Neural Correlates Associated with Suicide and Non-Suicidal Self-Injury in Youth

Abstract

There is no definitive neural marker of suicidal thoughts and behaviors (STB) or non-suicidal self-injury (NSSI), and relative to adults, research in youth is more limited. This comprehensive review focuses on magnetic resonance imaging (MRI) studies reporting structural and functional neural correlates of STB and NSSI in youth to: (i) elucidate shared and independent neural alternations, (ii) clarify how developmental processes may interact with neural alterations to confer risk, and (iii) provide recommendations based on convergence across studies. Forty-seven articles were reviewed (STB = 27; NSSI = 20), and notably, 63% of STB articles and 45% of NSSI articles were published in the previous 3 years. Structural MRI research suggests reduced volume in the ventral prefrontal and orbitofrontal cortices among youth reporting STB, and there is reduced anterior cingulate cortex volume related to STB and NSSI. With regard to functional alterations, blunted striatal activation may characterize STB and NSSI youth, and there is reduced frontolimbic task-based connectivity in suicide ideators and attempters. Resting state functional connectivity findings highlight reduced positive connectivity between the default mode network and salience network in attempters, and self-injurers exhibit frontolimbic alterations. Together, suicidal and non-suicidal behaviors are related to top-down and bottom-up neural alterations, which may compromise approach, avoidance, and regulatory systems. Future longitudinal research with larger and well-characterized samples, especially those integrating ambulatory stress assessments, will be well-positioned to identify novel targets that may improve early identification and treatment for youth with STB and NSSI.

Suicidal thoughts and behaviors (STB) and non-suicidal self-injury (NSSI) in youth are major public health concerns. Although STB and NSSI often co-occur (1), their etiologies are complex and only partially overlap, suggesting both shared and unique vulnerability factors that potentiate STB and NSSI (2, 3). Recent data show that 12.1% of adolescents think about suicide and 4.1% have made an attempt (4). This is concerning, as upwards of 70% of completers are first time attempters, with suicide deaths markedly higher among males (e.g., 5). By comparison, 15–20% of adolescents report lifetime NSSI engagement (6), disproportionately affecting females (7). These sex differences are important to consider, particularly as NSSI is often a precursor to suicidal behaviors in females, but many males attempting suicide report little or no NSSI history (8, 9). Critically, the motivation for engaging in suicidal (e.g., intolerance of psychological pain (10); perceived burdensomeness (11)) versus non-suicidal behaviors (e.g., self-punishment (12); social signaling (13)) also may differ, perhaps influencing why NSSI often desists in adulthood (14), whereas STB may persist (15, 16). Taken together, contemporary models underscore the importance of elucidating interactive, and potentially transactional, processes that may differentially shape the trajectories of STB and NSSI during the transition from adolescence to adulthood and accordingly, focus on identifying biological (e.g., neural correlates) and environmental (e.g., child abuse) factors in the STB and NSSI pathways (2, 3, 13, 17–20).

The diathesis-stress model provides a framework to understand vulnerability to self-injurious behaviors. Neural alterations are biological diatheses that may be a necessary but not sufficient predisposition to engage in self-injurious behaviors. Coupled with specific environmental exposures—particularly acute stress (e.g., peer victimization, interpersonal loss)—the likelihood of suicidal (21–23) and non-suicidal (24–26) behaviors may increase. Childhood and adolescence represent a unique developmental phase to unpack neural diatheses and stressors that may be separable from adults. From a neurodevelopmental perspective, this is a period characterized by asynchronicity between limbic and prefrontal systems (27). Although normative, this has a direct effect on motivated behavior, particularly as it relates to the neural circuitry of approach, avoidance, and regulatory systems. Within the triadic model of motivated behavior, the approach system is guided by the ventral striatum, the avoidance system is largely controlled by the amygdala, and the regulatory system—generally managed by the prefrontal cortex—is believed to balance approach and avoidance behaviors (28). Indeed, Casey and colleagues have long advanced that risky behaviors in youth stem from earlier maturation of subcortical regions relative to immature prefrontal systems (29). By contrast, STB and NSSI may manifest not when there is asynchronicity between limbic and prefrontal systems, but perhaps, due to deficiencies at each level—across approach, avoidance, and regulatory systems. Although failure with any one system may result in the emergence of debilitating symptoms or risky behaviors (e.g., (30, 31)), when each system is compromised, it undermines the capacity to engage in compensatory strategies (e.g., future-oriented thinking, cognitive flexibility) that would otherwise limit the tendency to engage in suicidal and non-suicidal behaviors. This is consistent with a recent model summarizing extant STB findings (32), showing that alterations within the extended ventral prefrontal cortex (VPFC) system may potentiate suicidal thinking given prominent roles in negative self-referential thinking and rumination (33, 34). Suicidal thinking may then be exacerbated by dorsolateral prefrontal cortex (DLPFC), inferior frontal gyrus (IFG), rostral prefrontal cortex (PFC), and dorsal anterior cingulate cortex (dACC) alterations, as this directly bears on effective cognitive control, flexibility, and decision-making (35). Perturbations in bottom-up and top-down connections between these extended systems may, for some, facilitate the transition from suicidal ideation to behavior. It remains unknown, however, whether this relates to STB in youth, and if this model is applicable to NSSI.

Neural alterations do not operate in isolation. Rather, they interact dynamically with stress (2, 3, 36, 37), a point underscored by the fact that stress is prominently featured in many leading STB (11, 20, 38) and NSSI (25, 39) theories. Childhood and adolescence reflect a unique developmental period whereby there is a progressive shift towards autonomy from parents and greater reliance on peers (40). Interpersonal stressors occur in great abundance during this life stage (23)—a byproduct of navigating puberty (41), fostering close friendships (42), exploring romantic relationships (43), and increasingly, using social media (44). Acute interpersonal stress may serve as proximal triggers that interact with neural diatheses, which we maintain in this review, may facilitate the transition from ideation to action—for both suicidal and non-suicidal behaviors.

Presently, there are no definitive neural markers of STB or NSSI. Several prior reviews have highlighted brain regions implicated in suicidal (32, 45–49) and non-suicidal behaviors (39, 50). This review focuses on extant magnetic resonance imaging (MRI) studies reporting structural and functional neural correlates of STB and NSSI in youth with the aim of: (i) elucidating shared and unique neural alternations, (ii) clarifying how developmental processes may interact with neural alterations to confer risk, and (iii) providing recommendations based on convergence across studies.

Method

A PubMed and Google Scholar literature search through January 15, 2020 was conducted to identify original research using the search terms: “MRI,” “magnetic resonance imaging,” “structural MRI,” “fMRI,” “functional MRI,” “functional magnetic resonance imaging,” “functional connectivity,” “resting state fMRI,” “resting state connectivity” in combination with “suicide,” “suicide attempt,” “suicidal ideation,” “suicidal behavior,” “non-suicidal self-injury,” “parasuicidal behaviors,” and “self-harm.” Articles were selected if they: (i) were published in an English-language peer-reviewed journal, (ii) included participants under age 26, (iii) included participants with lifetime STB and/or NSSI, and (iv) focused on structural, functional, or resting state MRI. There were no inclusion requirements based on STB or NSSI assessments (Tables S1–S3 for clinical measures) or MRI scanners. Several reviewed publications assess neural correlates of STB among youth with lifetime NSSI, and conversely, NSSI studies include STB youth (Tables S1–S3 for sample overlap). Extant research is summarized, but a pooled meta-analytic estimate of effects is not provided.

Results

Forty-seven articles satisfy inclusion criteria (STB = 27; NSSI = 20; Figure 1). The majority of participants in these studies reported lifetime major depressive disorder (MDD), bipolar disorder (BD), and/or borderline personality disorder (BPD) (Table 1–3 for sample characteristics). There is enormous heterogeneity in methodological approaches. Most STB studies compare suicide attempters versus a psychiatric or healthy comparison group. Few studies compare attempters versus ideators. NSSI samples are typically compared with either psychiatric or healthy youth, with some research testing correlates among NSSI youth only. The vast majority of research is cross-sectional.

Figure 1.

Suicidal thoughts and behaviors (STB; gray) and non-suicidal self-injury (NSSI; blue) neuroimaging studies included in this comprehensive review. Publications are divided into 3-year bins and calculated separately for STB and NSSI.

Table 1.

Structural alterations associated with suicidal thoughts and behaviors and non-suicidal self-injury

Publication	Sample	Age	Method	Select Findings
Suicidal Thoughts and Behaviors
Fan et al., 2019 (54)	BD+SA = 21 BD = 25 MDD+SA = 19 MDD = 18	14–25	Whole Brain, SPM8, GMV	1,4BD+SA, MDD+SA > BD, MDD ↓ Left VPFC (BA11); MDD+SA > MDD, BD+SA, BD ↑ Left VPFC (BA47)
Fradkin et al., 2017 (57)	MDD+SA = 29 HC = 29	12–19	Whole Brain, FreeSurfer/Qdec, Cortical Thickness, Surface Area	1,3MDD+SA ↓ VMPFC cortical thickness ↑ Motor impulsivity; MDD+SA ↑ Right paracentral lobule cortical thickness ↑ Non-planning impulsivity
Goodman et al., 2011 (59)	BPD+MDD = 13 HC = 13	13–17	ROI (Cingulate, PFC), GMV/WMV	1,3↑ SA ↓ BA24 volume, ↑ BA23 WMV
Ho et al., 2018 (58)	§Community Sample = 152	9–13	ROI (Caudate, NAcc, Putamen), FreeSurfer, GMV	1,3↓ Bilateral Putamen predicts ↑ Implicit SI
Huber et al., 2019 (55)	BD+SA = 15 BD = 18 HC = 25	13–21	ROI (OFC), FreeSurfer, Thickness/Volume	1,3BD+SA > HC ↓ Left OFC volume, ↓ Bilateral OFC thickness; BD+SA > BD ↓ Right OFC thickness; ↑ SA lethality ↓ Bilateral OFC volume; ↑ SI ↓ left OFC volume
Johnston et al., 2017a (53)	BD+SA = 26 BD = 42 HC = 45	14–25	Whole Brain, SPM5, GMV	1,4 BD+SA > BD ↓ Right OFC, Right Hippocampus, Bilateral Cerebellum GMV; BD+SA > HC ↓ Right Hippocampus GMV
Lippard et al., 2019 (56)	Future SA = 17 No Future SA = 29	13–25	Whole Brain, SPM12, GMV	1,3Future SA > No Future SA↓ VPFC GMV (Left BA11, Right Ventral BA10, Right BA47), ↓ Left DRPFC (BA10)
McLellan et al., 2018 (60)	MDD+SA = 14 MDD = 14 HC = 17	12–21	ROI (rSTG, Frontal, Temporal Regions), FreeSurfer, Cortical Thickness/Volume	1,3MDD+SA > HC ↓ rSTG volume
Pan et al., 2015 (61)	MDD+SA = 28 MDD = 31 HC = 41	12–17	Whole Brain, FreeSurfer/Qdec, Cortical Thickness/ GMV/WMV	1,3MDD+SA > HC ↓ rSTG GMV
Non-Suicidal Self-Injury
Ando et al., 2018 (64)	§NSSI+SA = 16 §NSSI = 13 HC = 21	*^NSSI = 15.9±1.3 *HC = 15.8±1.1	ROI (PFC, ACC, Insula, Thalamus, Hippocampus, Amygdala), FreeSurfer, GMV	1,5All NSSI > HC ↓ Insula GMV, ↓ ACC GMV; NSSI+SA> NSSI, HC ↓ ACC GMV
Beauchaine et al., 2019 (62)	§NSSI = 20 §No NSSI = 20	13–19	Whole Brain, SPM12, GMV	1,5NSSI > No NSSI ↓ Bilateral Insular GMV, ↓ Right IFG
Chanen et al., 2008c (66)	BPD = 20 HC = 20	*BPD = 17.3±1.1 *HC = 19.0±2.2	ROI (OFC, Amygdala, Hippocampus), Hand Tracing, GMV	1,4 No association between OFC GMV and number of episodes; 1,5 No association between Amyg or Hippocampal GMV and number of episodes  Sex: BPD Males > BPD Females, HC ↓ Amyg GMV
Takahashi et al., 2009b (65)	BPD = 20 HC = 20	*BPD = 17.3±1.1 *HC = 19.0±2.2	ROI (Insula), Hand Tracing, GMV	2,3No association between Insula GMV and number of episodes; Null findings between BPD with and without episodes in the past 6 months
Takahashi et al., 2009b (67)	BPD = 20 HC = 20	*BPD = 17.3±1.1 *HC = 19.0±2.2	ROI (AI, CSP), Hand Tracing, Length	1,4No association between AI or CSP length and number of episodes
Takahashi et al., 2010b (68)	BPD = 20 HC = 20	*BPD = 17.3±1.1 *HC = 19.0±2.2	ROI (STG), Hand Tracing, Volume	1,3No association between STG volume and number of episodes; Null findings between BPD with and without episodes in the past 6 months
Whittle et al., 2009c (69)	BPD = 15 HC = 15	*BPD = 17.4±1.15 *HC = 19.7±2.18	ROI (ACC), Hand Tracing, GMV	1,5↓ Left ACC GMV ↑ Number of episodes

Note. Bold = Strong methodological approach (i.e., clinical-control comparison, interview assessment of STB/NSSI, data analysis)

^*Age range not reported

^{^}Age range for all NSSI participants including the NSSI+SA group

^aAlso includes fMRI (Table 2)

^b,cSame sample

^§Comorbid disorders see Table S1 for inclusion and exclusion criteria

¹Age effects not reported

²Age effects null

³Sex effects not reported

⁴Sex effects null

⁵Female only sample Group.

BD = Bipolar Disorder; BPD = Borderline Personality Disorder; Future SA = Participant made a suicide attempt between baseline and follow-up; HC = Healthy Control; MDD = Major Depressive Disorder; No Future SA = Participant did not make a suicide attempt between baseline and follow-up; NSSI = Non-Suicidal Self-Injury; SA = Suicide Attempt

Method. GMV = Gray Matter Volume; ROI = Region of Interest; WMV = White Matter Volume

Region. ACC = Anterior Cingulate Cortex; AI = Adhesio Interthalamica; Amyg = Amygdala; BA = Brodmann Area; CSP = Cavum Septum Pellucidum; DRPFC = Dorsal Rostral Prefrontal Cortex; IFG = Inferior Frontal Gyrus; NAcc = Nucleus Accumbens; OFC = Orbitofrontal Cortex; PFC = Prefrontal Cortex; rSTG = Right Superior Temporal Gyrus; VMPFC = Ventral Medial Prefrontal Cortex; VPFC = Ventral Prefrontal Cortex

Outcome. SI = Suicidal Ideation

Table 3.

Resting state functional connectivity patterns related to suicidal thoughts and behaviors and non-suicidal self-injury

Publication	Sample	Age	Method	Select Findings
Suicidal Thoughts and Behaviors
Cao et al., 2015 (106)	SA = 19 HC = 20	*SA: 19.84 ± 1.61 *HC: 20.30 ± 1.72	Voxel-wise ReHo	2,4SA > HC ↑ MFG, Parietal, Precuneus ↓ Cerebellum, Fusiform, Hippocampus, Paraphippocampal, Angular, IFG, MFG
Cao et al., 2016a (107)	MDD+SA = 35 MDD = 18 HC = 47	15–25	Voxel-wise ALFF	2,4MDD+SA > MDD, HC ↑ MTG, STG, MOG
Cao et al., 2020a (103)	MDD+SA = 35 MDD = 18 HC = 47	15–25	ICA Networks	2,4MDD+SA > MDD ↓ aDMN-SN ↑ FPN-SN, FPN-aDMN; 2,4MDD+SA > HC ↓ aDMN-pDMN, aDMN-FPN ↑ FPN-SN
Cullen et al., 2014b (100)	MDD = 41 HC = 29	12–19	Seed-Based	2,4MDD-related AMYG seed did not relate to suicide severity
Ordaz et al., 2018c (104)	MDD = 40	14–17	ICA Networks	1,3↑ Lifetime ideation severity ↓ Coherence in ECN, aDMN, SN
Schreiner et al., 2019b (101)	MDD = 58	12–19	Seed-Based	1,3↑ Suicide severity ↑ Precuneus-Cerebellum, IFG; ↓ PCC-Cerebellum, Cingulate
Schwartz et al., 2019c (105)	MDD = 33	14–17	ICA Networks	1,36M Follow-Up: ↓ Ideation severity ↑ SN coherence
Zhang et al., 2016a (102)	MDD+SA = 35 MDD = 18 HC = 47	15–25	ICA Networks	1,3MDD+SA > MDD ↑ DMN-Cerebellum, Lingual ↓ DMN-Precuneus
Non-Suicidal Self-Injury
Cullen et al., 2020d (108)	§NSSI = 18	13–21	Seed-Based	1,5↓NSSI with NAC treatment = ↓ AMYG-SMA, ↑ AMYG-Inferior Frontal, ↓ NAcc-Superior Medial Frontal
Santamarina-Perez et al., 2019 (109)	§NSSI = 24 HC = 16	12–17	Seed-Based	1,3NSSI > HC ↓ AMYG-ACC, Insula; ↓mPFC-Precentral/Postcentral Gyri, Insula; ↓NSSI with treatment = ↓ AMYG-ACC, ↑ AMYG-Parahippocampal, ↓ mPFC-AMYG/Striatum,
Schreiner et al., 2017d,e (96)	§NSSI = 25 HC = 20	13–21	Seed-Based	1,5NSSI > HC ↓ AMYG-ACC/SMA, ↑ AMYG-Angular/Temporal Gyri

Note. Bold = Strong methodological approach (i.e., clinical-control comparison, interview assessments of STB/NSSI, data analysis, and/or longitudinal design)

^*Age range not reported

^aSame sample

^b,c,dOverlapping sample

^eAlso includes fMRI (Table 2)

¹Age effects not reported

²Age effects null

³Sex effects not reported

⁴Sex effects null

⁵Female only sample

^§Comorbid disorders see Table S3 for inclusion and exclusion criteria

Group. HC = Healthy Control; MDD = Major Depressive Disorder; NSSI = Non-Suicidal Self-Injury; SA = Suicide Attempt History

Method. ALFF = Amplitude of Low-Frequency Fluctuation; ReHo = Regional Homogeneity; ICA = Independent Component Analysis

Region. ACC = Anterior Cingulate Cortex; AMYG = Amygdala; IFG = Inferior Frontal Gyrus; MFG = Middle Frontal Gyrus; MOG = Middle Occipital Gyrus; mPFC = Medial Prefrontal Cortex; MTG = Middle Temporal Gyrus; NAcc = Nucleus Accumbens; PCC = Posterior Cingulate Cortex; SFG = Superior Frontal Gyrus; SMA = Supramarginal Gyrus; STG = Superior Temporal Gyrus

Network. aDMN = Anterior DMN; DMN = Default Mode Network; ECN = Executive Control Network; FPN = Fronto-Parietal Network; pDMN = Posterior DMN; SN = Salience Network

Medication. NAC = N-Acetylcysteine

Structural Alterations

STB and NSSI research probing structural differences employ a wide range of methodologies, including region-based and whole-brain (voxel- or vertex-wise) assessments of structural properties. Several studies rely on hand tracing, though the majority use automated methods (e.g., FreeSurfer) to quantify subcortical volumes as well as cortical thickness and surface area. Other studies use voxel-based morphometry (VBM) toolboxes (e.g., SPM) to quantify gray matter volume (GMV). Sample characteristics, structural method, and select findings (including age and sex effects) are summarized in Table 1.

Suicidal Thoughts and Behaviors

Structural alterations within the VPFC and orbitofrontal cortex (OFC), which are key to emotion inhibition, decision-making, and self-control (e.g., (51, 52)), are implicated. Among adolescents with MDD and BD (53, 54), suicide attempters show reduced VPFC and OFC GMV relative to MDD and BD non-attempters, respectively. There is evidence of reduced OFC thickness among BD suicide attempters, and reduced OFC volume negatively correlates with suicide ideation severity and attempt lethality (55). In a prospective study, reduced ventral and rostral prefrontal GMV at baseline differentiates adolescents and young adults with mood disorders who later attempt suicide (56). Studies on PFC differences, however, are complicated by inconsistent laterality effects (left (54, 55) versus right (53)). It also is unclear whether results reflect attempts specifically or the underlying mental disorders more broadly. MDD suicide attempters exhibit increased OFC GMV (left BA47) relative to MDD non-attempters as well as BD attempters and non-attempters (54). However, there is a negative correlation between ventromedial PFC thickness and impulsivity among MDD suicide attempters (57).

Outside of the PFC, BD suicide attempters have reduced hippocampal (53, 54) and bilateral cerebellum (53) GMV compared to BD non-attempters. Additionally, putamen volume is prospectively associated with implicit, but not explicit, suicide ideation (58). Research also implicates the cingulate cortex, as the number of suicide attempts in adolescents with comorbid MDD and BPD correlates with smaller averaged GMV and white matter volume (WMV) in the ventral anterior cingulate cortex (ACC), as well as greater WMV in the ventral posterior cingulate cortex (PCC) (59). Yet, it is not clear if this is driven by suicide attempts or overall mental disorder severity. There also is partial evidence linking temporal lobe structure to STB. MDD attempters have reduced right superior temporal gyrus (STG) GMV relative to healthy adolescents but not compared to MDD non-attempters (60, 61).

Non-Suicidal Self-Injury

Fewer studies examine structural alterations in youth exhibiting NSSI, and this work often: (a) assesses youth reporting parasuicidal behaviors—a term no longer commonly used, as it conflates suicidal and non-suicidal self-injury¹ and (b) compares NSSI youth to healthy controls (as opposed to psychiatric samples without NSSI).

There is limited work probing prefrontal differences related to NSSI. One study reports reduced GMV in the right IFG among NSSI adolescents compared to non-depressed individuals (62). Several studies examine the insula, which may modulate subjective emotional experience (63). In comparison to healthy controls, self-injurers show reduced insula GMV (62, 64), but this is not substantiated in BPD youth reporting parasuicidal behaviors (65). Other research found no relationship with the OFC (66), midline structures (67), or STG (68).

Several studies point to ACC alterations. Among adolescents with BPD, a greater number of lifetime parasuicidal behaviors in the past 6 months associates with smaller left ACC volume (69), consistent with findings comparing past-year self-injurers and healthy youth (64). Regarding this latter finding, most NSSI participants (55.2%) reported a lifetime suicide attempt, and given prior work linking anterior ACC volume to suicidal behaviors (59), we cannot determine whether effects are specific to NSSI.

Functional Neuroimaging

Across the STB and NSSI literature, investigators use a range of experimental paradigms (Table S4 for task descriptions). Several studies test task connectivity with psychophysiological interaction (PPI), quantifying condition-related differences in timeseries correlation between activity in a seed region-of-interest (ROI) and other brain regions (70, 71)). Sample characteristics, methodological approaches (e.g., whole brain, ROI), and select findings (including age and sex effects) are provided in Table 2.

Table 2.

Functional neural markers related to suicidal thoughts and behaviors and non-suicidal self-injury

Publication	Sample	Age	Method	Select Findings
Suicidal Thoughts and Behaviors
Alarcón et al., 2019a (75)	DEP+SA = 24 DEP+High SI = 27 DEP+Low SI = 31 HC = 38	11–18	PPI (Amyg seed) Emotional Self-Other Morph-Query Task  Contrast: Self vs. Other for Happy, Neutral, or Sad	1,3DEP+SA, DEP+High SI > DEP+Low SI ↑ PPI to DLPFC/dACC, DMPFC, Precuneus; DEP+High SI > DEP+SA, DEP+Low SI ↑ PPI to IPL; DEP+SA ↑ Left PPI to ACC; DEP+High SI ↑ Right PPI to ACC
Chase et al., 2019 (84)	MDD+SA = 19 MDD = 22 HC = 23	*MDD+SA = 15.6±1.5 *MDD = 16.0±1.4 *HC = 14.7±1.8	Whole Brain Dynamic Faces Task  Contrast: Face vs. Shape for Angry, Fearful, Happy and Sad	1,3Null findings between MDD+SA and MDD
Harms et al., 2019a (82)	3 Group Analyses: DEP+High SI = 45 DEP+Low SI = 42 HC = 39	11–18	Whole Brain Cyberball Task  Contrast: Exclusion or Inclusion vs. Practice	1,4DEP+High SI (including SA) > DEP+Low SI, HC ↓ Insula, Putamen across conditions; DEP+SA > DEP+High SI, DEP+Low SI ↑ ACC, SFG, MFG across conditions
	4 Group Analyses: DEP+SA = 26 DEP+High SI = 28 DEP+Low SI = 33 HC = 39
Johnston et al., 2017d (53)	BD+SA = 26 BD = 42 HC = 45	14–25	Bilateral Amyg seed connectivity Emotional Face Processing Paradigm  Contrast: Emotion vs. Fixation for Fearful, Happy, or Neutral	1,4BD+SA> BD ↓ PPI to Left Ventral PFC connectivity across neutral and happy conditions (fearful did not survive AlphaSim spatial extent thresholding) and ↓ PPI to Right Rostral PFC for neutral emotion; Among BD+SA ↑ SI ↓ Amyg-right rostral PFC and ↑ Attempt Lethality ↓ Amyg-left VPFC
Just et al., 2017 (86)	SI = 17 HC = 17	*SI = 22.9±3.6 *HC = 22.1±2.8	Whole Brain, Multi-Voxel Analysis Emotional Semantics Task Machine Learning	1,3Left Inferior Parietal Region and Left IFG predict SI group membership; Gaussian Naïve Bayes classifier correctly discriminated ideators from controls as well as attempters (subset of ideators) from ideators with high (85–94%) accuracy
Miller et al., 2018 (85)	SI = 14 No SI = 35	13–20	Whole Brain Emotion Regulation Task  Contrast: Reappraisal Negative vs. View Negative; View Negative vs. View Neutral	1,3SI > No-SI ↑ right DLPFC during cognitive reappraisal; SI > No-SI ↓ Right Thalamus, Left Cerebellum/Lateral Occipital Region, Right DLPFC, Right TPJ (only when controlling for depression symptoms) when viewing negative stimuli
Oppenheimer et al., 2020 (80)	ANX = 36	11–16	Whole Brain, ROI (AI and dACC) Chatroom Interact Task  Contrast: Rejection vs. Acceptance	2,4Peer victimization moderated Right AI activation and SI
Pan et al., 2011b (78)	MDD+SA = 15 MDD = 15 HC = 14	13–17	Whole Brain Go/No-Go Task	2,4MDD+SA > MDD ↓ Right Anterior Cingulate Gyrus during response inhibition blocks
Pan, Hassel, et al., 2013b (83)	MDD+SA = 14 MDD = 15 HC = 15	*MDD+SA = 16.2±0.8 *MDD = 15.9±1.6 *HC = 15.3±1.4	Whole Brain, PPI (Right Anterior Cingulate seed) Facial Emotion Processing Task  Contrast: Angry 50% Intensity vs. Fixation	2,4MDD+SA > MDD ↑ Attentional Control Circuitry (Right Anterior Cingulate Gyrus, Left DLPFC), Primary Sensory Cortex, Right MTG; MDD+SA > MDD, HC ↓ PPI to Bilateral Insula connectivity
Pan, Segreti, et al., 2013b (77)	MDD+SA = 15 MDD = 14 HC = 13	12–17	Whole Brain Iowa Gambling Task	2,4MDD > MDD+SA, HC ↑ Left Hippocampus during low-risk decisions (MDD > MDD+SA does not survive post-hoc correction for multiple comparisons)
Quevedo et al., 2016a (74)	DEP+HS = 43 DEP+LS = 39 HC = 37	11–18	Whole Brain Emotional Self-Other Morph-Query Task  Contrast: Self or Other vs. Baseline for Happy, Neutral, or Sad	1,4DEP+HS > DEP+LS, HC ↑ Bilateral Cuneus, MOG; DEP+HS, DEP+LS > HC ↑ Right DLPFC across all conditions
Non-Suicidal Self-Injury
Bonenberger et al., 2015 (99)	§NSSI = 14 HC = 16	*22.4±3.4	Whole Brain, ROI (Posterior, Medial, and Anterior Insula, Somatosensory Cortex) Electrical Stimulation Task	1,5NSSI > HC ↓ intensity-modulated activity in AI
Brown et al., 2017 (90)	MDD+BPD+NSSIAdult = 14 MDD+NSSIAdolescent = 13 HCAdolescent = 15 HCAdult = 17	*MDD+BPD+NSSI = 23.6±4.1 *MDD+NSSI = 15.5±2.0 *HCAdolescent = 14.5±1.7 *HCAdult = 23.2±4.4	Whole Brain Cyberball Task  Contrast: Exclusion vs. Inclusion	1,3MDD+BPD+NSSI > HCAdult and MDD+NSSI > HCAdolescent ↑ Ventral ACC; MDD+NSSI > MDD+BPD+NSSI, HCAdolescent ↑ Putamen
Demers et al., 2019c (97)	§NSSI = 25	13–21	Whole Brain Masked Emotional Face Task  Contrast: Masked Fearful vs. Masked Happy	1,5↑ Emotional awareness ↑ Right SMA and Right IFG activity
Groschwitz et al., 2016 (88)	MDD+NSSI = 14 MDD = 14 HC = 15	*15.2±1.8	Whole Brain, ROI (VLPFC, mPFC) Cyberball Task  Contrast: Exclusion vs. Inclusion	1,3MDD+NSSI > MDD ↑ mPFC, VLPFC, Parahippocampus
Osuch et al., 2014 (98)	§NSSI = 13 §PC = 15	16–24	Task Connectivity, Whole Brain Cold Stimulus Task  Contrast: Experimenter or Self, Instruction or Administration of Cool or Cold Stimuli vs. Fixation	1,3NSSI > PC ↑ Parahippocampal Gyrus, IFG, Amyg, Right Midbrain/Pons, Right Middle Frontal Gyrus across all conditions
Perini et al., 2019 (89)	§NSSI = 27 HC = 27	15–18	Whole Brain, Multi-Voxel Pattern Analysis Social Processing Task	1,5Whole Brain gray matter, multivariate analysis resulted in 68% accuracy for classification
Plener et al., 2012 (95)	§NSSI = 9 HC = 9	14–18	Whole Brain Emotional and Self-Injuring Image Task	1,5NSSI > HC ↑ OFC, Inferior Parietal Cortex, Inferior/Middle Frontal Cortex to NSSI stimuli
Poon et al., 2019 (93)	§NSSI-HR = 19 §NSSI-LR = 52	12–14	ROI (Caudate, Putamen, NAcc, vmPFC) Card Guessing Task  Contrast: Win vs. Neutral	2,4NSSI-HR > NSSI-LR ↑ Bilateral Putamen after reward receipt
Quevedo et al., 2016 (87)	DEP+NSSI = 50 DEP = 36 HC = 37	*14.8±1.6	Whole Brain Interpersonal Self-Processing Task  Contrast: Perspective vs. Baseline	1,3DEP+NSSI > DEP and HC ↑ Dorsal PFC, Precuneus, PCC, Superior Parietal Lobule, Left/Right Limbic Structures, Fusiform, MTG across all conditions; DEP+NSSI > DEP, HC ↑ Amyg, Hippocampus, Parahippocampus, Fusiform during mother’s perspective
Sauder et al., 2016 (92)	§Self-Injury = 19 HC = 19	13–19	ROI (Striatum, OFC, Amyg) Monetary Incentive Delay Task Contrast: Reward Cue vs. Neutral Cue	1,5Self-injury > HC ↓ Putamen, OFC, Amyg to reward cues
Schreiner et al., 2017c,e (96)	§NSSI = 24 HC = 17	13–21	Task Connectivity, PPI (Amy seed) Emotional Face Matching Task	1,5NSSI > HC ↑ Right Amyg connectivity clusters: (a) Right Lingual Gyrus, Occipital Pole, Occipital/Temporal Fusiform, (b) Right Lateral Occipital Cortex, Superior Parietal Lobule during entire task

Note. Bold = Strong methodological approach (i.e., clinical-control comparison, interview assessments of STB/NSSI, data analysis, large sample)

^*Age range not reported;.

^a,b,cOverlapping sample

^dAlso includes structural MRI (Table 1)

^eAlso includes resting state MRI (Table 3)

^§Comorbid disorders see Table S2 for inclusion and exclusion criteria

¹Age effects not reported

²Age effects null

³Sex effects not reported

⁴Sex effects null

⁵Female only sample Group.

ANX = Anxiety Disorders; BP = Bipolar Disorder; BPD = Borderline Personality Disorder; DEP = Depressive disorder including MDD, Dysthymia, or Depressive Disorder NOS; HC = Healthy Control; HR = High Risk; HS = High Suicidality, LS = Low Suicidality; MDD = Major Depressive Disorder; NSSI = Non-Suicidal Self-Injury; NSSI-HR = High risk for NSSI based on thoughts; NSSI-LR = Low risk for NSSI based on absence of thoughts; PC = Psychiatric Controls with no NSSI history; SA = Suicide Attempt History; Self-Injury = Self-inflicted injury including SA; SI = Suicide Ideation

Method. PPI = Psychophysiological Interaction; ROI = Region of Interest

Region. ACC = Anterior Cingulate Cortex; AI = Anterior Insula; Amyg = Amygdala; dACC = Dorsal Anterior Cingulate Cortex; DMPFC = Dorsomedial Prefrontal Cortex; DLPFC = Dorsolateral Prefrontal Cortex; IFG = Inferior Frontal Gyrus; IPL = Inferior Parietal Lobule; MFG = Middle Fontal Gyrus; MOG = Middle Occipital Gyrus; mPFC = Medial Prefrontal Cortex; MTG = Middle Temporal Gyrus; NAcc = Nucleus Accumbens; OFC = Orbitofrontal Cortex; PCC = Posterior Cingulate Cortex; PFC = Prefrontal Cortex; SFG = Superior Frontal Gyrus; SMA = Supramarginal Gyrus; TPJ = Temporoparietal Junction; VLPFC = Ventrolateral Prefrontal Cortex; VPFC = Ventral Prefrontal Cortex

Suicidal Thoughts and Behaviors

Self-Processing

Distorted self-referential thinking is a hallmark of adolescent MDD (72), and these biases potentiate suicidal thinking (73). Two recent studies focus on processing images of one’s own versus others’ faces—a subdomain of self-referential processing. Quevedo and colleagues compare adolescents reporting low and high suicidality—composite scores of ideation and attempts. High suicidality youth exhibit greater activation in bilateral cuneus and middle occipital gyrus compared to low suicidality and healthy participants across all conditions. Additionally, both suicidality groups show greater activation in the right DLPFC (74). Within the same sample, high ideators and attempters also show greater task-connectivity between the amygdala and the DLPFC, dACC, dorsomedial PFC, and precuneus compared to low ideators (75). Amygdala-seed PPI connectivity is elevated in the high ideators and attempters relative to the low ideation group in the DLPFC, dorsomedial PFC, and precuneus, and by contrast, connectivity with inferior parietal lobule is elevated in the high ideation group compared to low ideation and attempters. Greater connectivity with the rostral ACC during self versus other face trials also is found for the left amygdala in attempters (compared to all groups) but right amygdala in high ideators (relative to low ideators and controls) (75).

Impulsivity and Social Reward Processes

Impulsive decision-making and experiencing social rejection are known correlates of suicidal behaviors (21, 76). Although non-attempters show greater left hippocampal activation during low-risk (but not high-risk) decisions relative to healthy adolescents and suicide attempters during an Iowa Gambling Task (77), it does not survive post-hoc correction for multiple comparisons. Healthy and attempter youth do not differ. The investigators also administer a response inhibition task, and non-attempters exhibit greater activation within the right ACC relative to attempters but not healthy controls (78).

One study uses the Chatroom Interact Task (79) to probe neural response following social acceptance and rejection. Experience of peer victimization moderates the relationship between right anterior insula activation (rejection versus acceptance) and suicide ideation (80). A study using the Cyberball Task (81) finds that, compared to low ideators and controls, high ideators (including attempters) show reduced activation in the insula and putamen across conditions, but subsequent analyses indicate a pattern of increased activation in the ACC, superior frontal gyrus, and middle frontal gyrus among attempters versus ideators (82).

Emotion Processing

Pan and colleagues test response to face stimuli morphed to different levels of emotional intensity. The most consistent findings emerge during the 50% angry face condition; relative to non-attempters, attempters show greater activation in attentional control circuitry (right ACC, left DLPFC), primary sensory cortex, and right middle temporal gyrus (83). Group differences in PPI show reduced connectivity between a right ACC seed and the bilateral insula when viewing 50% angry faces versus baseline for depressed attempters relative to depressed non-attempters and healthy controls. Recent work from the same research group, however, fails to replicate these effects (84).

During a gender-judgement emotional face task, BD attempters exhibit reduced bilateral amygdala connectivity to the left VPFC (across happy and neutral conditions) and to the right rostral PFC for neutral faces versus baseline. Within the attempter group, suicidal ideation negatively correlates with amygdala-right rostral PFC connectivity, and attempt lethality negatively correlates with amygdala-left VPFC connectivity (53). Among adolescent ideators, there is evidence of greater right DLPFC activation during effortful regulation when viewing negative images compared to non-ideators. There is, however, reduced activation among ideators during passive viewing of negative stimuli within the right thalamus, left cerebellum/lateral occipital region, right DLPFC, and the right temporal parietal junction (85).

One study utilizes machine learning to establish a biological, neurocognitive basis for altered representations of suicide (e.g., death) and emotion (e.g., gloom) words. During the presentation of each word, participants are asked to actively think about the construct. Findings show distinct neural signatures in the left inferior parietal region and left IFG that are predictive of suicide ideator group membership (86). A similar classification approach also discriminates suicide ideators and attempters. A critical next step is to determine whether these neural signatures can identify ideators prior to the attempt.

Non-Suicidal Self-Injury

Self-Processing

Quevedo and colleagues’ task requires participants to use their own (direct) or others’ (indirect: mother, best friend, classmate) perspective when considering self-characteristics (87). Relative to a combined sample of depressed and healthy adolescents, depressed self-injurers exhibit greater activation across all perspectives in the dorsal PFC, precuneus, PCC, superior parietal lobule, left/right temporal limbic structures (amygdala, parahippocampus, hippocampus), fusiform, and middle temporal gyrus. The mother perspective condition (compared to baseline) yields similar findings among self-injurers in the amygdala, hippocampus, parahippocampus, and fusiform.

Social (Reward) Processing

Several NSSI studies probe neural response to social exclusion during the Cyberball Task. Relative to depressed adolescents, depressed self-injurers show increased activation in the medial PFC, ventrolateral PFC, and parahippocampus during exclusion versus inclusion trials; suggesting that self-injurers may view social exclusion more negatively (88) (c.f., (89)). In a subsequent study among the self-injuring youth, there is enhanced activation in the left putamen within the same contrast relative to BPD adults and healthy youth (90).

Research also explores alterations in incentive processing given known associations in adolescent MDD (91). Compared to healthy adolescents, there is blunted reward anticipation in the putamen, OFC, and amygdala in self-injuring youth (92). Among youth at high risk for NSSI (based on thoughts) but without any history of self-injurious behaviors, there is greater bilateral putamen activation following reward receipt compared to youth at low-risk for NSSI, suggesting hypersensitivity to reward (93). Although alterations in anticipatory and consummatory reward processing may underlie risk for NSSI, divergent associations between mental disorders and reward responsiveness (e.g., MDD (91) vs. ADHD (94)) may influence this effect.

Emotion Processing

Researchers use a variety of emotion processing tasks to investigate neural alterations in NSSI. Plener and colleagues show self-injurers and healthy youth IAPS and self-injury images (e.g., razors). Compared to healthy youth, self-injurers exhibit greater activation in OFC, inferior parietal cortex, and inferior/middle frontal cortex to NSSI stimuli (95). During an emotion face-matching task, there is no evidence of emotion-specific effects. However, NSSI and healthy youth show divergent amygdala task-connectivity, but it is not clear whether this reflects NSSI or MDD (or other comorbid disorders) (96). Finally, Demers and colleagues find associations in self-injurers’ emotional awareness and response to viewing masked fearful faces (relative to happy faces) in the supramarginal gyrus and right IFG (97).

Pain Processing

When comparing psychiatric controls to self-injurers, a painful cold compress is either self-administered (analogue to self-injury) or experimenter-administered. Compared to psychiatric controls, self-injurers show greater activation in the parahippocampal gyrus, IFG, and amygdala across conditions (98). Bonenberger and colleagues also probe pain via unpleasant electrical stimulation. Healthy controls exhibit anterior insula activation varying with increasing electrical stimuli intensity, but among self-injurers, this association is blunted (99).

Resting State MRI

The resting state MRI literature on adolescent STB and NSSI is smaller relative to structural and task-based fMRI, and interpretations are challenging given mixed methods and results (Table 3). Investigators primarily apply seed-based approaches, which focus on timeseries correlations between an ROI and the rest of the brain voxel-wise. Several studies apply data-driven approaches (e.g., independent components analysis [ICA]) to explore commonly assessed networks (e.g., default mode network [DMN]). Two studies use voxel-based approaches to characterize local connectivity or timeseries variability.

Suicidal Thoughts and Behaviors

Seed-based

Cullen and colleagues identify depression-related differences in amygdala seed-based resting state functional connectivity (RSFC), but this is not related to suicide severity (100). However, using an expanded sample of depressed youth, within-group analyses show that suicide ideation severity associates with greater connectivity between a right precuneus seed and cerebellum and right IFG, as well as lower connectivity between a left PCC seed and cerebellum and cingulate (101).

Network-based

Compared to psychiatric controls, attempters show increased connectivity between an ICA-derived DMN component and the cerebellum and lingual gyrus, and decreased connectivity with precuneus. Across ICA networks, however, there is reduced positive RSFC between the anterior portion of the DMN and the salience network (SN), less negative anterior DMN-right fronto-parietal network connectivity, as well as increased right FPN-SN and DMN-cerebellum connectivity in patients with versus without an attempt history (102, 103). Among depressed adolescents, lower average connectivity/coherence in the left fronto-parietal executive control network (ECN), anterior DMN, and SN associates with greater lifetime ideation severity (104). Within this same sample, increasing SN coherence predicts lower ideation severity over time (105).

Voxel-wise

Young adults with a lifetime attempt history but no current mental disorders show greater regional homogeneity (ReHo; measuring local synchrony) in the middle frontal gyrus (MFG), parietal lobe, and precuneus, and reduced ReHo in the cerebellum, fusiform, hippocampus, parahippocampal gyrus, angular gyrus, IFG, and MFG compared to healthy controls (106). Cao and colleagues reanalyzed data (102, 103) to examine fractional amplitude of low-frequency fluctuation (ALFF), which characterizes voxel-wise intensity of spontaneous BOLD signal fluctuations across a low-frequency band. Depressed patients with an attempt history show increased ALFF in the medial temporal, superior temporal, and middle occipital gyri compared to patients and healthy controls (107).

Non-Suicidal Self-Injury

Relative to healthy females, adolescents with an NSSI history exhibit reduced left amygdala-seed RSFC with a cluster spanning ACC and supramarginal gyrus (SMA), as well as increased RSFC with angular gyrus and a temporal gyrus cluster (96). These youth underwent an 8-week open-label N-acetylcysteine trial. Reductions in NSSI frequency after treatment associate with decreases in left amygdala-right SMA and right nucleus accumbens-left superior medial frontal cortex RSFC, and increases in right amygdala-right inferior frontal cortex RSFC (108). Finally, relative to healthy adolescents, youth with an NSSI history show reduced amygdala-seed connectivity with the ACC and a cluster spanning right planum temporale and right insula, as well as reduced medial PFC connectivity with the precentral/postcentral gyri and left insula. Reductions in NSSI frequency post-treatment associate with greater negative amygdala-ACC, greater amygdala-brainstem/parahippocampal gyrus, reduced positive medial PFC-amygdala/striatum, and more negative medial PFC-intracalcarine cortex RSFC. Medial PFC-amygdala connectivity also correlates with change in suicidal ideation but not suicidal behavior following treatment (109). Although limited, findings suggest that amygdala-seed connectivity is altered in adolescents with an NSSI history, particularly with the ACC (96, 109).

Discussion

Neural alterations in youth with lifetime STB and NSSI history show modest convergence. First, reduced VPFC and OFC volume associates with STB (53–55), and there is reduced volume in the ACC for youth reporting STB (59) and NSSI (64, 69). Second, blunted striatal activation characterizes STB (82) and NSSI (92), and there is reduced frontolimbic connectivity in suicide ideators and attempters (75). Last, RSFC shows reduced positive connectivity between the DMN and SN in attempters (102, 103), and for NSSI behaviors, growing evidence implicates frontolimbic alterations (96). As a whole, self-injurious behaviors are related to top-down and bottom-up neural alterations.

The overlap in neural correlates of suicidal and non-suicidal behaviors is not surprising given the high co-occurrence (1). However, the inability to elucidate specific neural circuitry may reflect: (i) sample overlap (i.e., NSSI youth in STB studies; STB youth in NSSI studies) and (ii) inconsistent reporting of both STB and NSSI histories (Table S1–S3 for overlap, reporting). Nonetheless, this review highlights tentative differences—focusing on research applying strong methodological approaches (Tables 1–3 bolded publications)—that may prove fruitful to explore. Consistent with Schmaal and colleagues (32), youth STB research is characterized by alterations within prefrontal (53–55) and limbic (62, 75, 102, 103) systems whereby a compromised regulatory system may potentiate suicidal thinking, and perturbations in top-down and bottom-up connections may lead to suicidal behaviors. A key difference is that disruptions across regulatory, approach, and avoidance systems may be even more pronounced in youth given ongoing PFC neuromaturation through young adulthood (30, 31). By contrast, NSSI research reveals ACC structural alterations (64, 69) as well as disruptions in amygdala and ACC connectivity (96), which contributes to hyperarousal and reduced impulse control—central features of NSSI (110, 111). Whether this neural pathway is unique to NSSI remains unclear, as STB research also shows ACC (59) and limbic (75) alterations, and NSSI research has not clarified which ACC (69) and amygdala (96) subregions are more closely linked to NSSI.

Neurodevelopmental Model of STB and NSSI

This review highlights promising neural diatheses related to STB and NSSI. These diatheses are shaped by distal factors occurring during a critical period of neuromaturation (30, 31). Specific genes (e.g., 5-HTTLPR; (112, 113)) influencing neuromaturation have known associations with self-injurious behaviors (114, 115). Trauma and maltreatment occurring during developmentally sensitive periods impact cortical and subcortical development (116), directly affecting top-down (e.g., ACC) and bottom-up (e.g., striatum) processes implicated in STB and NSSI (117). This may be compounded by environment (e.g., poverty (118)), parent (e.g., mental disorders (119)), and youth (e.g., stress (120); psychiatric symptoms (121)) factors—each affecting neuromaturation—and potentially, increasing STB and NSSI vulnerability.

In the context of proximal antecedents, these neural alterations may confer increased STB and NSSI risk (Figure 2). For example, the timing of puberty is important, as early onset for girls versus late onset for boys may increase risk for internalizing disorders (122). Earlier mental disorder onset shapes brain development (e.g., (123)) and relates to self-injurious behaviors (1). Additionally, adolescent interpersonal stress (e.g., loss (21, 25), peer victimization (22, 24))—may initially act on cortical regions by compromising cognitive flexibility and future oriented thinking (124), which over time, may potentiate suicidal and non-suicidal thinking (125, 126). Repeated and chronic stress may tax limbic systems that modulate arousal and approach behaviors, disrupt bottom-up and top-down connections, and potentially, facilitate the transition from ideation to behavior (32, 127).

Figure 2.

The neurodevelopmental model of STB and NSSI highlights distal and proximal processes that may potentiate risk for self-injurious behaviors during a critical period of development. There are a wide range of (A) distal risk factors that shape neuromaturation, including genetic, environment, parental, and youth factors. These distal factors occur in the context of ongoing pubertal and neuromaturation from childhood to young adulthood. This review highlights a number of brain regions and connectivity patterns (B) that show alterations potentially unique to suicidal thoughts and behaviors (STB; Teal) and non-suicidal self-injury (NSSI; Purple), or common across both (Yellow). In STB, these include connections between the default mode network (hub in the posterior cingulate cortex [PCC]) and salience network (hub in the anterior insula [AI]) as well as between the ventrolateral prefrontal cortex (VLPFC) and the amygdala (amyg). NSSI shows alterations in connections between the anterior cingulate cortex (ACC) and amygdala. Both STB and NSSI implicate structural and functional alterations in the striatum, amygdala, and ACC. (C) Coupled with acute stressors—particularly interpersonal stress—these distal neural markers may increase risk for engaging in suicidal and non-suicidal behaviors. Acute stress may directly impact brain development, and concurrently, may disrupt top-down cortical processes related to self-referential processing rumination, and future-oriented thinking, which may lead to suicidal and non-suicidal thinking. Stressors that become chronic in nature may tax limbic systems, which modulate arousal and approach behaviors. Disruptions to bottom-up and top-down connections may, for some, facilitate the transition from thinking to acting. More broadly, stress also elicits a range of negative emotions (e.g., sadness, anger), and in the absence of effective emotion regulation strategies, this may then lead to STB and/or NSSI. Presently, there is not sufficient evidence attributing sex differences in STB and NSSI to discrete neural circuitry. Epidemiological research, however, shows that NSSI thoughts and behaviors are more common in female (red) versus male (blue) adolescents, though estimates vary (e.g., (6, 7)). Similarly, suicidal thinking (15% vs. 9%) and behaviors (6% vs. 2%) are more common in females versus males (e.g., (4)). Although not often examined in large epidemiological cohorts, NSSI and STB is proportionally higher among transgender and gender-non-conforming youth (green) (140, 141). Accounting for stress exposure using interview and ambulatory approaches may shed key insights into shared and unique neural markers that lead to suicidal versus non-suicidal behaviors.

Although sex differences in STB and NSSI are well documented (4, 5, 7), there is insufficient evidence showing discrete neural circuitry underlying these effects. One possibility is that neural diatheses do not differ, but rather, stress exposure may differentially trigger underlying diatheses. For example, MDD females report a higher frequency of interpersonal stress (128) whereas males experience more achievement-related stressors (129). The neurobiology of MDD may be similar (91), but the antecedent trigger may differ. Research incorporating stress-based interviews (21) and ambulatory approaches (e.g., experience sampling (130), passive sensor data (131)) may clarify how the type, intensity, and timing of stress exposure impact self-injurious behaviors.

Recommendations for Future Research

Future research addressing key methodological limitations may be better positioned to elucidate unique neural pathways for suicidal and non-suicidal behaviors. Herein, we provide five core recommendations. First, reporting lifetime mental disorders and associated comorbidity will clarify whether effects can be attributed to the targeted behaviors or the underlying disorders. Interview-based assessments of mental disorders, STB, and NSSI are preferable. Self-reports are prone to misattributions and retrospective biases (132), and interviews ensure shared definitions of what constitutes a suicide attempt or an NSSI episode. Given the high co-occurrence of STB and NSSI (1), consistently reporting suicidal and non-suicidal behaviors is essential.

Second, comparing STB or NSSI samples with healthy individuals makes it challenging to attribute neural alterations to self-injurious behaviors versus mental disorders. Comparing suicide attempters versus ideators or adolescents reporting NSSI behaviors versus psychiatric controls may clarify whether there are neural mechanisms implicated in thinking versus engaging in self-injurious behaviors (133).

Third, small samples are ill-equipped to explore age, sex, and gender effects (2, 3; Table S1–S3 for sample sex characteristics). Larger, diverse samples are key given the inherent heterogeneity (e.g., frequency, method, underlying disorders, medication use) and is consistent with initiatives to replicate brain-based effects in representative youth samples (134).

Fourth, longitudinal research is scant. A substantial number of participants characterized by suicidal and non-suicidal thoughts will engage in self-injurious behaviors over time. Longitudinal research—particularly studies incorporating fine-grained ambulatory stress assessments (38)—will have an opportunity to identify distal neural markers that may facilitate the transition from thinking to acting in the context of proximal stressors. This methodological approach can clarify why certain individuals who ideate about suicide or initiate non-suicidal behaviors transition to suicidal behaviors (131).

Fifth, additional neuroimaging modalities—including diffusion tensor imaging (56, 135) and electrophysiology (136, 137)—may provide complementary information. There also may be innovative ways to integrate neuroimaging modalities (e.g., (138, 139) and apply novel data analytic strategies (86).

Summary

Initial research probing neural correlates of STB and NSSI has identified promising brain-based markers. From a neurodevelopmental perspective, an important next step is to elucidate pathways differentially leading to suicidal versus non-suicidal behaviors by incorporating multimodal neuroimaging approaches in the context of fine-grained stress assessments while accounting for wide ranging distal risk factors that influence neuromaturation. Longitudinal research following youth through this developmental period may reveal neural circuitry that potentiates risk for self-injurious behaviors and provide novel targets for treatment.