Thalamic Network Controllability Predicts Cognitive Impairment in Multiple Sclerosis

Abstract

Recent research suggests that individuals with multiple sclerosis (MS) and cognitive impairment exhibit more effortful and less efficient transitions in brain network activity. Previous studies further highlight the increased vulnerability of specific regions, particularly the thalamus, to disease-related damage. This study investigates whether MS affects the controllability of specific brain regions in driving network activity transitions across the brain and examines the relationship between these changes and cognitive impairment in patients. Resting-state functional MRI and neuropsychological data were collected from 102 MS and 27 healthy controls. Functional network controllability analysis was performed to quantify how specific regions influence transitions between brain activity patterns or states. Disease alterations in controllability were assessed in the main dataset and then replicated in an independent dataset of 95 MS and 45 healthy controls. Controllability metrics were then used to distinguish MS from healthy controls and predict cognitive status. MS-specific controllability changes were observed in the subcortical network, particularly the thalamus, which were further confirmed in the replication dataset. Cognitively impaired patients showed significantly greater difficulty in the thalamus steering brain transitions towards difficult-to-reach states, which are typically associated with high-energy-cost cognitive functions. Thalamic network controllability proved more effective than thalamic volume in distinguishing MS from healthy controls (AUC = 88.3%), and in predicting cognitive status in MS (AUC = 80.7%). This study builds on previous research highlighting early thalamic damage in MS, aiming to demonstrate how this damage disrupts activity transitions across the cerebrum and may predict cognitive deficits. Our findings suggest that the thalamus in MS becomes less capable of facilitating broader brain activity transitions essential for high-energy-cost cognitive functions, implying a potential pathological mechanism that links thalamic functional changes to cognitive impairment in MS.

Keywords: cognitive impairment; multiple sclerosis; network controllability; resting‐state fMRI; subcortical network; thalamus.

FIGURE 1

Overview of network controllability calculation. Based on rs‐fMRI data from 129 participants and using combined cortical and subcortical parcellations, we extracted regional mean time courses and calculated functional connectivity between pairs of regions to construct brain networks. We then calculated the most commonly used controllability measures to quantify how specific regions influence brain activity transitions throughout the brain. Specifically, average controllability measures how easily a brain region moves the brain into nearby or easily reachable states, which reflects the brain's capacity for low‐energy‐cost and frequent small adjustments in brain states. Modal controllability, on the contrary, measures a region's ability to move the brain into difficult or unstable states, which is important for executing high‐energy‐cost large transitions in brain activity (e.g., between resting and active states). Regional activation energy captures the feasibility or minimum energy required by the given region to induce a transition between brain states.

FIGURE 2

Controllability changes in the subcortical network in MS. Increased average controllability but decreased modal controllability and decreased activation energy in the subcortical network (averaged across the 54 subcortical ROIs) were observed in MS from the main dataset. Replicated increased average controllability in the subcortical network was observed in MS from the replication dataset. Brain network visualizations were generated using BrainNet Viewer (Xia et al. 2013) and GRETNA (Wang et al. 2015). The nodes and edges illustrate the connections of subcortical regions with other parts of the brain. Specifically, the nodes represent the subcortical regions and the regions to which they are connected. The edges depict the top 200 connections with the highest functional connectivity strength among the whole brain connections. SubCor = subcortical network; *p < 0.05, FDR corrected.

FIGURE 3

Controllability changes in the thalamus in MS. Increased average controllability but decreased modal controllability and decreased activation energy in the thalamus (averaged across the 16 thalamic ROIs) were observed in MS from the main dataset. Replicated increased average controllability in the thalamus was observed in MS from the replication dataset. Brain network visualizations were generated using BrainNet Viewer (Xia et al. 2013) and GRETNA (Wang et al. 2015). The nodes and edges illustrate the connections of thalamic regions with other parts of the brain. Specifically, the nodes represent the thalamic regions and the regions to which they are connected. The edges depict the top 200 connections with the highest functional connectivity strength among the whole brain connections. SubCor = subcortical network; *p < 0.05, FDR corrected.

FIGURE 4

Controllability changes in the subcortical network in CIMS and CPMS. Increased average controllability but decreased modal controllability and decreased activation energy in the subcortical network (averaged across the 54 subcortical ROIs) in both CIMS and CPMS when compared to HC. Brain network visualizations were generated using BrainNet Viewer (Xia et al. 2013) and GRETNA (Wang et al. 2015). The nodes and edges illustrate the connections of subcortical regions with other parts of the brain. Specifically, the nodes represent the subcortical regions and the regions to which they are connected. The edges depict the top 200 connections with the highest functional connectivity strength among the whole brain connections. SubCor = subcortical network; *p < 0.05, FDR corrected.

FIGURE 5

Controllability changes in the thalamus in CIMS and CPMS. Both CIMS and CPMS showed increased average controllability in the thalamus (averaged across the 16 thalamic ROIs) when compared to HC, while CIMS exhibited significantly greater changes than CPMS. Besides, CIMS showed additional decreases in modal controllability and activation energy in the thalamus compared to HC. Brain network visualizations were generated using BrainNet Viewer (Xia et al. 2013) and GRETNA (Wang et al. 2015). The nodes and edges illustrate the connections of thalamic regions with other parts of the brain. Specifically, the nodes represent the thalamic regions and the regions to which they are connected. The edges depict the top 200 connections with the highest functional connectivity strength among the whole brain connections. SubCor = subcortical network; *p < 0.05, FDR corrected.

FIGURE 6

Classification ROC curve derived from volumetric, controllability, combined, and random classifiers. In distinguishing MS from HC, classifiers based on the thalamic network controllability alongside thalamic volume achieved the best performance (AUC = 88.3%) among all four types of classifiers. In distinguishing CIMS from CPMS, classifiers based on the thalamic network controllability measures alone achieved the best performance (AUC = 80.7%) among all four types of classifiers. AUC, area under curve; ROC, receiver operating characteristic.