Detecting suicide risk in bipolar disorder from lymphoblastoid signatures (www.nature.com)

Researchers used RNA sequencing of lymphoblastoid cell lines (LCLs) from 20 bipolar disorder (BD) patients to train machine‑learning models that distinguish high vs. low suicide risk. Thirteen patients (6 who later died by suicide, 7 prospectively monitored non‑suicide) were used for training/validation and seven independent patients for testing. Differential expression analysis highlighted genes and pathways unexpectedly linked to brain phenotypes despite peripheral sampling: primary immunodeficiency, ion‑channel signaling and cardiovascular defects. Key genes identified as potential biomarkers include LCK (immune signaling), KCNN2 (small‑conductance Ca2+‑activated K+ channel) and GRIA1 (AMPA glutamate receptor). The authors report that ML classifiers achieved high accuracy separating high‑ and low‑risk patients and explored genetic overlap with other psychiatric disorders. Significance: the study suggests peripheral LCL transcriptomes can carry suicide‑risk signals in BD, pointing to immune, synaptic and ion‑channel mechanisms that could be developed into predictive biomarkers for early intervention. Key technical caveats include the very small, all‑Caucasian cohort (n=20), heterogenous medication exposure, and LCLs’ limitations as brain proxies; results therefore require replication in larger, multi‑ancestry and medication‑controlled cohorts plus functional validation before clinical application. Data and code are linked in a GitHub repository for further evaluation.