Nutrition labels influence value computation of food products in the ventromedial prefrontal cortex
Funding agencies: This work was partly supported by the BMBF conceptual phase for the competence-cluster DietBB (01EA1404) and the Heisenberg Grant (DFG We 4427/3–2).
Disclosure: The authors declare no conflict of interest.
Author contributions: LE and BW conceptualized the experiment, LE carried out experiments, analyzed data, and drafted the manuscript, PE provided technical support and revised the manuscript, YH provided critical input for the data analyses and on earlier versions of the manuscript, and BW supported data analyses, revised the manuscript, and approved the final manuscript as submitted.
Abstract
Objective
Prevalence of obesity is high in most industrialized nations, and therefore, it is crucial to understand contextual factors underlying food choice. Nutrition labels are public policy interventions designed to adequately inform consumers about nutritional value and overall healthiness of food products. The present study examines how different nutrition labels, namely a purely information-based label (guideline daily amount, GDA) and a more explicit traffic light (TL) label, influence product valuation and choice in a functional MRI setting.
Methods
Thirty-five healthy participants across different BMIs were instructed to valuate healthy and unhealthy food products in combination with one of the two labels and to state their willingness to pay (WTP) for the product.
Results
The labeling methods significantly influenced participants' WTP. Red TL signaling activated parts of the left inferior frontal gyrus/dorsolateral prefrontal cortex, a region implicated in self-control in food choice. This region, in the case of red signaling, and the posterior cingulate cortex, in the case of green signaling, showed increased coupling to the valuation system in the ventromedial prefrontal cortex.
Conclusions
Our results suggest that explicitly directing attention toward nutritional values using salient nutrition labels triggers neurobiological processes that resemble those utilized by successful dieters choosing healthier products.
Introduction
Primary diseases such as obesity and secondary diseases like hypertension place a high economic burden on society (1). Palatable foods high in sugar, fat, and calories have transformed food intake due to their abundance and constant availability in developed countries (2). The increasing obesity epidemic may be due in part to insufficient knowledge about the nutritional value of food products, even in health-conscious consumers (3). Nutrition labels serve a decisive role in informing consumers about food products, their composition, and healthiness, and an effective food labeling system may help to mitigate the prevalence of obesity by promoting healthier choices (3). The guideline daily amount (GDA) information displays both the amount and the percentage of recommended daily intake of several nutrients per serving. The traffic light (TL) label deploys a color-coded system, depicting the amount of nutrients per 100 g while also translating nutritional values into colors of the traffic light (4). Both labels provide very similar information content; for a rational decision maker, they should therefore similarly improve the estimation of a product's nutritional composition and ultimately foster healthier choices. However, several studies and a recent meta-analysis have provided evidence that a TL label, compared to a purely numeric GDA label, is superior in terms of facilitating interpretation of nutrition values and increasing the selection of healthier products (5). Thus far, underlying brain mechanisms that alter the subjective valuation of a product in the case of TL versus GDA signaling remain elusive. A more detailed understanding of these mechanisms may help to increase the effectiveness of nutrition label invterventions.
Valuation is a dynamic process and can be influenced by attention and exogenous cues (6, 7). The ventromedial prefrontal cortex (vmPFC) is a brain region consistently associated with value computations across task modalities (8). Previous studies showed that exogenously directing attention toward health aspects of food items led to healthier choices, possibly by a stronger inclusion of healthiness attributes into value computations within the vmPFC (7). Green TL signaling for rather healthy, and red TL signaling for rather unhealthy products, compared to the respective GDA labels, might exogenously direct attention toward nutrition aspects and alter valuation processes found in the vmPFC. Specifically, a green TL conveys positive health information and signals a delayed reward (i.e., better health consequences). Accordingly, it should increase valuation of the respective products, possibly accompanied by an alteration of the vmPFC valuation signal. For the selection of healthier options, a brain region of interest that modulates the vmPFC valuation signal should be involved in reward expectation and the encoding of delayed rewards, as a variety of positive health consequences do not occur immediately. Aptly, the posterior cingulate cortex (PCC) has been shown to fulfill both prerequisites (9, 10). It has been shown to be connected to the vmPFC, both structurally and functionally (11, 12). In contrast, a red TL conveys negative health information (i.e., possible negative health consequences) and should therefore decrease valuation of the respective products. For the rejection of unhealthier options, response inhibition and self-control might be indispensible. A higher body mass index (BMI) has been shown to be related to reduced inhibitory control in response to food images, (13) reduced self-control in the realm of intertemporal choice (14) and reduced activity in response to food logos in brain regions suggested to be involved in self-control (15). A previous fMRI study of dietary choice showed that successful self-controllers exhibit greater left IFG/dlPFC activation modulating the vmPFC valuation signal, (16) implying that activity in the IFG/dlPFC is required to incorporate long-term considerations into the valuation signal (7, 16). Therefore, the IFG/dlPFC serves as a region of interest that possibly modulates the vmPFC valuation signal more strongly in the case of red TL signaling compared to the respective GDA labels. Reductions in self-control and increased impulsivity observed in obese persons might be modifiable (17). Indeed, cognitive reappraisal strategies, such as thinking of the long-term costs of eating unhealthy foods, have shown to increase activation in inhibitory regions in response to palatable food stimuli (17). Exogenous health cues were shown to activate neural mechanisms of self-control similar to those employed by successful self-controllers (7, 16). Correspondingly, effective and salient nutrition labels might decrease impulsivity and increase self-control.
The present study examined brain activations in response to healthy and unhealthy food items along with a TL or a GDA label. To our knowledge, there are no prior studies examining neural correlates of the effect of nutrition labels. Thus, little is known regarding potential brain mechanisms that alter subjective valuation processes. We hypothesized increased activation in regions implicated in self-control in the case of red TL signaling, compared to the respective GDA labels, which in turn modulates the vmPFC valuation signal. Activity in areas important for reward expectation should modulate the valuation signal in the case of green TL signaling for healthy products.
Methods
A total of 35 healthy adults (19 females) were subjected to fMRI scanning. Standard exclusion criteria such as metal implants, neurological, or psychological disorders applied. All subjects had normal or corrected-to-normal visual acuity. Detailed participant characteristics can be found in Table 1. All participants received detailed information on fMRI, exclusion criteria, data handling, and task instructions. Prior to the experiment, subjects gave informed consent and received a brochure on both nutrition labeling methods to ensure common knowledge. The study was approved by the local ethic committee.
| N | Age | Height in cm | Weight in kg | BMI | |
|---|---|---|---|---|---|
| Behavioral data analysesa | 33 (18 females) | 25.8 (4.3); 19-37 | 172.5 (9.7); 150-193 | 69.5 (14.3); 44-109 | 23.5 (4.0); 17-33 |
| fMRI data analysesb | 25 (14 females) | 23.3 (4.4); 19-37 | 172.2 (9.5); 159-193 | 68.7 (14.5); 44-109 | 23.1 (3.4); 17-33 |
- Values in the four right columns denote means (standard deviations) and ranges. Participants were recruited via university mailing lists and flyers and consisted mostly of university students and PhD candidates.
- a For the behavioral analyses, data of two participants were excluded because their frequency of rejection (i.e., bids of zero cents) was two SD above the mean (mean rejection rate: 18.8 out of 100, SD = 22.3).
- b For the fMRI analyses, data of four additional participants were excluded due to excess translational and rotational head movement (> 3 mm and > 2.5 degrees, respectively), and four had to be excluded due to technical problems.
Experimental design
The GDA recommendation values were extracted from the website of the EU Food and Drink Confederation (18), the TL guidance values from a Food Standards Agency's information leaflet (19). Sample labels can be found in Figure 1A. For this experiment, products were classified as rather healthy or rather unhealthy. A typical variety of processed food products (e.g., chocolate, yogurt, frozen meals, including “light” and organic products in both categories) was selected for the experiment. Retail product prices were not significantly different for healthy versus unhealthy products (mean healthy €1.74, standard deviation (SD) = 0.81; mean unhealthy €1.52, SD = 0.80; t(98) = 1.37; P > 0.05). Subjects received €25 endowment for participation, which they could use for purchasing products. The Becker–DeGroot–Marschak auction was used as a model for market transactions in the laboratory (20, 21) in order to measure individual preferences and the exact willingness to pay (WTP) from each subject for every product; see Supporting Information for details on the procedure and product classification. Photographs of the products were obtained from the manufacturer's website and served as stimuli; labels were self-generated based on manufacturer's nutrition information. The sequences of pictures as well as the combination of labels and pictures were randomized across participants. Products were shown only once, in random combination with either a TL or a GDA label. The design was presented using Presentation© software (NeuroBehavioral Systems Inc.). Subjects saw the stimuli inside the scanner via video goggles (Nordic NeuroLab, Bergen, Norway) on a black background (resolution: 800 × 600 pixels). Subjects were instructed to observe the processed food items combined with their real nutritional values and state their WTP. In summary, this within-subjects experiment included 100 trials for each subject:
(a) Sample labels generated for the products. Left: color-coded traffic light (TL) label for a rather healthy (upper) and rather unhealthy (lower) product. Right: guideline daily amount (GDA) label for the same products as on the left. (b) Illustration of the trial setup. Products were shown on a white background (size: 378 × 315 pixels). Both labels had the same overall size and were also presented on a white background (size: 504 × 157 pixels). Participants saw 100 different products labeled either with a TL or a GDA label for 5 seconds (s). After a variable fixation time of 4–6 s, subjects were asked to state their willingness to pay for each product. Trials were separated by a variable intertrial interval of 4–6 s. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
- 25 healthy products with a (mainly green) TL label (H-TL)
- 25 unhealthy with a (mainly red) TL label (U-TL)
- 25 healthy with a GDA label (H-GDA)
- 25 unhealthy with a GDA label (U-GDA)
Figure 1B shows the timeline of one trial: After product presentation of 5 seconds (s), subjects saw a central fixation cross, for 4–6 s, followed by the instruction to state their WTP with an accuracy of up to 5 cents; the duration for entering the WTP was dependent on the subjects' individual speed. Trials were separated by an intertrial interval of 4–6 s. The whole fMRI experiment lasted approximately 30–40 minutes. Behavioral data was analyzed using R Studio (Version 0.97.551, Boston, MA) and SPSS (Version 20.0. Armonk, NY).
Imaging data acquisition
Scanning was performed in a 1.5T Siemens Avanto Scanner (Erlangen, Germany). An echoplanar imaging (EPI) sequence was used with a repetition time (TR) of 2.5 s, echo time (TE) of 45 ms, and a flip angle of 90°. 31 slices (regular-up, thickness: 3 mm, interslice gap: 0.3 mm) were acquired in an axial (AC-PC) orientation, with a field of view of 192 mm using a standard 8 channel head coil.
Imaging data analyses
Data analysis was performed using SPM8 (Wellcome Department of Imaging Neuroscience, Institute of Neurology, London, UK). Preprocessing steps included motion and slice time correction (with a reference time of TR/2 for regular-up acquisition), reslicing to a 3 × 3 × 3 mm voxel size, co-registration to the individual high-resolution T1-weighted structural image of each participant and transformation into the Montreal Neurological Institute template space. A high pass temporal filter with a filter size of 128 s was utilized. A final smoothing step with a Gaussian Kernel with full-width at half maximum of 8 mm was applied. For each participant, brain activation was estimated using a general linear model (GLM). The GLM was defined using the following regressors: (1) onset H-GDA, (2) onset U-GDA, (3) onset H-TL, (4) onset U-TL, (5) onset bid across all products modeled as a single regressor of no interest, (6-11) movement regressors (3 translation, 3 rotation regressors), and (12) a session constant. All regressors were convolved with a hemodynamic response function. The picture onset regressors were modeled using 5 s box-car functions, and the regressors for the bid responses were modeled using stick functions using individual reaction times. For each onset regressor, parameter estimates were generated. Contrasts of the onsets of the stimuli (1–4) versus baseline were calculated. Further, GLM 2 was used to identify areas correlating with WTP across products; see Supporting Information for details on the procedure.
Second-level random effects analysis
A flexible factorial analysis with the factors label (2 levels: GDA and TL) and category (2 levels: healthy and unhealthy) was performed while controlling for the subject factor (22). Brain regions were labeled according to the automated anatomic labeling tool implemented in the WFU pickatlas for SPM8 (23). A region of interest (ROI) analysis using small volume correction (SVC) using 10 mm-spheres around reported peak voxels was applied to relevant contrasts (at Puncorrected < 0.001, whole brain), while controlling for multiple comparisons only in voxels within the sphere using Gaussian random field theory (24) for small volumes as implemented in SPM. The IFG/dlPFC was chosen as an a priori hypothesis-derived ROI for the contrast U-TL versus U-GDA based on previous studies of self-control and food choice (7, 16) since a mainly red TL label was expected to trigger self-control, compared to the non-directive GDA label.
Psychophysiological interaction
Psychophysiological interaction (PPI) analyses were used as a method for investigating label-specific changes in brain connectivity (25). The vmPFC valuation signal was expected to be differentially modulated by regions involved in reward expectation (in the case of green TL signaling, compared to the respective GDA label) and by regions involved in inhibitory self-control (in the case of red TL signaling, compared to the respective GDA label). Therefore, two PPI analyses were carried out to test wether two a priori regions known to be important for the above-mentioned processes would modulate activity within the vmPFC. To analyze connectivity between areas implicated in reward expectation and valuation, the PCC was chosen as a seed region for PPI 1 in the contrast H-TL versus H-GDA. To analyze connectivity between areas implicated in self-control and valuation, the IFG/dlPFC was chosen as a seed region for PPI 2 in the contrast U-TL versus U-GDA. Details on seed definition can be found in the Supporting Information. An additional PPI analysis was used to analyze connectivity between the vmPFC and regions important for reward-guided learning across all products; see Supporting Information for procedure and results. PPI analyses were carried out for each subject individually, and the resulting images of contrast estimates were entered into a random effects group analysis using one-sample t-tests on the single-subject contrast coefficients, with a statistical threshold set to Puncorrected < 0.001 and a cluster size of at least 5 voxels (adjusted based on Random Field Theory).
Results
Behavioral data
A mixed effects regression analysis with subject as a first-level factor and WTP as dependent factor was performed. This analysis revealed a significant main effect of category but not label. The interaction between label and category was significant. Pairwise comparisons revealed higher WTP for healthy products. Further comparisons based on estimated marginal means showed that WTP was significantly higher for healthy products labeled with a TL compared to a GDA label, but not significantly different for unhealthy products labeled with a TL compared to a GDA label, see results in Table 2 and Figure 2.
Behavioral data: mean willingness to pay (WTP), split into label and category. Error bars indicate standard error of the mean, *P < 0.05. GDA: guideline daily amount; TL: traffic light. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
| Mixed-effects regression analysisa | |||
|---|---|---|---|
| Estimates of fixed effects | Test of fixed effects F (1, 3267) | Significance of fixed effects (P value) | |
| Main effect category (healthy versus unhealthy) | 16.99 (17.1) | 34.35 | <0.001 |
| Main effect label (TL versus GDA) | 3.33 (17.1) | 0.39 | >0.05 |
| Interaction label/category | 4.89 | <0.05 |
| Pairwise comparisons | ||
|---|---|---|
| Mean difference | Significance (P value) | |
| Healthy versus unhealthy | 12.34 (12.1) | <0.001 |
| TL label versus GDA label | 1.32 (12.1) | >0.05 |
| Healthy TL versus healthy GDA | 5.98 (17.1) | <0.05 |
| Unhealthy TL versus unhealthy GDA | -3.33 (17.1) | >0.05 |
- a The pairwise comparison values are measured in cents (€) based on estimated marginal means with standard deviations (SD) in parentheses.
- WTP varied from subject to subject, the between-subject variance in WTP was 914.6 cents (SD = 30.2 cents), and this was significantly different from zero, z = 3.9, P < 0.001. Gender was not a significant covariate (Z = 0.221, P = 0.83).
Functional MRI data
As shown in previous studies, (8, 21) WTP was correlated, among other regions, with activity in the vmPFC across all products (Figure 3 and Supporting Information Table S2). The contrast TL unhealthy > GDA unhealthy showed significantly increased activation in the left IFG/dlPFC for the TL label, without significant gender differences at P = 0.001 (two-sample t-test). Using a sphere around the peak voxel reported in the study by Hare and colleagues (16) confirmed a significant activity within the same region described by the authors to be important for the application of self-control in food choice (IFG/dlPFC; see also Supporting Information Tables S1 and S3). To investigate wether the kind of label information differentially influences the value signal in vmPFC, we performed functional connectivity (PPI) analyses. When a red TL label was shown along with an unhealthy product, compared to the respective GDA label, the IFG/dlPFC showed positive task-related connectivity with the vmPFC. When a green TL label was shown along with a healthy product, compared to the respective GDA label, the PCC showed positive task-related functional connectivity with the vmPFC. Figure 4 depicts an overview of the PPI results (see also Supporting Information Table S4).
Neural correlates of subjective valuation: activity in the ventromedial prefrontal cortex, anterior cingulate cortex, putamen, and nucleus caudatus was positively correlated with willingness to pay at the time of evaluation (i.e., stimulus onset), displayed at P < 0.001, uncorrected, on the mean 1.5T structural SPM8 template. Color bars indicate T values. vmPFC: ventromedial prefrontal cortex; ACC: anterior cingulate corte; Put: Putamen; NC: nucleus caudatus. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
Summary of the functional connectivity results. The posterior cingulate cortex (PCC) seed showed increased connectivity with the ventromedial prefrontal cortex (vmPFC) for green signaling (green arrow and green area within the vmPFC) and the left inferior frontal gyrus/dorsolateral prefrontal cortex (IFG/dlPFC) seed showed increased connectivity with the vmPFC in the case of red signaling (red arrow and red area within the vmPFC). Activity within the vmPFC that correlated with willingness to pay is highlighted in yellow. Data are shown using MRIcroGL (MNI template) and in radiological convention, with the right hemisphere on the left. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
Discussion
The present study investigated the neural processes underlying the influence of purely information-based compared to more salient nutrition labels on food valuation and choice. Traffic light signals influence the valuation of food products in that participants increased their subjective valuation of healthier food items compared to a purely information-based label. The fMRI data suggest similarities of the labeling effect to the exertion of self-control in food choice. Red labeling signals activate a region previously implicated in response inhibition and self-control in food choice, that is, the IFG/dlPFC (7, 16). This region also exhibits a stronger functional coupling to the vmPFC during red signaling. Green labeling on the other hand increases coupling between the PCC and vmPFC, suggesting an increase in reward expectation (9, 26).
We found that the vmPFC, among others, correlated with subjective valuation, that is, WTP. This is in line with a large amount of previous studies (8, 21) that have indicated that the vmPFC is important for value computations in simple choices. A red TL label was expected to be a cue directing attention toward nutritional values and long-term health goals, thereby exogenously triggering self-control. In line with this assumption, we found a stronger activation in the IFG/dlPFC, a region shown to be involved in the exertion of endogenously generated self-control (7, 16). Various studies have demonstrated the importance of this brain region in different domains of self-control, for example, in controlling impulses in intertemporal choices, (27) or in the compliance to social norms (28). The IFG/dlPFC has also been shown to be critical for response inhibition in Go/No-go tasks, (29) in motor as well as in affective self-control, (30) and has been shown to positively correlate with the strength of diet goals while viewing food cues (31). Disruption of the left lateral PFC increases choices of immediate over larger delayed rewards, providing evidence that this region is causally involved in self-control mechanisms (27). Our hypothesis was that the valuation signal in the vmPFC should be modulated by the IFG/dlPFC with red signaling, which was the case. A previous study on endogenous self-control in food choice found that the vmPFC was functionally connected to the IFG/dlPFC, and this connectivity was higher in self-controllers (16). Another study found that a very proximate region correlated with health ratings and exhibited increased functional connectivity with the vmPFC, suggesting that this region aids in integrating health information into the valuation processes within the vmPFC (7). Although the direction of information flow (i.e., IFG/dlPFC controls vmPFC or vice versa) cannot be clearly analyzed using PPI, (32) a prior TMS study suggested that the dlPFC actually influences the vmPFC (33). Our findings dovetail with prior studies showing that external cues can trigger brain processes similar to those employed by endogenous self-controllers, which in turn modulate valuation within the vmPFC (16). Colored nutrition labels might be feasible public policy interventions that reduce unhealthy food choice by triggering self-control and altering the vmPFC valuation signal. We further hypothesized that green signaling denotes beneficial long-term health consequences, which are integrated into the valuation signal. We observed a stronger coupling between the PCC and the vmPFC when the food products were associated with a green TL signal. The PCC has been consistently shown to be significantly more activated in response to positive compared to negative feedback (34) has been suggested to be part of a top-down attentional control system (35) and is thought to be part of a baseline cortical system important for behavior mediated by internal goals (10, 36). Beneficial health outcomes, as denoted with a green TL label, could be considered a ubiquitous internal goal that may trigger goal-directed behavior by increasing subjective valuation.
Several limitations of the present study and future directions should be noted. We did not specifically recruit obese subjects. It is of specific interest to investigate possible interaction effects of nutritional labels and nutritional status. Obese subjects might especially profit from salient nutritional signals, especially because work of Hare and colleagues has highlighted the importance of self-control in the integration of health-information in food valuation and choice (7). Future studies should elucidate how nutrition labels actually influence valuation of a specific product and how behavioral interventions may augment these effects. Across all labels, subjective valuation, that is, WTP, was higher for healthy compared to unhealthy products. This might be surprising in light of studies highlighting the rewarding value of high-calorie food products (2). However, subjects did not fast prior to the experiment, which might have decreased the rewarding value of high-calorie food products. Further, our finding is in line with a similar previous study of dietary choice, that found increased choices of healthy compared to unhealthy food products (7). While we did not elicit the cycle phase and use of hormonal contraceptives, it may affect the processing of labels. Although we did not find significant gender differences, future studies should further address this issue, especially investigating the altered brain responses to food cues depending on the menstrual cycle phase (37). Future studies should further address the effectiveness of nutrition labels in situations of reduced self-control, as for example after sleep deprivation, stress or time pressure. It would also be interesting to control for blood glucose and gut-hormone levels. On a more technical note, an important point of the PPI analyses in our design is that label and product are shown simultaneously, thereby activity within the IFG/dlPFC and vmPFC is temporally coupled, that is, potentially confounded. If the IFG/dlPFC indeed modulates the vmPFC and therefore changes valuation of a product, this effect might take place time displaced. Future studies are needed to identify the exact roles of other dlPFC subregions, the nature of possible modulators and whether indeed, the IFG/dlPFC is the first level of a top-down cascade triggering self-control processes which then in turn modulates the decision-making process. Studies like the present and previous ones may help to identify a neurobiological model of self-control and foster research on the pivotal role of differences in nutrition labeling.
A very interesting method to further elucidate the effects of nutrition labels on food choice is eye-tracking-based analyses of preferences applying attentional drift-diffusion models of simple choices (38). A previous study found that, although preferred, standard nutrition tables received little visual attention, compared to TL labels (39). Another study showed that participants' eye movements lack focus and healthiness ratings lack accuracy when confronted with a standard nutrition table, which was reversed in the case of TL labeling (40). More salient nutritional labels might influence the time spent on the respective information and the gathering of evidence for the respective items followed by a stronger influence on the subsequent choice than purely information-based nutritional labels (38).
To the best of our knowledge, this study is the first to demonstrate that salient traffic light labels influence the valuation of food products by a region implicated in endogenous and exogenous self-control and its connectivity with the vmPFC valuation system. Nutrition labels are widespread public policy interventions and should be designed to adequately inform and help customers in their decision process. They offer an opportunity to influence the type of food that is selected and the amount that is consumed. Our results and future studies can inform the current public crisis of obesity by shedding light on factors that positively influence food choice, and consequently reduce overeating and obesity.
