Attenuation of dopamine-modulated prefrontal value signals underlies probabilistic reward learning deficits in old age.

Lieke De Boer,Lars Bäckman,Marc Guitart-Masip,Peter Dayan,Jan Axelsson,Katrine Riklund,Lars Nyberg

doi:10.7554/elife.26424

Lieke De Boer, Lars Bäckman + Show 5 more

Open Access

https://doi.org/10.7554/elife.26424

Copy DOI

Abstract

Probabilistic reward learning is characterised by individual differences that become acute in aging. This may be due to age-related dopamine (DA) decline affecting neural processing in striatum, prefrontal cortex, or both. We examined this by administering a probabilistic reward learning task to younger and older adults, and combining computational modelling of behaviour, fMRI and PET measurements of DA D1 availability. We found that anticipatory value signals in ventromedial prefrontal cortex (vmPFC) were attenuated in older adults. The strength of this signal predicted performance beyond age and was modulated by D1 availability in nucleus accumbens. These results uncover that a value-anticipation mechanism in vmPFC declines in aging, and that this mechanism is associated with DA D1 receptor availability.

Highlights

In order to navigate an uncertain world successfully, humans and other animals are required to learn and update the values of available actions and switch between them appropriately
We did this by (1) assessing differences between age groups in the BOLD signal related to anticipatory expected value in the ventromedial prefrontal cortex (vmPFC), (2) assessing differences between age groups in the BOLD signal related to reward prediction errors (RPEs) in the nucleus accumbens (NAcc), and (3) investigating the de Boer et al eLife 2017;6:e26424
SD = 5.25; Myoung = 6.44, SD = 6.07; t(55) = 2.38; p=0.021). This difference in vmPFC value signal did not arise because of the difference in learning performance: when we restricted our analysis to high performers as defined by a median split (13 old, 15 young), a difference in performance was no longer significant (p=0.60), but the strength of expected-value signal in vmPFC was correlated with age (r(26) = À0.39, p=0.040) and we found a marginally significant difference between age groups (Mold = 4.21, SD = 4.81; Myoung = 8.29, SD = 5.72; t(26) = 2.03, p=0.054)

Summary

Introduction

In order to navigate an uncertain world successfully, humans and other animals are required to learn and update the values of available actions and switch between them appropriately. Older individuals are poor at probabilistic reward learning and subsequent optimal action selection (Eppinger et al, 2011; Mell et al, 2005). Previous studies reported lower correlations between RPEs generated from probabilistic reward learning tasks and nucleus accumbens (NAcc) BOLD signals in older compared with younger adults (Eppinger et al, 2013; Samanez-Larkin et al, 2014). By decomposing RPEs in a dynamic twoarmed bandit task into their two subcomponents: obtained reward (R) and expected value (Q), Chowdhury et al (2013) showed that, in older adults, neural activity in NAcc reflected just the de Boer et al eLife 2017;6:e26424.

Objectives

Methods

Results

Discussion

Conclusion