Cognitive reappraisal of food craving and emotions: a coordinate-based meta-analysis of fMRI studies

Abstract

Growing evidence supports the effectiveness of cognitive reappraisal in down-regulating food desire. Still, the neural bases of food craving down-regulation via reappraisal, as well as their degree of overlap vs specificity compared with emotion down-regulation, remain unclear. We addressed this gap through activation likelihood estimation meta-analyses of neuroimaging studies on the neural bases of (i) food craving down-regulation and (ii) emotion down-regulation, alongside conjunction and subtraction analyses among the resulting maps. Exploratory meta-analyses on activations related to food viewing compared with active regulation and up-regulation of food craving have also been performed. Food and emotion down-regulation via reappraisal consistently engaged overlapping activations in dorsolateral and ventrolateral prefrontal, posterior parietal, pre-supplementary motor and lateral posterior temporal cortices, mainly in the left hemisphere. Its distinctive association with the right anterior/posterior insula and left inferior frontal gyrus suggests that food craving down-regulation entails a more extensive integration of interoceptive information about bodily states and greater inhibitory control over the appetitive urge towards food compared with emotion down-regulation. This evidence is suggestive of unique interoceptive and motivational components elicited by food craving reappraisal, associated with distinctive patterns of fronto-insular activity. These results might inform theoretical models of food craving regulation and prompt novel therapeutic interventions for obesity and eating disorders.

activation likelihood estimation, cognitive reappraisal, rehabilitation, food craving regulation, emotion regulation

Introduction

Food craving can be defined as an affective state comprising the subjective experience of wanting a specific food and the strong appetitive motivation to approach and consume it (Giuliani and Pfeifer, 2015). Despite being often elicited by palatable energy-dense foods, food cravings are highly idiosyncratic (Rozin and Vollmecke, 1986) and can be distinguished from ordinary food choices in their intensity and specificity, arising independently from the momentary hunger state (Pelchat et al., 2004). There is substantial agreement on the multidimensional nature of food craving, including cognitive (e.g. attentional biases and intrusive thoughts), affective (e.g. intense desire to eat the craved food), motivational (e.g. actively seeking the desired food) and physiological (e.g. increased salivation and heart rate) components (Giuliani and Berkman, 2015).

The neural bases of food craving responses to palatable food have been widely investigated. According to prior meta-analyses (van der Laan et al., 2011; van Meer et al., 2015), the exposure to visual food stimuli consistently elicits neural activity in regions related to reward anticipation and hedonic processing, including the ventral striatum (VS), insula, amygdala, orbitofrontal cortex and ventromedial prefrontal cortex (vmPFC). In addition to this network coding for appetitive motivations (Canessa et al., 2011, 2013, 2017, 2022; Dagher, 2012), food perception reliably activates brain regions that support visual and somatosensory processing, such as the fusiform gyrus, primary gustatory cortex and primary somatosensory cortex (van der Laan et al., 2011; van Meer et al., 2015). In particular, the anterior insula has been suggested as a possible hub region integrating visual, olfactory and gustatory food-related information to prompt physiological and psychological responses to food stimuli (Huerta et al., 2014). Interestingly, such neural responses to food are known to be modulated by individual differences in body mass index (BMI). In detail, obesity has been associated with persistent food-related hyper-activation of brain regions related to reward and salience processing, as well as visual and gustatory perception, alongside reduced activity in areas supporting cognitive control and interoceptive awareness, such as the left dorsolateral prefrontal cortex (dlPFC) and insula (Brooks et al., 2013; Kennedy and Dimitropoulos, 2014; Pursey et al., 2014; Devoto et al., 2018). Altogether, these findings are relevant for clinical practice, as they are indicative of a hypersensitivity to interoceptive signals of hunger and increased incentive sensitization in obesity (Robinson and Berridge, 1993; Simmons and DeVille, 2017), paralleled by inadequate inhibitory control over food craving.

Indeed, despite being surrounded by palatable foods that might elicit intense craving states, individuals are endowed with the ability to engage in cognitive regulation and resist potentially harmful temptations that conflict with long-term health goals. Building upon an extensive literature on emotion regulation, it was suggested that one of the most effective ways to down-regulate food craving is cognitive reappraisal, i.e. the deliberate reinterpretation of an emotion-eliciting stimulus or event, aimed to change its affective meaning (Gross, 1998; Giuliani and Gross, 2009). Within the process model of emotion regulation (Gross, 1998), cognitive reappraisal constitutes an antecedent-focused regulation strategy, usually aimed at down-regulating the impact of negative emotions. However, reappraisal can likewise be employed with the opposite, yet complementary, goal of up-regulating emotional responses, mostly positive ones (McRae et al., 2012; Ochsner et al., 2012). In recent years, cognitive reappraisal has proved effective in down-regulating craving for various appetitive stimuli, such as nicotine (Kober et al., 2010a,b), alcohol (Naqvi et al., 2015; Suzuki et al., 2020) and food (e.g. Kober et al., 2010b; Hollmann et al., 2012; Giuliani et al., 2013). Analogous cognitive regulation techniques are indeed crucial components of cognitive-behavioural treatments for addiction, smoking cessation and eating disorders, which have yielded promising results in reducing craving and avoiding relapse over time (e.g. Ashton et al., 2009; Grilo et al., 2011). Similarly, in the context of eating behaviours, food craving might be effectively down-regulated via cognitive reappraisal strategies that can be flexibly generated depending on context-specific features and goals. For instance, individuals might override their momentary craving states by thinking about the long-term negative health consequences of repeatedly consuming the desired food (e.g. weight gain), also known as ‘costs of eating’ framing (Yokum and Stice, 2013). Otherwise, they might adopt the alternative ‘benefits of not eating’ tactic, by focusing on the health advantages of refraining from eating the craved food (e.g. weight loss). Cognitive reappraisal thus represents a promising avenue for therapeutic interventions, as it directly targets the motivation to engage in overeating of unhealthy foods rather than relying on externally mediated changes in eating habits (Giuliani et al., 2013).

The so-called Regulation of Craving (RoC) task (Kober et al., 2010a,b) has been often used to investigate the effects of cognitive reappraisal on food craving reduction. In this task, participants are instructed to observe the pictures of palatable, often energy-dense, foods and to respond to these images either by naturally admitting their current craving (i.e. ‘Look’ condition) or by down-regulating the desire to eat the depicted food via reappraisal strategies (i.e. ‘Decrease’ condition). At the behavioural level, cognitive reappraisal has proved effective at decreasing subjective food craving, leading to significantly lower self-reported craving intensity after the ‘Decrease’ vs ‘Look’ trials (e.g. Kober et al., 2010b; Giuliani et al., 2013, 2014; Yokum and Stice, 2013; Giuliani and Pfeifer, 2015). Parallel versions of the RoC task have included a third set of instructions to investigate the opposite goal of food craving up-regulation (i.e. ‘Increase’ condition) (e.g. Hutcherson et al., 2012; Scharmüller et al., 2012; Ferreira et al., 2019, 2021; Janet et al., 2021). Outside the laboratory, similar reappraisal strategies could therefore prompt long-lasting changes in eating behaviour not only by restraining the desire for unhealthy foods but also by enhancing craving and motivating consumption towards healthier food options. For instance, Boswell et al. (2018) showed that a brief training in cognitive reappraisal was able to produce a significant reduction in total caloric intake, mainly from unhealthy foods.

At the neural level, several meta-analyses have helped clarifying the neural correlates of emotion regulation. Overall, emotion down-regulation via reappraisal appears to consistently engage a (mostly left-hemispheric) frontoparietal network including the dlPFC, ventrolateral prefrontal cortex (vlPFC) and posterior parietal cortex (PPC), along with the supplementary motor area (SMA) and pre-SMA, insula and left posterior temporal cortex (Buhle et al., 2014; Frank, et al., 2014; Kohn et al., 2014; Morawetz et al., 2017, 2020). In this context, the frontoparietal control system likely underpins higher-order executive processes that support perceptual and semantic re-evaluation of emotional stimuli in temporal regions, which in turn indirectly modulate neural activity within the network generating appropriate emotional responses (Ochsner et al., 2012).

Concerning the down-regulation of food craving by means of reappraisal, instead, evidence from previous functional magnetic resonance imaging (fMRI) studies seems to converge on two major findings (Kober et al., 2010b; Hollmann et al., 2012; Giuliani et al., 2014, 2020; Giuliani and Pfeifer, 2015; Cosme et al., 2018; Janet et al., 2021). Firstly, the recruitment of frontoparietal regions involved in cognitive control—including the dlPFC, vlPFC, SMA/pre-SMA, anterior cingulate cortex and inferior parietal lobule (IPL)—during the down-regulation of food craving. Secondly, food craving reappraisal additionally reflects in decreased neural activity in brain regions related to reward and hedonic processing (e.g. VS and vmPFC), paralleling the behavioural outcome of craving reduction. This literature however shows some inconsistent findings across studies, likely related to between-experiment variability in task design and sample composition (Kober et al., 2010b; Siep et al., 2012; Yokum and Stice, 2013; Cosme et al., 2018; Ferreira et al., 2021). For instance, some studies reveal an extensive set of regions associated with the down-regulation of food craving via reappraisal (Giuliani et al., 2014; Cosme et al., 2018), whereas others fail to find any significant activation for this contrast (Ferreira et al., 2021). A synthetic overview of previous neuroimaging results is thus required to unveil the regions that are consistently associated with domain-specific food reappraisal.

Interestingly, the regions that have been associated with food craving reappraisal strikingly resemble those reported in prior meta-analyses of emotion reappraisal (Buhle et al., 2014; Morawetz et al., 2017). This would support conceptual definitions of food craving as an affective state, whose regulation can be operationalized within the extended process model of emotion regulation (Giuliani and Berkman, 2015). Nonetheless, food craving presents some distinctive elements, such as the intense motivational component to consume the craved food, which are not necessarily shared with other affective states. To date, the similarity between food craving and emotion reappraisal has been outlined only from a theoretical standpoint. A systematic quantification of the extent to which domain-specific food craving reappraisal relies on overlapping vs distinct sets of neural regions compared to emotion reappraisal is therefore needed to advance knowledge in this field.

Prior theoretical models are expected to suggest an extensive overlap across the neural bases of food craving and emotion down-regulation via reappraisal, particularly including frontoparietal regions underlying higher-order executive processes (e.g. dlPFC, vlPFC and IPL). Still, the distinctive hedonic and motivational features of food craving suggest the engagement of specific brain regions over and beyond those recruited by emotions. One promising candidate is the insula, especially its anterior portion. This region might mediate the greater integration of interoceptive and salience signals when generating reappraisals to regulate food craving compared with emotions, for instance by evaluating one’s future bodily states according to long-term health goals in order to translate visceral states into goal-directed action plans (Simmons and DeVille, 2017; Wager and Barrett, 2017; Han et al., 2018).

To test these hypotheses, we performed two coordinate-based meta-analyses of fMRI studies on the down-regulation of either food craving or emotions by means of cognitive reappraisal. An activation likelihood estimation (ALE) approach was used to identify the regions that have been consistently associated with either of the conditions and then to assess their overlap vs specificity via conjunction and subtraction analyses, respectively. We extended this approach to two exploratory meta-analyses aimed to assess (i) the neural bases of spontaneous food viewing compared to active regulation and (ii) the brain regions associated with the opposite goal of food craving up-regulation.

Materials and methods

Rationale of the meta-analytic approach

The aims of the present meta-analytic study were (i) to identify the brain regions consistently associated with down-regulating food craving by means of reappraisal and (ii) to assess the degree of overlap and specificity between the areas activated by domain-specific food craving down-regulation and domain-general emotion down-regulation via reappraisal. Concerning food processing, we also aimed to explore the regions reliably associated with (iii) spontaneous food viewing, compared to volitional regulation, as well as with (iv) food craving up-regulation. To pursue these goals, we opted for a coordinate-based ALE meta-analytic approach that employs coordinates of local maxima (i.e. foci) to synthetize and integrate published findings from multiple neuroimaging studies (Turkeltaub et al., 2002). This method overcomes the typical limitations of single fMRI studies, such as reliance on underpowered studies with small sample sizes, sensitivity to experimental and/or analytic procedures and lack of replication studies (Carp, 2012). These constraints are known to increase the likelihood of false negatives (Button et al., 2013), which in turn might promote the choice of analytic procedures that inflate false positive rates (Carp, 2012; Eklund et al., 2016; Müller et al., 2018).

We ran four separate ALE analyses to address the neural processing of food craving down-regulation, emotion down-regulation, food viewing compared to active regulation and food craving up-regulation in healthy individuals. Then, we performed conjunction and subtraction analyses to unveil both overlapping and specific activations across food craving and emotion down-regulation via cognitive reappraisal.

All the authors selected and approved the inclusion criteria and the keywords for literature search. To minimize the risk of selection bias and ensure the quality of inclusion, the entire procedure was carried out in a double-check fashion by independent investigators, as recommended by recent guidelines (Müller et al., 2018). The selection process began in May 2020, when M.G. and G.M. independently screened all articles retrieved for the ‘food craving down-regulation’ meta-analysis. This dataset was updated in September 2021 and approved by the other authors. For the ‘emotion down-regulation’ meta-analysis, C.M. provided the dataset of eligible articles from another meta-analytic study on emotion reappraisal and empathic perspective-taking (Morawetz et al., 2022).

Literature search and study selection

Food craving down-regulation and exploratory meta-analyses

We performed the literature search and study selection following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement (Moher et al., 2009; Page et al., 2021a,b) and the best-practice recommendations for neuroimaging meta-analyses (Müller et al., 2018). Firstly, we identified relevant keywords for our meta-analysis and determined the search string, which consisted of the following terms: (“reappraisal” OR “appraisal” OR “regulation” OR “control” OR “cognitive control” OR “suppression”) AND (“food” OR “appetite” OR “food craving”) AND (“fMRI” OR “BOLD” OR “MRI” OR “functional magnetic resonance imaging”).

We ran the same search string on five different databases: Embase (www.embase.com; date of consultation: 20 May 2020), PubMed (www.pubmed.ncbi.nlm.nih.gov; 22 May 2020), PsycINFO (www.apa.org/pubs/databases/psycinfo; 03 June 2020), Scopus (www.scopus.com; 03 June 2020) and Web of Science (www.webofscience.com; 03 June 2020). The search was constrained to studies on human subjects, written in English and published after 1990. Afterwards, on 8 September 2021, the database search was updated by narrowing the temporal filter from 2020 onwards, in order to retrieve the latest potentially relevant papers. The selection process was carried out on Rayyan (www.rayyan.ai), a web-based application for conducting systematic reviews and meta-analyses.

After duplicate removal, the literature search yielded a total preliminary pool of 4773 records. These records were screened first by title, then by abstract, by two independent investigators (M.G. and G.M.) in blind mode, leading to the exclusion of any articles that did not meet one or more of the following selection criteria:

• Published studies written in English;

• Empirical task-based fMRI studies. This criterion resulted in the exclusion of review and meta-analytic articles, behavioural studies and resting-state fMRI studies, as well as studies adopting other neuroimaging techniques such as positron emission tomography;

• Studies reporting whole-brain activation coordinates in standard anatomical reference space, either Talairach or Montreal Neurological Institute (MNI), leading to the exclusion of studies using regions of interest (ROIs) or small volume correction (SVC) analyses. A prerequisite for fMRI meta-analyses is indeed that convergence among experiments is tested against a null hypothesis of random spatial association across the whole brain, assuming that each voxel has the same a priori chance of being activated (Eickhoff et al., 2012; Müller et al., 2018). Hence, the inclusion of studies based on heterogeneous ROIs or SVC would violate this assumption, with the risk of an inflated significance for overrepresented brain regions.

• Studies including at least one group of non-clinical and drug-free participants, to avoid potential differences in brain activations related to pharmacological manipulations or co-occurring neuropsychiatric diseases;

• Studies with adult participants (age range: 18–60 years). Mixed samples ranging from childhood to adulthood were deemed acceptable for inclusion whenever the activation coordinates divided by age group could not be retrieved;

• Studies including at least one group of healthy weight participants (BMI range: 18.5–24.9 kg/m²) without co-occurring presence of eating disorders. Mixed samples with participants ranging from healthy weight to overweight or obesity (BMI > 25 kg/m²) were suitable for inclusion whenever the activation coordinates divided by weight group category could not be retrieved;

• Studies using a food craving reappraisal task, also known as RoC task. For the main meta-analysis, we selected studies performing the contrast Decrease > Look in the presence of visual food cues. The ‘Decrease’ condition comprises all trials in which participants are instructed to actively down-regulate their current desire for the depicted food via reappraisal strategies. The ‘Look’ condition includes all trials in which participants are required to passively watch the depicted food and naturally admit their craving. This selection criterion excluded tasks that do not entail any active regulatory effort (e.g. passive viewing of food cues vs fixation) or tasks using other food-related inhibitory control paradigms (e.g. Go/No-Go). To perform the two exploratory meta-analyses, we also examined whether each study reported the following contrasts:

  • Studies performing the inverse contrast Look > Decrease were included in the dataset for the ‘food viewing compared to active regulation’ meta-analysis;

  • Studies reporting the Increase > Look contrast were included in the dataset for the ‘food craving up-regulation’ meta-analysis. The ‘Increase’ condition comprises all trials in which participants are asked to up-regulate craving for the depicted food via reappraisal strategies.

After screening the preliminary pool of 4773 records by titles and abstracts, 96 papers were judged as potentially relevant, then retrieved in full-text format and carefully assessed for eligibility. Whenever a potentially eligible study did not report some of the required information (e.g. activation coordinates for the contrast of interest), the corresponding authors were contacted to obtain missing data. The selection procedure led to the exclusion of 1 review/meta-analytic article, 3 studies using other neuroimaging techniques, 1 study focused on clinical populations, 16 studies with passive viewing paradigms and 51 studies using inappropriate tasks and/or contrasts. Although a final pool of 24 studies fulfilled our inclusion criteria, we were unable to retrieve the activation coordinates for our contrast of interest (i.e. Decrease > Look in the presence of visual food stimuli) for 4 studies (Kober et al., 2010b; Siep et al., 2012; Miedl et al., 2018; Wilson et al., 2021) (see Supplementary Table S1 for further details on the specific reasons of exclusion from the quantitative ALE meta-analysis). Thus, 20 studies contributed to the quantitative ALE meta-analysis, while all 24 studies were included in the qualitative synthesis. In the attempt to expand the dataset, we conducted both forward and backward reference searches, i.e. examining papers quoting, or quoted by, each of the included articles. However, reference searching did not highlight further studies fulfilling our inclusion criteria. The detailed flowchart of literature search and selection, following the PRISMA 2020 guidelines (Page et al., 2021a,b), is depicted in Figure 1.

Fig. 1.

PRISMA 2020 flowchart of the literature search and selection process for the ‘food craving down-regulation’ ALE meta-analysis.

The 20 studies included in the ‘food craving down-regulation’ ALE meta-analysis (Table 1), corresponded to a total of 20 experiments (i.e. individual contrasts included in the analysis) with 856 subjects and 348 activation foci (20 out of mask). The detailed information regarding the four articles that contributed only to the qualitative synthesis is reported in the Supplementary Table S1.

Table 1.

Overview of the 20 studies included in the quantitative ALE meta-analysis on the neural bases of ‘food craving down-regulation’

N	Authors (year)	N° subjects (F/M ratio)	Age (years)	Weight status	Stimulus material	Task	Contrasts	Foci	Space
1	Cosme et al. (2018)	33 (16 F/17 M) (27 included in analyses)	Mean age (SD) = 18.12 ± 0.34		Images of personally craved foods	Craving regulation task (conditions: Look, Regulate)	(a) Regulate > Look	10	MNI
							(b) Look > Regulate	24
2	Cosme et al. (2020)*	166 (146 F/20 M)	Mean age (SD) = 30.0 ± 9.54	Mean BMI (SD) = 27.78 ± 8.05 kg/m2	Images of high-calorie (HC), either craved or not-craved, vs neutral low-calorie (LC) food	Craving regulation task (conditions: Look, Regulate)	(a) Regulate > Look	177	MNI
3	Dietrich et al. (2016)	43 (only F)	Mean age (SD) = 26.7 ± 3.5	Mean BMI (SD) = 27.5 ± 5.3 kg/m2	Images of HC foods	Food craving regulation task (conditions: Admit, Regulate)	(a) Regulate_tasty > Admit_tasty	24	MNI
							(b) Admit_tasty > Regulate_tasty	13
4	Ferreira et al. (2019)	14 (9 F/5 M) (control group)	Median (range) age = 23 ± 3.0	Healthy weight (no BMI data)	Images of unhealthy snacks	Regulation task (conditions: Natural, Distance, Indulge)	(a) Distance > Natural	1	MNI
							(c) Indulge > Natural	13
5	Ferreira et al. (2021)  *	14 (5 M/9 F) (control group)	Median (range) age = 30 ± 14.0	Mean BMI (SD) = 24.9 ± 3.7 kg/m2	Images of food items	Regulation task (conditions: Natural, Distance, Indulge)	(a) Distance > Natural	1	MNI
							(b) Natural > Distance	1
							(c) Indulge > Natural	1
6	Giuliani et al. (2014)	50 (33 F/17 M) (48 included in analyses)	Mean age (SD) = 21.77 ± 2.36	Mean BMI (SD) = 21.71 ± 2.8 kg/m2	Images of low- and high-energy-dense food, either craved or not-craved	Eating regulation task (conditions: Look, Regulate)	(a) Regulate > Look (all food)	12	MNI
							(b) Look > Regulate (all food)	3
7	Giuliani et al. (2020)	105 (73 F/32 M) (88 included in analyses)	Mean age (SD) = 39.2 ± 3.57	Mean BMI (SD) = 31.5 ± 3.9 kg/m2	Images of unhealthy craved and not-craved foods vs healthy foods	RoC task (conditions: Look, Regulate)	(a) Regulate_craved > Look_craved	8	MNI
							(b) Look_craved > Regulate_craved	16
8	Giuliani and Pfeifer (2015)	60 (only F)	Mean age (SD) = 16.66 ± 3.68		Images of low- and high-energy-dense food, either craved or not-craved	Eating regulation task (conditions: Look, Regulate)	(a) Regulate > Look (all food)	16	MNI
							(b) Look > Regulate (all food)	11
9	Hollmann et al. (2012)	20 (only F) (17 included in analyses)	Mean age (SD) = 25.3 ± 3.1	Mean BMI (SD) = 25.1 ± 3.5 kg/m2	Images of HC food items, rated either as tasty or not-tasty	Regulation task (conditions: Admit, Regulate)	(a) Regulate_tasty > Admit_tasty	11	MNI
10	Hutcherson et al. (2012)	26 (9 F/17 M)	Mean age = 22		Images of appetizing snack foods	Regulation task (conditions: Natural, Distance, Indulge)	(a) Distance > Natural	6	MNI
							(b) Natural > Distance	4
							(c) Indulge > Natural	7
11	Janet et al. (2021)	25 (12 F/13 M)	Mean age (SD) = 22.45 ± 3.88	Mean BMI (SD) = 21.74 ± 0.36 kg/m2	Images and odours of food items	Modulatory instructions task (conditions: Natural, Distance, Indulge)	(a) Distance > Natural	5	MNI
							(c) Indulge > Natural	5
12	Kohl et al. (2019)*	16 (12 F/4 M) (DLPFC group)	Mean age (SEM) = 29.25 ± 1.93	Mean BMI (SEM) = 31.63 ± 0.91 kg/m2	Images of HC and LC food items	Functional localizer task (conditions: Look, Down-regulation)	(a) Down-regulation > Look	11	MNI
13	Kruschwitz et al. (2018)	31 (16 F/15 M)	Mean age (SD) = 25.9 ± 3.34		Images of unhealthy snacks	Cognitive emotion regulation task (conditions: Now, Positive, Negative)	(a) Positive + Negative > Now	4	MNI
14	Demos Mcdermott et al. (2019)	30 (24 F/6 M)	Mean age (SD) = 42.3 ± 9.9	Mean BMI (SD) = 31.8 ± 3.5 kg/m2	Images of personally craved food items	Cognitive strategies task (conditions: Now, Later, Distract, Allow)	(a) (Later + Distract) > Now	14	MNI
							(b) Now > (Later + Distract)	15
							(c) Allow > Now	17
15	Scharmüller et al. (2012)	14 (only F) (control group)	Mean age (SD) = 25.6 ± 6.7	Mean BMI (SD) = 20.6 ± 1.3 kg/m2	Images of HC food items and neutral objects	Food craving regulation task (Watch, Decrease, Increase)	(a) Decrease_food > Watch_food	7	MNI
							(c) Increase_food > Watch_food	5
16	Silvers et al. (2014)*	105 (71 F/34 M)	Mean age (SD) = 14.27 ± 4.85		Images of unhealthy palatable food items	RoC task (conditions: Close, Far)	(a) Far > Close	8	MNI
							(b) Close > Far	2
17	Striepens et al. (2016)	31 (only F) (control group)	Mean age (SD) = 25.35 ± 4.37	Mean BMI (SD) = 22.26 ± 3.03 kg/m2	Images of palatable food items (sweets)	Craving regulation task (conditions: Now, Later)	(a) Later > Now	19	MNI
							(b) Now > Later	38
18	Tuulari et al. (2015)	14 (only F) (control group)	Mean age (SD) = 44.9 ± 11.9	Mean BMI (SD) = 22.6 ± 2.7 kg/m2	Images of food items	Appetite control task (conditions: Passive viewing, Inhibition, Imaginary eating)	(a) Inhibition > Passive view	3	MNI
							(b) Passive view > Inhibition	5
							(c) Imaginary eating > Passive view	2
19	Walter et al. (2020)	39 (18 F/11 M)	Mean age (SD) = 27.22		Images of unhealthy and personally craved snack foods	Craving regulation task (conditions: Imagine taste, Self-control)	(a) Self-control > Imagine taste	2	MNI
							(b) Imagine taste > Self-control	2
20	Yokum and Stice (2013)	21 (13 F/8 M)	Mean age (SD) = 15.2 ± 1.18	Mean BMI (SD) = 27.9 ± 5.16 kg/m2	Images of food items	Cognitive reappraisal task (conditions: Imagine Eating, Costs of eating, Benefits of not eating, Suppress craving)	(a) (Costs of eating + Benefits of not eating + Suppress craving) > Imagine eating	9	MNI
							(b) Imagine eating > (Costs of eating + Benefits of not eating + Suppress craving)	5

From left to right, the table reports the progressive study number (N), the authors and publication year of each study, the number of subjects (with female/male ratio) included in each study, alongside their age (in years) and weight status (i.e. BMI), as well as the stimulus material and type of task used (with conditions). Finally, the original contrasts included in each ALE meta-analysis are reported, together with their respective number of foci and standard anatomical space. The letters before each contrast refer to the ALE meta-analysis of reference: (a) refers to contrasts included in the ‘food craving down-regulation’ dataset; (b) refers to contrasts included in the ‘food viewing compared to active regulation’ dataset; (c) refers to contrasts included in the ‘food craving up-regulation’ dataset. The asterisk (*) indicates when data were not available in the original publication and were directly retrieved by contacting the corresponding author of the study.

Concerning exploratory analyses, the screening of the included articles for the inverse contrast Look > Decrease and the Increase > Look contrast led to the inclusion of 13 and 7 experiments (i.e. individual contrasts included in the analysis) in the ‘food viewing compared to active regulation’ and ‘food craving up-regulation’, datasets respectively. The former comprised 573 subjects and 139 activation foci (6 out of mask), whereas the latter corresponded to 164 subjects and 50 activation foci.

Emotion down-regulation

The ‘emotion down-regulation’ dataset was provided by C.M. from a recently published study (Morawetz et al., 2022), corresponding to the subset of emotion reappraisal studies included in the Decrease > Look individual meta-analysis. While the methodological selection criteria were the same as stated before (points 1–5, section 2.2.1), the ‘emotion down-regulation’ dataset included only studies employing reappraisal with the goal of down-regulating emotions in response to static emotion-eliciting visual stimuli. The ‘Decrease’ condition comprises all trials in which participants are asked to apply reappraisal tactics, such as reinterpretation or distancing, to down-regulate their current emotional response. Instead, the ‘Look’ condition includes trials in which participants are asked to pay attention to the pictures and respond naturally, without actively regulating their emotional response. Overall, this procedure resulted in the inclusion of 90 previously published studies in the ‘emotion down-regulation’ ALE meta-analysis, corresponding to a total of 128 experiments (i.e. individual contrasts included in the analysis) with 4022 subjects and 1262 activation foci (102 out of mask). For a detailed overview of the literature search and selection, together with the PRISMA flowchart and the list of included studies, please refer to Morawetz et al. (2022).

Activation likelihood estimation (ALE)

Four distinct ALE analyses, using the GingerALE 3.0.2 software (Eickhoff et al., 2009, 2012), were carried out to identify consistently activated brain regions for (i) food craving down-regulation, (ii) emotion down-regulation, (iii) food viewing compared to active regulation and (iv) food craving up-regulation. We followed the procedure based on the revised ALE algorithm by Eickhoff et al. (2012), as previously adopted by our group (Arioli and Canessa, 2019; Arioli et al., 2021a,b). Coordinates from all relevant contrasts of a study were pooled into one experiment to adjust for within-group effects (Turkeltaub et al., 2012; Müller et al., 2018). This procedure increases the validity of meta-analytic results, since including multiple contrasts from the same group of subjects—either within or between articles—can generate dependence across experiment maps. In line with current recommendations (Eickhoff et al., 2016; Müller et al., 2018), both our main meta-analyses included at least 17 experiments (20 and 128 for food craving and emotion down-regulation, respectively). Instead, both the additional meta-analyses on food viewing compared to active regulation and food craving up-regulation included <17 experiments (13 and 7, respectively). Being underpowered, these two latter meta-analyses should be considered exploratory.

In each ALE analysis, to determine whether the spatial convergence of foci across studies is larger than what expected by chance, all the activation foci from each study were treated as three-dimensional Gaussian probability distributions centred at the given coordinates (Eickhoff et al., 2009). Coordinates were reported in the MNI space or MNI converted using routines implemented in GingerALE. The ALE algorithm compensates for the spatial uncertainty associated with each individual coordinate by weighting each study according to sample size, to adjust the width of the probability distribution. The three-dimensional probability distributions of all reported activation foci in single experiments were merged for each voxel to obtain ‘modelled activation’ (MA) maps (Turkeltaub et al., 2012). Next, the union of probabilities of all individual MA maps was calculated to extract voxel-wise ALE scores representing the convergence of results across experiments at each voxel location in the brain (Turkeltaub et al., 2002). Finally, to distinguish locations of ‘true’ convergence across experiments from random convergence (i.e. noise), ALE scores were compared with an empirical null distribution representing a random spatial association between experiments (Eickhoff et al., 2012). On a computational level, this null distribution was derived by sampling a voxel at random from each MA map, recording the ALE scores derived under this assumption of spatial independence and finally iterating the permutation process several times until a sufficient sample of the ALE null distribution is reached. In line with recent guidelines (Eickhoff et al., 2016), the ‘true’ ALE scores were then tested against the ALE scores of the null distribution by adopting a cluster-level family wise error (cFWE)–corrected threshold of p < 0.05, while the cluster-forming voxel-wise threshold was set at P < 0.001 uncorrected. All voxels surviving this threshold were interpreted as revealing above-chance convergence between experiments in each analysis (Eickhoff et al., 2012), i.e. ‘true’ activation associated with a given process. Afterwards, to uncover common and specific activations across ‘food craving down-regulation’ and ‘emotion down-regulation’, the resulting maps were loaded for direct comparisons and conjunction analyses in GingerALE. The conjunction image, created with the voxel‐wise minimum value of the input ALE images, was computed to reveal the similarity between the two datasets. Conversely, two ALE subtraction images (i.e. direct contrast analysis between the maps generated by the ‘food craving down-regulation’ and ‘emotion down-regulation’ meta-analyses, meant as a ‘meta-contrast’ between the two resulting meta-analytic maps) were obtained by directly subtracting one input image from the other, with significant ALE subtraction scores converted into Z scores to aid their interpretation (Eickhoff et al., 2011). For the conjunction and subtraction analyses, we adopted a false discovery rate assuming independence or positive dependence (FDRpID) of 0.05, with a minimum volume size of 50 mm³.

Publication bias

Even though different forms of publication bias may influence meta-analytic findings, coordinate-based meta-analytical results are mostly hampered by the potential presence of unpublished studies with null results, which in turn increase the likelihood of positive, rather than negative, findings to be reported in the literature. This peculiar form of publication bias is referred to as ‘file drawer problem’, whereby studies that fail to obtain anticipated results (e.g. due to a lack of statistical significance) may remain unpublished. To further ensure the robustness of our ALE meta-analytic results against this form of publication bias, we carried out a fail-safe N (FSN) analysis for meta-analyses, following the procedure described by Acar et al. (2018). Compared to other methods for assessing the presence of publication bias, such as funnel plots and Egger tests in the context of seed-based d mapping, this procedure has been specifically designed for the ALE algorithm, providing insight not only into the stability of meta-analytic results and the robustness against publication bias but also into the contribution by a minimum number of desired studies on final meta-analytic results.

The FSN can be defined as the number of ‘noise studies’ (i.e. null-result experiments) that can be added to an original ALE meta-analysis before the spatial convergence of foci in a given cluster is no longer statistically significant (Acar et al., 2018). Using Acar et al.’s (2018) original code and guidelines, null-result experiments were created in R 3.6.1 (https://www.r-project.org), by matching the real experiments in terms of sample size and number of foci reported but with random distribution across the brain. These null-result experiments were then used to perform new ALE meta-analyses, addressing the FSN. This procedure requires assessing whether the FSN for each cluster lies between two pre-specified lower and upper boundaries, thus revealing that meta-analytic results are sufficiently robust against the file-drawer publication bias, as well as driven by at least the desired minimum of contributing studies (Acar et al., 2018). Conversely, an FSN below the lower boundary suggests non-robustness against publication bias, since the addition of a handful of null-result experiments renders a cluster statistically non-significant. An FSN above the upper boundary shows that few hyper-influential studies drive the results, which may therefore not be sufficiently representative of the available literature. Concerning the lower boundary, it has been estimated that the rate of publication bias could be up to 30% (i.e. up to 30 unpublished neuroimaging null studies per 100 published ones; Samartsidis et al., 2020). Using this estimate, the lower boundary was pre-specified at 30% of the real data. Regarding the upper boundary, this refers to the minimum percentage of studies that contribute to a given effect. Since we expected each cluster to be driven by at least 10% of the included studies, we calculate the upper boundary of the FSN for each cluster with the following formula: [(N studies contributing to a cluster/0.1) − (total N of studies included in ALE meta-analysis)]. Therefore, a cluster can be considered sufficiently robust against the potential presence of unpublished null results, as well as against the effects of few hyper-influential experiments, only if its actual FSN lies between these two pre-specified boundaries.

Results

Food craving down-regulation

Down-regulating food craving via cognitive reappraisal was consistently associated with activations in left-hemispheric regions of the prefrontal control system, i.e. the left vlPFC/inferior frontal gyrus (IFG), extending into the left anterior insula and adjacent frontal operculum, and the left dlPFC within the middle frontal gyrus (MFG). Moreover, food craving down-regulation involved large clusters of activation in the pre-SMA, as well as in the IPL (supramarginal and angular gyri) bilaterally. Further significant clusters involved a left temporal region located in the posterior middle temporal gyrus (MTG), alongside two separate clusters in the right-hemispheric anterior and posterior insula, with the latter extending into the right posterior superior temporal gyrus (STG) (Figure 2 and Table 2).

Fig. 2.

Brain activations consistently associated with food craving down-regulation via cognitive reappraisal (i.e. Decrease > Look contrast with food stimuli; 20 individual contrasts and 856 subjects included). All the reported clusters survived a cFWE-corrected statistical threshold of P < 0.05 (forming threshold: P < 0.001 uncorrected). The colour bar represents the voxel-wise ALE score, i.e. degree of non-random convergence of activation between experiments at any given voxel.

Table 2.

Summary of ALE results for food craving down-regulation via cognitive reappraisal (i.e. Decrease > Look contrast)

Cluster	Volume (mm3)	Brain region	x	y	z	ALE score
1	4520	Left IFG (BA 45)	−56	22	8	0.035
		Left IFG (BA 47)	−50	22	−6	0.031
		Left IFG (BA 45)	−48	32	−10	0.024
		Left insula (BA 13)	−44	12	−4	0.018
2	1976	Left MFG (BA 6)	−40	6	46	0.027
		Left MFG (BA 6)	−36	6	58	0.019
3	1840	Left IPL (BA 40)	−56	−52	42	0.023
		Left supramarginal gyrus (BA 40)	−58	−54	36	0.022
4	1720	Left superior frontal gyrus (BA 6)	−8	14	64	0.028
		Right superior frontal gyrus (BA 6)	6	12	68	0.020
5	1408	Right supramarginal gyrus (BA 40)	56	−50	30	0.025
		Right supramarginal gyrus (BA 40)	62	−46	32	0.025
		Right supramarginal gyrus (BA 40)	62	−46	36	0.025
6	1224	Left MTG (BA 21)	−56	−34	−6	0.028
		Left MTG (BA 21)	−66	−38	−2	0.017
		Left MTG (BA 21)	−64	−44	2	0.015
7	1016	Right insula/STG (BA 22)	48	−30	−6	0.034
8	1008	Right insula (BA 13)	44	12	−4	0.029

From left to right, the table reports the cluster size (in mm³), anatomical labelling (with corresponding Brodmann area), stereotaxic MNI coordinates of local maxima and ALE scores of the clusters that were consistently associated with down-regulating food craving via reappraisal. All the reported clusters survived a cFWE correction of P < 0.05 (forming threshold: P < 0.001 uncorrected), with 1000 permutations performed. IFG = inferior frontal gyrus; IPL = inferior parietal lobule; MFG = middle frontal gyrus; MTG = middle temporal gyrus; STG = superior temporal gyrus.

Regarding the publication bias analysis for the ALE results of food craving down-regulation via reappraisal, the FSN lied between the lower and upper boundary for five out of eight clusters (clusters 1, 3, 4, 6 and 8). By showing that the involvement of these clusters survived the inclusion of 30% of null studies and was supported by the required minimum number of studies, this evidence confirms their robustness against publication bias. Conversely, the FSN exceeded the upper boundary for the remaining three clusters (clusters 2, 5 and 7), revealing that only a subgroup of hyper-influential studies might have potentially driven the results, in contrast to desired minimum percentage of studies that should contribute to a given effect. In particular, Giuliani and Pfeifer (2015), Dietrich et al. (2016) and Cosme et al. (2020) might have exerted an extensive influence on the meta-analytic results for these three clusters, likely due to proportionally larger sample sizes compared with the other included studies. Detailed information about the FSN for every cluster is displayed in Supplementary Table S2.

Emotion down-regulation

In line with prior meta-analyses (e.g. Buhle et al., 2014; Morawetz et al., 2017, 2022), down-regulating emotions by means of cognitive reappraisal was consistently associated with widespread bilateral activations in the frontoparietal control system, including the vlPFC/IFG, dlPFC, SMA/pre-SMA, dorsal anterior cingulate cortex and IPL, alongside the left STG/MTG. For further details, see Supplementary Figure 1 and Supplementary Table S3.

Down-regulation of food craving vs emotion

To unveil the common and distinct brain structures across food craving and emotion reappraisal, we jointly assessed the ALE results of the corresponding individual meta-analyses. The conjunction analysis between food craving and emotion down-regulation via reappraisal revealed widespread common activations in the left vlPFC/IFG, left dlPFC and SMA/pre-SMA. Further overlapping activations were identified in the PPC bilaterally (i.e. IPL), as well as in the left MTG (BA 21) and right anterior insula (BA 13). Further details on the shared neural substrates for cognitive reappraisal across food craving and emotion are reported in Figure 3 (light blue) and Table 3A.

Fig. 3.

Common and distinct neural substrates across food craving down-regulation and emotion down-regulation via reappraisal in healthy individuals, resulting from the ALE analyses, are shown in different colours (see legend below: Fc = food craving down-regulation; Em = emotion down-regulation). All the reported clusters survived an FDRpID-corrected statistical threshold of P < 0.05 and minimum volume size of 50 mm³.

Table 3.

Common and specific brain regions across food craving down-regulation and emotion down-regulation via cognitive reappraisal

(A) Food craving reappraisal and emotion reappraisal
Cluster	Volume (mm3)	Brain region	x	y	z	ALE score
1	3680	Left IFG (BA 45)	−56	22	8	0.0347
		Left IFG (BA 47)	−50	22	−6
		Left IFG (BA 45)	−48	32	−10
2	1760	Left MFG (BA 6)	−40	6	46	0.0278
		Left MFG (BA 6)	−36	6	58
3	1488	Left IPL (BA 40)	−56	−52	42	0.0228
		Left supramarginal gyrus (BA 40)	−58	−54	36
4	1440	Left superior frontal gyrus (BA 6)	−8	14	64	0.028
		Right superior frontal gyrus (BA 6)	4	12	66
5	1192	Left MTG (BA 21)	−56	−34	−6	0.0276
		Left MTG (BA 21)	−66	−38	−2
		Left MTG (BA 21)	−64	−44	2
6	960	Right supramarginal gyrus (BA 40)	56	−50	30	0.0253
		Right supramarginal gyrus (BA 40)	62	−46	32
		Right supramarginal gyrus (BA 40)	60	−48	38
7	368	Right insula (BA 13)	46	16	−4	0.0238
8	8	Right superior frontal gyrus (BA 6)	6	12	64	0.0144
(B) Food craving reappraisal > emotion reappraisal
Cluster	Volume (mm3)	Brain region	x	y	z	ALE score
1	469	Right insula (BA 13)	45.8	11	−2.6
		Right insula (BA 13)	42	10	−8
2	224	Right insula/STG (BA 22)	45.7	−30	−6.7
3	216	Left IFG (BA 45)	−56	25	12
4	64	Left superior frontal gyrus (BA 6)	−5	36	54
(C) Emotion reappraisal > food craving reappraisal
Cluster	Volume (mm3)	Brain region	x	y	z	ALE score
No clusters found.

From left to right, the table reports the cluster size (in mm³), anatomical labelling (with Brodmann area), stereotaxic MNI coordinates of local maxima and ALE scores of the clusters that were commonly (top, A) and specifically (bottom, B or C) associated with the use of cognitive reappraisal strategies to down-regulate food craving and emotion. All clusters survived an FDRpID-corrected statistical threshold of P < 0.05 and minimum volume size of 50 mm³. IFG = inferior frontal gyrus; IPL = inferior parietal lobule; MFG = middle frontal gyrus; MTG = middle temporal gyrus; STG = superior temporal gyrus.

Direct comparisons, i.e. subtraction analysis, highlighted activations specific to food craving down-regulation, compared with emotion down-regulation, in the right anterior insula (BA 13) and in the right posterior insula/STG (BA 22), plus a cluster in the pars triangularis of the left IFG (BA 45) [see Figure 3 (red) and Table 3B]. Instead, we found no significant evidence of consistent specific activity for emotion down-regulation compared with food craving down-regulation via reappraisal (see Table 3C).

Exploratory meta-analyses

Food viewing compared to active regulation

In the first exploratory meta-analysis, naturally viewing food, compared to volitional regulation (i.e. Look > Decrease contrast), was associated with consistent activations in the left mid-posterior insula (i.e. primary gustatory cortex, BA 43) and in the left postcentral gyrus (BA 3) extending into the IPL. Additional clusters were observed in the left fusiform gyrus, right precentral gyrus/IFG (pars opercularis) and right dlPFC (see Figure 4 and Table 4).

Fig. 4.

Brain activations consistently associated with food viewing compared to active regulation (i.e. Look > Decrease contrast; 13 individual contrasts and 573 subjects included), resulting from the ALE meta-analysis. All the reported clusters survived a cFWE-corrected threshold of P < 0.05 (forming threshold: P < 0.001 uncorrected). The colour bar represents the voxel-wise ALE score, i.e. the degree of non-random convergence of activation between experiments at any given voxel.

Table 4.

Summary of ALE results for food viewing compared to active regulation (i.e. Look > Decrease contrast)

Cluster	Volume (mm3)	Brain region	x	y	z
1	2072	Left postcentral gyrus (BA 43)	−52	−18	18
		Left insula (BA 13)	−38	−18	16
		Left postcentral gyrus (BA 43)	−60	−16	20
2	1712	Left postcentral gyrus (BA 3)	−36	−24	52
		Left postcentral gyrus (BA 3)	−42	−22	64
		Left IPL (BA 40)	−46	−32	42
3	1560	Left fusiform gyrus (BA 37)	−50	−58	−8
4	1296	Right precentral gyrus (BA 6)	44	6	26
5	1136	Left insula (BA 13)	−42	−2	14
6	992	Right MFG (BA 46)	48	36	16

From left to right, the table reports the cluster size (in mm³), anatomical labelling (with corresponding Brodmann area), stereotaxic MNI coordinates of local maxima and ALE scores of the clusters that were consistently associated with spontaneous food viewing compared to volitional regulation. All clusters survived a cFWE correction of P < 0.05 (forming threshold: P < 0.001 uncorrected), with 1000 permutations performed. IPL = inferior parietal lobule; MFG = middle frontal gyrus.

Concerning the analyses on publication bias, the FSN always lied between the predefined lower and upper boundaries for all six clusters (see Supplementary Table S4). Therefore, these meta-analytic results are sufficiently robust against the publication bias and also supported by the desired minimum number of individual studies.

Food craving up-regulation

Since no cluster associated with food craving up-regulation (i.e. Increase > Look contrast) was found at the cFWE-corrected threshold of P < 0.05, in line with the exploratory nature of this meta-analysis, we adopted a more liberal threshold of P < 0.005 uncorrected. At this threshold, increasing food craving via reappraisal elicited activations in the left fronto-insular cortex, encompassing the left anterior insula/frontal operculum and IFG (Supplementary Figure S2 and Supplementary Table S5). However, the FSN for this cluster fell below the predefined lower boundary, as detailed in Supplementary Table S6. This indicates non-robustness of results against the file-drawer bias, which could be expected given the limited number of experiments included in this exploratory analysis. Therefore, results from this meta-analytic contrast should be interpreted with caution.

Discussion

We performed a coordinate-based meta-analysis of previous fMRI results to identify the brain regions consistently associated with food craving down-regulation via cognitive reappraisal. Given the conceptual similarity between food craving and other affective states, we also assessed the degree of overlap vs specificity across the neural bases of domain-specific food craving down-regulation and domain-general emotion down-regulation via reappraisal. Two additional exploratory meta-analyses on food-related stimuli were aimed to unveil the neural bases of (i) food viewing compared to active regulation and (ii) food craving up-regulation via reappraisal.

Common neural bases for food craving vs emotion down-regulation

In line with our initial hypotheses, a conjunction analysis confirmed that cognitive reappraisal for both food craving and emotions recruited consistent common activations in the dorsomedial prefrontal (SMA/pre-SMA), left lateral prefrontal (left dlPFC and vlPFC/IFG) and bilateral parietal (IPL) cortex, alongside additional nodes in the left posterior temporal region and right insula/STG (Figure 3, light blue). In keeping with prior related observations (Giuliani and Berkman, 2015; Giuliani and Pfeifer, 2015; Han et al., 2018), the present results provide novel quantitative evidence for the widespread overlapping neural bases of emotion and food craving down-regulation via reappraisal. The existence of a shared neural network for the flexible deployment of reappraisal across various types of stimuli and contexts supports Giuliani and Berkman’s (2015) definition of food craving as an affective state, which can be regulated using the framework of the extended process model of emotion regulation (Gross, 1998). These findings therefore highlight a possible neural basis of the similarity between domain-specific food craving reappraisal and domain-general emotion reappraisal, which has been so far outlined only in theoretical and conceptual terms.

Applying reappraisal strategies to down-regulate either food craving or affective states is indeed expected to engage common cognitive processes. In particular, the left dlPFC might sustain higher-order functions of selective attention, goal maintenance and working memory (Curtis and D’Esposito, 2003; Wager and Smith, 2003; Wager et al., 2004), required for flexibly manipulating the contents and goals of reappraisals regardless of the target being an emotional event or a craved food. This hypothesis fits with the previously reported involvement of the left dlPFC in the context of food craving, in which this region is considered to underpin the implementation of effective changes in real-life food choices, by favouring long-term health considerations over short-term temptations (McClure et al., 2004; Hutcherson et al., 2012; Wilson et al., 2021). For instance, habitual self-controllers have been shown to recruit the left dlPFC to a greater extent, compared with non-self-controllers, during successful self-control trials (Hare et al., 2009). Moreover, increased functional connectivity between the left dlPFC and vmPFC after neurofeedback training has been shown to predict the improvement in food intake control in individuals with overweight and obesity (Kohl et al., 2019). However, there is mixed evidence concerning the left-sided predominance of control-related regions, such as the dlPFC and vlPFC, in the down-regulation of food craving. For example, a dysregulation of right prefrontal regions—rather than left ones—for direct brain lesions or reduced cortical thickness has been previously associated with higher BMI and reduced cognitive control on food intake in obesity (Alonso-Alonso and Pascual-Leone, 2007; Vainik et al., 2018).

The loading on attentional processes for both food craving and emotion reappraisal also explains the common engagement of the IPL, previously associated with attentional reorienting and self-reflection processes in response to both aversive and appetitive stimuli (Igelström and Graziano, 2017; Langner et al., 2018). Specifically, the recruitment of the IPL during food craving down-regulation might reflect the reorienting of attention away from the tempting sensory attributes of food, e.g. towards more abstract long-term goals. In turn, the latter process likely represents a crucial prerequisite for shared processes of inhibitory control over initial goal-inappropriate responses, favouring the selection of goal-appropriate ones, that have been previously associated with other commonly activated regions such as the left vlPFC/IFG and SMA/pre-SMA (Nachev et al., 2008; Swick et al., 2008; Cohen et al., 2013; Limongi and Pérez, 2017). Moreover, the deployment of reappraisal to down-regulate both food craving and emotions was revealed to be associated with overlapping neural nodes in left posterior temporal and right anterior insular regions. While the former represents an intermediary station for semantic and perceptual re-evaluation of either food or emotion-eliciting stimuli (Ochsner et al., 2012), the latter might constitute a key node between lower-level bodily perceptions and higher-order cognitive functions such as attention switching and saliency processing (Menon and Uddin, 2010; Wager and Barrett, 2017).

Overall, this pattern highlights commonly activated posterior temporal, insular and frontoparietal regions, underlying the multifaceted contribution of semantic, attentional and executive neuro-cognitive processes supporting the deliberate down-regulation of either food craving or emotions via cognitive reappraisal.

Distinctive neural bases for food craving down-regulation

In addition, the present results revealed distinctive activations for domain-specific food craving down-regulation compared to emotion down-regulation via reappraisal but no significant clusters for the reverse comparison. Down-regulating food craving vs emotions specifically involved the right anterior insula (in a slightly more caudal sector than that highlighted by conjunction analysis), the left vlPFC/IFG (pars triangularis) and the temporo-opercular cortex encompassing the right posterior insula and STG (Figure 3, red). Although food craving is admittedly defined as an affective state (Giuliani and Berkman, 2015), down-regulating the desire for food via cognitive reappraisal presents some distinctive features that might help explaining the selective recruitment of these regions. In line with our initial predictions, the involvement of the right anterior insula might be better understood in the light of domain-specific strategies specifically applied to reappraise food stimuli. Firstly, the effective down-regulation of food craving entails an extensive integration of interoceptive information about current bodily states, such as signals of hunger or satiety, with salience information about food cues. Secondly, a common reappraisal tactic such as ‘costs of eating’ requires anticipating the negative consequences of consumption by mentalizing about future bodily states (e.g. stomachaches and weight gain) to ultimately promote goal-directed behavioural changes. Given its role in anticipating interoceptive states related to positive hedonic experiences (Naqvi et al., 2014), the right anterior insula is well suited as a hub for such food-specific reappraisal functions. However, the dorsal anterior insula is also involved in attentional and cognitive control (Menon and Uddin, 2010; Chang et al., 2013; Wager and Barrett, 2017), and its recruitment during food craving down-regulation may also reflect the stronger loading on working memory and attentional processes compared with domain-general emotion down-regulation.

Interpreting the specific role of the left vlPFC/IFG (pars triangularis) and right posterior insula/STG in food craving down-regulation via reappraisal is less straightforward. The cluster in left vlPFC/IFG might underlie the higher loading of food-related, compared with emotional-related, drives on top-down inhibitory control over unwanted urges and sensations (Swick et al., 2008). This interpretation is consistent with Giuliani et al.’s (2014) evidence that reappraising the value of personally craved vs not-craved foods relies on the selective activation of the left, rather than the right, IFG. This portion of the left vlPFC/IFG might thus support processes selectively involved in the domain-specific reappraisal of food craving, whereby individuals first need to override the strong motivational urge to consume the desired food, before being able to reinterpret its affective meaning and activate reappraisal strategies. Instead, the available literature does not provide cues regarding the putative specific role of the right posterior insula/STG in food craving reappraisal. This region might underpin a similar function in reappraisal as its left counterpart, i.e. perceptual re-evaluations of food-related stimuli, but through a stronger loading on right-lateralized attentional and spatial processes than left-lateralized verbal and semantic processes (Ochsner et al., 2012). This interpretation suggests a prominent engagement of attentional and spatial processes, underpinned by the right STG, when reappraising the value of tempting food stimuli (Shapiro et al., 2002; Ellison et al., 2004; Shah-Basak et al., 2018). Further empirical work is needed, however, before drawing strong conclusions about the distinctive contribution of this right temporo-opercular region to food craving vs emotion reappraisal.

Exploratory meta-analyses

Food viewing compared to active regulation

Our first exploratory meta-analysis showed that naturally viewing food cues compared to the deployment of deliberate regulatory effort (i.e. Look > Decrease) was associated with consistent activations in the primary gustatory cortex, primary somatosensory cortex, left fusiform gyrus, right dlPFC and right precentral gyrus (Figure 4). Supporting the robustness of these findings, all these clusters survived the publication bias analysis.

This reverse contrast was aimed at investigating the target regions of (down)regulation by reappraisal. Indeed, we found that deploying cognitive reappraisal to down-regulate food craving effectively modulates the activity in the same brain regions that are responsive to palatable foods, in analogy with the well-known modulation of amygdala reactivity during reappraisal in the emotion regulation literature (e.g. Buhle et al., 2014; Morawetz et al., 2022). In particular, this widespread set of regions is reminiscent of meta-analytic findings from previous food-cue reactivity paradigms, which contrasted the unregulated passive viewing of food with non-food pictures, or mere fixation (e.g. van der Laan et al., 2011; Brooks et al., 2013; van Meer et al., 2015; Devoto et al., 2018). Here, the perception of food cues consistently recruits regions associated with both exteroceptive (i.e. visual) and interoceptive (i.e. somatosensory and gustatory) processing of food-related information. In particular, while the fusiform gyrus is often associated with the heightened visual processing of emotionally salient food cues (van der Laan et al., 2011), neural activity in the gustatory and somatosensory cortices is thought to underlie the concurrent integration of other food-related sensory representations, such as taste, texture and smell. The present exploratory meta-analysis thus complements our main meta-analysis, by providing preliminary hints on the brain target of modulation by cognitive reappraisal in the context of food craving down-regulation. Accordingly, the above-mentioned involvement of frontoparietal regions in food craving down-regulation might modulate the neural processing of palatable food-related sensory attributes, in a similar way as the recruitment of cognitive control regions during emotion reappraisal is known to modulate neural circuits that underpin emotional reactivity and salience (Ochsner et al., 2012). Besides, our exploratory meta-analysis revealed a stronger involvement of the IPL and precentral gyrus during passive food viewing, relative to deliberate regulation. Given their respective role in bottom-up orienting of attention (Igelström and Graziano, 2017) and in coordinating approach tendencies towards craved food (Brooks et al., 2013; van Meer et al., 2016), these findings further suggest the effectiveness of reappraisal strategies in modulating attention- and motivation-related activation in response to food images (Yokum and Stice, 2013).

Altogether, on both theoretical and application levels, these results might also inform neurocognitive theories of obesity or eating disorders. The same food-cue reactivity regions that are effectively down-modulated by reappraisal in healthy weight individuals have been reported as hyper-active in individuals with obesity and in patients with bulimia nervosa, which might be related to individual differences in inhibitory control (Brooks et al., 2011, 2013; Kennedy and Dimitropoulos, 2014; Pursey et al., 2014; Devoto et al., 2018). Therefore, future empirical work is needed to unveil the precise neural mechanisms underlying the interaction between food-cue reactivity and food craving reappraisal substrates, in both healthy and clinical populations.

Food craving up-regulation

In our second exploratory meta-analysis, the use of reappraisal to ‘increase’, rather than decrease, food craving was associated with the recruitment of the left fronto-insular cortex (Supplementary Figure S2). However, this is a preliminary finding, resulting from a small subset of studies and with an uncorrected threshold, and should be therefore interpreted cautiously. Notwithstanding this limitation, such evidence provides novel insights into the possible role of the anterior insula and frontal operculum as an integrative hub for multiple processes contributing to enhance food craving (Frank et al., 2013), ranging from the lower-level perception of food cues across multiple modalities (van der Laan et al., 2011; Huerta et al., 2014) to food-specific hedonic and motivational processes. This hypothesis fits with the engagement of the left insular cortex that has been suggested to play a prominent role, compared with its right-hemispheric counterpart, in parasympathetic functions and approach behaviours (Craig, 2005; Frank et al., 2013). However, this cluster did not survive the publication bias analysis. While limiting any interpretation of the possible functional role of the observed region, this evidence shows that further research is required to draw robust conclusions about the neural bases of food craving up-regulation and inform translational efforts.

Limitations

While this is, to the best of our knowledge, the first meta-analysis of fMRI studies directly comparing food craving vs emotion reappraisal, some limitations must be acknowledged. First, the results of the exploratory analyses on ‘food viewing compared to active regulation’ and ‘food craving up-regulation’ should be interpreted with caution because of their limited sample size and underpowered nature. A similar concern might hold for the ‘food craving down-regulation’ dataset (20 experiments included) that, despite fulfilling current recommendations to include at least 17 studies (Eickhoff et al., 2016; Müller et al., 2018), is comparatively smaller than the ‘emotion down-regulation’ one (128 experiments included). Since this might have prevented the detection of smaller effects, further empirical work is required in this field to strengthen the conclusions of the present meta-analysis. Moreover, the limited number of studies in the ‘food craving down-regulation’ dataset might have also made the results more prone to the effects of publication bias, especially concerning over-influence of single experiments. An FSN analysis (Acar et al., 2018) indeed showed that, while our results were sufficiently robust against the file-drawer problem, three out of eight clusters from the ‘food craving down-regulation’ meta-analysis were driven by few over-influential studies. Moreover, more lenient inclusion criteria concerning participants’ characteristics were used for ‘food craving’, compared with ‘emotion’, reappraisal. Namely, when information about individual subgroups could not be retrieved from the original article or the corresponding authors, we included studies with mixed samples of participants, either in terms of age (Yokum and Stice, 2013; Silvers et al., 2014; Giuliani and Pfeifer, 2015) or BMI (Tuulari et al., 2015; Dietrich et al., 2016; Demos Mcdermott et al., 2019; Kohl et al., 2019; Giuliani et al., 2020). Since both age and BMI are known to modulate the neural reactivity to food cues, however, we cannot exclude the possible confounding effect of such mixed samples. Another limitation is inherent in the literature on cognitive reappraisal of emotion and food craving, as it encompasses an extremely heterogeneous category of tactics (e.g. reinterpretation vs detachment) that vary not only across separate studies but even within the same study (e.g. Hutcherson et al., 2012; Dietrich et al., 2016). Thus, even if the overall instructions were almost identical across studies, the variety of reappraisal tactics might have influenced our results, thus highlighting the need of further investigation.

Conclusions and future directions

We reported novel meta-analytic evidence of a widespread network of areas consistently associated with the down-regulation of food craving by means of cognitive reappraisal, showing both overlapping and distinctive components compared with domain-general emotion down-regulation. The present results indeed highlighted a considerable overlap across food craving and emotion down-regulation, suggesting the engagement of a shared, mostly left-lateralized, network for the flexible deployment of reappraisal across various types of stimuli and contexts. Direct comparisons highlighted no evidence for regions specifically underlying the down-regulation of emotion, compared with food craving, via reappraisal. Instead, a domain-specific network for reinterpreting the hedonic and emotional value of food cues was found to involve left-hemispheric frontoparietal regions (e.g. dlPFC, vlPFC and IPL) alongside distinct clusters in the right anterior and posterior insula. Overall, this pattern is suggestive of multiple processes specifically engaged by food craving reappraisal. These include interoceptive awareness towards current states of satiety or hunger, as well as the anticipation of future interoceptive states, as required by food-specific reappraisal strategies (right anterior insula), inhibitory control over the immediate motivational drive towards craved foods (left vlPFC/IFG) and right-lateralized attentional and spatial processes during the active reinterpretation of food stimuli (right posterior insula/STG).

This evidence for specific neural substrates associated with the cognitive reappraisal of food craving, but not emotions in general, might inform current definitions and models of food craving regulation. While it is largely recognized that food craving is an affective state comparable to other emotions, as it can be regulated using similar reappraisal techniques (Giuliani and Berkman, 2015), by definition it entails a unique motivational component to approach the desired food, which is not necessarily shared with other affective states. This feature should be emphasized in the theoretical frameworks for food craving down-regulation, because generating effective reappraisal strategies first requires counteracting this intense appetitive urge to consume the craved food.

The present findings might thus lay the foundations for improving therapeutic interventions for obesity and eating disorders, whereby altered interoceptive sensitivity and inadequate inhibitory control may predispose individuals to develop unhealthy eating habits (Simmons and DeVille, 2017). For instance, the distinctive contribution of the anterior insula in food craving down-regulation suggests that current treatments inspired by cognitive reappraisal, such as the RoC Training (Boswell et al., 2018) or the Minding Health training (Stice et al., 2015), might benefit from complementary mindfulness-based interventions targeted at enhancing interoceptive awareness of hunger and satiety (Simmons and DeVille, 2017). By leveraging the contributions of insular and frontoparietal regions to interoceptive awareness and cognitive control, respectively, a combined approach might improve the flexible application of reappraisal strategies to distinct contexts. Likewise, our results provide novel insights into promising targets for neuromodulation and neurofeedback techniques, such as the left dlPFC (Lowe et al., 2017; Kohl et al., 2019).

Supplementary data

Supplementary data is available at SCAN online.

Data availability

All the meta-analytic maps are publicly available at the following link https://identifiers.org/neurovault.collection:16102.

Contribution statement

Marta Gerosa: Investigation, Data curation, Formal Analysis, Visualization, Writing – original draft preparation. Nicola Canessa: Methodology, Software, Writing – review & editing. Carmen Morawetz: Data curation, Writing – review & editing. Giulia Mattavelli: Conceptualization, Methodology, Software, Formal analysis, Writing – review & editing, Project administration.

Funding

This research was partially supported by the ‘Ricerca Corrente’ funding scheme of the Italian Ministry of Health. The funding source had no role in any stage of the project.

Conflict of interest

The authors declared that they had no conflict of interest with respect to their authorship or the publication of this article.