Ventromedial prefrontal value signals and functional connectivity during decision-making in suicidal behavior and impulsivity

The study addresses whether impaired value-based decision-making and impulsivity in suicidal behavior share common neural mechanisms. Prior work links suicidal behavior to impaired lab-based decision-making and self-reported impulsivity, and shows disrupted vmPFC value encoding that scales with trait impulsivity. Impulsivity relates to reduced frontoparietal cognitive control and decreased connectivity with vmPFC during decision-making in the general population. Two non-mutually exclusive mechanisms are proposed for suicidal behavior: (1) disrupted vmPFC value representations and (2) altered vmPFC-frontoparietal connectivity associated with impulsivity. The study aims to determine whether these mechanisms are shared with general impulsivity or distinct in suicidal behavior, and to contrast vmPFC findings with striatal connectivity. It focuses on older adults, where suicidal behavior is more representative of suicide deaths and spans a wider impulsivity range. The authors examine neural encoding of value and its modulation by trait impulsivity (negative urgency) during a volatile reinforcement learning task, comparing suicide attempters, ideators, depressed controls, and nonpsychiatric controls.

Background work shows: (a) impaired value-based decision-making and computationally defined learning deficits in suicidal behavior, including difficulty choosing between similarly valued options and reduced responsiveness to recent reinforcement; (b) impulsivity is associated with suicidal behavior but does not robustly distinguish ideators from attempters based on self-report alone; (c) in healthy populations, higher impulsivity correlates with weaker connectivity among frontoparietal regions and reduced coupling between these control regions and vmPFC during value-based choice; (d) impulsivity has also been linked to altered striatal functioning and connectivity in various psychopathologies. Prior neuroimaging suggests ventral prefrontal abnormalities in suicide, though findings are mixed. These literatures motivate testing whether vmPFC value signals and vmPFC-frontoparietal connectivity—and their modulation by negative urgency—specifically characterize suicidal behavior beyond ideation and depression.

Design: Cross-sectional fMRI study during a volatile reinforcement learning (three-armed bandit) task with model-based fMRI and psychophysiological interaction (PPI) connectivity analyses, plus behavioral modeling. Participants: N=116 middle-aged and older adults (47–79 years): 35 suicide attempters (depression with past attempt(s) and recent attempt or strong current ideation), 25 suicide ideators (depressed, active ideation with specific plan, no history of attempt), 25 depressed non-suicidal controls (no lifetime self-injurious behavior, ideation, or attempts), and 31 nonpsychiatric controls (no lifetime psychiatric disorders). Recruitment sites included inpatient, outpatient, primary care, and community sources in Pittsburgh, PA. Informed consent obtained; IRB-approved. Inclusion/exclusion and verification of attempt status detailed in Supplementary Materials. Two participants (one depressed control, one ideator) excluded for non-engagement (>10 identical button presses in a row). Measures: Psychiatric diagnoses via SCID; depression severity via HRSD-17; impulsivity via UPPS (negative urgency, positive urgency, lack of premeditation, lack of perseverance). Additional cognitive measures (Executive Interview, Dementia Rating Scale, WTAR IQ) and clinical variables collected for sensitivity analyses. Task: 300-trial three-armed bandit during fMRI. Reward probabilities for three stimuli drifted independently across trials, requiring continuous learning. Reward magnitude (10, 25, or 50 cents) was manipulated independently and shown at trial start. Trial structure included stimulus/magnitude display until choice, feedback after jittered ISI (mean ~4000 ms), feedback display 750 ms, ITI mean ~2920 ms. Behavioral modeling: Choices fit with a reinforcement learning model (per Dombrovski et al., 2018). For each trial, maximum available value (best possible outcome) was computed using best-fitting subject parameters. Prior behavioral results in this sample showed reduced responsiveness to reinforcement and difficulty discriminating close-value options among attempters. Imaging acquisition and preprocessing: Details in Supplementary Materials. First-level GLM modeled expected value regressors. ROI definition: Meta-analytic ROI for value-related vmPFC (Bartra et al., 2013) used to extract beta values for expected value. Comparisons also conducted for left and right striatum ROIs from the same meta-analysis. Group comparisons: vmPFC value beta compared across groups (attempters vs ideators, depressed controls, nonpsychiatric controls). Whole-brain analyses controlled FWE via nonparametric cluster correction (p<0.05 corrected; cluster-forming threshold p<0.001) using AFNI 3dttest++ with -ClustSim. Connectivity (gPPI): Seed = vmPFC meta-analytic ROI. PPI modeled interaction between deconvolved vmPFC time course and feedback timepoint to identify voxels with greater coupling to vmPFC during feedback (value updating) versus other times. In nonpsychiatric controls, UPPS Negative Urgency added as subject-level regressor to test impulsivity moderation of connectivity. Regions showing impulsivity-related modulation in controls (threshold p<0.001 uncorrected) formed an independent mask. This mask was used to test, in the three patient groups, whether vmPFC-frontoparietal connectivity modulation by negative urgency varied by group (Group × Negative Urgency interaction). PCA on 13 clusters from this mask supported a single-component summary (69% variance); subsequent analyses used average beta across clusters. Parallel PPI analyses used left and right striatum seeds to test striatal value connectivity and its moderation by impulsivity. Brain–behavior analysis: Extracted individual average beta values from the control-derived impulsivity-modulated connectivity mask. Hierarchical linear mixed-effects regression (lme4) predicted the trial-wise expected value of participants’ actual choices (from the RL model) from the interaction of connectivity beta, group status, and previous-trial reinforcement (rewarded vs not). Model comparisons tested whether including the three-way interaction improved fit; follow-up contrasts focused on attempters vs depressed controls. To avoid circularity, these analyses were limited to patient groups (attempters, ideators, depressed controls). Sensitivity analyses excluded participants with possible brain damage, substance use, or antipsychotic treatment.

- vmPFC value signals: Suicide attempters showed reduced activation to expected value in vmPFC compared to nonpsychiatric controls (t64=2.24, p<0.05); trend vs ideators (t58=1.76, p<0.1); no difference vs depressed controls (t58=0.35, p>0.1). Overall group effect F3,112=2.22, p=0.08. Within attempters, vmPFC value signal did not vary by attempt lethality, age of onset, or impulsivity (all p>0.1) and was robust to excluding potential confounds. - vmPFC connectivity in controls: During value updating, vmPFC showed strong connectivity with limbic and frontoparietal networks (corrected p<0.05). Importantly, frontoparietal connectivity (inferior frontal gyrus; superior and inferior parietal lobules) was negatively moderated by impulsivity (UPPS Negative Urgency) in nonpsychiatric controls (PFWE<0.05). Limbic/medial cortical connectivity (striatal, hippocampal, PCC/precuneus) was not moderated by impulsivity. IQ and executive function did not moderate vmPFC connectivity. Striatal seeds also showed strong connectivity during value updating but without impulsivity moderation. - Group differences in impulsivity–connectivity relationship: Using the control-derived frontoparietal mask, the relationship between negative urgency and vmPFC-frontoparietal connectivity differed by group (ANOVA interaction F2,73=4.69, p=0.01). Suicide attempters showed a significantly reduced (abolished) modulation by impulsivity relative to depressed controls (t≈3.06, p<0.005). No analogous group moderation effects were found for striatal seeds (right striatum F2,73=0.628, p>0.5; left striatum F2,73=0.906, p>0.4). - Brain–behavior coupling: Allowing group to modulate the interaction between previous reinforcement and vmPFC-frontoparietal connectivity on subsequent choice value improved model fit (χ2=55.95, p<0.001). A three-way interaction (connectivity × group × previous reinforcement) was significant (χ2=37.68, p<0.001); attempters vs depressed controls: t=6.01, p<0.001. In depressed controls and ideators, higher vmPFC-frontoparietal connectivity predicted higher-value choices after both rewarded and non-rewarded trials. In attempters, higher connectivity predicted lower-value choices following rewards. Effects were robust to excluding participants with potential confounds. Findings were specific to vmPFC (not striatum).

The study shows that suicidal behavior is associated with disrupted value processing and connectivity in vmPFC during learning and decision-making. Consistent with prior work, attempters exhibited blunted vmPFC expected value signals relative to healthy controls, suggesting impaired representation or updating of learned value. In non-suicidal individuals, trait impulsivity (negative urgency) selectively weakened vmPFC coupling with frontoparietal cognitive control regions during value updating, while sparing limbic connectivity—consistent with reduced cognitive control over valuation with higher impulsivity. Critically, this normative impulsivity–connectivity relationship was absent in suicide attempters. Behaviorally, whereas stronger vmPFC–frontoparietal connectivity supported better, higher-value choices in comparison groups, in attempters it was associated with poorer choices following recent rewards, indicating abnormal use of reinforcement information despite intact connectivity magnitude. Collectively, results point to a neurobiologically distinct pattern in impulsive individuals with suicidal behavior: dysfunctional vmPFC–frontoparietal communication during value updating that fails to confer adaptive control over choices. This offers a potential mechanism by which mood-dependent impulsivity and valuation deficits could jointly increase suicide risk, particularly in volatile, high-stakes contexts resembling suicidal crises.

This work advances the understanding of neural mechanisms linking impulsivity and impaired decision-making to suicidal behavior. It demonstrates reduced vmPFC value signals in suicide attempters and an abnormal pattern of vmPFC–frontoparietal connectivity: in non-suicidal individuals, higher negative urgency weakens this connectivity, whereas in attempters this modulation is abolished and relates to maladaptive choices after rewards. The effects are specific to vmPFC rather than striatum. These findings suggest a neurobiologically distinct subtype of suicidal behavior characterized by aberrant value updating and dysfunctional coupling with cognitive control networks. Future research should employ longitudinal designs, larger samples to resolve differences among patient groups, and probe additional circuits (e.g., basal ganglia, amygdala) and contextual influences to refine biomarkers and inform targeted interventions.

- Cross-sectional design precludes causal inference about neural mechanisms and suicidal behavior. - Imaging protocol precluded inclusion of older and more cognitively impaired individuals, limiting generalizability to the full spectrum of late-life depression and suicidality. - Sample size may have limited power to detect differences between attempters and patient comparison groups in vmPFC value signals. - Task characteristics (volatile, continuously drifting contingencies without clear reversals) may attenuate some value-related signals observed in simpler tasks, reflecting a trade-off in sensitivity to individual differences in valuation versus cognitive control demands. - Potential unmeasured moderators may influence vmPFC value representation and connectivity.