Study of the effect of positive and negative correlations on functional connectivity disruption in MCI

INTRODUCTION Graph theory is a state of the art approach to assess functional connectivity (FC) disruption in neurological disorders. However, this methodology does not allow distinction between positive and negative FC correlations; negative correlations, or anticorrelations, are usually either discardes or taken as positive values, and therefore its influence is rarely studied. RESEARCH QUESTION In the present study, we used magnetoencephalography (MEG) to construct cross-frequency networks, and studied the positive and negative correlations independently and jointly, to evaluate the contribution of each to the FC changes in Mild Cognitive Impairment (MCI) patients. MATERIALS AND METHODS To this end, we performed eyes-closed resting-state MEG in 105 MCI patients and 172 controls. Cross-frequency networks were constructed by calculating the absolute value of the corrected amplitude envelope correlation between 1206 cortical sources in five frequency bands (delta, theta, alpha, beta and gamma), considering both intra- and inter-band connections. As a result, a single network (complete network) was constructed for each subject, with each source represented in all bands. Two additional networks were constructed: one just with the positive correlations (positive network), and one just with the negative ones (negative network). Subsequently, several centrality measures were calculated for the three networks of each subject, and the results were compared. RESULTS About four fifths of the links were positive; thus, the results for the complete and positive networks were similar: centrality was increased in the slow bands (delta and theta) in MCI patients with respect to controls and decreased in the fast bands. The negative network, on the other side, showed quite an opposite behavior: a centrality decrease in delta, and a centrality increase in theta, alpha and gamma. CONCLUSION The results for the complete network are consistent with previous literature. Furthermore, the dissimilarities found between the positive and negative networks highlight the need to study correlations and anticorrelations independently.