Background: Working memory deficits are a common feature in major depressive disorder and are associated with poor functional outcomes. Intact working memory performance requires the recruitment of large-scale brain networks. However, it is unknown how the disrupted recruitment of distributed regions belonging to these large-scale networks at the whole-brain level brings about working memory impairment seen in major depressive disorder. Methods: We used graph theory to examine the functional connectomic metrics (local and global efficiency) at the whole-brain and large-scale network levels in 38 patients with major depressive disorder and 41 healthy controls during a working memory task. Altered connectomic metrics were studied in a moderation model relating to clinical symptoms and working memory accuracy in patients, and a machine learning method was employed to assess whether these metrics carry enough illness-specific information to discriminate patients from controls. Results: Global efficiency of the frontoparietal network was reduced in major depressive disorder (false discovery rate corrected, p = 0.014); this reduction predicted worse working memory performance in patients with less severe illness burden indexed by Brief Psychiatric Rating Scale (β =–0.43, p = 0.035, t =–2.2, 95% confidence interval = [–0.043,–0.002]). We achieved a classification accuracy and area under the curve of 73.42% and 0.734, respectively, to discriminate patients from controls based on connectomic metrics, and the global efficiency of the frontoparietal network contributed most to the diagnostic classification. Conclusions: We report a putative mechanistic link between the global efficiency of the frontoparietal network and impaired n-back performance in major depressive disorder. This relationship is more pronounced at lower levels of symptom burden, indicating the possibility of multiple pathways to cognitive deficits in severe major depressive disorder.
