Gene-brain-behavior links revealing generative neurophysiological subtypes of suicide-related dysconnectivity in bipolar depression

Suicidal thoughts and behaviors (STBs) in bipolar depression are heterogeneous, but biologically coherent subtypes remain poorly defined. We aimed to identify neurophysiological subtypes of suicide-related dysconnectivity in bipolar depression and examine their gene-brain-behavior correlates, reproducibility and longitudinal clinical relevance. We analyzed multimodal data from 802 individuals, including cross-sectional resting-state functional MRI from 657 patients with bipolar depression (405 with current-episode STBs and 252 without STBs) and 103 healthy controls across one discovery and two independent replication cohorts, together with longitudinal follow-up in 42 patients. A semi-supervised clustering-generative adversarial network (Smile-GAN) was applied to identify imaging-defined subtypes referenced to patients without STBs. Subtypes were characterized using symptoms, cognition, transcriptomic and neurotransmitter maps, targeted sequencing of 61 suicide-related single nucleotide polymorphisms, and exploratory early treatment response to pharmacotherapy and repetitive transcranial magnetic stimulation (rTMS). Two reproducible neurophysiological subtypes were identified. A visual cortex-predominant subtype showed hyperconnectivity associated with greater anxiety and poorer cognitive flexibility and working memory. In this subtype, a 3-single nucleotide polymorphism serotonergic genetic risk score derived from rs1631327/rs6320 ( HTR5A ) and rs8066602 ( SLC6A4 ) was associated with current-episode STBs, and its effect was partially mediated by right peripheral visual dysconnectivity (portion of mediated = 11.4%, β = 1.94e-2, p <.05, 95% CI=[1.08e-03, 0.05]). A default mode network-central executive network (DMN-CEN)-predominant subtype was characterized by hyperconnectivity linked to brooding rumination. In exploratory analyses, among rs8066602 TC/TT carriers within the DMN-CEN subtype, patients with STBs showed greater 2-week symptom improvement following pharmacotherapy and rTMS than those without STBs ( β = 31.01, p <.01, 95% CI=[16.94, 45.08]). Subtype-specific dysconnectivity patterns replicated across independent cohorts (replication-1: r = .80/0.64; replication-2: r = .65/0.73, adjusted p spin <0.001) and covaried longitudinally with suicide-risk fluctuations. Our findings suggest that suicide-related dysconnectivity in bipolar depression may be organized into two reproducible neurophysiological subtypes with distinct gene-brain-behavior profiles. These findings support subtype-informed stratification of suicide risk, although the genetic and treatment-related signals remain preliminary and require prospective replication.