Maximizing Dissimilarity in Resting State detects Heterogeneous Subtypes in Healthy population associated with High Substance-Use and Problems in Antisocial Personality

Rajan Kashyap, Sagarika Bhattacharjee, B.t. Thomas Yeo, Shen‐Hsing Annabel ChenbioRxiv 2019 · DOI

Abstract

Resting-state fMRI (rs-fMRI) is a cornerstone of neuroimaging, yet scan-to-scan variability often obscures stable trait effects. Traditional methods average or concatenate scans, potentially losing vital information. This study utilizes Common-and-Orthogonal-Basis-Extraction (COBE) on 788 subjects from the Human Connectome Project to isolate stable individual brain signatures. By identifying a subset of subjects with Maximal-COBE-Dissimilarity (MCD), we reveal a distinct subtype within the "healthy" population characterized by reduced Default Mode Network (DMN) activity. This subtype is strongly associated with higher substance use (marijuana, alcohol, tobacco) and antisocial personality traits. These findings suggest that healthy control groups are significantly heterogeneous, necessitating more nuanced categorization in clinical research.

Introduction

In the field of neuroimaging, resting-state fMRI (rs-fMRI) serves as the primary tool for mapping the brain’s intrinsic functional architecture. However, the signal-to-noise ratio in a single scan is often low, influenced by transient mental states, head motion, and physiological noise. To mitigate this, researchers typically average multiple scans or concatenate them into a single time series. While these methods improve signal stability, they often wash out the nuanced, scan-to-scan variability that might contain meaningful biological information.

Furthermore, clinical neuroimaging research frequently relies on a "Case vs. Control" binary, assuming that the "Control" group represents a uniform, healthy baseline. This study challenges that assumption by investigating whether hidden "subtypes" exist within healthy populations—individuals whose brain signatures and lifestyle traits place them at the functional edges of the traditional spectrum.

Methodology: Common-and-Orthogonal-Basis-Extraction (COBE)

To address the limitations of scan averaging, we employed a signal extraction technique known as Common-and-Orthogonal-Basis-Extraction (COBE). This method decomposes multiple fMRI runs from a single subject into two distinct subspaces:

  1. The Common Subspace: Features that remain stable and are shared across all four rs-fMRI scans of an individual, representing the "trait" effect.
  2. The Run-Specific Subspace: Features unique to a specific scan, likely representing transient states or noise.

By extracting the "Common Component" for 788 subjects from the Human Connectome Project (HCP) S1200 release, we created a unique functional "fingerprint" for each participant. We then applied a dissimilarity-based classification strategy to isolate a subset of subjects (n=107) whose brain patterns were most divergent from the majority. This group was designated as the Maximal-COBE-Dissimilarity (MCD) group, while the remaining subjects (n=681) formed the COBE-Similarity (CS) group.

Results: The DMN-A Signature and Behavioral Correlates

The analysis revealed that the dissimilarity in the MCD group was not random but followed a specific neurobiological pattern. The most significant finding was a drastic reduction in weights within the Default Mode Network (DMN-A), particularly in the Posterior Cingulate Cortex (PCC). The PCC is a critical hub for self-referential thought, social cognition, and internal monitoring. The statistical significance of this difference was profound ($p < 1e^{-14}$).

These brain patterns were strongly correlated with real-world behavioral measures. Out of 69 behavioral and physiological measures analyzed, six specific traits related to substance use and personality differentiated the two groups with high statistical significance ($p < 1e^{-5}$):

  • Marijuana Use: 59% of the MCD group had used marijuana more than 1,000 times. In contrast, 53% of the CS group reported never having used it.
  • Alcohol Consumption: 39% of the MCD group reported consuming six or more drinks per week, compared to only 10.7% in the CS group.
  • Tobacco Use: While 87% of the CS group were non-users, only 49.5% of the MCD group fell into the same category.
  • Personality and Gender: The MCD group was predominantly male and scored significantly higher on measures of antisocial personality, including traits like aggressiveness and rule-breaking behavior.

Discussion: The Heterogeneity of "Healthy"

The existence of the MCD subtype suggests that "healthy controls" are far from a monolith. Even individuals who meet the criteria for a healthy population in large-scale datasets can possess brain architectures and behavioral profiles that mirror those found in clinical populations, such as those with substance use disorders or conduct disorders.

This finding has significant implications for how we design and interpret "Case vs. Control" studies. If a control group contains a significant percentage of individuals with "edge-case" brain patterns, the statistical power to detect differences between patients and controls may be diluted. Identifying these subtypes allows for a more refined understanding of the spectrum between health and pathology.

Limitations and Future Directions

While the link between DMN activity and substance use is clear, the causal direction remains an open question. It is unclear whether these resting-state patterns predispose individuals to substance use and antisocial behavior, or if the behaviors themselves have reshaped the brain's functional architecture over time. Additionally, the HCP dataset lacks granular information regarding the "last use" of substances prior to scanning, which could influence acute brain states.

Future research should expand beyond cortical mapping to include detailed parcellations of the cerebellum and subcortical structures. Applying COBE-based subtyping to clinical groups could also help identify more precise diagnostic categories, moving psychiatry toward a more personalized, data-driven approach.

Conclusion

By maximizing dissimilarity in resting-state features, we have identified a distinct subtype within a healthy population that is characterized by reduced DMN connectivity and elevated levels of substance use and antisocial traits. These results underscore the importance of accounting for population heterogeneity in neuroimaging and suggest that the "edge" of normal may provide the most valuable insights into the biological basis of behavior.

Summary updated

Abstract

Abstract Patterns in resting-state fMRI (rs-fMRI) are widely used to characterize the trait effects of brain function. In this aspect, multiple rs-fMRI scans from single subjects can provide interesting clues about the rs-fMRI patterns, though scan-to-scan variability pose challenges. Therefore, rs-fMRI’s are either concatenated or the functional connectivity is averaged. This leads to loss of information. Here, we use an alternative way to extract the rs-fMRI features that are common across all the scans by applying Common-and-Orthogonal-Basis-Extraction (COBE) technique. To address this, we employed rs-fMRI of 788 subjects from the human connectome project and estimated the common-COBE-component of each subject from the four rs-fMRI runs. Since the common-COBE-component are specific to a subject, the pattern was used to classify the subjects based on the similarity/dissimilarity of the features. The subset of subjects (n=107) with maximal-COBE-Dissimilarity (MCD) was extracted and the remaining subjects (n = 681) formed the COBE-similarity (CS) group. The distribution of weights of the common-COBE-component for the two groups across rs-fMRI networks and subcortical regions was evaluated. We found the weights in the default mode network to be lower in the MCD compared to the CS. We compared the scores of 69 behavioral measures and found 6 behaviors related to the use of marijuana, illicit drugs, alcohol, and tobacco; and including a measure of antisocial personality to differentiate the two groups. Gender differences were also significant. Altogether findings suggested that subtypes exist even in healthy control population and comparison studies (Case vs Control) need to be mindful of it.
BibTeX
@misc{kashyap2019maximizing,
  author = {Rajan Kashyap and Sagarika Bhattacharjee and B.T. Thomas Yeo and Shen‐Hsing Annabel Chen},
  title = {{Maximizing Dissimilarity in Resting State detects Heterogeneous Subtypes in Healthy population associated with High Substance-Use and Problems in Antisocial Personality}},
  year = {2019},
  doi = {10.1101/787606},
  howpublished = {bioRxiv},
}
Figure