Brain Scans Reveal How Mood Disorder Symptoms Map onto Cortical Structure

Depression and bipolar disorder have long been treated as separate categories of illness, yet the people who receive these diagnoses often share a bewildering overlap of symptoms: crushing low mood, loss of pleasure, restless energy, self-harm urges, insomnia, and racing thoughts. A new structural neuroimaging study, published in BMC Psychiatry by researchers at Renmin Hospital of Wuhan University, argues that the clearest picture of what is happening in the brain may come not from diagnostic labels but from the symptom patterns themselves. Working with one of the largest single-cohort clinical samples assembled for this kind of analysis — 1,023 patients diagnosed with major depressive disorder or bipolar disorder — the team set out to determine whether dimensionally defined symptom profiles, measurable across both diagnoses, correspond to distinct anatomical signatures in the cortex and deeper brain structures.

The methodological heart of the study lies in its commitment to data-driven phenotyping. Rather than assuming in advance which symptom clusters matter, the researchers applied K-means clustering to 174 individual clinical items drawn from validated instruments, including the Patient Health Questionnaire-9, the Generalized Anxiety Disorder-7, the Young Mania Rating Scale, the Hypomania Checklist-32, the Snaith-Hamilton Pleasure Scale, the Insomnia Severity Index, the Dysfunctional Attitudes Scale, and measures of non-suicidal self-injury and psychotic symptoms rated on the Brief Psychiatric Rating Scale. Clustering at the level of individual questionnaire items, rather than summed scale scores, allowed the algorithm to detect combinations of symptoms that cut across conventional diagnostic boundaries. The procedure produced two stable, non-overlapping transdiagnostic phenotypes, and the cluster solution held up under internal validation using the adjusted Rand index.

The first group was labeled the Negative Affect-Anhedonic phenotype. Its defining features were self-injurious tendencies together with depressive and anhedonic symptoms — the withdrawal, joylessness, and despair that most people associate with severe depression, but which in this sample appeared in patients carrying either diagnosis. The second group, the Manic-Activation phenotype, was characterized by manic and hypomanic activation: elevated mood, heightened drive, reduced need for sleep, and accelerated thinking and behavior. Crucially, each phenotype drew members from both diagnostic camps. A patient with a bipolar diagnosis could belong to the negative affect cluster, and a patient diagnosed with unipolar depression could show prominent manic-activation features. The researchers report that phenotype dominance only partially aligned with categorical diagnosis, revealing substantial cross-diagnostic overlap and equally substantial heterogeneity within each diagnosis.

With the symptom phenotypes established, the team turned to the brain. All participants underwent high-resolution 3 Tesla structural magnetic resonance imaging, and the scans were processed with FreeSurfer, the widely used reconstruction pipeline, to extract three complementary measures of neuroanatomy: cortical thickness, cortical surface area, and subcortical volumes. These metrics are biologically distinct. Cortical thickness reflects the width of the folded gray matter ribbon and is thought to relate to dendritic arborization and synaptic density, whereas surface area reflects the degree of cortical folding and the expansion of functional columns; the two show largely independent genetic determinants. Subcortical volumes capture structures such as the hippocampus, amygdala, and basal ganglia, regions repeatedly implicated in mood regulation. To relate these anatomical measures to the symptom phenotypes, the researchers used sparse canonical correlation analysis, a multivariate technique that identifies weighted patterns of variables on both sides of the equation that maximally covary with one another — an approach well suited to the complexity of brain-wide morphology paired with multi-item symptom profiles.

The results delivered a clear modality split. Both the Negative Affect-Anhedonic and Manic-Activation phenotypes showed statistically significant multivariate associations with cortical thickness and with cortical surface area, but neither was linked to subcortical volumes in the full sample. Statistical significance was established through permutation testing, and the stability of individual features was verified with bootstrapping, giving the team confidence that the brain-symptom relationships were not artifacts of a single analytic pass. The absence of subcortical findings is itself notable: despite decades of attention to the hippocampus and amygdala in mood disorders, the strongest transdiagnostic signals in this cohort lived in the cortex.

Perhaps the most striking technical finding concerns the spatial organization of the two cortical metrics. The associations involving cortical thickness were focal, concentrated in higher-order association regions — the areas of the cortex that integrate information across sensory modalities and support complex cognition, self-referential processing, and emotion regulation. By contrast, the associations involving cortical surface area were spatially distributed across much of the cortical sheet but appeared to be network-organized, suggesting coordinated maturational or structural variation along the lines of large-scale neural networks rather than isolated regional differences. This thickness-versus-area dissociation matters because the two metrics index different cellular and developmental processes, and the study’s authors interpret the pattern as evidence that transdiagnostic symptom phenotypes map onto partially dissociable, modality-specific structural brain signatures rather than a single undifferentiated anomaly.

The study also tackled a developmental question. Because the sample spanned adolescence and emerging adulthood — a window during which the brain undergoes dramatic cortical thinning and network refinement — the researchers tested whether the strength of brain-symptom associations varied continuously with age in 960 participants aged 10 to 29 years. Six models were examined, crossing each symptom phenotype with each structural modality. The verdict after correction for multiple comparisons was uniformly negative: no age-by-symptom interaction survived false discovery rate correction across the six tested models. The smallest nominal effect appeared in the Negative Affect-Anhedonic model with subcortical volumes, but the authors were explicit that this effect failed to survive correction and should be treated as exploratory only. In a field where developmental moderation claims are often made on nominal p-values, this restrained reporting is a methodological lesson in itself.

The clinical implications are layered. At a practical level, the findings lend empirical weight to dimensional frameworks such as the Research Domain Criteria initiative advanced by the United States National Institute of Mental Health, which encourages researchers and clinicians to study psychopathology along measurable dimensions rather than relying solely on DSM-5 categories. The Wuhan team’s data show that symptom-based phenotypes capture clinically meaningful heterogeneity that categorical diagnoses obscure — two patients with the same diagnostic label can sit in opposite symptom clusters, and two patients with different labels can share a cluster and, potentially, a common cortical profile. That said, the researchers caution that the phenotypes only partially aligned with diagnosis, so the categorical system has not been rendered obsolete; rather, it appears to be an incomplete lens. A multimetric framework that tracks both cortical thickness and surface area, alongside dimensional symptom measures, may better capture the shared and distinct biology of major mood disorders.

There are also important limits to keep in view. The study is cross-sectional, so it can demonstrate that symptom phenotypes and cortical anatomy covary, but not that one causes the other — cortical differences could predispose to particular symptom patterns, chronic illness could reshape the cortex, medications could contribute, or a third factor could drive both. The clustering solution, while internally stable, represents one data-driven cut through a vast symptom space, and alternative algorithms or item sets might yield different groupings. And the null developmental findings mean only that no moderation was detectable in this cohort with these methods, not that age is irrelevant to brain-symptom relationships. The study was registered prospectively at the Chinese Clinical Trial Registry (ChiCTR2500106094), received ethics approval from Renmin Hospital of Wuhan University, and was funded by the National Key R&D Program of China and the National Natural Science Foundation of China, among other sources.

Even with those caveats, the scale and rigor of the work make it a landmark in the effort to connect the messy, overlapping reality of mood disorder symptoms to measurable brain structure. By showing that data-derived symptom phenotypes — not diagnostic labels — align with reproducible, modality-specific patterns of cortical morphology in more than a thousand patients, the study offers a template for how large clinical cohorts, item-level machine learning, and multivariate neuroimaging statistics can be combined to move psychiatry beyond its century-old categories. The next step, the researchers and their colleagues in the field will likely agree, is longitudinal: following patients over time to see whether these cortical signatures predict symptom trajectories, treatment response, or the emergence of manic episodes in those who today carry only a depression diagnosis.

Subject of Research: Neuroanatomical correlates of transdiagnostic symptom phenotypes in major depressive and bipolar disorders

Article Title: Neuroanatomical correlates of transdiagnostic clinical phenotypes in major depressive and bipolar disorders

Article References: Lu, X., Hu, M., Luo, D., Hou, H., Feng, M., He, J., Wang, W., Yang, B. X., & Zong, X. (2026). Neuroanatomical correlates of transdiagnostic clinical phenotypes in major depressive and bipolar disorders. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08602-z

Image Credits: AI Generated

DOI: 10.1186/s12888-026-08602-z

Keywords: major depressive disorder, bipolar disorder, transdiagnostic psychiatry, structural MRI, cortical thickness, cortical surface area, K-means clustering, sparse canonical correlation analysis, FreeSurfer, Research Domain Criteria, anhedonia, mania