PERIPHERAL BLOOD CELLS UNVEIL NEURAL AND SEX-RELATED SUBTYPES OF DEPRESSION: AN UNSUPERVISED MACHINE LEARNING APPROACH

Abstract Background Heterogeneity in depression results from interactions of multiple biological and environmental factors. For instance, exposure to early-adverse events amplifies the cross-talk between inflammation and brain activity, resulting in depressive symptomatology [1]. This brain-immune relationship is particularly crucial for females who are more vulnerable to inflammation and its mood effects compared to males [2]. However, only 30% of patients exhibit an abnormal immune profile, suggesting that inflamed depression might represent a distinct subtype [3]. Data-driven approaches are promising in yielding novel insights into depression neurobiology while disentangling heterogeneity [4]. Aims & Objectives By applying an unsupervised machine learning approach, we aim to unveil immunophenotypes of depressed patients based on immune cell counts, also investigating sex, childhood trauma, and white matter differences. Methods Peripheral blood mononuclear cells (PBMC) from 145 depressed patients (92 Major Depressive Disorder [MDD], 53 Bipolar Disorder [BD]) were analysed using FlowJo v.10.8.1 (FlowJo LLC, Ashland, OR). An immune profile was generated on PBMC using a multiparametric flow cytometry. A 28-color flow cytometry panel was used to characterize lymphocyte phenotype and function. In a subsample of 108 patients (67 MDD, 41 BD), extracted tract-based fractional anisotropy and diffusivity indexes were computed from diffusion tensor images (TBSS, FSL). Exposure to childhood trauma was assessed through the Childhood Trauma Questionnaire (CTQ) [5], dichotomizing domains’ scores according to standard cut-offs [6]. A stability-based relative clustering validation approach [7] was applied to immune cell populations, considering K-means as clustering algorithm and support vector machine as classifier. Hyperparameters tuning was performed in a 2x10 repeated cross-validation, iterating over a number of clusters from 2 to 5. The solution that minimised the normalized stability (i.e., prediction error) were chosen as the best one. Statistical significance was assessed by 10,000 simulations from a single null Gaussian distribution. Differences in diagnosis, sex, extracted white matter diffusivity indices, and CTQ domains were assessed through chi-square and Mann-Whitney U tests. Results The unsupervised pipeline identified 2 clusters with a normalized stability of 0.28 (p<0.0001). While one cluster (N=53) showed a higher proportion of MDD patients (p=0.008) with higher i) lymphocytes; ii) CD4+ and CD8+CD69+; iii) NK; iv) CD4+CD25highFoxP3+CD39+; and v) CM, EM and TEMRA CD8+Perf-Grz+ cells count (pFDR<0.05), the other one (N=92) was characterized by higher F:M ratio (p=0.064) and higher i) CD4+CD25+; ii) CD4+CD25highFoxP3-CD39+; and iii) TEMRA, CM, EM CD8+Perf+GrzB- cells count (pFDR<0.05). Higher axial, radial, and mean diffusivity values in the genu of the corpus callosum and left cingulum were found in the first cluster (pFDR<0.05). Discussion & Conclusions Our results provided evidence of an immune cells-based stratification of depressed patients, suggesting the existence of 2 inflamed subgroups differentiated by distinct immune cell profiles and structural connectivity patterns. These findings foster our understanding of the neurobiological complexity of depression, which could have important implications in tailoring personalized treatments targeting the individual immune-biological profile. References [1]NUSSLOCK, R. &MILLER, G. E. 2016. Early-life adversity and physical and emotional health across the lifespan: A neuroimmune network hypothesis. Biological psychiatry, 80, 23-32. [2]DERRY, H. M., PADIN, A. C., KUO, J. L., HUGHES, S. &KIECOLT-GLASER, J. K. 2015. Sex differences in depression: does inflammation play a role? Current psychiatry reports, 17, 1-10. [3]CHAMBERLAIN, S. R., CAVANAGH, J., DE BOER, P., MONDELLI, V., JONES, D. N., DREVETS, W. C., COWEN, P. J., HARRISON, N. A., POINTON, L. &PARIANTE, C. M. 2019. Treatment-resistant depression and peripheral C-reactive protein. The British Journal of Psychiatry, 214, 11-19. [4]LYNALL, M.-E., TURNER, L., BHATTI, J., CAVANAGH, J., DE BOER, P., MONDELLI, V., JONES, D., DREVETS, W. C., COWEN, P. &HARRISON, N. A. 2020. Peripheral blood cell– stratified subgroups of inflamed depression. Biological psychiatry, 88, 185-196. [5]BERNSTEIN, D. P., STEIN, J. A., NEWCOMB, M. D., WALKER, E., POGGE, D., AHLUVALIA, T., STOKES, J., HANDELSMAN, L., MEDRANO, M. &DESMOND, D. 2003. Development and validation of a brief screening version of the Childhood Trauma Questionnaire. Child abuse & neglect, 27, 169-190. [6]MACDONALD, K., THOMAS, M. L., SCIOLLA, A. F., SCHNEIDER, B., PAPPAS, K., BLEIJENBERG, G., BOHUS, M., BEKH, B., CARPENTER, L. &CARR, A. 2016. Minimization of childhood maltreatment is common and consequential: results from a large, multinational sample using the childhood trauma questionnaire. PloS one, 11, e0146058. [7]LANDI, I., MANDELLI, V. &LOMBARDO, M. V. 2021. reval: A Python package to determine best clustering solutions with stability-based relative clustering validation. Patterns, 2, 100228.