0.001, cluster-level p < 0.05)

Adjusted mean GMV values are presented as estimated marginal means ± standard errors (SE). P values represent Bonferroni-corrected post hoc comparisons between the MDD/NSSI and MDD/SA groups. Partial η² is reported as a measure of effect size

Abbreviations: AAL, Automated Anatomical Labeling; MNI, Montreal Neurological Institute; GMV, gray matter volume; SE, standard error; MDD/NSSI, major depressive disorder with non-suicidal self-injury; MDD/SA, major depressive disorder with suicide attempts

Sensitivity analyses

To further evaluate the stability of the primary between-group GMV findings, two sensitivity analyses were conducted. First, after excluding the nine participants in the MDD/NSSI group with a lifetime history of SA, the overall pattern of GMV alterations remained largely unchanged. Significant differences between the MDD/NSSI and MDD/SA groups were still observed in the right SFG/MFG, bilateral caudate, right STG/MTG, right MCG, and left SFG. However, the difference in the left MCG was no longer significant after exclusion (p = 0.142). The direction of GMV differences and the overall pattern of results remained consistent with the primary analysis. These results are presented in Supplementary Table 3.

Second, an alternative GLM model was conducted by replacing years of education with illness duration as a covariate and further including HAMD scores and medication load index as additional covariates. Under this model, similar patterns of between-group GMV differences were observed. Significant GMV reductions in the MDD/SA group relative to the MDD/NSSI group persisted in the right SFG/MFG, bilateral caudate, right STG/MTG, right MCG, and left SFG. The left MCG no longer remained statistically significant after covariate adjustment (p = 0.125), whereas the left caudate remained significant, although with reduced statistical significance (p = 0.034). These results are presented in Supplementary Table 4. Overall, the direction and spatial pattern of GMV alterations remained stable across analyses.

Overall, these sensitivity analyses suggested that the observed GMV alterations were relatively stable and were unlikely to be solely attributable to lifetime suicide attempt history within the MDD/NSSI group, depression severity, or medication exposure.

Independent associations of MDD, SA, and NSSI With regional GMV

To further examine the independent contributions of MDD, SA, and NSSI, a supplementary general linear model including these variables as simultaneous predictors was performed (Supplementary Table S2). After mutual adjustment, NSSI showed significant positive associations with GMV across all examined brain regions, including the right rectus gyrus, bilateral caudate, right MTG, bilateral MCG, and right STG (all p < 0.01). The corresponding effect sizes ranged from η²p = 0.042 to 0.101. In contrast, SA was not significantly associated with GMV in any region after adjustment for MDD and NSSI (all p > 0.05). MDD showed a significant association only in the right STG (p = 0.049, η²p = 0.017), with no significant effects observed in the remaining regions. These results are presented in Supplementary Table 2. Overall, NSSI demonstrated the most consistent and robust independent associations with regional GMV, whereas the independent effects of SA and MDD were limited.

Correlation analysis between GMV and clinical variables

Exploratory Pearson correlation analyses were performed between altered GMV regions identified in the MDD/SA and MDD/NSSI comparisons and clinical variables. In the sMDD group (n = 68), the GMV of the right caudate was negatively correlated with HAMA scores (r = − 0.347, p = 0.004). However, this correlation did not survive FDR correction (pFDR > 0.05). No other significant correlations were observed between GMV values and HAMD, HAMA, or NGASR scores. In the combined sample of the three patient groups (MDD/SA, MDD/NSSI, and sMDD; n = 172), the GMV of the right caudate was negatively correlated with HAMD scores (r = − 0.197, p = 0.010) and HAMA scores (r = − 0.161, p = 0.035). In addition, the GMV of the left MCG was negatively correlated with HAMD scores (r = − 0.158, p = 0.038) and HAMA scores (r = − 0.190, p = 0.012), while the GMV of the right MCG was negatively correlated with NGASR scores (r = − 0.165, p = 0.030). However, none of these correlations remained significant after FDR correction (all pFDR > 0.05), as shown in Supplementary Fig. 1.

Discussion

This study systematically explored the structural differences in GMV in early adulthood depression patients with NSSI and SA using VBM. The results revealed that, compared to MDD/NSSI patients, MDD/SA patients exhibited significant reductions in GMV in brain regions including the right Rectus gyrus, right caudate, right SFG, right STG/MTG, left caudate, and bilateral MCG. These findings suggest that self-harm subgroups within MDD may exhibit different patterns of brain structural alterations. However, supplementary GLM analyses indicated that NSSI-related characteristics showed more consistent associations with regional GMV than SA history, suggesting that the observed group differences may reflect clinical heterogeneity between self-harm subgroups.

The structural changes in the ventral prefrontal cortex (VPFC) and orbital frontal cortex (OFC) are crucial for emotional regulation, decision-making, and self-control [35]. In adolescent patients with MDD and BD, those with a history of SA exhibit significantly smaller GMV in the VPFC and OFC compared to those without SA [36, 37]. Furthermore, a prospective study found that reduced GMV in the baseline VPFC and rostral prefrontal cortex (PFC) may increase the risk of future suicide attempts in adolescents with mood disorders [6]. These findings suggest that frontal lobe GMV alterations have been associated with suicidal behavior in previous studies. However, the present findings should not be interpreted as evidence that the observed group differences directly contribute to suicidal behavior, particularly given the cross-sectional design and the supplementary GLM results. Previous postmortem studies have shown that individuals who died by suicide exhibit atrophy in structures such as the frontal lobe, hippocampus, caudate, thalamus, and anterior cingulate gyrus (ACG) [38]. In studies of suicidal behavior in MDD, the ENIGMA-MDD working group’s meta-analysis reported reduced overall subcortical volume in MDD patients with suicide attempts, while no robust and consistent regional volume reductions were identified [23]. In non-clinical populations with NSSI, reduced gray matter volume was observed in the insula and ACG [39]. In the present study, MDD/NSSI patients showed relatively increased GMV in the SFG compared with MDD/SA patients, whereas MDD/SA patients showed reduced GMV in several frontal and subcortical regions. The SFG is a crucial node in the fronto-parietal network, involved in decision-making, reward processing, emotional regulation, and cognitive control [40]. However, given the cross-sectional nature of the present study, these findings should not be interpreted as evidence of compensatory, adaptive, or protective structural changes. Instead, they may reflect structural heterogeneity associated with different self-harm phenotypes within MDD. In addition, because the NSSI group was younger than the SA group, developmental factors may also contribute to the observed differences.

Furthermore, compared with the MDD/NSSI group, the MDD/SA group exhibited lower GMV in the bilateral MCG, caudate, and SFG. These regions are key components of fronto-parietal executive and reward-related networks and have previously been implicated in cognitive control, reward processing, and self-harm–related neurobiological models [41–45]. However, supplementary GLM analyses indicated that NSSI-related characteristics showed more consistent independent associations with regional GMV, whereas SA history was not independently associated with GMV after mutual adjustment. Therefore, the observed group differences are more appropriately interpreted as reflecting clinical and neurobiological heterogeneity across self-harm subgroups rather than SA-specific structural abnormalities. Although the present findings do not establish causal relationships, they suggest that structural differences involving fronto-parietal and reward-related regions may contribute to the neurobiological heterogeneity observed across self-harm subgroups in patients with MDD.

In contrast, NSSI is widely conceptualized as a maladaptive coping strategy aimed at obtaining temporary relief from negative affect or interpersonal distress, rather than an intent to die. Such behaviors are thought to serve affect-regulation and interpersonal functions, distinguishing them from suicidal behavior [11, 12, 17, 18].

Based on our findings, patients with MDD/NSSI may exhibit a relatively distinct structural pattern characterized by increased GMV in the left SFG compared with the MDD/SA group. Previous studies have implicated the SFG in cognitive control and emotion regulation processes [46]. Therefore, the observed structural differences may reflect heterogeneity in neural characteristics underlying different forms of self-injurious behaviors in MDD. However, given the cross-sectional nature of the present study, these interpretations should be considered preliminary.

Previous studies have also reported that suicidal behavior in MDD is associated with structural and functional alterations in frontal and limbic regions involved in emotional processing, reward-related functions, and cognitive control [26]. The present findings may be broadly consistent with previous reports of structural differences among clinical subgroups characterized by different self-harm behaviors, although direct mechanistic inferences cannot be established based on the current structural imaging results alone.

In the exploratory correlation analyses, several associations were observed between regional GMV alterations and clinical symptom severity; however, none survived FDR correction and therefore should be interpreted cautiously. Specifically, in the sMDD group, reduced GMV in the right caudate showed a trend-level association with greater anxiety severity. Previous studies have implicated the caudate in reward processing, motivational regulation, and affective functioning [47–50], suggesting that the observed structural alterations may be broadly consistent with previous neuroimaging findings in MDD.

Similarly, in the combined sample, lower GMV in the right caudate and left MCG showed trend-level negative correlations with HAMD and HAMA scores, whereas reduced GMV in the right MCG was associated with higher NGASR scores prior to correction. These regions have previously been implicated in emotional regulation, cognitive control, and behavioral inhibition [51–54]. Nevertheless, given the lack of statistical significance after multiple-comparison correction, these findings should be regarded as exploratory and hypothesis-generating rather than definitive evidence of specific neurobiological mechanisms.

This study observed group differences in the right STG/MTG, with MDD/SA patients showing reduced GMV compared to MDD/NSSI patients. Previous studies have reported reduced temporal lobe volume in adolescents with suicide attempts [54], and temporal structures have been suggested to differentiate suicidal from non-suicidal individuals [55]. The STG/MTG is implicated in auditory processing and language-related functions and has been linked to social cognition and affective processing in MDD [56, 57]. However, supplementary GLM analyses provided important clarification regarding the independent contributions of clinical variables. After mutually adjusting for MDD, SA, and NSSI, NSSI showed the most consistent positive associations with regional GMV across multiple frontal–temporal–striatal regions, whereas SA was not independently associated with GMV in any examined region. MDD showed only limited associations, primarily in the right STG. Taken together, these findings suggest that the observed group differences in the STG/MTG are more likely to reflect clinical heterogeneity between self-harm subgroups rather than SA-specific structural alterations. Therefore, these findings should be interpreted cautiously in the context of clinical subgroup composition.

Several limitations of the present study should be acknowledged. First, the cross-sectional design precludes causal inference regarding the relationship between GMV alterations and self-harm behaviors. Longitudinal studies are needed to clarify whether the observed structural differences reflect pre-existing vulnerability or consequences of illness progression. Second, potential confounding and sampling bias should be considered. Although age and sex were included as covariates in all analyses, the MDD/SA and MDD/NSSI groups differed in demographic composition, and residual confounding related to sex-related variability in brain structure cannot be fully excluded. In addition, no formal comparison was conducted between included and excluded participants; therefore, potential selection bias related to data completeness and imaging quality may limit the generalizability of the findings. Third, medication effects cannot be completely ruled out. Although a medication load index was included as a covariate to account for current pharmacological exposure, this measure may not fully capture cumulative medication history, treatment duration, dosage variability, or long-term neurobiological effects of antidepressant use. Therefore, residual pharmacological confounding may still be present. Fourth, structural MRI reflects relatively stable neuroanatomical characteristics and may not capture the temporal dynamics of self-harm behaviors. Given that both NSSI and SA are clinically fluctuating phenomena, the observed GMV alterations may reflect trait-related vulnerability rather than state-dependent neural correlates of recent behavior. Finally, the relatively modest sample size should be considered when interpreting the findings. Structural neuroimaging studies of major depressive disorder generally yield small-to-moderate effect sizes. Consequently, although the present sample size is comparable to that of many previous voxel-based morphometry studies, it may still have provided limited statistical power to detect subtle group differences, particularly after correction for multiple comparisons. Therefore, some true effects may have remained undetected, increasing the possibility of Type II errors, and null findings should be interpreted with caution. Although sensitivity analyses supported the robustness of the observed results, residual heterogeneity cannot be completely excluded. Future large-scale, longitudinal, and multimodal neuroimaging studies with greater statistical power are needed to improve reproducibility, enhance generalizability, and further elucidate the neurobiological mechanisms underlying self-harm behaviors in major depressive disorder.

Conclusions

In conclusion, this study identified group-related differences in gray matter volume among early-adulthood patients with major depressive disorder who exhibited suicide attempts, non-suicidal self-injury, or no self-harm behaviors. Compared with patients with non-suicidal self-injury, those with suicide attempts showed lower gray matter volume in frontoparietal, cingulate, striatal, and temporal regions. However, supplementary general linear model analyses indicated that NSSI-related characteristics showed stronger independent associations with regional gray matter volume, whereas suicide attempt history was not independently associated with gray matter volume after mutual adjustment. These findings suggest that observed group differences may reflect clinical heterogeneity across self-harm subtypes rather than SA-specific structural abnormalities. By comparing multiple MDD subgroups within a single framework, this study highlights the importance of distinguishing NSSI and suicide attempts in neuroimaging research on depression. Further longitudinal and multimodal studies are needed to clarify the clinical and developmental significance of these findings.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MDD

Major depressive disorder

NSSI

Non-suicidal self-injury

SA

Suicide attempt

VBM

Voxel-based morphometry

GM

Gray matter

GMV

Gray matter volume

MRI

Magnetic resonance imaging

MDD/NSSI

Major depressive disorder with non-suicidal self-injury

MDD/SA

Major depressive disorder with suicide attempts

sMDD

Major depressive disorder without self-harm

HCs

Healthy controls

HAMD-17

17-item Hamilton Depression Rating Scale

YMRS

Young Mania Rating Scale

DSM-5

Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition

CRSNSSI

Clinician-Rated Severity of Non-Suicidal Self-Injury

HAMA

Hamilton Anxiety Rating Scale

NGASR

Nurses’ Global Assessment of Suicide Risk

BOLD

Blood oxygen level–dependent

TR

Repetition time

TE

Echo time

FOV

Field of view

SPM

Statistical Parametric Mapping

FWHM

Full width at half maximum

ANOVA

Analysis of variance

GRF

Gaussian random field

SFG

Superior frontal gyrus

MFG

Middle frontal gyrus

STG

Superior temporal gyrus

MTG

Middle temporal gyrus

FDR

False discovery rate

VPFC

Ventral prefrontal cortex

OFC

Orbitofrontal cortex

BD

Bipolar disorder

PFC

Prefrontal cortex

MCG

Middle cingulate gyrus

ACG

Anterior cingulate gyrus

Author contributions

Mengzhi Zhang: Analysis of data and interpretation of result. Conception of idea. Drafting of paper. Aihua Zhou: Analysis of data. Interpretation of data. Meijun Liu: Analysis of data. Drafting of paper. Yibo Wang: Analysis of data. Acquisition of data. Jiabo Shi: Revision of paper. Yu Chen: Analysis of data. Lingling Hua: Analysis of data. Rui Yan: Revision of paper. Zhijian Yao: Revision of paper and final approval of the submission. Design of the study. Revision of Paper. Qing Lu: Interpretation of data. Revision of Paper.

Funding

This project was supported by the National Natural Science Foundation of China (62571115, 82571749, 82271568, 82301718); the Jiangsu Medical Innovation Center for Mental Illness (CXZX202226); the Jiangsu Provincial Key Research and Development Program (BE2019675, BE2023667); the Key Project of Science and Technology Innovation for Social Development in Suzhou (2022SS04); Jiangsu Provincial Natural Science Youth Fund (BK20230154).

Data availability

The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Nanjing Brain Hospital (Approval No. 2016-KY010). Informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.