ReviewWhat is reinforced by phasic dopamine signals?
Introduction
When trying to discover the role of a specific component in a complicated system it is helpful to have some understanding of the system's overall function. The nigro-striatal and mesolimbic dopamine (DA) projections are important components of the basal ganglia, which are one of the brain's fundamental processing units. Thus, in this article we will start by considering briefly some prevalent ideas on basal ganglia involvement in action selection and reinforcement learning. Within this context we will proceed to evaluate the specific contribution of the phasic response of DA neurones to the processes of reinforcement learning. Although it is clear that DA projections ascending from the ventral midbrain target numerous structures, including intrinsic nuclei of the basal ganglia, frontal cortex, amygdala, hippocampus, septal area, several thalamic nuclei, and the habenula (Lindvall and Bjorklund, 1974), for the purpose of the present article we will restrict ourselves to discussion of how striatal function may be influenced by phasic DA input. Hopefully, a better understanding of this comparatively well characterised system will provide important clues concerning the role of phasic DA input to other target structures. This paper is an extension of ideas expressed in our recent perspectives article (Redgrave and Gurney, 2006).
Section snippets
Basal ganglia functions
In humans, basal ganglia dysfunction has been associated with numerous debilitating conditions including Parkinson's disease, Huntington's disease, Tourette's syndrome, schizophrenia, attention-deficit disorder, obsessive–compulsive disorder (Crossman, 2000, Fuxe et al., 2006, Ring and SerraMestres, 2002), and many of the addictions (Everitt and Robbins, 2005). However, as with investigating the roles of specific components, to understand and correctly interpret such malfunctions it would be
The source of short-latency visual input to the ventral midbrain
Since most experiments analysing the sensory properties of DA neurones have used visual stimuli (Morris et al., 2004, Morris et al., 2006, Ravel and Richmond, 2006, Satoh et al., 2003, Schultz, 1998, Schultz, 2006, Takikawa et al., 2004), we will concentrate on the potential sources of visual afferents to DA containing regions of the ventral midbrain.
In recent reviews of cortical visual processing (Rousselet et al., 2004, Thorpe and Fabre-Thorpe, 2001) Thorpe et al. indicated that signals
Visual perception in the dorsal midbrain
Electrophysiological responses of neurones in the mammalian superior colliculus are characterised by an exquisite sensitivity to spatially localised luminance changes in the retina (Grantyn, 1988, Sparks, 1986, Stein and Meredith, 1993, Wurtz and Albano, 1980). Such changes typically signify the appearance, disappearance or movement of something in the visual field. While there may be cell-types that can perform broad species specific stimulus classifications (de Gelder and Hadjikhani, 2006,
Visual perception in the ventral midbrain
Given that the superior colliculus is specialised to inform its efferent targets where something has occurred in the visual field, not what has occurred, it would be appropriate to pause and consider how DA neurones seem able to perform the detailed visual processing required to discriminate the complex conditioned visual stimuli that have been used to signal different reward magnitudes and probabilities (Fiorillo et al., 2003, Tobler et al., 2003, Tobler et al., 2005). Careful reading of
Agency determination and discovery of novel actions
The following discussion will extend our recent proposal (Redgrave and Gurney, 2006) that sensory driven phasic DA signals reinforce the discovery of events for which the animal, or more generally the agent, is responsible. Reinforcement of relevant signals within the presumed selection architecture of the basal ganglia (Fig. 2B) could enable the system to converge on aspects of context and behavioural output responsible for eliciting initially unpredicted sensory events.
The fog lifts a little?
Having identified a specific reinforcing role for DA in a connectional architecture whose function is to determine sources of agency and discover novel actions, we will now consider several hitherto puzzling issues.
Summary and conclusions
This paper is an elaboration of our recent proposal that sensory driven DA responses provide reinforcement signals required for the brain to discriminate the sensory events for which it is responsible (Redgrave and Gurney, 2006). As part of this process new responses required in specific circumstances to make events happen are discovered. A critical feature is that the proposed architecture requires a minimum of sensory processing to generate the necessary reinforcing signals. All the system
Acknowledgments
The authors would like to acknowledge the contributions of Paul Overton, Veronique Coizet, John McHaffie and Terry Stanford throughout the progressive evolution of these ideas. This review was written while the authors were in receipt of research funding from The Wellcome Trust (PR), BBSRC (PR), EPSRC (KG, PR), and the Marsden Fund of the Royal Society of New Zealand (JR, PR).
References (165)
- et al.
Space, time and dopamine
Trends Neurosci.
(2007) - et al.
Parallel incentive processing: an integrated view of amygdala function
Trends Neurosci.
(2006) - et al.
Information processing, dimensionality reduction and reinforcement learning in the basal ganglia
Prog. Neurobiol.
(2003) - et al.
Midbrain dopamine neurons encode a quantitative reward prediction error signal
Neuron
(2005) - et al.
Novel-object place conditioning: behavioral and dopaminergic processes in expression of novelty reward
Behav. Brain Res.
(2002) - et al.
Reinforcement delay of one second severely impairs acquisition of brain self-stimulation
Brain Res.
(1985) - et al.
Dopamine-mediated regulation of corticostriatal synaptic plasticity
Trends Neurosci.
(2007) - et al.
Nociceptive responses of midbrain dopaminergic neurones are modulated by the superior colliculus in the rat
Neuroscience
(2006) - et al.
The role of prelimbic cortex in instrumental conditioning
Behav. Brain Res.
(2003) - et al.
Prediction error during retrospective revaluation of causal associations in humans: fMRI evidence in favor of an associative model of learning
Neuron
(2004)