Network-based approaches towards understanding the emergence of neurodevelopmental disorders

Abstract

Many psychiatric disorders are rooted in neurodevelopmental processes. However, achieving a mechanistic understanding of the developmental sensitivity of mental illness---and how disruptions to early life processes shape psychiatric symptomatology---remains a central challenge in psychiatry. This challenge is compounded by the complexity of neurobiological organization, spanning genes to brain systems, and by the interacting influence of environmental exposures. This thesis addresses these challenges using two complementary, network-based approaches under a translational framework: (1) mapping macroscale brain networks in a rat model to study the causal impact of early life stress (ELS) on normative neurodevelopment, and (2) characterizing microscale gene co-expression networks in postmortem human brain tissue to link genetic risk to transcriptomic signatures of psychiatric disorders. In Chapter 1, I review the literature linking environmental and genetic factors to psychiatric outcomes, with a focus on how these influences interact across development. I begin with evidence for the developmental sensitivity of psychiatric risk, then describe the role of early life stress and how its type and timing shape neurobehavioral phenotypes. I parallel this with an overview of genetic contributions to psychiatric outcomes, using schizophrenia as a case study given its high heritability and well-characterized genetic architecture, and introduce gene-by-environment interactions as a framework for integrating these domains. To address the complexity of these multiscale interactions, I introduce network-based approaches---specifically, macroscale magnetic resonance imaging (MRI)-derived brain networks and microscale gene co-expression networks. Finally, I argue for the utility of animal models, particularly rats, in experimentally dissecting the causal pathways linking environmental exposures to brain and behavioral phenotypes. Chapters 2 and 3 present an in vivo structural MRI study of the impact of ELS on the developing rat brain using morphometric inverse divergence (MIND), a novel network-based method for estimating within-subject cortical similarity based on morphometric feature distributions. I analyzed data from two independent cohorts: a normative developmental cohort scanned at four postnatal stages (PND 20 [weanling] to 290 [mid-adulthood]), and an experimental stress cohort in which half the animals experienced repeated maternal separation (RMS) while the other half were reared normally (Control). All animals were scanned in early adulthood (PND 63) and in later adulthood (~PND 300) following an additional adult stressor. In \textbf{Chapter 2}, I define the normative MIND network using magnetization transfer ratio (MTR) and show that it exhibits complex topological properties and aligns with independent measures of cortical cytoarchitectonics, mouse spatial transcriptomics, and tract-tracing. The network was highly reproducible across cohorts. In Chapter 3, I characterize how network similarity changes through development and aging, identifying a phase of fronto-hippocampal convergence in early development that diverges in aging. RMS exposure appeared to accelerate normative neurodevelopmental trajectories, highlighting the embedding of early stress in the dynamic rat brain network, with limited additional impact of adult stress. These changes were not captured by univariate MTR or volumetric features, demonstrating the unique insights offered by the network-level analysis. This work provides novel tools for systems-level study of the rat brain that can now be used to understand network-based underpinnings of complex lifespan behaviors and experimental manipulations enabled by this model organism. Chapters 4 and 5 transition to network-based analyses of bulk transcriptomic data from the subgenual anterior cingulate cortex (sgACC) of individuals with schizophrenia, bipolar disorder, major depressive disorder, and non-psychiatric controls. In Chapter 4, I apply weighted gene co-expression network analysis (WGCNA) to identify modules of co-expressed genes and assess their associations with psychiatric diagnosis and toxicological exposures. While several modules were linked to diagnostic status, many were confounded by environmental variables such as medication. To address this, Chapter 5 employs group regularized canonical correlation analysis (GRCCA), a multivariate method that integrates gene co-expression structure while adjusting for confounds. GRCCA identifies a schizophrenia-specific transcriptomic signature, enriched for genes implicated in common variant risk and characterized by opposing patterns of immune and neuronal expression. This analysis outperformed traditional differential expression analyses in both biological enrichment and alignment with known genetic architecture, thus demonstrating the utility of multivariate approaches for integrating genetic and genomic signals in legacy data. Finally, in Chapter 6, I summarize the technical and conceptual contributions of this work, which may converge on glial-mediated plasticity as a unifying mechanism linking neurodevelopmental patterns to psychiatric outcomes. I outline future directions for integrating across species, modalities, and biological scales to advance our understanding of the developmental origins of psychiatric disorders.