Multivariate representation of food preferences in the human brain
Introduction
In classical economics, preferences are usually determined by a revealed preference approach, e.g. by performing choices between alternatives (Von Neumann & Morgenstern, 1944). Related to this, humans facing a binary choice are often assumed to make their decisions based on the computation of subjective values (SV). SVs are proposed to serve as a common currency that allow the comparison of complex and qualitatively different alternatives on a common scale (Bartra et al., 2013, Kahneman and Tversky, 1979, Samuelson, 1937). To be more precise, during decision making a scalar subjective value is assumed to be computed for each alternative first, and the one with the greatest SV will be chosen subsequently. One major goal in decision neuroscience is to investigate the underlying neural computations and processes of subjective values and revealed preferences.
Functional MRI studies using a univariate approach have consistently revealed the medial prefrontal cortex (mPFC) to be highly involved in the computation of value related and preference related signals (Bartra et al., 2013, Chib et al., 2009, FitzGerald et al., 2009, Kim et al., 2011, Lebreton et al., 2009, Levy and Glimcher, 2011, Lin et al., 2012, Peters and Büchel, 2010, Smith et al., 2010). Furthermore, these studies revealed an overlap of value related activation for different types of stimuli (e.g. consumer goods, monetary rewards, social rewards), especially in the ventral parts of mPFC which has led to the hypothesis that a common value representation for objects from different categories might be computed in the ventro medial prefrontal cortex (vmPFC). Multi voxel pattern analysis approaches (MVPA) were used to provide further evidence for this hypothesis (Pereira, Mitchell, & Botvinick, 2009). The multivariate approach enabled to investigate whether preferences for products/objects/activities belonging to identical or different categories can be predicted significantly above chance from the coordinated activation of multiple voxels in certain regions of interest.
The first MVPA study in this field, conducted by Tusche, Bode, and Haynes (2010), revealed that consumer choices for different types of cars can be predicted from voxel clusters in mPFC (dorsal and ventral parts), occipital areas and the insula. However, this study allowed only inferences about the involvement of certain brain regions in case of a within category prediction because only one product class (i.e. cars) was used. Another MVPA study, conducted by McNamee, Rangel, and O’Doherty (2013), investigated whether a classifier cannot only decode stimulus/category dependent value representations but also across category value patterns by using a monetary valuation/bidding paradigm. To test for the presence of across category value signals, the classifier was trained to decode SVs from samples drawn from one of three stimulus categories (i.e. food, money, noncomestible consumer items) and tested in predicting the SV of the stimuli from the remaining two categories. In order to test for category dependent value codes, the classifier was trained and tested on stimuli of the same category. Data analysis revealed across category value patterns in the inferior parts of vmPFC which is also referred to as medial orbitofrontal cortex (mOFC) by some authors. In contrast, category dependent value codes were found in the superior regions of the vmPFC (i.e. the area above the mOFC). The finding of value related signals in the OFC was also confirmed by a MVPA study conducted by Kahnt, Park, Haynes, and Tobler (2014), which primarily focused on disentangling neural representations of value and salience. The most recent MVPA study in this field, conducted by Gross et al. (2014), dealt with the question whether individual preferences can be predicted across fundamentally different categories (i.e. snack foods, engaging activities) in the absence of monetary evaluation. To that end, subjects were presented with different activities and snack foods in written form during fMRI scanning and instructed to imagine the pleasure they would derive from eating the snack item or engaging in the activity. Data analysis located voxels carrying value signals across categories in the anterior and dorsal parts of mPFC (bilateral) as well as in the anterior cingulate cortex (ACC). Most interestingly, no cluster was found in the ventral parts of the mPFC.
The finding that preferences can be predicted for different types of objects/stimuli by functional MRI is very promising for clinical and neuro commercial applications. Especially the domain of predicting food preferences might be very useful from a clinical point of view in the domain of obesity research/prevention and should also be of a high interest in the domain of consumer neuroscience. However, previous MVPA studies did not focus on the prediction of food related value/preference signals per se but rather on more global aspects of value computation and preference prediction across utterly different stimulus categories. Because foods constitute primary rewards being linked to gustatory perception and serve the intake of nutrients and electrolytes, there might be some domain specific characteristics in neuronal processing that might not emerge in case of across category predictions beyond the domain of food products.
Thus, in the present study we wanted to investigate whether similar patterns of brain activity can be observed in case of different food products that are of more similar category than the classes of stimuli used in previous studies. To that end we used a multivariate whole brain searchlight approach (Kriegeskorte, Goebel, & Bandettini, 2006): In a first step participants were asked for their preferences regarding 20 different chocolate bars and 20 different salty snacks by means of a computer based pretest. In a second step, a classifier was trained to decode product preferences from fMRI data recorded during the presentation of preferred and non-preferred chocolate snacks. In a third step, we assessed the classifier’s performance in predicting preferences for products of the same category (within category prediction) as well as for products belonging to a very similar, although different, category (i.e. salty snacks; across category prediction). Voxel clusters having a predictive power above chance level in case of the within category prediction are assumed to reflect category dependent value/preference signals whereas predictive clusters in case of the across category classification indicate category independent preference signals, as put forward by previous studies (Gross et al., 2014, McNamee et al., 2013). However, we also applied an even stronger criterion for identifying category independent preference signals by selecting only those voxels showing a predictive power significantly above chance in case of the within and the across category prediction (i.e. voxels predicting preferences in case of both conditions significantly above chance no matter which stimulus class the classifier was initially trained on).
Section snippets
Participants
Sixteen healthy subjects (i.e. no neurological, psychiatric or gastrointestinal diseases, no eating disorders) participated in the experiment (6 female; mean age = 23.93; SD = 4.13). The mean body weight across all participants was 70.18 kg (SD = 11.96). The average body height was 1.74 m (SD = 0.09). This resulted in a mean body mass index (BMI) of 22.99 (SD = 2.23). All participants were neither vegans nor did they report any food allergies or dietary restrictions. The experiment was conducted with the
Computer based preference ranking
For both product categories we found a high negative correlation between reaction time and difference in rank (chocolate/snacks: ρ = −0.96/−0.90, p = 1.71 ∗ 10−11/9.96 ∗ 10−8). Thus, choices between products with similar preference ranks required longer reaction times, than choices between products which were more distinct in their ranking.
Product ratings in the scanner
As expected, there was a significant difference between the averaged ratings of preferred products (chocolate/snacks: M = 1.3/1.3, SD = 0.31/0.14) and non-preferred
Discussion
With the present MVPA study we wanted to investigate whether similar patterns of brain activity can be observed while subjects evaluate food products from different categories. To that end we used a multivariate searchlight approach in which a linear support vector machine (l-SVM) was trained to distinguish preferred from non-preferred chocolate bars and subsequently tested its predictive power in case of chocolate bars (within category prediction) and salty snacks (across category prediction).
Conflict of interest
The authors declare no competing financial interests.
Acknowledgments
LP is funded by a doctoral scholarship from the Studienstiftung des deutschen Volkes.
BW is funded by a Heisenberg Grant from the Deutsche Forschungsgemeinschaft (We 4427/3-2).
FM is funded by a Lichtenberg Grant from the Volkswagen Foundation and by a Grant from the Deutsche Forschungsgemeinschaft (Mo 930/4-1).
References (54)
- et al.
Neural activity in the primate prefrontal cortex during associative learning
Neuron
(1998) - et al.
The valuation system: A coordinate-based meta-analysis of BOLD fMRI experiments examining neural correlates of subjective value
NeuroImage
(2013) The multiple-demand (MD) system of the primate brain: mental programs for intelligent behaviour
Trends in Cognitive Sciences
(2010)The structure of cognition: Attentional episodes in mind and brain
Neuron
(2013)- et al.
The impact of certain methodological choices on multivariate analysis of fMRI data with support vector machines
NeuroImage
(2011) - et al.
Transcranial direct current stimulation of the prefrontal cortex modulates the desire for specific foods
Appetite
(2008) - et al.
An automatic valuation system in the human brain: Evidence from functional neuroimaging
Neuron
(2009) - et al.
Effect of expected reward magnitude on the response of neurons in the dorsolateral prefrontal cortex of the macaque
Neuron
(1999) - et al.
The neural dissociation of subjective valuation from choice processes in intertemporal choice
Behavioural Brain Research
(2012) - et al.
Neural correlates of behavioral preference for culturally familiar drinks
Neuron
(2004)
Valid conjunction inference with the minimum statistic
NeuroImage
Neural representations of subjective reward value
Behavioural Brain Research
The general linear model and fMRI: Does love last forever?
NeuroImage
Heterogeneous reward signals in prefrontal cortex
Current Opinion in Neurobiology
Attention enhances multi-voxel representation of novel objects in frontal, parietal and visual cortices
NeuroImage
Sparse kernel machines
Neuronal encoding of subjective value in dorsal and ventral anterior cingulate cortex
Journal of Neuroscience
Repetitive transcranial magnetic stimulation over the right dorsolateral prefrontal cortex decreases valuations during food choices
The European Journal of Neuroscience
Evidence for a common representation of decision values for dissimilar goods in human ventromedial prefrontal cortex
Journal of Neuroscience
An adaptive coding model of neural function in prefrontal cortex
Nature Reviews Neuroscience
The role of human orbitofrontal cortex in value comparison for incommensurable objects
The Journal of Neuroscience
Improved assessment of significant activation in functional magnetic resonance imaging (fMRI): Use of a cluster-size threshold
Magnetic Resonance in Medicine
Value signals in the prefrontal cortex predict individual preferences across reward categories
The Journal of Neuroscience
PyMVPA: A python toolbox for multivariate pattern analysis of fMRI data
Neuroinformatics
Monkey prefrontal neuronal activity coding the forthcoming saccade in an oculomotor delayed matching-to-sample task
Journal of Neurophysiology
Statistical analysis: Basic analyses
A functional hierarchy within the parietofrontal network in stimulus selection and attention control
The Journal of Neuroscience
Cited by (12)
Social, self, (situational), and affective processes in medial prefrontal cortex (MPFC): Causal, multivariate, and reverse inference evidence
2019, Neuroscience and Biobehavioral ReviewsCitation Excerpt :MVPA cross-classification across reward categories would seem to be one of the holy grails of true valuation. Four MVPA studies have now examined the common scale notion (Gross et al., 2014; Howard et al., 2015; McNamee et al., 2013; Pogoda et al., 2016). Somewhat surprisingly, these studies consistently point to AMPFC rather than VMPFC as central to valuation independent of stimulus class.
A decade of decoding reward-related fMRI signals and where we go from here
2018, NeuroImageCitation Excerpt :Supporting the idea of a general value signal in the vmPFC, significant cross-category decoding of value was found in this region (Fig. 3C). Several studies have replicated category-independent value signals in the medial PFC using different reward categories (Chikazoe et al., 2014; Gross et al., 2014; Pegors et al., 2015; Pogoda et al., 2016), and similar approaches have been used to show that perceived and inferred emotional states share representations in this area (Skerry and Saxe, 2014). These studies have in common that they use different reward identities with varying levels of subjective value in one category in order to define a value-coding activity pattern.
Thoughts for food in cognitive neuroscience: An introduction to the special issue
2016, Brain and CognitionThe determinants of food choice
2017, Proceedings of the Nutrition Society