Altered triple network model connectivity is associated with cognitive function and depressive symptoms in older adults

Abstract

Introduction: Late-life cognitive impairment and depression frequently co-occur and share many symptoms. However, the specific neural and clinical factors contributing to both their common and distinct profiles in older adults remain unclear.

Methods: We investigated resting-state correlates of cognitive and depressive symptoms in older adults (n = 248 and n = 95) using clinical, blood, and neuroimaging data. We computed a connectivity matrix across default mode, executive control, and salience networks. Cross-validated elastic net regression identified features reflecting cognitive function and depressive symptoms. These features were validated on a held-out dataset.

Results: We discovered that white matter hyperintensities and nine overlapping nodes spanning all three networks are associated with both cognitive function and depressive symptoms, including left amygdala, left hippocampus, and bilateral ventral tegmental area.

Discussion: Our findings reveal intertwined neural nodes influencing cognitive impairment and depressive symptoms in late life, offering insights into shared characteristics and potential therapeutic targets.

Highlights: Resting-state neuroimaging markers are associated with symptoms of cognitive decline and late-life depression. Symptom-associated connectivity alterations were present across three major brain networks of interest, including the salience, default mode, and executive control networks. Some regions of interest are associated with both cognitive function and depressive symptoms, including the left amygdala, left hippocampus, and bilateral ventral tegmental area.

Keywords: Alzheimer's disease; depression; machine learning; resting state connectivity.

FIGURE 1

(Step 1) We chose target nodes from the default mode, executive control, and SNs and computed their ROI–ROI connectivity matrix based on resting‐state scans. (Step 2) We included 343 participants with high‐quality clinical, demographic, neuroimaging, and blood samples, dividing them into train, test, and validate datasets. The data was used in an elastic net model with 5‐fold cross‐validation and a grid search for hyperparameter optimization. (Step 3) Using the glmnet package in R, we performed two 5‐fold repeated cross‐validation procedures with 500 repetitions each, training the model and (Step 4) iteratively assessing its performance on the test dataset. (Step 5) The model was then validated using a separate validation dataset. The red line in steps 3 through 5 represents the idealized line of perfect agreement between actual and predicted values in the scatterplot. fMRI, functional magnetic resonance imaging; MADRS, Montgomery–Asberg Depression Rating Scale; MMSE, Mini‐Mental State Examination; ROI, region of interest; SNs, salience networks

FIGURE 2

A, R and MAE and MSE values for the train, test, and validate datasets in the model predicting MMSE. B, Standardized coefficients from the elastic net regression used to predict MMSE scores. C, A circular plot demonstrating with and between network connectivity selected by the elastic net. Sample seeds used for the ROI–ROI connectivity are shown above the outer layer of the graph. B INS, bilateral insula; B IPG, bilateral inferior parietal gyrus; B PCG, bilateral posterior cingulate gyrus; B PreC, bilateral precuneus; B STG, bilateral superior temporal gyrus; B VTA, bilateral ventral tegmental area; B, bilateral; L AMY, left amygdala; L Hipp, left hippocampus; L ITG, left inferior temporal gyrus; L, left; MAE, mean absolute error; MMSE, Mini‐Mental State Examination; MSE, mean squared error; R IOFG, right inferior orbital frontal gyrus; R ITG, right inferior temporal gyrus; R, right; ROI, region of interest

FIGURE 3

A, R and MAE and MSE values for the train, test, and validate datasets in the model predicting MADRS. B, Standardized coefficients from the elastic net regression used to predict MADRS scores. C, A circular plot demonstrating with and between network connectivity selected by the elastic net. Sample seeds used for the ROI–ROI connectivity are shown above the outer layer of the graph. B INS, bilateral insula; B IPG, bilateral inferior parietal gyrus; B MFG, bilateral middle frontal gyrus; B MidCIN, bilateral median cingulate; B PCG, bilateral posterior cingulate gyrus; B PreC, bilateral precuneus; B SMTA, bilateral supplementary motor area; B STG, bilateral superior temporal gyrus; B VTA, bilateral ventral tegmental area; B, bilateral; L AMY, left amygdala; L HIPP, left hippocampus; L IOFG, left inferior orbital frontal gyrus; L SNpr, left substantia nigra pars reticulata; L, left; MADRS, Montgomery–Asberg Depression Rating Scale; MAE, mean absolute error; MSE, mean squared error; R AMY, right amygdala; R ITG, right inferior temporal gyrus; R MTG, right middle temporal gyrus; R SNpr, right substantia nigra pars reticulata; R, right

FIGURE 4

A, Overlapping symptoms between late‐life depression and mild or MNCD as described by previous research. B, rsFC linked to cognitive function and depression severity, based on elastic net results. Seed color indicates network belonging; blue: SN, cyan: DMN, orange: ECN, yellow: multiple networks. Red lines suggest higher connectivity was associated with higher MMSE/MADRS. Blue lines suggest lower connectivity was associated with lower MMSE/MADRS. C, Shared versus unique nodes with rsFC associated with cognitive function and depressive severity. DMN, default mode network; MADRS, Montgomery–Asberg Depression Rating Scale; MMSE, Mini‐Mental State Examination; MNCD, major neurocognitive disorder; rsFC, resting state functional connectivity; SN, salience network