A pattern of cognitive resource disruptions in childhood psychopathology

Abstract

The Hurst exponent (H) isolated in fractal analyses of neuroimaging time series is implicated broadly in cognition. Within this literature, H is associated with multiple mental disorders, suggesting that H is transdimensionally associated with psychopathology. Here, we unify these results and demonstrate a pattern of decreased H with increased general psychopathology and attention-deficit/hyperactivity factor scores during a working memory task in 1,839 children. This pattern predicts current and future cognitive performance in children and some psychopathology in 703 adults. This pattern also defines psychological and functional axes associating psychopathology with an imbalance in resource allocation between fronto-parietal and sensorimotor regions, driven by reduced resource allocation to fronto-parietal regions. This suggests the hypothesis that impaired working memory function in psychopathology follows from a reduced cognitive resource pool and a reduction in resources allocated to the task at hand.

Keywords: Cognition; Fractals; Hurst exponent; Psychopathology; Working memory; fMRI.

Figure 1.

Hypothesis 1. Time series with H close to 1 have smooth-looking temporal fluctuations (top). Such time series can also be described as scale-free or fractal in time. In contrast, time series with more random temporal fluctuations have H closer to 0.5 (bottom). We hypothesized that higher H would be associated with lower extracted factor scores for a general factor of psychopathology, p. Though we characterize individuals as having lower H or higher H, there is still variability in H across brain regions.

Figure 2.

Hypothesis 2. While some individuals differ in overall, whole-brain patterns of H (see Figure 1), we expect relatively higher H in cognitive brain networks to be indicative of fewer resources directed toward the task at hand and poorer task performance, and vice versa. Here we represent this hypothesis by showing the distribution of H throughout the brain for individuals with both high H and low H performing a task that requires recruiting frontal regions. Relative to the whole brain tendency of high or low H, we hypothesize that individuals who perform poorly would have higher H in frontal regions. This indicates a failure to properly engage and direct cognitive resources to those regions. In contrast, we hypothesize that individuals who perform well would have lower H in frontal regions. This indicates that those individuals engage those regions and direct cognitive resources toward them, resulting in good performance on the task.

Figure 3.

A Task-induced pattern relating scale-free brain activity as measured by H to childhood psychopathology. (A) The covariance matrix between H and extracted bifactor scores was decomposed with PLS to find maximally covarying latent-variables. (B) A single latent variable, which explained 49% of the cross-block covariance was significant. This LV describes a Hurst-psychopathology pattern in which lower H is associated with both higher general and ADHD extracted factor scores. (C) Mean H for brain regions with absolute value bootstrap ratios > 3 is positively correlated with both in-scanner (left) and out-of-scanner (middle) and future in-scanner (right) working memory performance (note: there were no regions with positive bootstrap ratios greater than this threshold). This indicates that individuals with lower H tend to have higher extracted general and ADHD bifactor scores and also worse working memory performance and that these deficits persist over time (years) and across tasks (the in-scanner and out-of-scanner memory tasks were different, i.e., EN-back task vs. the List Sorting working memory task).

Figure 4.

HPP scores in the HCP sample are significantly correlated with scores on the ASR DSM-Oriented Attention Deficit/Hyperactivity Problems and Avoidant Personality Problems scales. Spatial correlations between the HPP and each subject’s H map assess the degree to which each subject exemplifies the HPP. These were correlated with ASR DSM-Oriented scales and only significant correlations after correction for multiple comparisons across all six scales were retained.

Figure 5.

The Hurst-psychopathology pattern defines a functional activation axis associated with task performance and psychopathology. (A) To understand the relationship between the HPP and functional activation we first computed a pattern score for each subject by correlating their spatial patterns of H to the HPP. Next, we correlated these pattern scores across participants to functional activation in each brain parcel. (B) Higher general factor of psychopathology and ADHD extracted factor scores and lower H are associated with decreased fronto-parietal activation and increased occipital, medial-temporal, and sensorimotor activation on 2-back versus 0-back blocks. Areas where pattern scores are significantly correlated with functional activation after multiple comparison correction are outlined in black.

Figure 6.

The Hurst-psychopathology pattern defines a psychological axis associated with task performance and psychopathology. (A) We correlated the HPP with 116 Neurosynth term association maps showing the probability of activation with multiple psychological terms (shown for the terms “spatial attention,” “memory retrieval,” “listening,” and “face recognition”) to identify which terms had spatial patterns of activation most similar to the HPP. (B) Gray indicates nonsignificance based on 1,000 parametric spatial permutation tests (Benjamini-Hochberg correction, α = .01). On the x-axis, terms are ranked by the magnitude and sign of correlations. (C) Terms that are positively correlated with the HPP are the positive term set and terms that are negatively correlated with the HPP are the negative term set. (D) The positively correlated terms include task-relevant cognitive processes. The negatively correlated terms included processes involved in planning and executing responses to task cues. Surface maps of these two sets were created by taking the maximum value across term maps included in each set.

Figure 7.

Correlations (blue) were computed across all choices of z-score threshold used to define the task-cognition and cue-response ROIs from Neurosynth meta-analysis probabilistic activation maps. The significance of the correlations was assessed at all choices of threshold (gray). The p = 0.05 level is shown as the gray horizontal line. We found evidence of a correlation between higher H and higher HPP scores only for task-cognition areas. Insets show correlations between block level 2- versus 0-back H contrast and HPP scores for all choices of z-score threshold. The bar plot shows the correlation for the choice of z = 2. Error bars represent standard deviations from 1,000 bootstrap resamples.