TY  - THES
AU  - Chen, Ji
TI  - Modeling the psychopathology in schizophrenia: symptom dimensions, subtypes, brain connectivity patterns, and molecular architecture
PB  - Heinrich-Heine Universität
VL  - Dissertation
M1  - FZJ-2020-03308
SP  - 147
PY  - 2020
N1  - Dissertation, Heinrich-Heine Universität, 2020
AB  - Summary Schizophrenia is a severe, debilitating, and heterogeneous mental disorder. Disentangling the psychopathological heterogeneity in schizophrenia from its underlying dimensions and the related neurobiology remains a challenge. Although ample efforts have been devoted to their study, symptom dimensions and subtypes, as well as neurobiological substrates and differentiations in schizophrenia remain unclear. In my project, I implemented machine-learning frameworks aiming to develop a new method to robustly and reliably conceptualize the psychopathology of schizophrenia at both symptom and brain levels in a data-driven fashion. First, an orthonormal and projective variant of non-negative matrix factorization (OPNMF) was employed to identify the latent dimensions of the well-established Positive and Negative Syndrome Scale (PANSS). This method is capable of learning compact and homogeneous factors which can be readily generalized to novel patients. By evaluating OPNMF-derived factor models within a large, homogeneous schizophrenia dataset and then cross-validating the yielded models with an independent multi-site sample recruited from Europe, Asia, and the United States, a structure with four dimensions representing negative, positive, affective and cognitive symptoms was identified as the most stable and generalizable. This four-dimensional structure showed higher internal consistency than the original PANSS subscales and previously proposed factor models. Based on the identified dimensions, fuzzy-clustering was employed to derive symptomatically well-separated schizophrenia subtypes. Two core subtypes of schizophrenia patients were identified, with one featuring prominent negative and affective symptoms while the other featuring positive symptomatology. This positive-negative dichotomy was longitudinally stable in about 80% of the repeatedly assessed patients. Neurobiological divergence of the identified subtypes was assessed using classification analysis of resting-state functional MRI measurement with cross-validation in a subset of the multi-site sample. Individual subtypes could be well-discriminated using resting-state functional connectivity (rsFC) profiles of the ventromedial frontal cortex, temporoparietal junction, and precuneus, with the highest classification accuracy of 70%. Individual expression of the four symptom dimensions were predicted using relevance vector machine based on rsFC within 17 meta-analytically defined task-activation networks. A strict validation procedure including 10-fold cross-validation, leave-one-site-out experiments, and generalization to independent samples was conducted to derive robust symptom-network associations. Finally, the significant and robust symptom-predictive networks were spatially correlated with whole-brain density maps of nine receptors and transporters from prior molecular imaging in healthy populations to reveal the molecular architecture related to these networks. The theory-of-mind and the extended socio-affective default networks, which are implicated in social cognition and affective processes, were identified as significantly and robustly predictive of the cognitive symptom dimension. Moreover, node importance of these two networks showed a spatial pattern positively co-varying with D1 dopamine receptor and serotonin reuptake transporter densities as well as presynaptic dopamine capacity.The current work provides a systematic modeling framework of schizophrenia from symptomatology to neurobiology. Together the proposed hybrid dimensional-categorical conceptualization of symptomatology and the revealed intrinsic neurobiological processes and molecular architecture further disentangle the heterogeneity in schizophrenia, possibly allowing for the development of more specifically targeted treatments.
LB  - PUB:(DE-HGF)11
UR  - https://juser.fz-juelich.de/record/884904
ER  -