Cognitive effort devaluation and the salience network: a computational model of amotivation in depression

Amotivation in depression is linked to impaired reinforcement learning and effort expenditure via the dopaminergic reward pathway. To understand its computational and neural basis, we modeled incentive, temporal and cognitive burden effects, identifying key components and brain networks of cost-benefit valuation.

Data from 43 psychotropic-free individuals (31 non- or minimally depressed individuals), including Beck Depression Inventory (BDI), Apathy Evaluation Scale (AES), n-back task performance, and resting-state fMRI, were analyzed. Cost-benefit valuation was modeled using loss aversion, learning, temporal, and cognitive effort discounting factors. Model fitting and comparison (two-learning rate vs. two-temporal discounting) were performed. Principal Component Analysis and linear regression identified factors predicting amotivation severity. Correlations of estimated factors with nucleus accumbens and anterior insular cortex (AIC) functional connectivity were analyzed.

Overall, greater 2-back than 0-back accuracy occurred in longer, positively incentivized tasks. Non- or minimally depressed individuals showed accuracy difference by N-back load at higher rewards, with divergence between reward and loss tasks at higher incentive and longer lengths. The two-temporal discounting model best explained these results. Cognitive effort discounting specifically predicted amotivation scores, derived from BDI and AES, and correlated with AIC-anterior mid cingulate cortex (aMCC) functional connectivity.

Our findings demonstrate amotivation is specifically associated with cognitive effort devaluation in a cost-benefit analysis incorporating loss aversion, incentive learning, temporal discounting, and cognitive effort discounting. Modulation of effort valuation via the AIC-aMCC network suggests a potential treatment target.