Computerized physical and cognitive training improves the functional architecture of the brain in adults with Down syndrome: A network science EEG study

Abstract

Understanding the neuroplastic capacity of people with Down syndrome (PwDS) can potentially reveal the causal relationship between aberrant brain organization and phenotypic characteristics. We used resting-state EEG recordings to identify how a neuroplasticity-triggering training protocol relates to changes in the functional connectivity of the brain's intrinsic cortical networks. Brain activity of 12 PwDS before and after a 10-week protocol of combined physical and cognitive training was statistically compared to quantify changes in directed functional connectivity in conjunction with psychosomatometric assessments. PwDS showed increased connectivity within the left hemisphere and from left-to-right hemisphere, as well as increased physical and cognitive performance. Our findings reveal a strong adaptive neuroplastic reorganization as a result of the training that leads to a less-random network with a more pronounced hierarchical organization. Our results go beyond previous findings by indicating a transition to a healthier, more efficient, and flexible network architecture, with improved integration and segregation abilities in the brain of PwDS. Resting-state electrophysiological brain activity is used here for the first time to display meaningful relationships to underlying Down syndrome processes and outcomes of importance in a translational inquiry. This trial is registered with ClinicalTrials.gov Identifier NCT04390321.

Keywords: Adaptive neuroplasticity; Cognitive training; Down syndrome; Electroencephalography; Network science indices; Physical training.

© 2020 Massachusetts Institute of Technology.

Figures

Figure 1.

Cortical connectivity between post- and pre-intervention networks and for each time point (t value > 3.52). The color scales represent t values. (A) Circular graph depicting the cortical reorganization in the Down syndrome (DS) brain (cyan: left hemisphere; lime: right hemisphere; blue: frontal lobe; green: occipital lobe; purple: parietal lobe; orange: temporal lobe; maroon: limbic lobe). The cortical reorganization is characterized by the strengthening of direct connections within (i) the left hemisphere, from nodes in the occipital and temporal lobe to the frontal lobe, and from the frontal lobe to the parietal lobe; (ii) the right hemisphere, from nodes in the occipital and temporal lobe to the frontal lobe; and between (i) nodes of the left fusiform gyrus, and inferior temporal lobe and the right frontal lobe; (ii) the right middle occipital lobe, subgyral, parahippocampal and fusiform gyrus to the left frontal lobe; (iii) the left inferior temporal lobe to the right anterior cingulate; and (iv) the right frontal lobe to the left postcentral and middle frontal gyrus. (B) Depiction of pre- and post-intervention networks in comparison to the null hypothesis. The t values indicate that the post-resting-state network has significantly stronger connections than the pre-network. (C) Post vs. Pre. Significant post-pre connectivity differences. Information direction is depicted through line arrows. The visualized networks are significant at a level of p < 0.05, corrected for multiple comparisons via the nonparametric NBS method with an internal threshold of 0.8. The difference in nodal size depicts the increase in the node degree centrality; the nodes with the most increased connectivity are located at the left parietal lobe.

Figure 2.

Depiction of the shifts in the clustering coefficient (CC) of nodes. The colormap indicates significance at an alpha value (α < 0.05) coupled with the direction of the shift. The black nodes have no significant shifts in CC values post- and pre-intervention, blue nodes show a significant increase at α < 0.05, while magenta nodes show a significant decrease at α < 0.05.

Figure 3.

Illustration of our theoretical proposal. Changes in connectivity and graph-theory characteristics, as an outcome of the adaptational neuroplasticity in the DS brain, characterize the post-intervention DS network (B → C) as a transitional state from the random-like organization of the pre-intervention DS network (Ahmadlou et al., 2013) (B) toward a healthier functional structure. Random networks (Erdös–Rényi, random graph) (A) are characterized by high global efficiency (GE; overintegration) and low clustering (under-segregation) and exemplify the general abilities of general intelligence (low intelligence level). Small-world (SW) networks (C) feature the functional organization of typically developed brain networks and incorporate characteristics of both random and regular networks, achieving an optimal balance between global and local characteristics. SW networks (C) are associated with the broad abilities’ component of general intelligence (higher intelligence level). The pre-intervention DS network (B) showed a random-like, simplified architecture, with impaired segregation and integration, as evidenced by the low CC (random network characteristic) and decreased GE and increased characteristic path length (CPL), respectively. This is in line with previous literature (Ahmadlou et al., ; Anderson et al., ; Pujol et al., ; Vega et al., 2015). The DS pre-network’s (B) integration-related characteristics (lower GE, higher CPL) are not common in random networks. Hence, the DS pre-network (B) is classified as random-like and not entirely random, maintaining an equal distance from random (A) and SW (C) networks. The DS post-network (B → C) exhibits an increase in integration (random network characteristic), as well as segregation (regular network characteristic), so it is interpreted as a step toward an SW-like architecture that highlights a healthier brain organization.

Figure 4.

Down Syndrome–Long Lasting Memories (DS-LLM) Care design and flow of participants with DS. CT, cognitive training; PT, physical training.

Figure 5.

EEG data analysis schematic. Preprocessing (blue): EEG data were interpolated and artifact corrected, visually inspected, high-pass, bandpass, and low-pass filtered. Independent component analysis (ICA), as well as visual inspection, were used to reject artifactual data. Fifteen segments of 4 seconds were randomly selected. Source reconstruction (orange): the data were processed within 0.53–35 Hz frequency range, source reconstructed (4-D LORETA, for all time points), and a previously used 863-node atlas (Paraskevopoulos et al., 2015) was applied to extract the time series of every voxel from every segment per subject, which were used to estimate 1,000 surrogate time series per subject/group. Functional connectivity (green): functional connectivity was computed for every segment of every subject, using the phase transfer entropy metric (PTE), and the 15 matrices of every subject were averaged into one. PTE networks were estimated for every surrogate time series of each subject to create a null model for the rejection of noise in the real data. A network science approach was taken for the computation of graph measures per subject. Group average statistics were calculated to identify the statistically significant differences between post- and pre-intervention networks.