Overlap across psychotic disorders: A functional network connectivity analysis

Hossein Dini, Luis Emilio Bruni, Thomas Z. Ramsøy, Vince D. Calhoun, Mohammad S.e. SendibioRxiv 2022 · DOI

TL;DR

This study leverages dynamic functional network connectivity (dFNC) from resting-state fMRI to uncover overlapping neural patterns across schizophrenia (SZ), bipolar disorder (BP), and schizoaffective disorder (SAD). By identifying distinct brain states and their associated features, the research provides a more nuanced understanding of these disorders' shared and unique neural underpinnings, which can predict clinical outcomes.

What Problem Does This Paper Address?

Distinguishing between psychotic disorders like schizophrenia, bipolar disorder, and schizoaffective disorder is notoriously challenging due to significant symptom overlap. Traditional diagnostic approaches often struggle to capture the full complexity of these conditions. While static functional network connectivity (sFNC) has shown promise in differentiating patient groups from healthy controls, it may not fully capture the dynamic nature of brain activity. This paper aims to address this by investigating the overlap and distinctiveness of these disorders using dynamic functional network connectivity (dFNC), hypothesizing that dFNC can reveal more sensitive markers of these overlapping neural mechanisms.

Methods

The researchers employed a multi-stage analytical pipeline:

  1. Independent Component Analysis (ICA): They first used group ICA to identify 53 reliable independent components (ICs) representing distinct functional brain networks from resting-state fMRI (rs-fMRI) data.
  2. Dynamic Functional Network Connectivity (dFNC) Calculation: For each participant, dFNC was computed by applying a sliding window approach to the time-series of these ICs. This involved segmenting the rs-fMRI data into short windows and calculating the Pearson correlation coefficient between the ICs within each window. This process generated time-varying connectivity matrices.
  3. Static Functional Network Connectivity (sFNC) Calculation: As a comparison, sFNC was also calculated by averaging connectivity across the entire rs-fMRI session.
  4. Clustering of dFNC States: The concatenated dFNC data from all participants were then clustered using k-means clustering to identify distinct, recurring brain states. The optimal number of states was determined to be three.
  5. Feature Extraction: From the identified states, five key features were extracted for each participant: Occupancy Rate (OCR - the proportion of time spent in each state), Hidden Markov Model (HMM) features, number of transitions to a state, total number of transitions, and traveled distance (a measure of how much the connectivity patterns changed between windows).
  6. Statistical Analysis: One-way ANOVA was used to compare these extracted features within and between diagnostic groups (SZ, BP, SAD, and their relatives, plus healthy controls). Statistical significance was adjusted using the Benjamini-Hochberg false discovery rate (FDR) correction. Partial correlation analysis was employed to investigate the link between dFNC features and clinical scores (PANSS).

Datasets

The study utilized data from the Bipolar–Schizophrenia Network on Intermediate Phenotypes (B-SNIP) cohort. This dataset included:

  • Patients with schizophrenia (SZP, N=102)
  • Patients with bipolar disorder (BPP, N=102)
  • Patients with schizoaffective disorder (SADP, N=102)
  • Relatives of patients with schizophrenia (SZR, N=102)
  • Relatives of patients with bipolar disorder (BPR, N=102)
  • Relatives of patients with schizoaffective disorder (SADR, N=102)
  • Healthy Controls (HC, N=118)

The total sample size was 730 participants. The data was collected from six different sites.

Key Insights and Ideas

  • Dynamic Connectivity Reveals Overlap: dFNC, unlike sFNC, is more sensitive to the dynamic nature of brain activity and effectively captures overlapping functional connectivity patterns across SZ, BP, and SAD.
  • Distinct Brain States: The analysis identified three distinct brain states, each characterized by unique patterns of functional network connectivity. These states represent different modes of brain organization.
  • State-Specific Biomarkers: Features like Occupancy Rate (OCR) within specific states proved to be discriminative. For instance, OCR in state 2 differentiated SZ patients from their relatives and healthy controls, and also separated SZ from BP patients.
  • Complementary Nature of SZ and BP: While SZ and BP may not have identical neural patterns, their activity patterns appear complementary, suggesting a complex interplay rather than a simple overlap.
  • Predictive Power for Clinical Outcomes: dFNC features, particularly OCR and traveled distance, were found to significantly predict clinical severity scores (PANSS Total and General), suggesting their potential as biomarkers for treatment response or disease progression.

Key Results

  • Within-Disorder Differences: Significant differences in OCR were found within SZ (SZP vs. SZR, SZP vs. HC in state 2), BP (BPP vs. HC in state 1), and SAD (SADP vs. HC in state 2).
  • Between-Disorder Differences: OCR in state 2 differentiated SZP from BPP, and in state 3, it differentiated BPP from SADP.
  • Clinical Prediction:
    • OCR from BPP in state 2 showed a significant relationship with general PANSS scores (r=0.269, p=0.043).
  • OCR and distance features from SZP significantly linked to general PANSS (r=0.257, p=0.015) and total PANSS (r=0.233, p=0.027).
  • Traveled distance from BPP significantly related to PANSS General (r=0.243, p=0.027).
  • Combined distance features from all patients could predict general PANSS (r=0.159, p=0.009) and total PANSS (r=0.143, p=0.020).
  • dFNC vs. sFNC: dFNC captured significant differences between regions that sFNC did not, highlighting its superior ability to reveal nuanced connectivity patterns.

Limitations and Future Work

  • Noisy Clinical Scores: The recorded clinical scores were noted as potentially noisy, with some missing data. This was addressed by removing corresponding features, but it remains a limitation.
  • Multi-site Data: Data was collected from multiple sites, which can introduce variability. While considered in the analysis, it's a factor to acknowledge.
  • Psychiatric Classification: The classification was based on DSM-V criteria, not the International Classification of Diseases (ICD), which could influence findings.
  • Future Directions: The study highlights the potential of dFNC as a powerful tool for understanding psychotic disorders and their overlaps. Future work could explore these dynamic connectivity patterns in more detail, potentially leading to improved diagnostic tools and personalized treatment strategies. The predictive power of dFNC features for clinical outcomes also warrants further investigation.

Summary updated

Abstract

Abstract Functional network connectivity has previously been shown to distinguish patient groups from healthy controls (HC). However, the overlap across schizophrenia (SZ), bipolar (BP), and schizoaffective disorder (SAD) is not clear yet. This study focuses on finding overlap across these three psychotic disorders using dynamic FNC (dFNC) and compares it with static FNC (sFNC). We used resting-state fMRI, demographics, and clinical information from the Bipolar– Schizophrenia Network on Intermediate Phenotypes cohort. The data includes three groups of patients with schizophrenia (SZP, N=102), bipolar (BPP, N=102), and schizoaffective (SADP, N=102), their relatives SZR (N=102), BPR (N=102), SADR (N=102), and HC (N=118) groups. After estimating each individual’s dFNC, we put them into three identical states. We estimated five different features, including occupancy rate (OCR), number of transitions, the total number of transitions, and the total distance traveled. Finally, the extracted features are tested statistically across patients and HC groups. In addition, we explored the link between the clinical scores and the extracted features. We found that the OCR difference between SZP and SZR in state2, between BPP and HC in state1, and between SADP and HC in state2. Also, state2 OCR separates SZP from BPP, state 3 OCR separates BPP from SZP and SADP. Moreover, the OCR and traveled distance feature extracted from SZ and BP could significantly predict PANSS Total and PANSS General scores. Finally, combined distance features of all disorders showed a significant relationship to PANSS Total and PANSS General scores.
BibTeX
@misc{dini2022overlap,
  author = {Hossein Dini and Luis Emilio Bruni and Thomas Z. Ramsøy and Vince D. Calhoun and Mohammad S.E. Sendi},
  title = {{Overlap across psychotic disorders: A functional network connectivity analysis}},
  year = {2022},
  doi = {10.1101/2022.03.13.484190},
  howpublished = {bioRxiv},
}
Figure