2026 ECNP Congress

PS02-2127

Structural–diffusion brain topology reveals affective and sleep-related subtypes in bipolar disorder

Emma Tassi — G. Torino, G. Conte, F. Colombo, B. Bravi, T. Cazzella, S. Poletti, C. Colombo, C. Fabbri, F. Benedetti, B. Vai, P. Brambilla, E. Maggioni · Fondazione IRCCS Ca Granda Ospedale Policlinico di Milano, Department of Neurosciences and Mental health, Milano, Italy

2026 ECNP Congress

Background

Bipolar disorder (BD) is characterized by marked clinical heterogeneity across depressive symptom dimensions [1]. Identifying markers of symptom-based subtypes is essential for personalized approaches. Topological Data Analysis (TDA), when paired with graph-based community detection, enables identifying subgroups with distinct clinical profiles [2, 3]. This study developed a TDA-based pipeline integrating clustering, feature ranking, and graph-based mapping - whereby new samples are projected onto a pre-constructed graph. We identified BD subgroups with homogeneous depressive symptom profiles for Hamilton Depression Rating Scale (HDRS) items comparing feature sets of demographics, environmental (ENV), structural MRI (sMRI), diffusion tensor imaging (DTI), and their combinations.

Methods

We analyzed 133 BD subjects, experiencing current depressive episodes (training n=109, test n=33), recruited at San Raffaele Hospital of Milan. Feature sets included ENV, sMRI, DTI, demographics, and their combinations. Outcomes included HDRS items.

For each set, a TDA network was built on the training set using the Mapper algorithm [4], with UMAP dimensionality reduction. The space was divided into overlapping intervals, and DBSCAN identified clusters forming nodes; shared subjects defined edges. Community detection identified coherent patient structures; only core individuals (belonging uniquely to one community) were retained. HDRS differences across communities were assessed, and significant items used to derive community symptom severity profiles based on the prevalence of low- vs. medium/high symptom severity. Features were ranked by UMAP correlations and community differentiation.

Test subjects were mapped onto training communities of best-performing graph (i.e., showing most significant HDRS differences) via multivariate Kernel Density Estimation assignment using top-ranked features, assigning each to the community with highest likelihood. Pattern preservation was quantified as consistency of low- vs. medium/high symptom severity distributions between training and test sets.

Results

sMRI+DTI communities (n=6) were associated with significant HDRS differences for late insomnia (p=0.003), middle insomnia (p=0.025), somatic anxiety (p=0.024), and suicidality (p=0.041).

One community emerged as highest-severity subgroup for somatic anxiety and suicidality, with all subjects showing moderate-to-high somatic anxiety and increased suicidality (33.3%), but balanced severity distributions for insomnia.Two communities showed high insomnia prevalence (93%) but low suicidal ideation (<7%), suggesting independent symptom dimensions. Another community showed low symptom severity across significant outcomes.

Feature ranking highlighted DTI measures, with radial diffusivity in the left corticospinal tract as the top-ranked feature. The highest-severity subgroup for somatic anxiety and suicidal ideation exhibited elevated RD (mean=0.415), whereas insomnia-dominant communities showed reduced values (mean=−0.425 and −0.692). In the test set, severity patterns were consistent: highest-severity subgroups remained elevated for somatic anxiety (70%) and suicidal ideation (40%), while the low-severity community maintained lower symptoms. Insomnia-dominant communities retained the highest prevalence of middle (85.7%) and late insomnia (80-85.7%). Overall, high- and low-severity subgroups showed the strongest train-test stability.

Conclusions

TDA-based stratification identified biologically grounded BD symptomatologic subtypes, with multimodal imaging providing robust and generalizable patterns. Results suggest somatic anxiety and suicidality may reflect independent dimensions with respect to sleep disturbance, consistent with evidence linking white matter microstructure to anxiety [5] and sleep disruption [6] in mood disorders. Future studies should also address the potential effect of concurrent pharmacotherapy.

References

  1. Vieta, E., Berk, M., Schulze, T.G. et al. Bipolar disorders. Nat. Rev. Dis. Primers 4, 18008 (2018). https://doi.org/10.1038/nrdp.2018.8
  2. Singh, G., Mémoli, F. & Carlsson, G.E. Topological methods for the analysis of high dimensional data sets and 3D object recognition. Eurographics Symp. Point-Based Graphics, 91–100 (2007). https://doi.org/10.2312/SPBG/SPBG07/091-100
  3. Skaf, Y. & Laubenbacher, R. Topological data analysis in biomedicine: a review. J. Biomed. Inform. 130, 104082 (2022). https://doi.org/10.1016/j.jbi.2022.104082
  4. van Veen, H.J., Saul, N., Eargle, D. & Mangham, S.W. Kepler Mapper: a flexible Python implementation of the Mapper algorithm. J. Open Source Softw. 4(42), 1315 (2019). https://doi.org/10.21105/joss.01315
  5. Jenkins, L.M. et al. A meta-analysis on the uncinate fasciculus in depression. Psychol. Med. 53(10), 4679–4693 (2023). https://doi.org/10.1017/S0033291723000107
  6. Benedetti, F., Melloni, E.M.T., Dallaspezia, S. et al. Night sleep influences white matter microstructure in bipolar depression. J. Affect. Disord. 218, 380–387 (2017). https://doi.org/10.1016/j.jad.2017.05.008