Altered Domain Functional Network Connectivity Strength and Randomness in Schizophrenia

Victor M. Vergara, Eswar Damaraju, Jessica A. Turner, Godfrey D. Pearlson, Ayşenil Belger, Daniel H. Mathalon, Steven G. Potkin, Adrian Preda, Jatin G. Vaidya, Theo G.m. van Erp, Sarah McEwen, Vince D. CalhounFrontiers in Psychiatry 2019 · DOI

TL;DR

This study introduces a "domain-based" approach to analyze functional brain connectivity in schizophrenia, moving beyond individual brain networks to examine groups of functionally related networks. They found that individuals with schizophrenia exhibit altered connectivity strength and randomness across these domains, suggesting a widespread rewiring of brain networks that may underlie cognitive deficits.

What Problem Does This Paper Address?

Schizophrenia is characterized by disruptions in brain function, and functional connectivity (how different brain regions communicate) is a key area of investigation. Previous research has identified abnormal functional connectivity in schizophrenia, often by looking at individual brain networks (Resting State Networks or RSNs). However, these studies sometimes show similar patterns across many RSNs, suggesting that the problem might be more systemic, affecting groups of related networks. The challenge is to find a way to analyze connectivity at a level that captures these broader patterns without being too coarse (like whole-brain analysis) or too fine-grained (like individual RSNs). This paper aims to address this by proposing and testing a "domain-based" approach, where RSNs are grouped into functional domains, and then analyzing connectivity between and within these domains.

Methods

The researchers employed a novel "domain functional connectivity" (FNC) approach. Instead of analyzing individual RSNs or the entire brain's connectivity, they grouped RSNs into seven predefined functional domains: sub-cortical (SBC), auditory (AUD), visual (VIS), sensorimotor (SEN), cognitive control (COG), default-mode network (DMN), and cerebellum (CER).

They then analyzed two key measures of connectivity within and between these domains:

  1. Connectivity Strength: This is a straightforward measure, calculated as the average of all FNC values (correlations between RSN time courses) within a specific domain pair.
  2. Randomness: This is a more novel measure, adapted from random matrix theory. It quantifies the unpredictability and lack of a discernible pattern in the connectivity within a domain submatrix. A lower randomness value suggests a more structured and predictable pattern, while a higher value indicates more randomness or unpredictability.

They used group independent component analysis (gICA) to identify RSNs from resting-state fMRI data and then calculated FNC matrices. These matrices were then partitioned into domain-specific submatrices for analysis. Statistical comparisons were made between healthy controls (HC) and individuals with schizophrenia (SZ) using t-tests and regression models, with corrections for multiple comparisons.

Datasets

The study utilized a dataset comprising resting-state fMRI data from 163 healthy controls (HC) and 151 schizophrenia patients (SZ). This data was collected from seven different institutions in the United States. For statistical analyses, a subset of 119 SZ patients with available medication data was used.

Key Insights and Ideas

  • Domain-level analysis reveals broader patterns: The study demonstrates that analyzing connectivity at the domain level, rather than just individual RSNs, can reveal more consistent and potentially more fundamental disruptions in brain networks in schizophrenia.
  • Schizophrenia shows altered connectivity "strength" and "randomness": The core finding is that individuals with schizophrenia exhibit significant differences in both how strongly their brain domains are connected and how predictable those connections are.
  • Less randomness = more structure (potentially maladaptive): A key insight is that lower randomness (higher L values) in connectivity patterns is associated with greater structure. In schizophrenia, this can manifest as less random connectivity in some cognitive domains, suggesting a potentially rigid or altered network organization.
  • Trade-offs in brain rewiring: The observed changes in connectivity strength and randomness suggest that the brain in schizophrenia might be undergoing a "rewiring" process. This rewiring may involve trade-offs, where some connections become stronger or more structured, while others weaken or become more chaotic, potentially impacting cognitive function.
  • DMN's sensitivity: The default mode network (DMN) appears particularly sensitive to these disruptions, showing altered connectivity strength and randomness with several other domains.

Key Results

  • Whole Brain: Compared to controls, individuals with schizophrenia showed:
  • Lower connectivity strength (p < 4e-14).
  • Less randomness (higher L values, p < 2e-5), indicating more structured connectivity.
  • Higher modularity (p < 8e-13), suggesting a more organized, albeit potentially rigid, network structure.
  • Domain-Specific:
    • Connectivity Strength: Schizophrenia patients showed lower connectivity strength in several domains, particularly within and between auditory-visual and sensorimotor (AVSN) domains, as well as between cognitive-auditory and DMN-sensorial domains. Conversely, they showed higher connectivity strength in subcortical-auditory, subcortical-visual, and DMN-cerebellum domains.
  • Randomness: Significant differences in randomness were observed. For example, connectivity between cognitive and default mode domains showed less randomness (more structure) in schizophrenia, while connectivity between default mode-sensorial areas showed more randomness (less structure).

Limitations and Future Work

  • Domain aggregation masks individual RSN effects: A primary limitation is that by aggregating RSNs into domains, specific effects within individual RSNs might be masked. This is a trade-off for increased detection power at the domain level.
  • Short fMRI scan duration: The study used a 5-minute fMRI scan, which is relatively short and could be a limitation, although the authors argue it likely avoids sleep states.
  • Spatial resolution: The 4 mm slice thickness of the fMRI data may limit the resolution of signals from small brain structures.
  • Future Work: The authors suggest that future work could explore the mathematical underpinnings of the inverse relationship between connectivity strength and randomness/modularity. They also highlight the need for further investigation into how these domain-level connectivity changes specifically impact cognitive functions in schizophrenia. The study also points to the potential of randomness analysis to uncover patterns not detectable by modularity measures due to mathematical restrictions.

Summary updated

Abstract

Functional connectivity is one of the most widely used tools for investigating brain changes due to schizophrenia. Previous studies have identified abnormal functional connectivity in schizophrenia patients at the resting state brain network level. This study tests the existence of functional connectivity effects at whole brain and domain levels. Domain level refers to the integration of data from several brain networks grouped by their functional relationship. Data integration provides more consistent and accurate information compared to an individual brain network. This work considers two domain level measures: functional connectivity strength and randomness. The first measure is simply an average of connectivities within the domain. The second measure assesses the unpredictability and lack of pattern of functional connectivity within the domain. Domains with less random connectivity have higher chance of exhibiting a biologically meaningful connectivity pattern. Consistent with prior observations, individuals with schizophrenia showed aberrant domain connectivity strength between subcortical, cerebellar, and sensorial brain areas. Compared to healthy volunteers, functional connectivity between cognitive and default mode domains showed less randomness, while connectivity between default mode-sensorial areas showed more randomness in schizophrenia patients. These differences in connectivity patterns suggest deleterious rewiring trade-offs among important brain networks.
BibTeX
@inproceedings{vergara2019altered,
  author = {Victor M. Vergara and Eswar Damaraju and Jessica A. Turner and Godfrey D. Pearlson and Ayşenil Belger and Daniel H. Mathalon and Steven G. Potkin and Adrian Preda and Jatin G. Vaidya and Theo G.M. van Erp and Sarah McEwen and Vince D. Calhoun},
  title = {{Altered Domain Functional Network Connectivity Strength and Randomness in Schizophrenia}},
  booktitle = {Frontiers in Psychiatry},
  year = {2019},
  doi = {10.3389/fpsyt.2019.00499},
}
Figure