The researchers combined a Japanese discovery sample from the Strategic Research Program for the Promotion of Brain Science, comprising 824 controls and 229 people with major depressive disorder, with a German test sample from the Marburg–Münster Affective Disorders Cohort Study, comprising 923 controls and 694 patients. The discovery data had been collected at 13 Japanese sites using eight scanner types, whereas the German data came from two scanners at universities in Marburg and Münster. Applying a pattern learned in one population to a separately recruited population with different clinical ascertainment, scanner environments and acquisition protocols tested whether the depression-related signal could survive substantial methodological heterogeneity.

Both cohorts underwent resting-state functional MRI, and each brain was parcellated into 114 cortical regions. The investigators estimated 6,441 possible functional connections by calculating the correlation between the activity time series of each pair of regions. For the Japanese discovery cohort, they compared every connection in patients and controls while accounting for age, sex, head motion and scanner site.

The resulting edge-wise effect sizes formed a map of depression-associated connectivity differences. For every participant in the German cohort, PCSMDD was calculated as a weighted average between that map and the participant’s own connectivity matrix. Higher scores indicated a closer resemblance to the connectivity pattern associated with MDD in the discovery sample.

The external diagnostic result was statistically significant: patients had higher PCSMDD values than controls, with z = 5.09, P < 0.001, an AUC of 0.69 and an additional 1.3% of diagnostic variance explained beyond covariates. For context, split-half analyses within the Japanese dataset yielded an AUC of 0.83 and incremental R² of 3.5%. The lower cross-cohort performance showed that some signal was lost when the metric was transferred to an independent population, but the association remained detectable.

The score was not explained simply by medication or the presence of another mental disorder. Clinical state was also associated with the measure. Acute patients had the highest scores and partially remitted patients also scored significantly higher than controls, with Hedges’ g values of 0.32 and 0.27, respectively.

The results therefore associate the connectivity pattern with differences in current clinical state, but the cross-sectional design cannot show whether connectivity changes precede, accompany or follow symptoms, recovery or relapse. Analyses indicated that the diagnostic information was spread across thousands of connections rather than concentrated in a small anatomical hotspot. Performance was highest when all 6,441 connections were retained.

As progressively stricter statistical thresholds reduced the map to between 1,120 and six connections, predictive performance declined, reaching 0.5% explained variance when only six remained. The independently derived score also explained more variance than 99.97% of 6,441 models based on individual connections, despite those single-edge models having the advantage of being fitted directly in the German dataset. Permuting the weights while preserving their overall distribution reduced performance, showing that the information depended on the spatial arrangement of the connectivity changes rather than only on their magnitude.

The pattern was also more informative than average within-network and between-network connectivity. PCSMDD continued to explain diagnostic variance after those broad summaries were included in the model, while the global measures were no longer significant in the full analysis. The result supports a distributed but spatially organized signature that cannot be reduced to average connectivity within or between large-scale networks.

The study’s principal analyses used connectivity after regression of the global mean signal, which isolates relative changes across the connectome. Separate analyses without that correction showed the broad group difference and produced stronger symptom associations for some measures, highlighting the sensitivity of neuroimaging results to preprocessing choices. Within the German MDD group, higher PCSMDD was associated with more severe current symptoms measured by the Beck Depression Inventory.

The adjusted correlation with the total score was r = 0.09, with a 95% confidence interval of 0.01 to 0.16 and P = 0.022. Among individual symptoms, loss of pleasure, or anhedonia, was the only association that remained significant after false-discovery-rate correction, with r = 0.12 and adjusted P = 0.041. Associations with agitation, indecisiveness, loss of appetite and suicidal ideation were nominally significant but did not survive correction for multiple comparisons.

A clinician-rated Hamilton Depression Rating Scale analysis pointed in the same direction but did not reach statistical significance, with r = 0.07 and P = 0.056. The depression score showed limited relationships with previous disease course. A nominally significant correlation with the number of earlier depressive episodes, r = 0.085 and P = 0.049, did not remain significant after multiple-comparison correction, and there were no significant associations with age at onset or number of hospitalizations.

Comparisons with scores derived for bipolar disorder and schizophrenia also indicated relative specificity for current depressive symptoms. The bipolar and schizophrenia scores contained diagnostic information in the MDD sample, and the schizophrenia score tracked remission status, but neither was significantly associated with current depressive symptom severity or individual symptoms after correction. The MDD score was more strongly associated with symptom severity than the bipolar score; its advantage over the schizophrenia score was directionally similar but not statistically conclusive.

These comparisons do not establish that the underlying connectivity pattern is biologically unique to MDD. Several limitations constrain the findings. Effect sizes were small, the study was not preregistered, and the authors describe the results as exploratory pending independent replication.

Cross-population differences in MRI acquisition and clinical recruitment may have attenuated the signal, while medication regimens, comorbid disorders and global signal regression could have influenced the findings. Only resting-state functional connectivity was examined, leaving open whether structural networks or other biological measurements contain similar or complementary information. One successful cross-population replication also cannot establish generalizability to all populations.

With no evidence that PCSMDD can support individual diagnosis or prognosis, the researchers position it as a research-level measure for characterizing a shared neurobiological pattern in depression rather than as a test ready for clinical use.

Subject of Research: Cross-population resting-state MRI analysis of connectome-wide dysconnectivity and current clinical status in major depressive disorder

Article Title: Connectome-wide dysconnectivity signature tracks clinical status in major depressive disorder

Article References: Gruber, M., Libedinsky, I., Ahrens, K. F., Rehm, P., Staub, H., Fischer, S., Timm, F., Edwin Thanarajah, S., Flinkenflügel, K., Borgers, T., Winter, N. R., Goltermann, J., Leehr, E. J., Meinert, S., Hahn, T., Opel, N., Thomas-Odenthal, F., Stein, F., Jamalabadi, H., … Repple, J. (2026). Connectome-wide dysconnectivity signature tracks clinical status in major depressive disorder. Nature Mental Health. https://doi.org/10.1038/s44220-026-00747-w

Image Credits: AI Generated

DOI: 10.1038/s44220-026-00747-w

Keywords: major depressive disorder, resting-state fMRI, functional connectivity, connectome dysconnectivity, polyconnectomic score, anhedonia, remission status, cross-cohort replication, depression neuroimaging, Connectome-wide, dysconnectivity, signature