Decoding different roles for vmPFC and dlPFC in multi-attribute decision making
Introduction
Successful decision making requires precise anticipatory representations of the reward values that can be obtained from choosing specific behavioral options. However, in everyday life most decision alternatives consist of multiple reward-related attributes. For instance, different attributes of a fruit–size, shape, color and surface texture–signal parts of its nutritional value. To make an optimal choice i.e. to pick the fruit with the highest expected value, the reward predictions of all attributes need to be integrated into a combined value. To describe such decision processes, Multi-Attribute Utility Theory (MAUT) was developed by behavioral decision researchers in the 1970s (Slovic et al., 1977, von Winterfeldt and Fischer, 1975).
Neuroscience has mainly focused on decisions regarding single-attribute options (Daw et al., 2006, Glascher et al., 2009, Hampton et al., 2006, Kim et al., 2006, O'Doherty et al., 2003b). Studies on multiple attributes have typically directly investigated the trade-off between two attributes such as taste vs. health (Hare et al., 2009), amount of money vs. delay (Kable and Glimcher, 2007) and pleasure of acquisition vs. price (Knutson et al., 2007). Furthermore, studies on decisions between real-life objects (comprising multiple attributes) typically did not address their multiple-attribute character explicitly (Chib et al., 2009, FitzGerald et al., 2009, Hare et al., 2009, Knutson et al., 2007, Plassmann et al., 2007). One study aimed to identify brain regions involved in experimentally controlled multi-attribute decisions (Zysset et al., 2006). In this study, however, only the attribute-wise similarity between alternatives i.e. the difficulty of the decision was examined. Taken together, although single- and two-attribute decisions have been studied, no study has moved beyond two attributes and comprehensively investigated how such multi-attribute objects are represented in the brain.
Each attribute of a multi-attribute object can have its own predictive information for reward. Importantly, different attributes of the same object can signal different or even conflicting reward values. For instance, for one object all attributes could signal an intermediate value, whereas for another object different attributes could signal high and low values. Thus, although both objects have the same combined value (i.e. intermediate) the multi-attribute objects would differ considerably in the variability of the rewards predicted by their individual attributes (i.e. low vs. high). Hence, unlike single-attribute objects, different multi-attribute objects can differ not only in their expected value but also in the variability of the rewards predicted by their attributes. In order to understand how decisions are made on the basis of multi-attribute objects we investigated how the combined value and the variability of the rewards predicted by the individual attributes are represented in the human brain.
Recently we have shown that expected values can be decoded from distributed fMRI patterns in the ventromedial prefrontal cortex (vmPFC) (Kahnt et al., 2010). This distributed coding is consistent with reports from single-unit recordings showing that different neural populations in value sensitive cortex increase and decrease their firing rate with increasing reward value, respectively (Kennerley et al., 2009, Kobayashi et al., 2010, Morrison and Salzman, 2009, Padoa-Schioppa and Assad, 2006, Schoenbaum et al., 2007). Previous experimental and theoretical work on the human visual system has revealed that applying multivariate pattern analysis (MVPA) techniques to fMRI data is specifically suited to extract information encoded in distributed neural populations (Haynes and Rees, 2005, Haynes and Rees, 2006, Kamitani and Tong, 2005, Norman et al., 2006). Similarly, information about cognitive and decision processes has been shown to be encoded in distributed fMRI patterns in the prefrontal cortex (PFC) (Hampton and O'Doherty, 2007, Haynes et al., 2007, Soon et al., 2008). Thus, it might be expected that distributed fMRI patterns contain more information about the combined value of multi-attribute objects and the variability of rewards predicted by individual attributes than the average fMRI signal. Hence, here we used MVPA techniques (Haynes and Rees, 2006, Norman et al., 2006) to decode information about these two variables.
Section snippets
Participants
Sixteen right-handed subjects (8 female, mean age = 26.4 ± 1.06 years SEM) participated in the experiment. Subjects had normal or corrected-to-normal vision and gave written informed consent to participate. The study was approved by the local ethics review board of the Charité-Universitätsmedizin Berlin.
Classical conditioning session
In all experiments we used objects that could vary in three visual attributes shape, color and coherence of moving dots with three levels per attribute. In the days prior to scanning (mean 3.19 ± 0.31
Results
In this section we first report the behavioral results and show how the combined value and the variability of the multi-attribute objects are reflected in subjects' behavior. We then proceed to show how the combined value and variability are encoded in the brain. We find that non-overlapping brain regions, the ventromedial PFC (vmPFC) and the dorsolateral PFC (dlPFC), respectively, contain information about these variables that characterize multi-attribute objects.
Discussion
In the current experiment we demonstrate that two variables characterizing multi-attribute objects are encoded in different brain regions, the vmPFC and the dlPFC. Whereas the combined value is represented in the vmPFC, the variability of the reward predictions of the individual attributes is encoded in the dlPFC. A previous study identified a network of brain regions including the medial PFC and the dlPFC that was involved in multi-attribute decision making (Zysset et al., 2006). Based on the
Acknowledgments
This work was funded by the Bernstein Computational Neuroscience Program of the German Federal Ministry of Education and Research (BMBF Grant 01GQ0411), the Excellence Initiative of the German Federal Ministry of Education and Research (DFG Grant GSC86/1-2009) and the Max Planck Society.
References (55)
- et al.
Midbrain dopamine neurons encode a quantitative reward prediction error signal
Neuron
(2005) - et al.
Decoding sequential stages of task preparation in the human brain
NeuroImage
(2009) - et al.
Functional imaging of neural responses to expectancy and experience of monetary gains and losses
Neuron
(2001) - et al.
Amygdala tractography predicts functional connectivity and learning during feedback-guided decision-making
NeuroImage
(2008) - et al.
Reading hidden intentions in the human brain
Curr. Biol.
(2007) - et al.
Neural predictors of purchases
Neuron
(2007) - et al.
Temporal prediction errors in a passive learning task activate human striatum
Neuron
(2003) - et al.
Neural economics and the biological substrates of valuation
Neuron
(2002) - et al.
Beyond mind-reading: multi-voxel pattern analysis of fMRI data
Trends Cogn. Sci.
(2006) - et al.
Temporal difference models and reward-related learning in the human brain
Neuron
(2003)