Linking food-induced brain responses to
short- and long-term measures of eating
behavior: a review
January, 2015
Femke de Boer, 10287280
Master: Brain and Cognitive Sciences
Track: Cognitive Neuroscience
University of Amsterdam
Supervisor:
Nynke van der Laan
Image Sciences Institute, University Medical Center Utrecht
Co-assessor:
Susanne la Fleur
Abstract
Unhealthy eating and obesity are often attributed to disadvantageous decision-making. Food choice, the decision what to eat, and food intake, the decision when to stop eating, are both important determinants of energy intake. Therefore, these short-term eating behaviors might ultimately contribute to long-term weight change, either natural or due to an intervention. Since decision-making processes converge in the brain, we here review neuroimaging studies that investigated the neural correlates of food choice, food intake, natural weight change and weight change due to an intervention. Our aim was to assess the neural underpinnings of these four measures of eating behavior and to explore whether these overlap or differ. To this end, we included thirty-seven studies that related the neural response to food stimuli to one of the eating measures. We found that there was considerable overlap in brain regions identified in the studies from the different categories of eating behaviors, although some brain areas were more frequently identified in a specific category than in another. Many of the identified regions are involved in multiple food-related processes, such as food anticipation and self-control. Regions encountered in each category were the insula, anterior cingulate cortex and nucleus accumbens. This implies that these regions have an important role in both short- and long-term measures of eating behavior. Studies combining neuroimaging with several eating behavior measures are currently lacking. Thus, future research should assess the neural substrates of food choice and intake employing a longitudinal design, to obtain greater insight into the neural background of repeated disadvantageous eating decisions which might result in overweight in the long-term.
1. Introduction
Obesity is a major health problem, with high worldwide prevalence rates and serious physical, emotional and social consequences1. In the Netherlands, the prevalence of overweight has increased
from 33% to 48%, and the prevalence of obesity doubled from 5% to 12%, over the past thirty years2. Overweight and obesity are the result of a long term positive energy balance. This positive energy balance is thought to arise primarily from an overconsumption of food rather than a decrease in physical activity3. Overconsumption originates from complex interactions between the body and the brain and can therefore be investigated from different perspectives. From a neuroeconomic point of view, obesity can be seen as a decision-making disorder4. Eating patterns constitute decisions about what to eat, when to start eating and when to stop eating, and all these decisions are taken in the brain. Thus, ultimately, overweight is caused by a series of disadvantageous food choices.
Although ample of research has investigated what neural mechanisms are related to overweight, many earlier studies are limited in the ability to determine which decision process goes awry, as they often cross-sectionally compare overweight versus healthy weight individuals5.
Therefore, they are not able to provide more knowledge on the development of overweight and its neural determinants. In neuroscience, several short and long term measures of eating behavior have been investigated: food choice and food intake (short-term) and natural weight change and weight change due to an intervention (long-term). The neural correlates of food choice, or the decision what to eat, have been relatively well-investigated by studies that let participants decide whether they would like to have a certain food item. Food intake, or the decision when to stop eating, has been less well investigated. The few studies that did investigate this, let participants execute a task in a MRI scanner, and then correlated these brain measures with subsequent (ad libitum) food intake. The neural determinants of long-term natural weight change have been studied by scanning individuals at baseline, and following them up after a substantial period of time later (such as one year) to see what brain activation on baseline predicted their weight change. Finally, several intervention studies employed a quite similar design, but then applied a certain weight-loss intervention in between, to see whether brain activation was able to predict whether individuals would lose weight or not.
Currently, it is poorly understood whether the neural determinants of short-term eating behaviors are similar to long-term brain mechanisms. Logically, it might be expected that the short- and long-term neural determinants agree, however, there could also be differences, because the short-term measures of eating behavior during an experiment are single events, while these repeated events in real life might be more complex and noisy. Therefore, this review aims to establish the relation between food-induced brain responses and short- and long-term measures of eating behavior, and compare the neural determinants. We will first describe the neural determinants for every type of measure, and then describe their similarities or differences. We will end with a discussion, directions for future research and a concluding remark.
2. Methods
We were interested in studies investigating the relationship between food-induced brain responses and one of the four different eating behavior measures (food choice, food intake, natural weight change or weight change due to an intervention). Inclusion criteria for all the studies were that studies: 1) were original and empirical (no reviews), 2) were published in peer-reviewed, English-language journals, 3) employed functional human neuroimaging methods, such as functional magnetic resonance imaging (fMRI), electroencephalography (EEG), magnetoencephalography (MEG), functional near-infrared spectroscopy (fNIRS) and regional cerebral blood flow positron emission tomography (PET), and 4) used a task with food cue exposure (viewing, smelling, or tasting).
Additionally, for the food choice studies applied that they should investigate food choice with a real consequence. This implies that food preference studies were excluded, since the mere expression of preference or a hypothetical choice does not have any consequence for participants, which might lead to a bias and might thus be weakly related to real food decisions6,7. Similarly, we only included food intake studies that investigated real food intake, as opposed to hypothesized food intake.
PubMed, Google Scholar and Scopus were searched using the following key words: (“neuroimaging” OR “brain activity” OR “neural” OR “fMRI” OR “EEG” OR “MEG” OR “fNIRS” or “PET”) AND “predict” AND (“purchase” OR “choice” OR “food” OR “food choice” OR “food intake” OR “weight” OR “intervention”). Additional studies were found by examining the lists of references of retrieved articles. Overweight was defined as a body mass index (BMI) greater than 25 and obesity as a BMI greater than 30. Activation in the ventromedial prefrontal cortex (vmPFC) and medial orbitofrontal cortex (mOFC) was taken together in this review, since they are anatomical proximal, are thought to be involved in the same function and this is most often done in literature8.
3. Results
Thirty-seven studies were identified that used food-induced brain responses to predict measures of eating behavior on short-term (food choice and intake) and long-term (natural weight change and weight change due to an intervention) (Table 1). We will first discuss the findings of these four categories separately, starting with studies that investigated short-term measures of eating behavior, followed by the long-term studies. For every category, we will discuss the brain regions that were most frequently found across the different studies. Finally, to integrate these findings, we will explore the similarities and differences between the different categories.
3.1 Short-term measures of eating behavior
3.1.1 Food choice
Food choice, or what to eat, is an important aspect of eating behavior. We live in an obesogenic environment where food cues are continuously triggering food choices and dilemmas, also giving rise
to many potential food choice failures. Whereas many people aim to eat more healthy food, they are often not successful in achieving this46,47. Also, overweight and obese individuals choose fruit and vegetables less often than healthy weight individuals48. Furthermore, interventions that aim to change
food choice have been found effective in reducing weight, which suggests that food choice has a critical role in achieving weight loss49,50. To explore which brain regions are crucial for food choice,
we here discuss sixteen fMRI studies and one EEG study that investigated the neural correlates of food choice9-25. We will first consider several of the commonly found brain regions across the fMRI studies, and then examine the findings of the EEG study.
Brain activity in the vmPFC and mOFC was most frequently found to be associated with food choice, that is in eleven of sixteen fMRI studies11,13,15,17,18,20-25. The studies that identified this region, varied in study characteristics and designs (e.g., choice versus purchase, contrasts investigated). Despite these differences, all studies found that activation in this region was positively associated with the likelihood of choosing an item. This is in accordance with the proposed role of the vmPFC/mOFC, being a hub region for decision making, and studies indicating that activity in this region is indicative of decision strength21,51-52. During choice, this region integrates the different attributes of a food item, compares them, and then computes a overall value (the subjective value) for that item53,54. Basic attributes that
determine the value of a food, such as tastiness, might always be reflected in the subjective value calculated by the vmPFC. Therefore, vmPFC/mOFC activation is also related to the rewarding properties of food55-57. Notably, its activation has been also been found to differ between obese and healthy-weight individuals58,59. In spite of these important features of the vmPFC/mOFC, we did not
observe its predictive activation in five studies. One reason for this could be that the vmPFC/mOFC is susceptible to artifacts resulting from the air/tissue interface that occurs there60. Several studies (but not all) employed tilted acquisition to correct for this12,13,15,17,20,21,23,25. To summarize, the
vmPFC/mOFC was most robustly found to activate in response to chosen food items, likely because of its role in the integration of different food features that determine the subjective value of a food item.
Another brain structure that plays a decisive role in food choice, as demonstrated by its activation that was related to food choice, is the striatum11,12,15,16,18,19,21,22,24. Most of the studies that found the striatum employed tasks where participants could plainly choose whether they would like to have that food item or not (single item studies), but some also used a binary choice task, where people could choose between two items12,16,21. All studies showed that activity in this region increased when decision strength, the desirability of the food item or the willingness to pay for a food item, increased as well. The striatum is part of the limbic system and is a key node in the appetitive network61. It is widely held that the striatum couples motivation with action, which is also necessary for eating, since food is critical for survival62. The results of our included studies support this idea, since we found higher striatal activity for the chosen items, and are also in accordance with a meta-analysis that showed that
the striatum codes reward valence, among others63. Furthermore, elevated activation of the striatum
has been observed during the craving (intense desire) of personally favorite foods, which was related to BMI as well64,65. Thus, based on the results of the studies included in this review, we might
conclude that the striatum is of great importance for food choice, possibly because of its role in linking motivation to subsequent actions by highlighting the rewarding properties of that food item.
Several studies found that different parts of the inferior parietal lobule (IPL) to be correlated with choice, such as the inferior parietal gyrus, supramarginal gyrus and angular gyrus9,12,13,15,16,18,19,21,24.
These studies adopted various study designs with different features: high- and low-calorie food items, single and binary choices and choosing or bidding on food items. Although some studies did not explicitly report the direction of the relationship between IPL activation and choice13,19, most studies showed that higher IPL activation was positively associated with choice (yes or strong yes) for a food
item12,15,16,18 or with self-regulatory success during the choice period9,21. The IPL is traditionally seen
as the association cortex, but its role in other processes, such as the translation of intentions into actions, is emerging66-69. Reward anticipation, which contributes to the intention to engage in eating
behavior, engages IPL activation, as meta-analyses have shown that the viewing of food images (as opposed to nonfood images) induces activation in the inferior parietal lobule63,66,70. Next, other studies
have reported that the intraparietal sulcus, which lies in between the superior parietal lobe and IPL, stores the evidence regarding a particular decision possibility, and then when the decision is made, passes this information to the motor system51,71. Therefore, we putatively suggest that the IPL activation found in the studies included in this review might be indicative of the transformation of the anticipation for certain food item into the consequential choice action for that item.
Next, a brain region that was often found to correlate with food choice was the insula11,12,13,15,19,20,22,24.
Features of these studies varied greatly, although most studies investigated single food choice.In spite of some studies not describing the direction of the association13,19, nearly all studies demonstrated that
higher activity in the insula was positively correlated with the choice for food items. Except for Knutson and colleagues24, who found that insula activation was predictive of rejection of that item.
However, this might again originate from fact that their task used items with prices that the participants could choose to buy or not. Here, the insula became active when excessively high prices were asked for a preferred item, which was then associated with rejection. The insula plays a pivotal role in interoception, which is defined as the sense someone has regarding his or hers physiological condition72. Therefore, it could be that the sense of whether someone is hungry or satiated is important
for food choice. Additionally, the anterior insula encompasses the primary taste cortex, and is implicated in multimodal sensory integration73. As food consists of multimodal sensory features, such
as the sight, smell, olfaction and taste of food, which might all play a role in a choice for a certain food item, it is not surprising that insula activation is important for food choice61,74. Furthermore, the insula
activates during the anticipation of reward, as for instance happens during the craving for food63,64.
Summarizing, insula activation during the choice for a food item might be related to perception of internal bodily states, integration of different features of the food, and anticipation of reward.
Another region that multiple studies found to be related to food choice is the anterior cingulate cortex (ACC)13,15,18,20,21,23,25. Features like task design or stimuli considerably differed between all the studies.
Higher ACC activation was linked to a confirmative decision for a food item. Although none of the studies that observed the ACC discusses this finding, the ACC is often observed during the viewing, tasting and smelling of food59,75 and during the processing of rewards54,57,63. Similarly, anticipation of food such as imagining the taste or craving have been found to increase ACC activity13,76. Another
important role of the ACC is the detection of errors and conflict77. It has been hypothesized that the ACC might activate during choices that induce a conflict between the desire to eat or losing weight78.
Two studies included in this review investigated how this conflict is processed in the brains of dieters9,21. When analyzing the self-control trials, or the food decisions favoring long-term goals,
neither of these studies obtained the ACC. This might suggest that the ACC is apparently not involved in the conflict detection of short- versus long-term eating goals, but it should be noted that these two studies only investigated individuals with a healthy weight. As this conflict might be greater for individuals with an unhealthy weight, more research should investigate the role of the ACC in healthy eating conflicts in overweight and obese individuals. Altogether, we suggest that the predictive ACC activation during food choice might be primarily related to food anticipation.
Furthermore, we found that decision strength was correlated in six studies with the middle frontal gyrus15,19-22,24 and the dorsolateral prefrontal cortex (dlPFC)13,15,19-21,25. For both regions applied that higher activation was predictive of the decision to choose a food item. The middle frontal gyrus is implicated in several executive functions, such as attention, which might be important for food choice because people fixate longer on the item that they choose79. However, it is difficult to translate the
attentional role of the middle frontal gyrus to the findings of our study sample, since most studies used single food item choices, where only one food item was shown. Furthermore, middle frontal gyrus activation in response to visual food cues has been positively correlated to self-reported hunger during a fasted state in obese individuals80. This is in accordance with the current results, because nearly all
studies that obtained the middle frontal gyrus instructed participants to refrain from eating for two up till four hours (except for Knutson et al.24). The dlPFC plays an important role in self control and is thought to process more abstract values of food, such as healthiness53,81-83. This agrees with our results,
as the dlPFC was found in studies that employed more complex tasks with different conditions (e.g., tasks where self-control or emotion regulation should be executed, or food items with or without an organic label) instead of just simply choosing for a food item. Other regions found to be predictive of food choice are shown in Table 1.
As mentioned before, only one study investigated brain activity during choice with an EEG paradigm14. Since EEG has a relatively poor spatial resolution, this technique does not inform about the location of activation, but could investigate differences in laterality of the neural response for food choice. Ravaja and colleagues14 found that increased left frontal activation (relative to right activation) during the viewing of items predicted an affirmative decision. This is in line with the approach-withdrawal motivational model of emotion, which states that increased left frontal activity expresses a bias to approach a stimulus, whereas increased right frontal activity expresses a bias to withdraw from a stimulus84. Notable, this shift in frontal laterality has also been observed in overweight and obese
individuals85,86. In these groups, higher self-reported appetitive responsivity, hunger and disinhibition were associated with greater left-sided prefrontal cortex activation86. Additionally, response inhibition
has also been proposed to be right-lateralized87. Thus, food choice for appetitive items might also be determined by increased responsivity to food (presumably higher left-lateralized prefrontal activation) and decreased response inhibition (lower right-lateralized prefrontal activation)88. However, there is also evidence against the lateralisation of mechanisms involved in food choice89. We explored whether
this laterality was also present in the reviewed food choice fMRI studies. We could not confirm this: activation in the vmPFC/mOFC, middle frontal gyrus and dlPFC was nearly as often found left as right.
3.1.2 Food intake
Food intake, or how much we eat, is another important aspect of eating behavior. The decision to stop eating determines how much we eat from an earlier chosen product. For instance, one can choose some high-calorie food, but eat not too much of this. Furthermore, the current obesity epidemic seems to arise primarily from overeating in the absence of physiological hunger (also called hedonic eating)90. Because hedonic eating might be primarily triggered by palatable food (cues), which can
overrule physiological homeostatic signals, neural responsivity to food cues might be able to predict food intake91. It has been shown that optimal portion size can be predicted by both liking of the food
and expected satiety that will be obtained by the food92. Some of the brain regions that are involved in anticipatory responses to food might also be involved in the perception of satiety and thus the determination of meal intake. We found five studies that aimed to predict food intake on the basis of BOLD responses to food stimuli26-30.
The brain region that most often predicted food intake (two of five studies) was the nucleus accumbens (NAcc), such that higher activation in the NAcc predicted higher food intake26,27. These
studies both used a passive viewing paradigm together with a ROI approach to detect BOLD response differences in the NAcc between food versus non-food images. Lawrence and colleagues27 gave
participants a bowl of their favorite crisps while filling in questionnaires after scanning, and they found that increased NAcc activation predicted higher crisps intake. Lopez and colleagues26 observed
a similar relationship between NAcc activation and food intake during a week after scanning. During this week, participants engaged in experience sampling, which is defined as the acquisition of information on the experiences of individuals within the flow of daily life93. Seven times a day,
smartphones of the participants randomly signaled them to answer several questions. They were asked whether they had a food desire the past half hour, how strong it was, their resistance to it, whether they acted upon that desire and if so, what the amount was they had eaten. People with greater NAcc activity had more intense food desires, were more likely to also give in to their temptations and reported higher food amounts eaten26. These results agree with the role of the NAcc in reward-seeking
behavior and food-cue associations94. However, Mehta and colleagues28 applied an ad libitum buffet after scanning, and did not detect an association between NAcc activation and total caloric intake. They found a positive relationship between its activation and the percentage fat consumed and a negative relationship with the percentage of carbohydrates intake. The same relationships were also observed for the left amygdala, left insula and medial OFC, but again no effect on total calorie consumption. This finding could be due to the fact that Mehta and colleauges used a contrast investigating high-calorie versus low-calorie foods, with high caloric foods such as hamburgers, pizza, and French fries. Thus, NAcc activation identified in this contrast might be related to responsivity to high-calorie food, and is therefore only a predictor of higher fat intake. Another possible explanation for this null-finding might be that the participants were all normal-weight, while obese individuals are known to display higher NAcc activation in response to visual food cues5. Low inter-individual
variability in NAcc activation might therefore have limited the possibility to find a relationship with food intake. Indeed, participants of Lawrence et al.27 were normal-, over- and underweight and found
relatively big inter-individuals differences in NAcc activity.
As various lines of research propose an important role for self-control in eating behavior, this is increasingly topic of investigation95,96. For instance, as described above, Lopez and colleagues26 explored the relationship between self control and food intake using experience sampling. During scanning, they employed a Go/No-Go task including food and non-food images (which were not specifically divided over Go and No-go trials) to assess overall response inhibition. The authors found that individuals with greater IFG activation during response inhibition were able to ignore their desires more frequently: these individuals ate less when confronted with food desire. This result is in accordance with the fact that the IFG is crucial for response inhibition87 and the results of Hare et al.17
showing greater IFG activation during the consideration of health aspects of food. Additionally, this latter study investigated functional connectivity and demonstrated that IFG activation influenced the dlPFC, which consequently modulated vmPFC activation. As Lopez and colleagues found the IFG and the NAcc in two different tasks, they could not investigate the connectivity between those two structures. Still, it could be hypothesized that IFG activity modulates NAcc activity to decrease food
desire or intake, as a meta-analysis showed that these structures can indeed engage in functional connectivity with each other97, but further studies need to investigate this.
It should be taken into account that all of the above discussed studies used a ROI approach. However, one of the food intake studies analyzed brain activation across the whole brain during a passive food viewing task and did not find any of the above obtained regions30. However, this study
did identify activity in another region to be predictive to food intake: the dlPFC. Higher dlPFC activity (during the contrast of viewing of high-calorie, attractive foods versus non-foods) after an overnight fast was associated with lower food intake over three days. This in accordance with the self-control role of the dlPFC in food intake, and with the studies of Hare et al.17,21 which found that dlPFC-activation is increased to execute self-control over food choice. Interestingly, both studies that looked at food intake over a few days, instead of immediately after scanning, found regions that are involved in self-control26,30. It could be hypothesized that food intake after scanning might be experienced by
the participants as a single, unique event and therefore does not require self-control, whereas repeated food intake over days in real life might require self-control.
Next, Spetter and colleagues29 investigated the response to taste cues in satiated participants.
Satiation was accomplished by giving participants a juice preload, which they also tasted in the scanner, and could drink ad libitum after scanning. Investigating brain activation using regions of interests (ROIs), they found the ACC to be negatively related to subsequent ad libitum intake of the juice29. Thus, higher ACC activation was related to lower food intake. This prompts the authors to
conclude that the ACC reflects satiation, which ends food intake. This finding is in line with previous research that has demonstrated increased activation of the ACC after eating98-100. Additionally,
injection of the physiological satiety signal peptide YY is known to increase activity in the ACC101.
Besides its role in satiation, the ACC is found in response to taste stimuli, as demonstrated by a recent meta-analysis75 and is thought to process the (reward) value of taste, which can range from pleasant to
aversive98,102. Since the preload, taste cues and ad libitum intake were all of the same juice, Spetter et al.29 suggest that the ACC might be involved in food-specific satiation, the juice becoming more
aversive when drinking more of it, ultimately leading to intake termination. This might also be one of the reasons that the other food intake studies did not find the ACC, as they did not uses taste paradigms with one specific food.
Overall, we tentatively speculate that the brain region predictive of immediate food intake (i.e. ACC) might be more involved in satiation, which is the feeling that stops eating during an eating episode, as this study measured food intake and its determination only one time. However, the regions that were predictive of food intake over several days (i.e. IFG and dlPFC) might be more involved in satiety, which refers to the experience of fullness that lasts after eating to inhibit the initiation of eating, since these studies investigate multiple times of food determination over days. Indeed, it has been proposed that satiation mainly involves sensory processes, whereas satiety, which occurs later in time, requires more cognitive processes103. This might especially apply for the studies included in this
review, as these studies that investigated immediate food intake27,29 let participants eat only one kind
of food item (crisps and a particular drink respectively), possibly inducing sensory-specific satiety. The studies exploring food intake over several days26,30, however, investigated all different kinds of
food that were eaten over the days. The brain regions involved in these two processes might differently moderate other brain regions and interact with for example gut signals to achieve the satiation or satiety state. As the NAcc was found for both short- and long-term food intake, this region might be involved in both processes26,27. However, since these are merely speculations, future studies could explore this more elaborately.
3.2 Long-term measures of eating behavior
3.2.1 Natural weight change
Although a lot of people try to eat healthy, still approximately 50% of the Dutch population is overweight2. Apparently, a single meal or food decision is not necessarily related to long-term weight and therefore it might be very difficult to maintain a healthy eating pattern in the long term104.
Currently, it is poorly understood what the long-term neural determinants of this undesirable weight change are. Despite the numerous cross-sectional studies that have demonstrated differences in neural activation between obese versus lean people, it is not known whether these differences have arisen from obesity itself, or whether these differences predisposed these individuals to obesity. Thus, it would more informative to look at brain activation on baseline and then investigate weight change longitudinally to discover causality. We identified eight studies which used baseline brain activation in response to food cues to predict natural weight change after six months, one year or two years31-38.
Activation in striatal structures (i.e. caudate, NAcc and putamen) was most often found to predict weight change31-33,35,37,38. These studies used different kinds of paradigms (viewing food images, food
commercials, or tasting food stimuli) and predicted weight change on different time scales ranging from six months to two years. Moreover, studies used a ROI or whole brain approach, or both, and some included only females, whereas others also included males. Despite these divergent study characteristics, most studies found higher activity in the striatum to be predictive of weight gain, thus demonstrating an apparent robust effect. This is in line with studies showing that the striatum is involved in different kinds of valuation systems (Pavlovian, habitual and goal-directed) and in several steps of the decision-process (from action selection to learning) which are all important for eating behavior acting on different timescales4,105. Furthermore, the striatum processes both ‘wanting’ (motivational side) and ‘liking’ (hedonic side) of food106,107.
However, it should be noted that two of the included studies did not find a main effect of striatal activation on weight change, but found that possessing different variants of the TaqIA1 allele of the DRD2 gene moderated this relationship37,38. For individuals with the A1 allele of this gene there was a negative relationship between putamen or caudate activation and weight change over one year,
while individuals without this allele showed a positive association. Since the A1 allele of this gene is associated with lower D2 striatal receptor availability, the authors proposed that individuals with a genetic risk for low striatal responsivity might gain weight. However, since these two studies (and nearly all other studies) included both normal weight and overweight participants, a history of overeating in the latter group might have explained this effect. This is also in accordance with Yokum et al.32, who state that although baseline BMI did not significantly moderate the relationship between
striatal activation and weight gain, there were some positive interactions between baseline BMI and striatum responsivity, indicating that the relationship between striatal activation and one year follow-up weight gain might be stronger for heavier individuals. Thus, these studies could not provide greater insight into the causal processes which underlie the change from healthy to overweight.
Therefore, Stice and colleagues partially repeated their experiment of 2008, using the same paradigm but now with baseline healthy-weight participants, a bigger sample and including males34.
Here, they again did not find a main effect of striatal activity on weight change, but were also not able to replicate their findings regarding the predictive striatal activity depending on allele variants. Thus, it seems that striatal activation does not explain why healthy individuals gain weight. Alternatively, the authors state that it might be needed to observe these baseline healthy weight individuals longer than one year. No other study investigated weight gain in healthy weight individuals resulting in overweight, since all studies included both weight groups in their baseline, and thus evidence for this remains inconclusive.
A recent study of Burger & Stice31 again employed a similar tasting paradigm, with both healthy and overweight participants. However, this time they investigated changes of the BOLD response when repeatedly exposing individuals to the food stimuli during this paradigm, because the incentive-sensitization model poses that individual differences in these kinds of learning might determine weight108. The shift in BOLD response in the caudate nucleus over repeated food receipts
(food reward habituation) was able to predict weight change over two years, such that individuals displaying the largest decrease in caudate responsivity to the food receipts having greater weight gain. Interestingly, the authors mention in the discussion section that the average BOLD response over all the trials, which analysis is similar to the ones used in Stice et al.38 and Stice et al.31 did not show a
relationship with future weight. Thus, it might be that rather than the striatal overall BOLD response to food stimuli, changes in this response after repeated exposure are predictive of weight gain. Yet, it should be noted that a part of the habituation effect was explained by the great NAcc response to the initial taste, and that another tasting study did actually find an overall effect, while this study used two different tastes for the milkshakes (strawberry and chocolate) to prevent habituation effects33.
The activation of another part of the basal ganglia, the ventral pallidum, was also able to predict weight change in two studies that both used a taste paradigm31,33. The former study of Burger and Stice explored cue-reward learning by repeatedly exposure of individuals to cues (geometric shapes) that
precede the taste delivery, using trials where no liquid was delivered, to ensure that it was not confounded by actual receipt of the taste and to possibly augment incentive salience attribution33. Now, the authors found an effect for the ventral pallidum opposite to the one of the NAcc: the greater increase of ventral pallidum activation over the repeated cue trials, the greater the increase in BMI at follow-up33. This is in accordance with models that provide a role for the ventral pallidum in the
learning hippocampal-VTA loop, which is involved in the processing of novel stimuli109. Thus, these
new cues seem to become more attractive and desirable stimuli over repeated exposure and are linked with taste, as processed by the ventral pallidum. However, Geha et al.33 did not use cues in their
paradigm, and did also not investigate shifts in responds repetition of the stimuli, but still obtained the ventral pallidum. Nevertheless, it is not surprising that both these studies found the ventral pallidum, since there is an opioid (hedonic) hotspot located in the ventral pallidum, which is involved in the ‘liking’ reactions to sweet tastes110,111. Moreover, stimulation of the ventral pallidum does not only
enhance ‘liking’, but also ‘wanting’, and is thus pivotal for reward learning112. This is probably also because of its many connections with regions implicated in reward such as the striatum, being described as a convergent point for limbic behaviors112. Although the ventral pallidum possesses all
these important functions in the processing of food cues, it has not been identified by many studies, presumably because of its relative small size which is challenging to detect with fMRI113. In brief,
increased activation of the ventral pallidum in response to food tastes and cues might cause future weight gain, because of the ventral pallidum’s important role in both ‘liking’ and ‘wanting’.
Next, thalamus activation was found by two studies to be positively associated with future weight change32,33. These two studies differed in various features, such as paradigm (tasting milkshakes
versus viewing food commercials), age of the participants (adults versus adolescents), but they had in common (similar to many other long-term studies) that the baseline mean BMI of the participants was above 25. Both food anticipation, such as the viewing of appetitive food images as done in Yokum et al.32 and food receipt such as done in Geha et al.33 have been earlier found to elicit activity in the
thalamus114-116. Thalamus activation has been related to both ‘wanting’ and ‘liking’ ratings117 and is part of the cortico-striatal–thalamocortical circuit, which is implicated in instrumental conditioning including goal-directed behavior118,119. Indeed, a previous study showed that the thalamus is activated during regulation of desire for food65. On the one hand, it could be argued that the increased thalamic
activation observed here is probably not related to regulation since Geha and colleagues33 and Yokum
and colleagues32 only investigated food cue reactivity, instead of regulation. On the other hand, all participants were overweight, thus implicating that they might indeed need to recruit self-control when viewing food images. Besides, higher thalamic activation predicting weight increase could be related to another cognitive function: episodic memory. The study of Geha et al.33, which offered chocolate
and strawberry milkshakes interchangeably to the participants, found the anterior thalamus in particular, which has been linked to episodic memory including recognition (what) and temporal order
(when)120-122. Since memorized enjoyment of a food item is important for the liking at the time of
eating, tasting of milkshakes might also evoke memories to the last time individuals drank a milkshake and alter the current taste experience123,124. It could be hypothesized that people with greater thalamus
activity have more pleasant memories of the food presented, or are less able to inhibit these motivating memories. Summarizing, higher thalamus activation might be predictive of weight gain through increased food memories and altered motivation/learning processes that promote greater food ‘wanting’ and ‘liking’.
An additional region of which activity was found to predict long-term weight change was the middle temporal gyrus32,34. These studies had in common that both their samples consisted of adolescents, that
their participants fasted five hours before MRI scanning and that they investigated future BMI after one year. However, the studies differed on baseline BMI of their samples (overweight vs. lean participants respectively), number of participants and the paradigm used. As already mentioned, Stice et al.34 used a taste (milkshake) paradigm, whereas Yokum et al.32 employed a viewing (food
commercials) paradigm, and the 162 participants of the former study were healthy weight at baseline, whereas the thirty participants of the latter study were normal weight, overweight and obese. Stice and colleagues34 chose this study sample because they aimed to determine which brain regions could
predict the transition from healthy to overweight. They found that activation in the middle frontal gyrus negatively correlated with weight change: lower baseline activation in response to milkshake receipt (versus tasteless solution) in this region predicted greater weight gain. However, the study of Yokum and colleagues32 observed a relationship in the opposite direction: greater activation of the
middle temporal gyrus was associated with increased BMI at one year follow-up. This contradictory result might arise from the different paradigms: Yokum et al.32 investigated the neural response to food anticipation (viewing), whereas the study of Stice et al.34 studied the neural response to food
consumption (tasting). Following this line of reasoning, both increased food anticipation, as indicated by higher activation in the middle temporal, and decreased reward after actual consumption, as marked by reduced activation in the middle temporal gyrus, are predictive of weight gain. This is consistent with the idea of Dagher61 who proposed that precisely this combination of excessive reward
anticipation and comprised in reward experience during consumption might underlie obesity. However, it is remarkable that this differentiation takes place in the temporal gyrus, instead of in regions which are involved in motivation and reward such as the striatum or vmPFC/mOFC. Another potential explanation for the seemingly discrepant results could be derived from the different baseline weights of the two study samples. One could imagine that the increased responsivity to food cues in the Yokum et al.32 study might be driven by the overweight and obese individuals in this study sample. Indeed, previous literature comparing responsivity to food cues in obese and normal weight individuals found that obese individuals display higher activity in the middle temporal gyrus125,126. Research showed that the middle temporal gyrus is also activated when viewing manipulable objects
(such as a food item), and its activation has been linked to the retrieval of the possible actions that are associated with the objects (such as picking it up and eating it)127,128. Thus, food commercials in the study of Yokum et al.32 might evoke action planning of food purchases or consumption, to some
degree comparable to how drug cues elicit middle temporal gyrus activation in drug users129. However, middle temporal gyrus activation has not only been detecting during viewing, but also during the evaluation of stimulus characteristics (e.g., taste and pleasantness) during the tasting of food cues130. Thus, one could imagine that people who have increased tendencies to act upon food cues, based on the food items features, are more likely to gain weight. In sum, activation in the middle temporal gyrus is predictive of changes long-term weight, possibly because of its differential engagement in food anticipation and outcome, and because of its role in future planning of actions with food.
3.2.2 Weight change due to an intervention
Besides natural weight change, many people also try to actively reduce their weight by interventions. However, although weight reduction interventions are found to have short-term effects, the long-term effects are often unknown131. Also, these interventions might work for some, but not all individuals.
Launching national campaigns to reduce weight without understanding whether they will work is a waste of money. Using baseline neural responses, we might be better able to predict for which individuals an weight reduction intervention will work (or not) and how to generalize these results to bigger populations such as the whole society132,133. We identified seven intervention studies which
used baseline neural activity (or changes from baseline to post-intervention in neural activity) to predict changes in weight due to an intervention, with all studies having an overweight or obese sample39-44. Most studies aimed to reduce energy intake by a behavioral or lifestyle
intervention39,42,43,45. For example, in the study of Deckersbach and colleagues39 the purpose was to decrease the associations between unhealthy food and reward alongside with increasing associations between healthy food and reward. Others focused on exercise40,44 and one study investigated weight loss after laparoscopic adjustable gastric banding41. The durations of the interventions ranged from
twelve weeks to forty-five weeks, and most studies obtained the follow-up measurements of weight loss immediately after the intervention. Murdaugh et al.45 decided to do a longer-term follow-up (nine
months after the intervention had ended) next to the immediate measurement after the intervention, to see whether the effect of the intervention lasted. This study45, and the study of Cornier et al.44 also
measured brain activation at follow-up compare the difference in brain activation between before and after the intervention.
First of all, the insula emerged most frequently from our intervention study sample to predict weight change42-45. The direction of the prediction was inconsistent across the studies. One study found that
weight loss43. However, another study demonstrated that individuals who lost more weight due to the
intervention actually had higher pre-intervention insula activation as compared to individuals who did not respond to the intervention42. In line with this, the two studies that investigated the difference in
neural activation from baseline to after treatment showed that a bigger decrease in activation was related to greater weight loss, partially explained by higher baseline insula activation44,45. The distinct
result of Weygandt et al.43, who found that lower baseline insula activation was associated with more
successful weight loss, might originate from the different task and analysis employed in this study. In this study, brain activation was measured while participants made choices between a small immediate portion of a meal, or a larger delayed portion, and the researchers then analyzed which brain activation reflected subjective value. Subjective value here refers to the personal value for the food portion displayed when taking into account the time when it could be consumed (immediately or delayed). So, the activation of the insula that Weygandt et al.43 found represents time-dependent subjective value.
Because of this, and the fact that the insula is involved in many processes, such as emotional awareness, (time) perception and food-related processes, it is hard to interpret the results of Weygandt et al.43,134,135. Furthermore, it should be taken into account that a recent meta-analysis showed that
obese individuals actually have lower insula activation in response to food cues59. Thus, it could be hypothesized that the people with the highest weight, who are able to lose most weight, are also the ones with the least insula activation. Indeed, Weygandt et al.43 correlated brain activation with the difference in BMI, but did not consider baseline BMI.
Of the three studies showing that a higher baseline insula activation, or bigger decrease in activation, predicted successful weight loss, two used a food viewing paradigm (contrast high-calorie food versus non-foods), and the other used a one-back memory task with both food and non-food images intermixed42,44,45. This is a noticeable finding, given the role of the insula in for instance craving, which would suggest that higher insula activation is related to more craving for the food. However, it has been proposed that the (anterior) insula together with the insula might play an important role in self-regulation136. It could be that the people with higher baseline activity of the
insula also experience higher emotional awareness, and are thus better able to decrease their feelings that earlier motivated them to eat (due to the intervention). However, since the intervention of Cornier et al.44 only involved exercise, this explanation does probably not apply to this study. Studies have shown that exercise before scanning increases activation of the insula to low-calorie food images and that the time per week spent on exercise correlates with insula activation in response to high versus low calorie food images137,138. Thus, exercise seems to decrease reactivity to high-calorie food, and increase responsivity to low-calorie food, resulting in weight loss.
The next region that was found in three out of seven studies to predict intervention success was the
IFG41,42,45. The studies differed on several features, and the direction of the prediction was also
versus non-food images and during a one-back memory task respectively) to be predictive of successful weight loss after six months. In contrast, Murdaugh and colleagues45, who both explored baseline neural activation and after the intervention in response to high calorie food versus non-food images, observed that greater IFG activation on baseline, or greater attenuation in IFG activation from baseline to after the intervention, predicted long-term weight maintenance (after one year).
One explanation for these distinct findings might be that although all these studies report that they found the IFG, they might refer to different parts of the IFG or even different regions. The coordinates of the IFG reported by these studies are dissimilar (MNI coordinates: Murdaugh et al.45: x
=−48, y=26 z=14; Ness et al.41: x=43, y=49, z=-1; Hege et al.42: x=54, y=4, z=20). Reviewing these coordinates in the Automated Anatomical Labeling (AAL), the coordinates of Murdaugh et al.45 refer
to the IFG pars triangularis (BA 45), while the coordinates of Ness et al.41 make up the IFG pars orbitalis (BA 47), and the coordinates of Hege et al.42 are actually part of the precentral gyrus. This
latter finding might be because of the fact that Hege et al.42 was actually a MEG study, which is less able to infer information about regions. It is well-known that different parts of the IFG also execute different functions, for instance, the more dorsal part of the IFG might be involved in attentional control, whereas the ventral part of the IFG is thought to be implicated in inhibitory control139. Thus, the anatomical heterogeneity of the IFG findings might explain the heterogeneous directions found for the association between IFG activation and weight loss.
If we still would like to interpret the results in the light of the main function of the IFG (response inhibition), we could hypothesize that the participants of Ness at al.41 and Hege et al.42 did not apply response inhibition because they were not specifically asked for it during the viewing and one-back memory task. However, it could still be that when necessary in real life, these participants are able to recruit the IFG and execute self-control. Hege and colleagues42 interpret their results on the basis of a compensation mechanism: the people with high baseline IFG activation (who were not able to lose weight due to the intervention) might have to engage the IFG more to achieve the same level of performance and executive control during the one-back task, which might be ultimately related to diminished cognitive control over their eating behavior. On the contrary, the high baseline activation of the IFG, and its decrease after the intervention, being predictive of successful weight loss found by Murdaugh et al.45, could be seen as greater executive control or response inhibition. Interestingly, a recent study showed that training of inhibitory control actually decreases activation of the IFG during response inhibition140. Thus, the weight-loss intervention (which was behavioral in this study) might
possibly be seen as a training in cognitive control, reducing response inhibition related activity to food cues after the intervention. However, these results are still highly contradictory, so future studies should investigate the role of the IFG and its subparts in weight loss better.
The hippocampus was one of the other regions which activity was found to be predictive of weight change after interventions, although only in two of the seven intervention studies42,45. Both studies
investigated behavioral interventions and the weight losses after the interventions was for also approximately similar (around 2-4% of the baseline weight). However, the tasks, moment of follow-up and the measurement of brain activation was different. As mentioned before, Hege et al.42 investigated
brain activation using a one-back memory task (with both food and non-food images intermixed) with the contrast responders versus non-responders, while Murdaugh et al.45 employed a passive viewing
task and contrasted high-calorie foods versus non-foods. Furthermore, the former study of Hege et al.42 used baseline activation before the intervention to predict weight change after the interventions, whereas Murdaugh et al.45 here investigated brain activation after the intervention to predicted
long-term weight loss nine months after the intervention had ended. Possibly because of these differences, the direction of the hippocampal activation on weight change was dissimilar for the two studies. Hege et al.42 found that individuals who responded to the intervention showed higher hippocampal activation at baseline, whereas Murdaugh et al.45 demonstrated that lower activation after the
intervention was predictive of successful weight maintenance nine months later.
The results of Hege and colleagues42 suggest that individuals who lost weight are better able to
recruit the hippocampus during a memory task at baseline, which might also be useful during the intervention because the hippocampus is pivotal for learning new associations and goal-directed behavior4,141. Furthermore, it has been hypothesized that the hippocampus is involved in interoception
and inhibition of appetitive behavior, and it is important for the retrieval of memories about the previous eaten meal142,143. It has been shown that participants who are being remembered to the
previous they ate, actually eat less during the current meal123. Thus, people with higher hippocampal activation might be better able to learn from the intervention, have improved memory for previous meals and have enhanced perception of internal hunger and satiety signals, all resulting in less food intake and thus weight loss.
Murdaugh et al.45 failed to find predictive hippocampal activation during the baseline session,
but showed that higher hippocampal activation after the intervention was predictive of less successful weight maintenance in the nine months after this scan session. Although it is known that many weight-loss interventions do not have an effect, or even a opposite effect, on the long-term, this is actually the only study that acquired follow up data to predict weight maintenance after an intervention. Their results could be interpreted by the findings of a meta-analysis that showed that the hippocampus is implicated in the processing of (positive) rewards, which might suggest that individuals who do not lose weight experience greater desire when exposed to food images63. Indeed, ratings of craving for
favorite foods are correlated with hippocampal activation in obese persons126. Another study, which explored differences between obese, healthy-weight and postobese individuals (defined as persons who currently have a healthy, stable weight for at least three months but were obese before), showed that hippocampal response to tasting food in postobese individuals is still similar to that of obese individuals144. Furthermore, the hippocampus activation in described during the cue-trigged relapse in drug addiction145. Thus, following these lines of research we might conclude that higher hippocampal
activation to food cues is linked to higher craving and that although obese individuals might have lost weight, their hippocampus still responds the same as in obese individuals, which might induce relapse (to original weight or even weight gain).
Altogether, the results and the potential conclusions from these two studies are disparate, which might emerge from two aspects that differed between the studies: the task and the timing of the measurements. Maybe higher hippocampal activation during a general memory task (including both food and non-food images) is beneficial for weight loss (which is also supported by the fact that executive functioning is an important determinant of weight96), but increased responsivity to
high-calorie foods (versus contrast) in particular not. Similarly, greater hippocampal activation before the intervention might be useful during the intervention, while elevated hippocampal activity after the intervention might denote a ‘withdrawal syndrome’ from the weight loss during the intervention, that increases the risk of relapse146. Future studies on weight loss interventions should investigate baseline
and post-intervention hippocampal activity further, to provide more conclusive evidence to determine which direction this relationship has.
Finally, a shift in activation of the putamen from baseline measurement to after the intervention was predictive of weight change, although the direction of this change was not exactly similar for the two studies39,45. Murdaugh et al.45 found a decrease in putamen activity from baseline to post-intervention during the viewing high-calorie foods versus non-food to be predictive of weight loss. In the study of Deckersbach and colleagues39 was the direction of the change in activation depended on the different subregions of the putamen and the contrast investigated: an increase in ventral putamen activation in response to low calorie food (versus non-food) images and a decrease in dorsal putamen activation for high calorie food (versus non-food) images, were associated with more successful weight loss. The authors interpret this finding as a decrease in reward anticipation for high-calorie foods accompanied by elevated reward anticipation for low-calorie foods. Interestingly, the interventions had in common that they both focused on behavioral change, with topics such as goal-setting and motivation. That these behavioral-change studies observed a shift in putamen activity is in accordance with the role of the striatum (where the putamen belongs to) which is implicated in the translation from motivation into behavior62. Furthermore, the putamen itself is thought to be primarily involved in habit learning147, and also connects with regions like the mPFC and amygdala during the imaging of
achieving a goal148. Thus, the putamen might be involved in the change of reward anticipation that was
learned during the intervention, also changing habitual behavioral responses towards food, resulting in successful weight loss after the intervention. However, it should be noted that Deckersbach et al.39
used a very lenient threshold (p < 0.05 and p < 0.01, not corrected for multiple comparisons), rapidly increasing chances for false positive findings, so replication is warranted.
3.3 Comparing across the four measures of eating behavior
In the preceding chapters we summarized the brain regions that play a role in food choice, food intake, natural weight change and weight change due to an intervention. To get a grasp of the bigger picture we here combine and compare the results from these different measures of eating behavior. To date, no study has compared the brain regions involved in the different measures of eating behavior to assess whether they actually overlap. Moreover, it is unknown whether the neural processes underlying food choice and food intake also play a role in long-term weight change (natural or due to an intervention). This is crucial to know because eating behavior is determined by a series of single food decisions, and we can only influence long-term proxies of eating behavior by changing these decisions. Thus, to keep a healthy weight, it is necessary to apply long-lasting and persisting adjustments in food decisions, and not just the rejection of high-calorie food once. To gain more insight into this, we here compared neural activation across the different categories.
First, we made quantitative pairwise comparisons between the neural correlates of the different eating measures, to explore which neural correlates of eating measures agreed most. We observed that the food choice and intervention studies obtained most similar results, with twenty-seven brain regions that were both found in the food choice and intervention studies (Table 2). This suggests that interventions predominantly work via influencing the neural mechanisms of food choice. Indeed, several weight-reduction interventions included strategies (e.g., behavioral or lifestyle interventions) to change food choices (low-calorie instead of high-calorie food)39,43,45. Next, food
choice and natural weight change were also very similar (eighteen corresponding brain regions). Food intake and the long-term eating measures had least in common, which might indicate that the choice of food is more important for long-term weight changes than food intake. However, part of this finding might stem from the fact that we included more food choice studies than food intake studies, and that only 24% of the food choice studies used a ROI approach, while 60% of the food intake studies adopted this approach. Since this might underestimate actual occurrences, these results should be interpreted with caution.
To investigate which regions are similar or dissimilar between the short and long-term eating measures, we compared food choice and food intake with natural weight change and due to an intervention. We identified no brain region that was specific for a short or long-term eating measure exclusively. There was great overlap between the proximate and more distant measures of eating behavior. Most brain regions were found in multiple or all categories. Therefore, we decided to explore which brain regions were observed across all four eating behaviors. There were three brain regions that were described in the results of studies of all measures of eating behavior.
First, fourteen of all studies included in this review found the insula as region predictive of eating behavior (8/17 food choice studies11,12,13,15,19,20,22,24; 1/5 food intake studies28; 1/8 natural weight
change studies32; 4/7 intervention studies42-45). Although some of the directions of the relationship between insula activation and the prediction of eating behaviors were not completely consistent, it
does show that the insula has an important and complex function in all the proxies of eating behavior. Firstly, the insula includes a part of the primary taste cortex and is implicated in the integration of different senses, such as smell, taste and view70,75. Next, the insula is thought to be involved in
evaluating the value of food, for example, highly craved foods elicit stronger insula activation64. Moreover, insula activation correlates with basic characteristics of food, which suggests that it has a role in sensing the texture and viscosity of food, and even the amount of fat, sugar or calories in
food149,150. Additionally, another function of the insula is interoception72. Recently, it has been shown
that overweight and obese individuals have decreased insula activation (in response to food cues)59
and that they have weakened interoception151. This might cause greater food intake, as people with lower interoceptive awareness drink a higher amount of water during an ad libitum paradigm152.
Finally, it has been demonstrated that insula activation is influenced by feelings of hunger153 and that it is related to gastric distension and insulin responses following stomach distention154,155. It should be
noted that the insula consists of different subregions (anterior, middle and posterior), which might all have distinct functions. Future literature could differentiate between these parts of the insula to understand their different roles in eating behaviors better. Because of the many proposed functions of the insula, it is difficult to draw conclusions about its specific role in eating behavior. Considering the diversity of cognitive domains in which insula is involved, its involvement in the different categories might be due to low-level cognitive processes72,156,157.
Secondly, the ACC was found in ten studies: in seven out of seventeen food choice studies13,15,18,20,21,23,25, one out of five food intake studies29, one out of eight natural weight change studies35 and one out of seven intervention studies45. The food choice studies demonstrated a positive
relationship between ACC activation and the decision for a food item, as well as the long-term weight change studies (natural and due to an intervention), which showed that higher activation on baseline was predictive of greater weight gain or less successful weight loss. The ACC has been found to be activated during the mere viewing and smelling of food and during reward anticipation of the food
item57,75. This might suggest that the ACC activation seen during food choice is indicative of food
anticipation, and that increased food anticipation predisposes individuals to gain weight, which might ultimately lead to obesity. Indeed, a recent meta-analysis showed that obese individuals display more ACC activation during the viewing of food as compared to healthy controls59. However, the study that investigated food intake found a negative relationship between ACC activation and food intake: the higher ACC activity, the lower the food intake (or, in other words, the higher the satiation)29. This
could be because ACC activation was here measured after a preload, instead of after hours of fasting. Thus, lower activation after eating might reflect lower satiation, which is in accordance with a study that showed that overweight individuals show a decreased response in the ACC to food images during satiation116. Summarizing, we could state that higher ACC responsivity before eating, and possibly
