Introduction

The food addiction model revolves around the resemblance between addictions and severe forms of overeating behaviors. In essence, food addiction proposes that hyper-palatable food might lead to loss of control over eating and continued food intake despite adverse consequences in some individuals [1]. Moreover, it suggests the hypothesis that food addiction and drug addictions share similar etiological mechanisms [2]. Despite its increasing popularity, some aspects of the model remain highly debatable [3,4,5].

Food addiction can be used with two different meanings. The first one implies that some foods contain addictive agents able to trigger compulsive eating. In this vein, it has been shown that intermittent fasting can trigger compulsive intake of sugar in rodents [6]. However, nowadays there is not enough evidence to conclude that sugar or other nutrients have an addictive potential [7, 8]. Failure to identify the addictive components of food has led some researchers to re-name the phenomenon as “eating addiction” [8]. This alternative also emphasizes the trans-diagnostic similarities with non- substance addictions such as pathological gambling [8].

Food addiction can also be used to describe that some individuals suffer severe compulsive overeating. This symptom would be comparable to the losses of control observed in individuals with addictions [1]. Pathological overeating can occur in response to or as a prevention of distress [9]. Moreover, participants with high scores in food addiction tend to show higher impulsivity and reward sensitivity, suffer from frequent social isolation and stigmatization, and report the experience of intense food cravings [10]. In this review, we have adopted this second meaning to refer to food addiction, as it highlights a set of behavioral traits that can be operationalized for research or clinic purposes. Nevertheless, food addiction has not been included in the Diagnostic and Statistical Manual of Mental Disorders (DSM) and, as such, cannot be considered an accepted diagnostic concept. Having this in mind, a common approach to measure food addiction is the use of standardized psychometric instruments, such as the Yale Food Addiction Scale (YFAS) or its newer version, the YFAS version 2 (YFAS 2.0) [1, 11]. The scales were built based on the DSM criteria for substance-related disorders, and provide a symptom and a diagnostic score of food addiction [12].

High scores in food addiction are commonly observed in the context of eating disorders, namely, in bulimia nervosa (BN) and binge eating disorder (BED). Both psychiatric diagnoses are characterized by episodes of loss of control over eating [12]. Moreover, patients with BN show purging or other compensatory behaviors to avoid weight gain [12]. About 95% of individuals with BN meet criteria for food addiction, as measured with the YFAS 2.0 [13], while the correlation between binge eating episodes and YFAS 2.0 scores is approximately 60% [11]. Food addiction and BED might also be regarded as the higher ends of a spectrum in uncontrolled eating [14]. This way, uncontrolled eating can reflect difficulties in the regulation of food ingestion [14]. Uncontrolled eating is a construct captured by several eating-related questionnaires. Among them, the Reward-based Eating Drive (RED) scale incorporates items from different eating-related questionnaires reflecting lack of satiety, preoccupation with food, and lack of control over eating [15, 16]. As such, the RED scale can also be used to measure of food addiction traits.

The presence of analogous neural alterations between addictions and compulsive eating is often highlighted in neurobiology research [17]. Specifically, both food addiction and substance addictions seem to show anatomical and functional alterations in mesocorticolimbic and prefrontal areas [18]. These alterations might account for imbalances in reward sensitivity, cognitive control, and emotion regulation [19]. Evidence that high food addiction scores are associated with functional and structural brain differences, however, is mixed. Moreover, results supporting the food addiction model get often better visibility than the ones that do not align with it, even within the same paper.

To balance this, the present review is aimed at providing a critical overview of the available evidence on human neuroimaging research in food addiction. We first describe the relation between food addiction and obesity and refer to findings from structural neuroimaging. Next, we identify the three main neurobiological mechanisms associated with food addiction and present an overview of neuroimaging studies addressing these mechanisms. We finally suggest new avenues in the study of food addiction and how they can help overcome current limitations of neuroimaging research on this topic.

Disentangling Food Addiction and Obesity

Food addiction seems to share common neurobehavioral characteristics with obesity. These include high arousal response to potential rewards, lower cognitive control, and emotion dysregulation, and they can predispose some individuals to overeating when exposed to an obesogenic environment [19]. Although lean individuals might also show symptoms of food addiction [20], high food addiction scores are consistently associated with higher adiposity [20,21,22,23]. For example, the correlations between food addiction scores (measured with YFAS or RED scale) and BMI range somewhere between 0.23 and 0.36 [11, 16, 23, 24]. Moreover, symptoms of binge eating during adolescence seem to predict the prospective appearance of overweight and obesity [25]. This indicates that food addiction and obesity present a certain collinearity that should be taken into account in neuroimaging research.

Obesity has been associated with lower gray matter volume in brain areas that include the medial prefrontal cortex (extending to orbitofrontal cortex), cerebellum, and left temporal pole [26,27,28]. There is some evidence that food addiction is also associated with gray matter differences in these areas. Specifically, two studies suggest that both food addiction and obesity might present a certain overlap in the orbitofrontal cortex. Beyer et al. examined the neuroanatomical correlates of food addiction and BMI in a population-based sample of 625 adult participants [29•]. In a whole-brain analysis, YFAS scores did not show associations with cortical thickness. However, an additional regions of interest (ROI)–based analysis showed that higher BMI and food addiction symptoms were both independently associated with lower cortical thickness in the right lateral orbitofrontal cortex [29•]. Another study on 91 adolescents reported that, relative to their lean counterparts, individuals with obesity showed lower gray matter volume in the orbitofrontal cortex [30].

Higher scores in uncontrolled eating across participants, moreover, were associated with higher BMI and lower gray matter volume in the orbitofrontal cortex [30]. Early studies on the neural correlates of BN are notably inconsistent, and their methodological limitations have been noted somewhere else [31]. More recent results, however, show that BN symptoms are negatively correlated with cortical thickness in the orbitofrontal cortex [32, 33].

The cross-sectional nature of these studies prevents us from concluding upon the longitudinal relations between food addiction, obesity, and their brain correlates. However, these works suggest the involvement of the orbitofrontal cortex in uncontrolled eating. Moreover, they highlight the need to examine food addiction and obesity both separately and in relation to each other.

Neuroimaging Correlates of Food Addiction

In previous publications [14, 19], we have delineated three neurobehavioral mechanisms associated with food addiction and uncontrolled eating: (i) increased reward sensitivity; (ii) decreased cognitive control; (iii) increased stress reactivity. This conceptualization presents similarities to other works [34]. In the following, we structure findings from human neuroimaging studies on food addiction around these three elements.

Reward Sensitivity

The dopamine system plays a central role in the processing of rewards. Dopamine mediates desire for food and other rewards, usually referred to as “wanting” [35]. It has been hypothesized that dopaminergic dysfunctions increase the susceptibility to addictive behaviors and unhealthy weight gain via alterations in reinforcement learning, motivation, and self-regulation [17]. In this vein, three interrelated mechanisms have been suggested to explain the link between dopamine alterations, food addiction, and weight gain (Fig. 1). First, tendencies to pursue immediate rewards along with hyper-activity of the reward system might facilitate problematic overeating and weight gain [36, 37]. This way, increased responsivity in the reward system constitutes a risk factor for the development of food addiction. Second, studies on drug dependence have shown that the repeated exposure to addictive substances produces a “sensitization” of the dopamine circuits [38, 39]. This sensitization seems to play a key role in the escalation from drug use to drug abuse [38, 40]. In a similar vein, the incentive sensitization model of obesity suggests that some types of obesity are associated with increased cue-triggered desire (wanting) for foods. These increases in food wanting might be regarded as consequences rather than causes of obesity and food addiction [41]. As such, this model delineates neurobehavioral mechanisms involved in the maintenance of obesity and food addiction patterns. Finally, in substance addictions, the persistent abuse of drugs can lead to a state of dependency characterized by compromised tonic and phasic dopamine function [38]. In these severe phases of addiction, individuals might use psychoactive substances to compensate for their dopaminergic deficits [42]. This hypothesis has also been translated to the field of food addiction and obesity and it is often referred to as the reward deficiency theory [17, 43]. According to this theory, individuals with blunted in dopaminergic transmission might show compulsive overeating to balance their dopamine deficits. The reward deficiency theory might also help to explain why food addiction can become a chronic condition.

Fig. 1
figure 1

Three theoretical mechanisms that link food addiction and obesity with alterations in the mesolimbic system. The first mechanism hypothesizes that hyperactivity of reward system constitutes a risk factor for the development of food addiction and obesity [36]. The second theory proposes that a sustained hyper-sensitivity to rewards underlies the maintenance of addictive-like behaviors in obesity [41]. Finally, the reward deficiency theory suggests that individuals with deficits in dopaminergic transmission show compulsive overeating to compensate for their deficits [43]. A problem in the field is that these theories allow for any empirical finding on the mesolimbic system to be overinterpreted as an altered signature of food addiction or obesity. The pre-registration of hypotheses and statistical analyses might help overcome this problem in the future

Using fMRI, several studies have addressed the hypothesis that greater reactivity in the reward system precedes food addiction and weight gain. In an influential study, Stice et al. reported that striatal fMRI activity in response to foods is associated with longitudinal weight gain, and that this relationship was dependent on the DRD2 Taq1 A1 allele which influences DRD2 availability [44]. The relationship between striatal activity and prospective weight gain has also been shown in other studies [45, 46]. Similar to this, another paper reported that functional connectivity in the striatum is associated with food cravings and with longitudinal changes in BMI [47]. However, the same authors that initially described this phenomenon have questioned its reliability [48•]. Stice and Yokum tested the relationship between fMRI response to food intake and prospective weight changes on 135 adolescents, a larger sample size than previous studies. In a whole-brain analysis, the pre-supplementary motor area was the only region associated with prospective weight changes (i.e., weight loss). Next, the authors focused on the striatum in an ROI analysis. None of the striatal structures showed any associations with longitudinal weight changes [48•]. Two additional re-examinations of this same sample of adolescents yield mixed support to the hypothesis that increased striatal activity is associated with weight gain. In the first of them, the authors compared fMRI response with 4 different milkshakes that combined sugar (high and low) and fat (high and low) content with a tasteless solution. They compared participants at high risk of obesity (defined as having 2 parents with overweight) and individuals at low risk of obesity (defined as having at least one lean parent). A contrast between all 4 milkshakes versus the tasteless solution showed that participants at high risk had higher fMRI activity in the postcentral gyrus, orbitofrontal cortex, precentral gyrus, and insula. A contrast between high-sugar milkshake and the tasteless solution showed that participants with high risk of obesity showed higher activity in the superior temporal gyrus, central operculum, juxtapositional lobe, thalamus, and caudate. All the other contrasts yielded no group differences [49]. In a second paper, the authors compared fMRI response with milkshake and with food pictures during baseline and after a 2–3 follow-up period. Participants who gained weight, compared with participants who remained stable, showed longitudinal reductions in fMRI activity in the postcentral gyrus, middle frontal gyrus, and medial prefrontal cortex when comparing the receipt of a high-fat/low-sugar milkshake and a low-fat/low-sugar milkshake. Moreover, participants who gained weight showed longitudinal decreases in middle temporal gyrus activity and increases in cuneus activation compared with weight-stable participants in response to appetizing versus unappetizing food pictures [50]. Considering all these findings together, it is fair to say that nowadays there is no solid evidence of association between fMRI striatal activity and weight gain. Other studies have also used food-related fMRI paradigms to examine neural differences associated with food addiction. In response to anticipated receipt of food, participants with food addiction showed higher fMRI activity in the dorsolateral prefrontal cortex (DLPFC) and caudate along with lower activity in the orbitofrontal cortex [51]. Another study reported subtle fMRI differences in participants with food addiction during a visual food study that contrasted processed versus unprocessed food stimuli [52]. Participants with food addiction showed high activity in the superior frontal gyrus (within the DLPFC) in response to processed food and low activity in response to unprocessed food. The pattern was opposite in participants without food addiction [52].

Complementing these findings, some studies have performed correlations between continuous measurements of food addiction and fMRI activity. Among them, at least two investigations using a whole-brain approach have reported null correlations between food addiction scores and fMRI differences [53, 54]. One of these studies, however, included an additional ROI analysis and found that food addiction symptoms showed a positive correlation with activity in the left amygdala in response to high- versus low-calorie food cues [53]. Finally, another study examined the effects of food craving during an fMRI task that presented appetizing versus plain food images. Food craving was positively associated with activity in the DLPFC in overweight adolescents [55]. Overall, these studies suggest that food addiction is related to either small or null differences during food processing.

Some studies have examined monetary reinforcement learning in individuals with binge eating symptoms. Participants with BED seem to show decreased fMRI activity in the striatum [56] and in the ventromedial prefrontal cortex [57] relative to control participants. These findings suggest that individuals with binge eating might fail to incorporate learning signatures, this way compromising flexible behavioral adaptation [57]. In BN, a double-blind crossover study tested whether catecholamine depletion induced by alpha-methyl-paratyrosine (AMPT) affected performance in a monetary-incentive delay (MID) task in participants with BN in remission and control participants. In the placebo condition, control participants earned significantly more money than individuals with BN in remission. Both groups, however, showed equivalent performance under AMPT [58••]. These results suggest that catecholamine depletion has a smaller effect on reward processing in BN than in control participants. The authors hypothesized that this lesser effect could be due to pre-existing dopamine deficits in BN [58••], in line with the reward deficiency theory of food addiction.

Another piece of evidence linking dopamine signaling to uncontrolled eating is the occurrence of compulsive eating and weight gain in patients treated with dopamine agonists (for Parkinson’s disease or restless legs syndrome) [59, 60]. These drugs are also associated with other behavioral addictions such as problem gambling [60]. Moreover, preliminary evidence suggests that genetic variations in dopamine receptors, such as the DRD2/ANKK1 Taq1A polymorphism, might be associated with uncontrolled eating [61] and with delay discounting, a measure reflecting economic impulsivity [62]. At the same time, at least two meta-analyses have concluded that there is no consistent evidence supporting an association between DRD2/ANKK1 Taq1A polymorphism and obesity [63] or addictions (in this case, smoking) [64].

Finally, results from nuclear imaging studies in food addiction seem to align with the incentive sensitization model. For instance, high scores in uncontrolled eating have been associated with increased availability of one type of dopaminergic receptors (dopamine receptor D2 (DRD2)) in the lateral striatum [65]. Another study found that, among obese individuals, there was a positive correlation between DRD2 receptor availability in the striatum and greater delay discounting [66]. At the same time, preliminary data suggests that participants with BN might have lower dopamine release in the putamen [67]. Moreover, with regard to obesity, it is unclear whether a high BMI is associated with higher [68] or lower [43] dopamine D2 receptors. Severity of obesity [69] and age [70•] are some of the factors that might mediate this association.

Overall, the idea that food addiction and weight gain arise as a result of the alteration of dopamine mesolimbic circuits is intuitive and provides an interesting narrative. However, there is a marked distance between this theory and the empirical findings from human neuroimaging. Results from studies on reward-related fMRI activity and weight gain are inconsistent and difficult to integrate with each other. Studies including measures of food addiction mostly suggest the existence of null and positive correlations between fMRI activity in response to food reward and food addiction scores. On the other hand, studies on monetary reinforcement learning show decreased fMRI activity in BED patients compared with controls. One potential reason for variability in findings is the methodological heterogeneity across studies—with regard to the age of participants, the severity of food addiction or binge eating, or task-related parameters such as valence or arousal of the stimuli. A number of fMRI studies, moreover, have been performed with small sample sizes (i.e., less than 30 participants in the group of interest). This decreases the statistical power of the study and increases the margin of error.

Cognitive Control

Decreased general cognitive control is another important hallmark of addiction [19]. Such decreases have also been previously observed in food addiction, in obesity, and in individuals with binge eating symptoms [19, 71, 72]. It shows that, also in this respect, those conditions might be similar to each other. Generally, cognitive control consists of an implementation system and performance monitoring system, which are related to the prefrontal cortex and the anterior cingulate cortex, respectively [19, 73, 74]. In the context of obesity, decreased cognitive control is related to decreases in the prefrontal cortical activity and thickness [74,75,76,77,78,79]. In the food addiction context, one study found similar results [71]. Here, food addiction was related to decreased performance monitoring on a flanker task, a test of cognitive control. This was paralleled by lower error-related negativity (less negative amplitude) in EEG measurements in frontal and central electrodes (Fz and Cz) during task performance, suggesting that such lower performance monitoring in food addiction is related to deficient frontal cortical activity. Additionally, error-like positivity on Cz electrode correlated negatively with food addiction severity (as measured by YFAS) corroborating the interpretations that individuals with food addiction show lower performance monitoring. In the binge-eating context, Oliva and colleagues tested normal-weight individuals with and without binge eating episodes on Go/No-Go and Stop Signal Reaction time tasks [80]. Here, individuals with binge eating episodes showed lower activation of the right middle frontal gyrus in the non-food no-go trials [80]. Moreover, individuals with binge eating episodes showed higher activation in the left middle frontal gyrus in food stop trials during Stop Signal Reaction Time task performance. The Go/No-Go task results are in line with a study showing similar decreases in brain activity (as measured by MEG) in the right middle frontal gyrus for the BED group in the food no-go condition [81]. Additionally, the Stop Signal Reaction Time task results were corroborated in a study by Bartholdy and colleagues that used data from the IMAGEN cohort (2019). They showed that, in adolescents, development of binge eating and purging behavior is related to greater recruitment of the medial prefrontal cortex and anterior cingulate cortex as measured by fMRI during failed versus successful Stop Signal Reaction Time stop trials [82•]. In general, authors interpret those findings as reflecting decreased inhibitory control in individuals with binge eating episodes. The directionality of those alterations, however, might be interpreted in two ways. Lower activity of the right middle frontal gyrus might mean decreased overall inhibitory control. Higher activity in the left middle frontal gyrus during Stop Signal Reaction Time task, on the other hand, might mean increased demands for inhibitory control during the cancellation of an already initiated action. However, the validity of these hypotheses remains to be established. Further evidence for the role of prefrontal areas in food addiction comes from the field of neuromodulation. Increases in the excitability of the DLPFC seems to reduce food cravings [83]. This effect is substantially heterogeneous across studies, as highlighted in a meta-analytic review [83]. In fact, a recent preregistered report showed that transcranial direct current stimulation over the bilateral DLPFC does not affect craving, food consumption, or response inhibition on a Go/No-Go task [84••]. This last study, however, used a sample of normal-weight individuals and did not measure food addiction or binge eating symptoms.

Studies on resting-state fMRI might also answer the question of whether food addiction/BED is associated with alterations in cognitive control areas. In an unpublished study, Oliva and colleagues found that normal-weight individuals with binge eating episodes have lower degree centrality in the right middle frontal gyrus during resting-state fMRI as compared with normal-weight controls [85]; same sample as in [80]). Two other studies, however, did not find resting-state differences between individuals with BED and control groups [86, 87].

Overall, these studies suggest that decreased cognitive control in food addiction is related to altered function of the prefrontal cortex. However, the directionality of such alterations might be interpreted in two ways and it remains to be established whether such interpretations are correct. It is also possible that the individual mindset/current goals might have an influence on cognitive control [88].

Stress Reactivity

Research in addiction has suggested that stress and negative emotionality can trigger compulsive behaviors (such as compulsive overeating) as a means to deal with internal and external pressures [34, 89]. Accordingly, negative mood increases the risk of binge eating episodes in individuals with eating disorders [90, 91]. The application of emotion regulation strategies, on the other hand, reduces the incidence of binge episodes [91].

Chronic stress (and chronically elevated glucocorticoids) can lead to increased food intake by enhancing the preference for the so-called comfort foods [92]. Elevated glucocorticoids also tend to facilitate visceral fat accumulation [92].

Although differences in stress reactivity are likely to play a role in the neurobiology of food addiction and binge eating symptoms, no studies to date have thoroughly examined this relation. A preliminary study on stress compared fMRI activity between food and non-food photographs in participants with BED and control participants. Participants with BED showed lower activity in response to food stimuli in the superior frontal gyrus and insula. The authors did not report a main effect of the stress condition [93]. As such, these findings do not allow any conclusions on stress reactivity and BED.

Conclusions

The food addiction model proposes that compulsive overeating is associated with brain alterations that are similar to the ones found in drug addiction. However, as the number of human neuroimaging studies in food addiction has increased, evidence has become more inconsistent. Here we have provided an overview of human neuroimaging studies on the topic. In the following, we suggest new avenues in the study of food addiction and how they can help overcome the limitations of current research.

Neuroanatomical studies have suggested that food addiction is associated with subtle gray matter/cortical thickness differences in the orbitofrontal cortex [29•, 30]. In one of these studies, moreover, these differences appeared to be somewhat independent from obesity [29•]. We suggest that future studies should test both the independent and cumulative effects of food addiction and BMI on brain differences. Related to this, although some papers have collected obesity-related longitudinal measures [48•], the vast majority of research in food addiction is cross-sectional. In this vein, longitudinal designs are a promising research avenue in the identification of factors that might lead to food addiction and obesity.

fMRI studies in food addiction have provided limited consistency or replicability. While some findings suggest that food addiction is associated with greater activity in prefrontal areas during food reward tasks [52], evidence for null results also exists, and results are not generalizable across different reward paradigms [57]. fMRI studies using cognitive control tasks in food addiction are scarce. Some studies point at higher activity in the DLPFC during the inhibition of an already initiated response in food addiction/binge eating [80, 82•]. However, studies using different tasks have also provided evidence for the opposite direction (decreased prefrontal activity in binge eating) [80].

One of the reasons for this might be the intrinsic heterogeneity that food addiction presents. Food addiction is characterized by losses of control over food consumption, repeated food ingestion despite negative consequences, and food craving, among other symptoms [10, 94]. However, each of these traits might tap into different neuropathological mechanisms. For example, the DLPFC might play an important role in food cravings, as suggested by some studies on neuromodulation [83, 95]. Conversely, the hypothalamic-pituitary-adrenal axis might drive compulsive overeating in response to stress [9]. The examination of each of the behavioral traits related to food addiction separately might provide a clearer definition of the brain circuits involved.

Another problem in the field is that it allows for any finding to be interpreted as an abnormal neurobehavioral signature of food addiction. For example, findings of enhanced activation in obesity or food addiction in response to reward are generally interpreted as supportive of the dopamine sensitization hypothesis [96]. Conversely, decreased mesolimbic engagement in obesity or food addiction has been linked to dopamine hypofunctionality, as described in severe phases of drug addiction [43]. With regard to cognitive control, if individuals with food addiction/binge eating show higher brain activity in cognitive control areas, it can be concluded that they require an additional engagement of resources in order to exert self-regulation [82•]. Conversely, when participants with food addiction show lower brain activity in cognitive control areas, researchers conclude that they show deficits in the recruitment of cognitive control regions [97]. These conclusions might arise from the pre-existing assumptions of the researchers rather than from the robustness of the methodological procedures.

Relatedly, functional neuroimaging results are often presented together with findings of null behavioral differences in food addiction/binge eating during cognitive control tasks. In those particular cases, describing neuroimaging differences as “alterations,” “dysfunctions,” or “deficits” should be avoided, since it might contribute to stigmatization. Moreover, these labels are not warranted from a medical standpoint, since the differences obtained might not represent a deficit. Finding positive neuroimaging results along with null behavioral results might reflect that neuroimaging measures are more sensitive to detect subtle group differences. Alternatively, it might also mean that the mere neural difference observed does not have behavioral consequences.

We also believe that the field of food addiction will benefit from practices aimed at improving reproducibility and transparency [98]. This is especially so since neuroimaging data allows substantial analysis flexibility and multiple testing opportunities. Specifically, we propose that future studies in food addiction adopt open science practices, such as pre-registration of study hypothesis and methods. In fact, some studies have already implemented such strategies [84••]. Additionally, the use of large sample sizes (n > 100) is becoming a standard in neuroimaging research. We think that it would be beneficial if food addiction literature also followed this trend to increase replicability and achieve satisfactory statistical power.

In conclusion, neuroimaging studies have examined whether food addiction is associated with neuroanatomical and functional brain changes. These studies, however, have provided mixed results. This gives one the impression of skating on thin ice when attempting to draw conclusions based on them. To overcome this, we propose that future research should address four main points: (a) examine food addiction and obesity separately and in relation to each other; (b) discriminate between causes and consequences of food addiction; (c) take into account the heterogeneity of food addiction; (d) adopt research practices that prevent overinterpretation and promote replicability.