Introduction
According to the Centers for Disease Control and Prevention (CDC) WISQARS Leading Causes of Death Reports, suicide was the 11th leading cause of death in the United States in 2022 and exhibited an increasing trend since 2020 (Garnett & Curtin, Reference Garnett and Curtin2024). The incidence of suicide is complex and multifactorial, influenced by biological, psychological, and socio-environmental factors (Klonsky, Saffer, & Bryan, Reference Klonsky, Saffer and Bryan2018; Turecki et al., Reference Turecki, Brent, Gunnell, O’Connor, Oquendo, Pirkis and Stanley2019). Prevailing theories attempt to conceptualize suicide attempts (SAs), such as the three-step theory and interpersonal theory (Klonsky et al., Reference Klonsky, Saffer and Bryan2018; Klonsky & May, Reference Klonsky and May2015; Van Orden et al., Reference Van Orden, Witte, Cukrowicz, Braithwaite, Selby and Joiner2010). Self-harm, defined as self-inflicted injury to one’s own body, regardless of intent, is a significant risk factor for subsequent suicide. Although self-harm behaviors are strongly linked to suicide risk, the underlying mechanism and biological evidence remain insufficient (Barredo et al., Reference Barredo, Bozzay, Primack, Schatten, Armey, Carpenter and Philip2021; Klonsky & May, Reference Klonsky and May2015; Olfson et al., Reference Olfson, Wall, Wang, Crystal, Gerhard and Blanco2017). Recent studies have reported neurobiological alterations in suicidal behaviors, such as monoamine dysregulations and stress-related hypothalamic–pituitary–adrenal axis disturbances (Turecki et al., Reference Turecki, Brent, Gunnell, O’Connor, Oquendo, Pirkis and Stanley2019). Blood-oxygen-level-dependent (BOLD) signals, extracted from resting-state functional magnetic resonance imaging (rs-fMRI), provide a safe and effective means to observe potential neuronal activity (Fox & Raichle, Reference Fox and Raichle2007). A 2019 review summarized structural and functional neuroimaging studies from the past two decades, highlighting the critical role of the cortico-striatal limbic system in suicidal behaviors (Schmaal et al., Reference Schmaal, van Harmelen, Chatzi, Lippard, Toenders, Averill, Mazure and Blumberg2020). Despite convergent findings in neuroimaging, larger sample studies are required to validate reliability.
While alterations in resting-state functional connectivity (FC) between individuals with self-harm behaviors have been reported compared to diagnostic controls in different groups, few studies have directly investigated FC differences between SA and NSSI (Huang et al., Reference Huang, Rootes-Murdy, Bastidas, Nee and Franklin2020; Lai et al., Reference Lai, Jiang, Xu, Luo, Hao and Li2021; Schmaal et al., Reference Schmaal, van Harmelen, Chatzi, Lippard, Toenders, Averill, Mazure and Blumberg2020). Over the past two decades, suicidal behaviors have been associated with lower FC in the default mode network and reward circuit, and greater FC in the limbic network (Kang et al., Reference Kang, Na, Choi, Kim, Son and Lee2017; Kim et al., Reference Kim, Kim, Myung, Han, Fava, Mischoulon, Papakostas, Seo, Cho, Seong and Jeon2017; Ordaz et al., Reference Ordaz, Goyer, Ho, Singh and Gotlib2018; Schmaal et al., Reference Schmaal, van Harmelen, Chatzi, Lippard, Toenders, Averill, Mazure and Blumberg2020). For example, FC alterations in the reward circuit and prefrontal cortex have been identified in individuals with self-harm behaviors and psychiatric disorders (Dimick et al., Reference Dimick, Hird, Sultan, Mitchell, Sinyor, MacIntosh and Goldstein2023; Guo et al., Reference Guo, Lu, Ou, Huang, Li, Cui, Li, Zheng, Liang, Qiu and Liu2024). Regarding SA and NSSI differences, Li et al. (Reference Li, Liang, Li, Xu, Yang, Liu, Zhang, Wang, Zhang and Zhou2024) investigated FC changes in depressed adolescents with SA and NSSI, indicating divergent inter-network connectivity within visual, limbic, and subcortical networks (Li et al., Reference Li, Liang, Li, Xu, Yang, Liu, Zhang, Wang, Zhang and Zhou2024). In this study, we hypothesized that alterations in FC exist in individuals with lifetime SA, particularly in the reward circuit and limbic system, compared to those with lifetime nonsuicidal self-directed violence (NSSDV), with SA and NSSDV defined according to the CDC’s self-directed violence (SDV) nomenclature recommendations (Crosby, Ortega, & Melanson, Reference Crosby, Ortega and Melanson2011). Additionally, most resting-state studies uncovering the biological mechanisms of suicidal behaviors have been limited to small sample sizes (Huang et al., Reference Huang, Rootes-Murdy, Bastidas, Nee and Franklin2020). Participants in this study were therefore selected from the UK Biobank, providing a more comprehensive and robust cohort. The aims of this study were (1) to investigate alterations in FC between lifetime SA and lifetime NSSDV, and (2) to explore FC changes in relation to the frequency and recency of SDV. By expanding understanding of rs-fMRI phenotypes, this research may provide critical insights into the neurobiological distinctions underlying suicidal behaviors, facilitating more precise risk identification and informing targeted early intervention strategies.
Methods
Participants
The UK Biobank is a nonprofit, population-based biomedical study from the United Kingdom that includes more than 500,000 participants and a wide range of mental health, lifestyle, and neuroimaging information (Rusk, Reference Rusk2018). It was made available to registered researchers in 2019 and recruited volunteers aged 40–69 years (Miller et al., Reference Miller, Alfaro-Almagro, Bangerter, Thomas, Yacoub, Xu, Bartsch, Jbabdi, Sotiropoulos, Andersson, Griffanti, Douaud, Okell, Weale, Dragonu, Garratt, Hudson, Collins, Jenkinson and Smith2016; Tang, Reference Tang2018). Demographic data and medical records were collected at the time of enrollment, and mental health web-based questionnaires were regularly administered online (Davis et al., Reference Davis, Coleman, Adams, Allen, Breen, Cullen, Dickens, Fox, Graham, Holliday, Howard, John, Lee, McCabe, McIntosh, Pearsall, Smith, Sudlow, Ward and Hotopf2020, Reference Davis, Coleman, Adams, Breen, Cai, Davies, Davies, Dregan, Eley, Fox, Holliday, Hübel, John, Kassam, Kempton, Lee, Li, Maina, McCabe and Hotopf2025; Ollier, Sprosen, & Peakman, Reference Ollier, Sprosen and Peakman2005). Questionnaire H was designed by the UK Biobank working groups to assess details of self-harm behaviors. Participants who answered ‘Yes’ to question H3 in the online questionnaire – ‘Have you deliberately harmed yourself, whether or not you meant to end your life?’ (UK Biobank field ID 20480) – were defined as individuals with lifetime SDV. A total of 6,861 individuals were included in this research. After quality control of available imaging data (n = 5,741) and exclusion of participants with incomplete clinical baseline records (n = 50), 1,070 individuals were selected for image analysis. Supplementary Figure S1 presents the flowchart of participant selection. To address potential nesting bias due to overlapping psychiatric conditions, participants with psychiatric comorbidities were excluded based on the main ICD-10 diagnosis codes from hospital inpatient records (UK Biobank field ID 41202). Detailed diagnosis definitions are provided in Supplementary Table S1.
Measurement of frequency and recency of SDV, and classification of suicide attempts
Participants who answered ‘Yes’ to question H5: ‘Have you harmed yourself with the intention to end your life?’ (UK Biobank field ID 20483) were classified into the SA group, while those who answered ‘No’ were classified into the NSSDV group. The frequency of SDV was derived from question H3a: ‘How many times have you harmed yourself?’ (UK Biobank field ID 20482), with selectable responses: ‘1’, ‘2’, ‘3 or more’, and ‘Prefer not to answer’. The recency of SDV was obtained from question H3b: ‘Have you harmed yourself in the last 12 months, whether or not you meant to end your life?’ (UK Biobank field ID 20481), with selectable responses: ‘No’, ‘Yes’, and ‘Prefer not to answer’. The definition of SDV, SA, and NSSDV followed the CDC’s SDV nomenclature recommendations (Crosby et al., Reference Crosby, Ortega and Melanson2011). The UK Biobank Patient Health Questionnaire-9 (PHQ-9) data were collected from mental health web-based Questionnaire B (UK Biobank field ID 20514, 20510, 20534, 20519, 20511, 20507, 20508, 20518, and 20513). Recent thoughts of suicide or self-harm, assessed by PHQ-9 – Question 9 (UK Biobank field ID 20513) – were used for comparison with SDV recency results. Further details regarding the UK Biobank questionnaires can be accessed through the UK Biobank mental health web-based questionnaires protocol (https://bio-bank.ctsu.ox.ac.uk/crystal/refer.cg-i?id=22) on the internet.
MRI acquisition and preprocessing
Image acquisition and quality control of structural T1-weighted magnetic resonance imaging (T1-weighted MRI) and rs-fMRI were conducted by the UK Biobank team. The images were acquired using the Siemens Skyra 3T running VD13A SP4 with a standard Siemens 32-channel head coil (Alfaro-Almagro et al., Reference Alfaro-Almagro, Jenkinson, Bangerter, Andersson, Griffanti, Douaud, Sotiropoulos, Jbabdi, Hernandez-Fernandez, Vallee, Vidaurre, Webster, McCarthy, Rorden, Daducci, Alexander, Zhang, Dragonu, Matthews and Smith2018). All images were anonymized using a defacing mask to remove facial structures by the UK Biobank teams. T1-weighted MRI was acquired using 3-D magnetization prepared rapid gradient echo imaging (3D MPRAGE) (with inverse time (TI)/echo time (TE) = 880/2.01 ms, field of view (FOV) = 208 × 256 × 256 matrix, resolution = 1.0 × 1.0 × 1.0 mm, and flip angle = 8°). rs-fMRI was acquired using gradient echo-planar imaging with ×8 multislice acceleration (GE-EPI with ×8 multislice acceleration) (with repetition time (TR)/ echo time (TE) = 735/39 ms, field of view (FOV) = 88 × 88 × 64 matrix, resolution = 2.4 × 2.4 × 2.4 mm, flip angle = 52°, scan duration = 6 min, time point = 490) (Alfaro-Almagro et al., Reference Alfaro-Almagro, Jenkinson, Bangerter, Andersson, Griffanti, Douaud, Sotiropoulos, Jbabdi, Hernandez-Fernandez, Vallee, Vidaurre, Webster, McCarthy, Rorden, Daducci, Alexander, Zhang, Dragonu, Matthews and Smith2018). Details of the image acquisition are provided in the UK Biobank data acquisition and preprocessing protocol (https://www.fmrib.ox.ac.uk/ukbiobank/protocol/V4_230920-14.pdf) and its associated documents (https://biobank.ctsu.ox.ac.uk/crystal/crystal/do-cs/brain_mri.pdf).
Preprocessing was completed using the Data Processing Assistant for rs-fMRI (DPARSF) (Yan, Wang, Zuo, & Zang, Reference Yan, Wang, Zuo and Zang2016) and Statistical Parametric Mapping 12 (SPM12) (Friston, Reference Friston2007) in the Data Processing & Analysis for Brain Image (DPABI) V4.2_190919 toolbox, an extension in MATLAB (MathWorks, Natick, MA, USA). Preprocessing of T1-weighted MRI included reorientation along the anterior commissure-posterior commissure line (AC-PC line), regularization to the European template, normalization to MNI 152 space (Montreal Neurological Institute 152 space) (Mazziotta et al., Reference Mazziotta, Toga, Evans, Fox, Lancaster, Zilles, Woods, Paus, Simpson, Holmes, Collins, Thompson, MacDonald, Iacoboni, Schormann, Amunts, Palomero-Gallagher, Geyer, Parsons and Pike2001), skull stripping after reorientation, and tissue-type segmentation. Preprocessing of rs-fMRI included the removal of the first 10 time points to correct for initial signal instability, slice timing correction, realignment, reorientation according to the AC-PC line, co-registration to the T1-weighted MRI, normalization to MNI 152 space, head motion correction using the Friston 24-parameter model (Friston et al., Reference Friston, Williams, Howard, Frackowiak and Turner1996), regression of nuisance covariates to remove cardiac and respiratory effects, as well as mean white matter and mean CSF signals (Hallquist, Hwang, & Luna, Reference Hallquist, Hwang and Luna2013), and band-pass filtering was performed at 0.01–0.1 Hz.
Functional connectivity analysis
After preprocessing, the rs-fMRI data were parcellated into 90 brain regions using the automated anatomical labeling atlas (AAL-90) (Tzourio-Mazoyer et al., Reference Tzourio-Mazoyer, Landeau, Papathanassiou, Crivello, Etard, Delcroix, Mazoyer and Joliot2002). The time series was extracted from the mean of the blood-oxygen-level-dependent (BOLD) signals in each brain area across the 480 time points, with the interval set as the repetition time (TR = 0.735 s). Functional connectivity, assessed by calculating Pearson’s correlation between BOLD signals, was defined as the correlation coefficient (r score) for each region pair. Connectivity matrices (resulting in a 90 × 90 matrix) were constructed, followed by Fisher r-to-z conversion to standardize the r scores to a standard normal distribution (Finn et al., Reference Finn, Shen, Scheinost, Rosenberg, Huang, Chun, Papademetris and Constable2015; Rosenberg et al., Reference Rosenberg, Finn, Scheinost, Papademetris, Shen, Constable and Chun2016).
Statistical analysis
Two-sample t tests were conducted for continuous variables in the demographic tables. Chi-square tests were conducted for categorical variables, and Fisher’s exact test was used when the expected cell counts in any cell were <5 (Table 1 and Supplementary Table S2). The association between SA, frequency, and recency was analyzed using a general linear regression model (GLM). To compare the differences in FC across the 90 brain regions between the SA and NSSDV groups, the GLM was performed 4,005 times
$ \Big({C}_2^{90} $
= 4,005), with a statistical significance level set at p < 0.05. The same calculation steps were applied to compare the FC alterations related to the frequency and recency of SDV. Covariates included in the analysis were age, gender, BMI, and mean framewise displacement (FD; Jenkinson method), based on prior studies (Alfaro-Almagro et al., Reference Alfaro-Almagro, McCarthy, Afyouni, Andersson, Bastiani, Miller, Nichols and Smith2021; Satterthwaite et al., Reference Satterthwaite, Ciric, Roalf, Davatzikos, Bassett and Wolf2019; Smith & Nichols, Reference Smith and Nichols2018). False discovery rate (FDR) correction was applied for multiple comparison using the Benjamini–Yekutieli procedure with the false discovery rate controlled at α = 0.05 (Benjamini & Yekutieli, Reference Benjamini and Yekutieli2001; Shen et al., Reference Shen, Cox, Adams, Howard, Lawrie, Ritchie, Bastin, Deary, McIntosh and Whalley2018).
Demographic and clinical characteristics for the overall participants, divided by groups with a history of SA and NSSDV
Table 1. Long description
SA, suicide attempt; NSSDV, non-suicidal self-directed violence; SD, standard deviation; BMI, body mass index; SDV, self-directed violence
Note: BMI, body mass index; NSSDV, nonsuicidal self-directed violence; SA, suicide attempt; SD, standard deviation; SDV, self-directed violence.
a Psychiatric comorbidities were defined based on main ICD-10 diagnosis codes from hospital inpatient records (UK Biobank field ID 41202). Details are provided in Table S1.
To assess the base rate of significant connections between each brain region and the remaining 89 brain regions, 5,000 permutation tests were performed to generate a permutation-derived null distribution of the number of significant connections (Winkler et al., Reference Winkler, Ridgway, Webster, Smith and Nichols2014). To quantify the probability of obtaining the observed number of significant connections under the null distribution, the permutation p-value was calculated as
$ {P}_{perm}=\frac{b+1}{m+1} $
, where b represents the number of permutations yielding a number of significant connections greater than or equal to the observed value, and m represents the total number of permutations performed (Phipson & Smyth, Reference Phipson and Smyth2010).
Results
Demographic data
A total of 1,070 participants were included in this study, all of whom had engaged in self-harm at least once. The average age of the participants was 51.82 years (SD = 7.08), with 72.71% being female. The average BMI indicated that participants were overweight (mean = 26.59 kg/m2, SD = 4.81). Additionally, 97.66% of participants identified as non-Hispanic White. Regarding self-harm frequency, approximately half of the participants (n = 532) reported engaging in self-harm only once. In terms of timing, 91.78% (n = 982) indicated that their most recent self-harm incident had occurred more than 12 months ago. Here, 54.11% (n = 579) reported self-harm with the intention of ending their lives, classifying them as the SA group in this study. Demographic characteristics and questionnaire analysis are presented in Table 1. After excluding participants with psychiatric comorbidities, a total of 1,030 participants were included in this study. The demographic characteristics are presented in Supplementary Table S2.
Functional connectivity associated with SA
Figure 1 illustrates alterations in FC in the SA group compared to the NSSDV group after Benjamini–Yekutieli FDR correction (FDR controlled at α = 0.05; uncorrected p-value threshold = 0.0495). Individuals with SA exhibited negative correlations in FC between the bilateral amygdala and other brain regions, while positive correlations were observed between the right caudate and other brain regions (Figure 1a and Supplementary Figure S2). The bilateral amygdala and right caudate were the top three brain regions with the most significantly altered FC in the SA group (Figure 1b and Supplementary Tables S3 and S4). FC between the left amygdala and right parahippocampal gyrus was negatively correlated with SA when statistical significance was set at p < 0.0005 (Figure 1c). A total of 14 brain regions exceeded the 95th percentile of the permutation-derived null distribution (
$ {P}_{\mathrm{perm}} $
< 0.05). A sensitivity analysis excluding participants with psychiatric disorders identified 12 brain regions that exceeded the 95th percentile of the permutation-derived null distribution (Supplementary Tables S3 and S4). The top 50 significant FC links correlated with SA are presented in Supplementary Tables S11 and S12.
The significant FC correlation in the SA group compared to that in the NSSDV group. Five hundred and five models remained significant after Benjamini–Yekutieli FDR corrections (FDR controlled at α = 0.05; uncorrected p-value threshold = 0.0495). (a) Negative t-value (cold color spectrum) represents SA is associated with lower FC, while positive t-value (warm color spectrum) represents an association with higher FC. (b) Top three brain regions with the most statistically significant GLM models, which implied a shift in BOLD dynamics within these regions. (c) FC that negatively correlated with SA when setting the statistical significance level at p < 0.0005. Note: FC, ‘functional connectivity’; SA, ‘suicide attempt’; NSSDV, ‘nonsuicidal self-directed violence’; FDR, ‘false discovery rate’; GLM, ‘general linear regression model’; BOLD dynamics, ‘blood-oxygen-level-dependent dynamics’.
Figure 1. Long description
(a) Negative t (cold color spectrum) value represents SA is associated with lower FC, while positive t value (warm color spectrum) represents the association of higher FC. (b) Top three brain regions with the most statistically significant GLM models, which implied a shift in BOLD dynamics within these regions (c) FC that negatively correlated to SA when setting statistical significance level at p < 0.0005.
FC, functional connectivity; SA, suicide attempt; NSSDV, non-suicidal self-directed violence; FDR, false discovery rate; GLM, general linear regression model; BOLD dynamics, Blood-oxygen level dependent dynamics.
Functional connectivity associated with frequency of SDV
Figure 2 illustrates the alterations in FC associated with SDV frequency after FDR correction (FDR controlled at α = 0.05; uncorrected p-value threshold = 0.05). As frequency increased, positive correlations in FC were observed. The results showed clusters of positive correlations between the bilateral middle frontal gyrus, orbitofrontal cortex (OFC), cuneus, and occipital lobes (Figure 2a and Supplementary Figure S3). The left inferior orbitofrontal gyrus, right cuneus, bilateral inferior occipital gyrus, and left anterior cingulate and paracingulate gyri were the top three brain regions with the most significantly altered FC following increased SDV frequency (Figure 2b and Supplementary Tables S5 and S6). FC between the right cuneus and right middle OFC was positively correlated with SDV frequency when statistical significance was set at p < 0.0001 (Figure 2c). A total of four brain regions exceeded the 95th percentile of the permutation-derived null distribution (
$ {P}_{\mathrm{perm}} $
< 0.05). A sensitivity analysis excluding participants with psychiatric disorders identified 14 brain regions that exceeded the 95th percentile of the permutation-derived null distribution (Supplementary Tables S5 and S6). The top 50 significant FC links correlated with SDV frequency are presented in Supplementary Tables S13 and S14.
The correlation between frequency of SDV and FC. 262 models remained significant after FDR corrections (FDR controlled at α = 0.05; uncorrected p-value threshold = 0.0500). (a) Positive t-value (warm color spectrum) represents frequency of SDV is associated with higher FC, while negative t-value (cold color spectrum) represents an association with lower FC. (b) Top three brain regions with the most significant GLM model, which implied a shift in BOLD dynamics within these regions. (c) FC that positively correlated with SDV frequency when setting the statistical significance level at p < 0.0001. Note: SDV, ‘self-directed violence’; FC, ‘functional connectivity’; FDR, ‘false discovery rate’; GLM, ‘general linear regression model’; BOLD dynamics, ‘blood-oxygen-level-dependent dynamics’.
Figure 2. Long description
(a) Positive t value (warm color spectrum) represents frequency of SDV is associated with higher FC, while negative t value (cold color spectrum) represents the association of lower FC. (b) Top three brain regions with the most significant GLM model, which implied the shift in BOLD dynamics within these regions (c) FC that positively correlated to SDV frequency when setting statistical significance level at p < 0.0001.
SDV, self-directed violence; FC, functional connectivity; FDR, false discovery rate; GLM, general linear regression model; BOLD dynamics, Blood-oxygen level dependent dynamics.
Functional connectivity associated with recency of SDV and self-harm thoughts
Figure 3 depicts alterations in FC associated with the SDV recency after FDR correction (FDR controlled at α = 0.05; uncorrected p-value threshold = 0.05). The results showed that individuals who engaged in SDV within the past 12 months exhibited negative correlations in FC between the bilateral middle frontal gyrus, OFC, rectus gyrus, and temporal lobe (Figure 3a and Supplementary Figure S4). The right middle orbitofrontal gyrus, right superior orbitofrontal gyrus, left rectus gyrus, right precuneus, and left temporal pole were the top three brain regions with the most significantly altered FC in individuals with SDV within the past 12 months (Figure 3b and Supplementary Tables S7 and S8). FC between the right lingual gyrus and left calcarine cortex was negatively correlated with SDV recency when statistical significance was set at p < 0.001 (Figure 3c). However, none of the brain regions exceeded the 95th percentile of the permutation-derived null distribution both before and after excluding participants with psychiatric disorders (
$ {P}_{\mathrm{perm}} $
< 0.05) (Supplementary Tables S7 and S8). The top 50 significant FC links correlated with the SDV recency are presented in Supplementary Tables S15 and S16.
The correlation between recency of SDV and FC. One hundred and fifty-two models remained significant after FDR corrections (FDR controlled at α = 0.05; uncorrected p-value threshold = 0.0500). (a) Negative t-value (cold color spectrum) represents SDV recency was associated with lower FC, while positive t-value (warm color spectrum) represents an association with higher FC. (b) Top three brain regions with the most significant GLM models, which implied a shift in BOLD dynamics within these regions. (c) FC that negatively correlated with SDV recency when setting the statistical significance level at p < 0.001. Note: SDV, ‘self-directed violence’; FC, ‘functional connectivity’; FDR, ‘false discovery rate’; GLM, ‘general linear regression model’; BOLD dynamics, ‘blood-oxygen-level-dependent dynamics’.
Figure 3. Long description
(a) Negative t value (cold color spectrum) represents SDV recency was associated with lower FC, while positive t value (warm color spectrum) represents the association of higher FC. (b) Top three brain regions with the most significant GLM models, which implied the shift in BOLD dynamics within these regions (c) FC that negatively correlated to SDV recency when setting statistical significance level at p < 0.001.
SDV, self-directed violence; FC, functional connectivity; FDR, false discovery rate; GLM, general linear regression model; BOLD dynamics, Blood-oxygen level dependent dynamics.
Supplementary Figure S5 depicts alterations in FC associated with the self-harm thought recency based on PHQ-9 Question 9 after FDR correction (FDR controlled at α = 0.05; uncorrected p-value threshold = 0.05). The results demonstrated negative connectivity in the inferior frontal gyrus correlated to increased self-harm thought recency within the past 2 weeks (Supplementary Figures S5 and S6). The right triangular part of the inferior frontal gyrus, left caudate nucleus, and right amygdala were the top three brain regions with the most significantly altered FC following self-harm thought recency. Only the right triangular part of the inferior frontal gyrus exceeded the 95th percentile of the permutation-derived null distribution both before and after excluding participants with psychiatric disorders (
$ {P}_{\mathrm{perm}} $
< 0.05) (Supplementary Tables S9 and S10). The top 50 significant FC links correlated with the self-harm thought recency are presented in Supplementary Tables S17 and S18.
Discussion
This study explored FC differences in suicidal behaviors among adults. The key FC difference identified in this large-sample database study reflects altered reward circuits and limbic systems associated with clinical manifestations in individuals with SDV. First, widespread FC alterations in the bilateral amygdala and right caudate were observed in the SA group compared to the NSSDV group. Second, a significant increase in FC between the OFC and occipital lobes was observed with higher SDV frequency. Third, decreased FC between the OFC and rectus gyrus, as well as the OFC and temporal lobe, was found in individuals who engaged in SDV within the past 12 months.
The connectivity alterations identified in this study support neurofunctional distinctions between SA and NSSI. Although both SA and NSSI are often driven by a desire to relieve distress, they differ in suicidal intent, lethality, and interpersonal factors (Hamza, Stewart, & Willoughby, Reference Hamza, Stewart and Willoughby2012; Klonsky & May, Reference Klonsky and May2015). Previous studies suggest that the DMN, limbic network, and reward network play key roles in SA among adults (Kang et al., Reference Kang, Na, Choi, Kim, Son and Lee2017; Kim et al., Reference Kim, Kim, Myung, Han, Fava, Mischoulon, Papakostas, Seo, Cho, Seong and Jeon2017; Ordaz et al., Reference Ordaz, Goyer, Ho, Singh and Gotlib2018; Schmaal et al., Reference Schmaal, van Harmelen, Chatzi, Lippard, Toenders, Averill, Mazure and Blumberg2020), while neuroimaging findings specific to NSSI alone have been less reported (Domínguez-Baleón, Gutiérrez-Mondragón, Campos-González, & Rentería, Reference Domínguez-Baleón, Gutiérrez-Mondragón, Campos-González and Rentería2018). The observed FC alterations align with prior studies suggesting limbic network dysfunction and reward circuit differences (Huang et al., Reference Huang, Rootes-Murdy, Bastidas, Nee and Franklin2020; Kang et al., Reference Kang, Na, Choi, Kim, Son and Lee2017; Myung et al., Reference Myung, Han, Fava, Mischoulon, Papakostas, Heo, Kim, Kim, Kim, Kim, Seo, Seong and Jeon2016).
While sample size is crucial for reliability, analytic methodology also influences connectivity results. Thompson et al. (Reference Thompson, Ghahramanlou-Holloway, Murphy, Perera, Benca-Bachman, Palmer and Gray2023) examined triple-network differences between individuals with SA and those with self-directed violence using the same dataset, finding no significant differences in connectivity in extracted ICA components (Jenkinson, Bannister, Brady, & Smith, Reference Jenkinson, Bannister, Brady and Smith2002; Thompson et al., Reference Thompson, Ghahramanlou-Holloway, Murphy, Perera, Benca-Bachman, Palmer and Gray2023). ICA requires manual identification of neuroanatomical relevance (Fox & Raichle, Reference Fox and Raichle2007), and the limited number of components may overlook key systems. In contrast, our study accounted for physical artifacts through nuisance regression and used the AAL atlas for region-based connectivity analysis.
Altered functional connectivity in individuals with suicide attempt
A negative FC association in the bilateral amygdala and a positive association in the right caudate was observed in individuals with SA compared to NSSDV. The amygdala is a crucial structure within the limbic network, involved in processing emotions such as fear, anxiety, and aggression (Torrico & Abdijadid, Reference Torrico and Abdijadid2025). Dysregulation in amygdala neurotransmission has been strongly linked to anxiety disorders (Babaev, Piletti Chatain, & Krueger-Burg, Reference Babaev, Piletti Chatain and Krueger-Burg2018), a well-established personality trait associated with suicidal behavior (Turecki et al., Reference Turecki, Brent, Gunnell, O’Connor, Oquendo, Pirkis and Stanley2019). Previous studies have reported decreased left amygdala FC with the left lateral occipital cortex in adolescents with bipolar disorder and self-harm (Dimick et al., Reference Dimick, Hird, Sultan, Mitchell, Sinyor, MacIntosh and Goldstein2023), as well as increased amygdala connectivity with the insula, superior OFC, and middle temporal cortex in adults with major depressive disorder (MDD) and SA (Kang et al., Reference Kang, Na, Choi, Kim, Son and Lee2017).
The caudate contains the nucleus accumbens, a core hub of the reward circuit, and plays a key role in reward processing, motivation, and emotion (Driscoll, Bollu, & Tadi, Reference Driscoll, Bollu and Tadi2025; Haber & Knutson, Reference Haber and Knutson2010). Research has highlighted the reward circuit as a key component in impulsive regulation, with dopaminergic dysregulation implicated in reward deficiency syndromes (Basar et al., Reference Basar, Sesia, Groenewegen, Steinbusch, Visser-Vandewalle and Temel2010; Blum et al., Reference Blum, Braverman, Holder, Lubar, Monastra, Miller, Lubar, Chen and Comings2000). A review found evidence of blunted striatal activation in individuals with suicidal thoughts, behaviors, and NSSI (Auerbach et al., Reference Auerbach, Pagliaccio, Allison, Alqueza and Alonso2021). Given the stronger association between suicidal behavior and impulsive-aggressive traits (Turecki et al., Reference Turecki, Brent, Gunnell, O’Connor, Oquendo, Pirkis and Stanley2019), the increased connectivity in the caudate observed in this study may reflect dysregulation of the reward circuit in individuals with SA.
Functional connectivity in OFC and visual cortex associated with SDV frequencies and recency
A positive association of FC in the left OFC was observed with increased SDV frequency, while trends of negative association in the right OFC were found in individuals with a shorter recency since their last SDV. The OFC plays a key role in decision-making, emotion regulation, and response inhibition and is functionally connected to both the limbic network and the reward circuit (Haber & Knutson, Reference Haber and Knutson2010; Kringelbach, Reference Kringelbach2005; Rolls, Reference Rolls2015). Self-harm is often viewed as a maladaptive coping strategy when individuals experience stress (Hamza et al., Reference Hamza, Stewart and Willoughby2012; Turecki et al., Reference Turecki, Brent, Gunnell, O’Connor, Oquendo, Pirkis and Stanley2019). Dysfunctional OFC activity may lead individuals to resort to self-harm as an ineffective stress-relief strategy by impairing positive thinking and response inhibition. An rs-fMRI study reported increased FC between the OFC and sensorimotor network and ventral attention network, along with decreased FC between the OFC and frontoparietal network, in individuals with SA and MDD compared to MDD controls (Wen et al., Reference Wen, Zhang, Wei, Wu, Zhang, Zhang and Hou2025). Task-based fMRI studies have identified increased FC between the OFC and precuneus, and supplementary motor areas during reward processing in individuals with NSSI (Lin et al., Reference Lin, Liu, Qiu, Yang, Yang, Tian, Wang, Zhang, Bai and Xu2024). Guo et al. (Reference Guo, Lu, Ou, Huang, Li, Cui, Li, Zheng, Liang, Qiu and Liu2024) found a negative correlation between FC in the right OFC and right insula and NSSI frequency in adolescents with depression, consistent with the findings of this study (Guo et al., Reference Guo, Lu, Ou, Huang, Li, Cui, Li, Zheng, Liang, Qiu and Liu2024). Notably, the inferior frontal cortex, rather than the OFC, was associated with the recency of self-harm thoughts, suggesting a shift in the dominant brain regions following the progression of the ideation-to-action framework.
The increased FC associated with SDV frequency and visual-related regions, including cuneus and inferior occipital gyrus, has been less frequently discussed in previous neuroimaging studies on suicidal behavior (Schmaal et al., Reference Schmaal, van Harmelen, Chatzi, Lippard, Toenders, Averill, Mazure and Blumberg2020). Regions in the occipital lobes contribute to the perception of emotional information through pathways from the visual cortex to the OFC (Fusar-Poli et al., Reference Fusar-Poli, Placentino, Carletti, Landi, Allen, Surguladze, Benedetti, Abbamonte, Gasparotti, Barale, Perez, McGuire and Politi2009; Rolls, Reference Rolls2015). The positive correlation between FC and SDV frequency may suggest the role of visual stimuli in the neurocircuitry underlying SDV. A magnetoencephalography (MEG) study found increased activation and hyper-inter-network connectivity between visual regions and the limbic system in the low gamma band in individuals with MDD and a history of SA (Dai et al., Reference Dai, Zhang, Wang, Chen, Zhang, Hu, Yao and Lu2023). A facial expression recognition study reported that adolescents with NSSI rated emotional stimuli as more unpleasant and arousing compared to controls (In-Albon, Ruf, & Schmid, Reference In-Albon, Ruf and Schmid2015). In rs-fMRI research, divergent alterations in visual network were observed in depressed adolescents engaging in NSSI and SA (Li et al., Reference Li, Liang, Li, Xu, Yang, Liu, Zhang, Wang, Zhang and Zhou2024).
Limitations
Several limitations should be considered when interpreting these findings. First, while UK Biobank provides extensive and valuable data, distinguishing subgroups within the psychiatric population at high risk of suicidal behaviors remains challenging. The demographic analysis of our study included participants with mental illness, but only 5% were diagnosed with MDD and 5% with BD under the ICD-10 classification. Up to 6,000 participants without a psychiatric diagnosis engaged in self-directed violence, highlighting the underdiagnosis of mental illness in the general population. According to the CDC, more than half of individuals who died by suicide had no formal psychiatric diagnosis at the time of their death (Centers for Disease Control and Prevention, 2018). Second, there is inherent heterogeneity in medical history and clinical profile, including factors such as early life adversity and substance use. However, this heterogeneity reflects a real-world population, enhancing the generalizability of the findings. Third, the cross-sectional design of this study prevents causal inference between FC alterations and suicidal behaviors. Longitudinal studies are needed to establish causality, though they face challenges due to the complex and unpredictable nature of suicide risk factors (Klonsky & May, Reference Klonsky and May2015).
Conclusions
This study leveraged the large-sample UK Biobank to investigate differences in FC between individuals with lifetime SA and those with lifetime NSSDV. The main findings revealed that SA was associated with decreased FC in the bilateral amygdala and right putamen, alongside increased FC in the right caudate. Higher SDV frequency correlated with increased FC between the OFC and visual-related occipital lobes, while recent SDV within the past 12 months was linked to decreased FC in the OFC, particularly with the rectus gyrus and temporal lobe. These observations support the notion of neurobiological dysregulation in suicidal behaviors, especially in limbic system–associated brain regions. Further research is needed to examine how predisposition factors (e.g. early-life adversity and family history) influence functional neuroimaging findings. Overall, this study offers valuable insights into the neurobiological mechanisms underlying suicidal behaviors and may aid in risk stratification, ultimately informing public suicide prevention efforts.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726106114.
Data availability statement
Data supporting the findings of this study are available through the UK Biobank project. Additional information about registration and application to UK Biobank is available at the following link: https://www.ukbiobank.ac.uk/.
Acknowledgments
This study used data from UK Biobank under application number 91995. The authors would like to thank UK Biobank Access and the epidemiological and data analyst team for their tireless work to help researchers access the data. The authors would particularly like to thank the 500,000 participants in the UK Biobank study for their enormous generosity and altruism and their continued interest, support, and involvement.
Author contribution
Conceptualization: J.-Y.L., I-J.C., and A.C.-C.Y. Investigation: J.-Y.L. and I-J.C. Methodology: J.-Y.L. and J.-D.Z. Formal analysis: J.-Y.L. Visualization: J.-Y.L. and I-J.C. Writing – original draft: J.-Y.L. Writing – review and editing: J.-Y.L., I-J.C., J.-D.Z., and A.C.-C.Y. Project administration: J.-Y.L., J.-D.Z., and A.C.-C.Y. Resources: J.-D.Z. Supervision: A.C.-C.Y. Funding acquisition: A.C.-C.Y.
Funding statement
This work was supported by grants from the National Science and Technology Council, Taiwan (grant number NSCT 115-2321-B-A49-010, 115-2321-B-A49-002, 114-2321-B-A49-003, and 114-2634-F-A49-006). ACY was also supported by the Mt. Jade Young Scholarship Award from the Ministry of Education, Taiwan, as well as the Brain Research Center, National Yang Ming Chiao Tung University, the Ministry of Education (Aim for the Top University Plan), Taipei, Taiwan, and the Yin Shu-Tien Foundation Taipei Veterans General Hospital-National Yang Ming Chiao Tung University Excellent Physician Scientists Cultivation Program, No. 115-V-B-073. The authors gratefully acknowledge financial support from the UKB-RAP Transition Credits Programme provided by the UK Biobank Access team.
Competing interests
The authors declare no competing interests.
Open access
