Hunger and BMI modulate neural responses to sweet stimuli: fMRI meta-analysis

Abstract

Objective

Consuming sweet foods, even when sated, can lead to unwanted weight gain. Contextual factors, such as longer time fasting, subjective hunger, and body mass index (BMI), may increase the likelihood of overeating. Nevertheless, the neural mechanisms underlying these moderating influences on energy intake are poorly understood.

Methods

We conducted both categorical meta-analysis and meta-regression of factors modulating neural responses to sweet stimuli, using data from 30 functional magnetic resonance imaging (fMRI) articles incorporating 39 experiments (N = 995) carried out between 2006 and 2019.

Results

Responses to sweet stimuli were associated with increased activity in regions associated with taste, sensory integration, and reward processing. These taste-evoked responses were modulated by context. Longer fasts were associated with higher posterior cerebellar, thalamic, and striatal activity. Greater self-reported hunger was associated with higher medial orbitofrontal cortex (OFC), dorsal striatum, and amygdala activity and lower posterior cerebellar activity. Higher BMI was associated with higher posterior cerebellar and insular activity.

Conclusions

Variations in fasting time, self-reported hunger, and BMI are contexts associated with differential sweet stimulus responses in regions associated with reward processing and homeostatic regulation. These results are broadly consistent with a hierarchical model of taste processing. Hunger, but not fasting or BMI, was associated with sweet stimulus-related OFC activity. Our findings extend existing models of taste processing to include posterior cerebellar regions that are associated with moderating effects of both state (fast length and self-reported hunger) and trait (BMI) variables.

Introduction

Subjective experiences associated with food consumption can be strongly context dependent. For example, the perceptual and rewarding aspects of sweet chocolate eaten after a long fast can be considerably different from those associated with the same chocolate eaten after a heavy meal. These moderating effects could confound attempts to use functional neuroimaging to study the neural responses to sweet tastes, as the neural activity resulting from any particular sweet stimulus may be substantially moderated by uncontrolled contextual influences. Although it is plausible that activity associated with sweet stimuli might be modulated by combinations of state (hunger, hours fasted) or trait (body mass index (BMI) variables, these effects are rarely controlled or modeled in neuroimaging experiments. As such, their aggregate varying influences could result in low reliability when quantitatively reviewing the neural systems responsible for sweet stimulus processing. These contextual influences could have regional specificity, with their modulatory effects during food consumption occurring in different spatially distributed and interacting sensory, inhibitory control, homeostatic regulatory, and reward processing systems [1].

Quantitative image meta-analysis is a useful tool to investigate consistency of findings across studies. Previous quantitative reviews of taste processing used Activation Likelihood Estimation (ALE) to evaluate neural responses to salty, acid, bitter, sweet, and umami stimuli, comparing stimulus responses to water or tasteless solutions [2–5]. While these meta-analyses provide strong evidence for bilateral involvement of insula and postcentral gyri in processing sweet stimuli, they found less consistent support for other regions associated with reward processing, including the amygdala, orbitofrontal cortex (OFC), and anterior cingulate cortex [2–5]. The observed inconsistences may have resulted from variations in participant state, as hours fasted, hunger state, and BMI all may influence responses to sweet stimuli [6, 7].

To address these questions, we conducted meta-analyses examining neural responses to sweet stimuli. Restricting the analyses solely to sweet stimuli, while modeling variations in fast duration, hunger, and BMI, allowed investigation of the neural organization of sweet stimulus processing in different contexts. To examine continuous effects of contextual variables, we used meta-regression to investigate effects of hours fasted, self-reported hunger, and BMI on responses to sweet stimuli.

Materials and methods

Study eligibility criteria and rationale

We began by searching PubMed for human studies using [(“fMRI” and “PET”)] AND [(“taste”)] and [“fMRI” and “PET”] AND [“food consumption”] terms from 3/1989 through 08/2019. Additional studies were found by examining the reference lists from those articles, along with previous functional magnetic resonance imaging (fMRI) meta-analyses and systematic reviews. Inclusion and exclusion criteria were fMRI publications: (a) in a peer reviewed journal, (b) in English, (c) reporting Montreal Neurological Institute (MNI) or Talairach anatomical coordinates, (d) using whole-brain analysis, (e) reporting t, z, r, or p values describing responses to consuming sweet solutions or foods (although not alcohol) compared with a baseline condition consisting of water or a tasteless solution mimicking saliva, (f) where time since the last meal was reported, and (g) including adults in the healthy to overweight BMI range of 18.5–29.9 and obese Class I BMI range of 30 to <35 [8] and children and adolescents in the 5th to <95th percentile BMI for age and weight categories [9]. We excluded adults in Class II and III because BMIs in these ranges are associated with more numerous and severe medical problems associated with brain pathology [10].

Study selection

The PubMed search yielded 962 records. Seventeen additional records were identified from the reference lists of those records. Of the total 979 records, 44 duplicates were discarded. After initial screening of the abstracts using the above mentioned inclusion and exclusion criteria, 770 were excluded because they: (a) included animals (21 abstracts), (b) included patient samples with medical conditions and psychological disorders, including cancer, stroke, anorexia, bulimia nervosa, epilepsy, dementia, lesions, autism, alcohol abuse, or patients who had undergone bariatric surgery or were in the obese Class II or III BMI class [8, 9] (140 abstracts), (c) were reviews, commentaries, letters or meta-analyses (98 abstracts), or (d) were non-English language articles (34 abstracts), (e) were case studies (22 abstracts), (f) covered an irrelevant topic, such as pet food or food policy (92 abstracts), (g) did not utilize an fMRI task assessing response to food (306 abstracts), instead using tasks assessing olfaction or food picture responses, (h) did not administer sweet stimuli (34 abstracts), (i) did not use whole-brain analysis (14 abstracts), (j) did not report the activity coordinates (7 abstracts), or (k) were unclear about whether participants were fasted or fed, and if fasted, for how long (2 abstracts).

The full texts of the remaining 165 articles were then screened, with 135 excluded for: (a) not reporting whether participants were fasted or fed (28 articles), (b) not reporting the number of hours fasted (4 articles), (c) not reporting standardized coordinates for contrasts of interest (23 articles), (d) using patient samples (6 articles), (e) being methodology papers not using human fMRI (14 articles), (f) not using whole-brain analysis (32 articles), (g) describing neural responses due to olfaction (7 articles), (h) utilizing a food receipt task outside rather than inside the scanner (8 articles), (i) not presenting response to sweet foods (6 articles), or (j) using duplicate samples (7 articles) (Fig. 1).

Fig. 1.

Preferred reporting items for systematic reviews and meta-analyses flow diagram, from ref. [101].

Study characteristics

The selection process identified 30 fMRI articles examining responses to sweet stimuli compared with water or a tasteless solution in healthy to overweight participants (Table 1). The total sample included 995 participants (633 females) and 773 foci, with 6 papers studying 339 children and adolescents.

Table 1.

Studies and samples included in the meta-analyses.

#	Experiment sample	Contrast	Hours fasted	N	BMI m (sd)	# F	Age m (sd)	# foci	P
Fasted
1	Burger and Stice (2011) [21]	ms > ts	7.6	39	24.50 (5.35)	39	15.5 (0.94)	45	un. p <0.001
2	Burger and Stice (2013) [15]	ms > ts	5	155	20.80 (1.90)	80	15.8 (1.0)	24	un. p <0.001
3	Burger and Stice (2014) [24]	Soda > ts	4.1	27	22.80 (4.40)	12	15.2 (0.8)	12	cor. p < 0.05
4	Chen et al. (2017) [102]	ms > ts	8	34	30.37 (4.93)	34	40.68 (11.77)	58	un. p <0.001
5	Doornweerd et al. (2018) [103]	choc m > bs	12	32	26.4 (3.3)	32	49.80 (9.80)	14	FWE cor. p < 0.05
6	Eiler et al. (2018) [11]	sucr > wt	4	37	27.28 (5.66)	20	23.35 (0.99)	23	FWE cor. p < 0.05
7	Eiler et al. (2018) [11]	sucr > wt	4	37	24.12 (3.61)	22	22.2 (0.31)	16	FWE cor. p < 0.05
8	Galvan and McGlennen (2013) [104]	sucr > wt	3	30	Healthy	26	21.9a	17	cor. p < 0.05
9	Green and Murphy (2012) [12]	sucr > wt	12	12	27.13 (6.20)	7	23.9 (3.3)	24	cor. p < 0.05
10	Green and Murphy (2012) [12]	sucr > wt	12	12	25.03 (5.60)	7	23.0 (2.3)	15	cor. p < 0.05
11	Haase et al. (2009) [35]b	sucr > wt	12	18	23.73a	9	20.9a	37	cor. p < 0.0005
12	Jacobson et al. (2010) [13] olderb	sucr > wt	12	19	27.51 (2.88)	10	72.2a	26	cor. p <0.0015
13	Jacobson et al.(2010) [13] youngb	sucr > wt	12	19	24.45 (3.63)	9	23.9a	17	cor. p <0.0015
14	Nakamura et al. (2019) [25]	diff > ts	3	35	20.8 (2.1)	19	17.2 (1.9)	2	FWE cor. p < 0.05
15	Stice et al. (2011) [14] low risk	ms > ts	5	25	20.07 (1.80)	30c	15 (2.9)c	19	un. p <0.001
16	Stice et al. (2011) [14] high risk	ms > ts	5	35	20.64 (1.67)	–c	15 (2.9)c	14	un. p < 0.001
17	Stopyra et al. (2019) [19]b	Glucose	16	26	21.71 (1.46)	26	24.84 (5.11)	7	un. p < 0.001
18	Sun et al. (2014) [105]	ms > ts	5	32	25.30 (4.40)	18	25.5 (5.7)	11	FWE cor. p < 0.05
19	Thanarajah et al. (2019) [20]	ms > ts	10.5	13	25.57 (2.41)	0	56.0 (9.5)	9	cor. p < 0.05
20	Uher et al. (2006) [106]	choc m > ts	24	18	22.50 (2.80)	10	28.4 (8.4)	3	un. p < 0.005
21	Van den Bosch (2014) [107]	sucr > wt	3	34	21.67 (1.92)	22	24.81 (7.37)	14	un. p < 0.001
22	Van Rijn et al. (2015) [18]b	sucral > wt	3	30	22.60 (1.40)	30	22 (3)	78	un. p = 0.049
Fed
23	Boutelle et al. (2015) [26]	sucr vs. wt	0	23	21.90 (0.75)	10	10.15 (0.3)	13	FDR cor. p < 0.05
24	Felsted et al. (2010) [108]	ms > ts	2	26	28.59 (1.79)	24	24.62 (1.26)	13	cor. p < 0.01
25	Frank et al. (2012) [109]	sucr > wt	1	23	21.49 (1.42)	23	24.78 (5.64)	37	cor. p < 0.05
26	Geha et al. (2013) [110]	ms vs. ts	2	27	27.10 (1.60)	22	25.1 (1.1)	10	cor. p < 0.05
27	Haase et al. (2009) [35]b	sucr > wt	0	18	23.73a	9	20.9a	14	cor. p < 0.0005
28	Jacobson et al. (2010) [13] olderb	sucr > wt	0	19	27.51 (2.88)	10	72.2a	42	cor. p <0.0015
29	Jacobson et al. (2010) [13] youngb	sucr > wt	0	19	24.45 (3.63)	9	23.9a	53	cor. p <0.0015
30	Kishi et al. (2017) [111]	sucr > wt	2	22	Healthy	0	29.0 (6.0)	6	un. p <0.01
31	McCabe et al. (2012) [112]	choc liq > ts	1	25	22.00 (2.00)	16	19.2 (1.2)	14	FWE cor. p < 0.05
32	Murray et al. (2014) [113]	choc liq > ts	1	20	23.09 (1.80)	10	22.8 (4.6)	5	FWE cor. p < 0.05
33	Nakamura et al. (2012) [114]	sucr > ts	2	20	Healthy	10	24.2 (2.7)	1	FWE cor. p < 0.05
34	Oberndorfer et al. (2013) [115]	sucr > wt	0	14	22.60 (1.50)	14	27.4 (5.5)	4	un. p < 0.005
35	Simmons et al. (2013) [16]	ju + s vs. ts	1.75	21	22.00a	13	26, 23–29d	12	FDR cor. p < 0.05
36	Spetter et al. (2010) [116]	sucr > wt	2	15	22.00 (1.50)	0	23.3 (1.7)	3	un. p < 0.005
37	Stice and Yokum (2018) [17]	ms > bs	0	40	25.80a	29	21.6 (4.0)	30	un. p <0.001
38	Stopyra et al. (2019) [19]b	Water	0	26	21.71 (1.46)	26	24.84 (5.11)	8	un. p <0.001
39	Van Rijn et al. (2015) [18]b	sucr > wt	0	30	22.60 (1.40)	30	22.0 (3.0)	80	un. p = 0.049

# experiment number, # F number of female participants, bs baseline, choc m chocolate milk, choc liq chocolate liquid, glucose glucose infused into the stomach, ms milkshake, sucr sucrose, sucral sucralose, wt water, ts tasteless, ju + s apple juice + sucrose, diff choice of orange juice/lemonade/strawberry milk or sports drink or yogurt-flavored drink, un. uncorrected, water water infused into stomach, cor. corrected, FDR false discovery rate, FWE family-wise error.

^aNo standard deviation.

^bSame sample scanned twice.

^cOf total N = 60.

^dMedian.

The articles reported results from 39 separate experiments, including groups of diet-soda drinkers or non-diet soda drinkers, samples at high-risk or low-risk for obesity, samples selected given age, or with parents with and without a family history of alcoholism [11–14]. We included one study of adolescents “at-risk” for obesity and adolescents not “at-risk” for obesity. Being “at-risk for obesity” was defined by having biological parents who were both in the overweight or obese range. Being “not at-risk for obesity” was defined by having two lean parents [14]. For the experiments included, the average age was 25.7, sd = 12.22, with seven experiments with a mean age in middle childhood (9–11 years) or adolescence (12–18 years). Studies conducted with children or adolescents relative to those conducted with adults did not differ in the number of hours that participants were fasted (t = −0.47, df = 17.53, p = 0.64).

Eight of 39 experiments had females only, 3/39 had males only, and 29/39 were mixed. We requested and received additional unpublished results for nine articles [11, 14–21].

Eighteen of the 39 experiments presented a visual cue during presentation of the sweet stimuli. Most experiments (19/39) used a sucrose solution compared with baseline with 18/19 experiments using a water contrast, and 1/19 used a tasteless solution. Thirteen of 39 used chocolate milkshake or chocolate milk vs. tasteless solution contrasts. Less frequently used stimuli included chocolate milkshake or chocolate milk vs. crosshairs (2/39), flavored soda vs. tasteless solution (1/39), juice vs. tasteless solution (1/39), a choice of sweet drink, including orange juice, lemonade, strawberry milk, sports drink or yogurt drink (1/39), and sucralose solution vs. water (1/39). Although sucralose induces somewhat different neural responses from sucrose, we chose to include this study because sucralose is a regularly consumed sugar substitute [22, 23]. Note that 1/39 studies involved the gastric infusion of glucose relative to water. Four of 30 studies did not report the nutrient breakdown of the food administered.

In the 39 selected experiments, we used meta-regression to explore the effects of hours fasted as a continuous predictor of neural responses to sweet stimuli. Average hours fasted was 5.05 h, sd = 5.49 across all experiments.

We explored the effects of hunger on responses to sweet stimuli in the 20/39 experiments that included ratings of self-reported hunger. As pre-imaging self-reported hunger ratings used various scales, ratings were rescaled from 0 to 100 where 0 referred to “not at all hungry” and 100 to “very hungry”.

Our final meta-regression used 24/39 experiments to explore the effects of BMI as a predictor of sweet stimulus response. For this analysis, because child and adolescent BMI estimates are not equivalent to adult BMI estimates, we removed seven experiments from these six studies [14, 15, 21, 24–26]. In addition, we did not include experiments not reporting mean sample BMI (3/39) and included participants only in the fasted state if data reported the same participants in both the fasted and fed state (5/39). 10/24 experiments had a mean BMI in the overweight range with 1/24 with a mean BMI in the Obese Class I range [8]. The mean BMI was 24.6 (sd = 2.5) for these 24 experiments.

Study findings reported in Talairach space were converted to MNI coordinates and all model results are presented in MNI space [27].

Meta-analysis approaches

For our meta-analyses using categorical predictor models, we used the ALE method implemented in GingerALE 2.3.6 (http://www.brainmap.org/ale/). ALE estimates the most likely location for task-related activity to occur given the studies included and allows comparison with previous ALE fMRI meta-analyses. ALE creates a likelihood map for each peak coordinate by convolving an isotropic kernel with each peak and then modeling the likelihood of task-based activity in that area as a normally distributed Gaussian probability distribution [28–30]. Isotropic kernel values assume Euclidean distances between voxels and peaks. Comparison between the modeled activity map that represents the maximum of all the Gaussian distributions for all the experiment foci and the ALE null distribution yields a 3D volume for each probability level [30]. A random effects model was used.

In order to compare our sweet stimuli vs. baseline results with findings of previous meta-analyses [2–5], we used two different critical thresholds, both known to give good protection against false positives. The first used FDR-pID < 0.05 with a minimum cluster size of 8 mm³. The second employed a family-wise error (FWE) critical threshold with a cluster-level FWE rate of 5% was used, with a cluster-forming threshold of p = 0.001 and 1000 random permutations [31].

For continuous predictor models exploring contextual and individual difference effects, we used an effect size signed differential mapping method, implemented in Seed-Based d Mapping (AES-SDM) SDM5.141 (http://www.sdmproject.com/), yielding bidirectional meta-regression parameter estimates [32–34]. This approach estimates the effect size rather than the likelihood that a region is active during a task using anisotropic kernels. Unlike ALE meta-analyses, SDM uses anisotropic kernels to assign different values to neighboring voxels based on the spatial correlation among them [34]. Significant study coordinates and t, z, r or p values, converted to t values, were the model inputs. Effect size estimates and variance maps from each study, weighted by the average effect size, were combined and then randomly permutated 100 times. These maps were then submitted to a meta-analytic random effects general linear model, using the recommended uncorrected threshold p < 0.005, cluster extent of ten voxels, and SDM-Z ≤ ±1.0 [31]. Because meta-regression may be affected by study heterogeneity, we report the heterogeneity of each analysis [32].

Results

Brain regions responding to sweet stimuli

The ALE analysis using a FWE p < 0.05 corrected critical threshold revealed greater activity in: left inferior frontal gyrus, pars opercularis, left rolandic operculum, bilateral postcentral gyri, bilateral claustra, right insula, left thalamus, bilateral globus pallidus, and bilateral amygdalae. The ALE analysis using FDR-pID < 0.05 correction for sweet stimuli compared with baseline contrasts revealed greater activity in the left frontal pole, bilateral medial OFC, right anterior cingulate gyrus, right inferior frontal gyrus, right medial frontal gyrus, bilateral postcentral gyri, left inferior parietal lobule, right lingual gyrus, left insula, right thalamus, right putamen, bilateral caudate head, right caudate body, right globus pallidus, and bilateral cerebellar Lobule VI (Table 2, Fig. 2).

Table 2.

Activation Likelihood Estimation meta-analysis of sweet stimuli relative to a basal condition using 39 experiments (30 studies).

Label	Side	x	y	z	Volume (mm3)	Z	ALE
Frontal pole	L	−12	64	−2	8	3.449	0.020
Medial orbitofrontal cortex	L	−24	36	−16	32	3.652	0.021
Medial orbitofrontal cortex	R	22	32	−18	208	4.260	0.026
Anterior cingulate gyrus	R	8	10	46	40	3.521	0.020
Inferior frontal gyrus pars opercularis	L	−58	4	22	2896	3.244a	0.019
Inferior frontal gyrus	R	56	8	20	8	3.391	0.019
Medial frontal gyrus	R	8	4	54	32	3.459	0.020
Rolandic operculum	L	−54	−12	18	952	4.099a	0.025
Postcentral gyrus	R	64	−6	14	24	3.543	0.020
Postcentral gyrus	L	−56	−8	24	952	4.377a	0.027
Postcentral gyrus	R	58	−12	28	1552	6.575a	0.046
Postcentral gyrus	L	−50	−14	44	112	3.878	0.023
Inferior parietal lobule	L	−60	−20	28	64	3.860	0.023
Lingual gyrus	R	22	−80	6	48	3.555	0.020
Claustrum	L	−32	16	4	776	5.246a	0.034
Claustrum	R	34	12	6	840	5.522a	0.036
Claustrum	R	40	6	−10	2720	5.244a	0.034
Claustrum	L	−36	4	−12	2152	4.803a	0.030
Claustrum	L	−36	−6	8	1256	6.119a	0.042
Insula	R	38	−4	10	1976	6.444a	0.045
Insula	L	−40	−4	−4	8	3.393	0.019
Insula	L	−44	−8	16	8	3.393	0.019
Thalamus	L	−10	−16	8	768	5.219a	0.034
Thalamus	R	12	−16	4	584	5.092	0.033
Putamen	R	32	−4	−10	2720	3.936	0.023
Caudate head	R	14	22	−12	288	4.277	0.026
Caudate head	L	−10	10	−4	440	4.509	0.028
Caudate body	R	20	0	30	48	3.734	0.022
Globus pallidus	R	14	4	−6	2720	4.740	0.030
Globus pallidus	L	−10	−2	−8	32	3.508a	0.020
Globus pallidus	R	22	0	−12	2720	5.664a	0.038
Amygdala	R	32	−4	−10	3224	3.812a	0.023
Amygdala	L	−24	−4	−14	2152	5.974a	0.041
Cerebellum lobule VI	L	−20	−64	−26	48	3.695	0.021
Cerebellum lobule VI	R	20	−64	−26	336	4.666	0.029

The critical threshold to FDR-pID < 0.05 with a minimal volume of 8 mm³.

^aDenotes analyses using a critical threshold cluster-level family-wise error rate of 5%, with a cluster-forming threshold p = 0.001 with 1000 random permutations. Montreal Neurological Institute coordinates.

Fig. 2. Activation likelihood estimation meta-analysis of sweet stimuli relative to a basal condition.

Increased activity is in RED/YELLOW. The critical threshold was p = 0.05, Family-wise error cluster corrected, with a cluster-forming threshold of p = 0.001. Montreal Neurological Institute coordinates.

Five experiments from four studies reported reduced activity in response to sweet stimuli [13, 17, 26, 35]. An ALE analysis of these experiments revealed reduced activity in response to sweet stimuli compared with a basal baseline in one cluster of 169 mm³ in the right medial superior frontal gyrus at x = 2, y = 60, z = 12, with a maximum value of Z = 1, and ALE value of 0.0089. This analysis used the cluster-level FWE rate of 5%.

Brain regions sensitive to hours fasted

Treating hours fasted as a continuous predictor in a meta-regression model revealed that hours fasted was positively associated with responses to sweet stimuli in the left precentral gyrus, right superior parietal lobule, left middle occipital gyrus, left anterior cingulate cortex, right putamen, left thalamus, left amygdala, and bilateral cerebellar lobule VI. Hours fasted was negatively associated with activity in the left supplementary motor area (SMA), left lateral OFC, left supramarginal gyrus, and right fusiform gyrus (Table 3a, Fig. 3a). The results for increased and decreased activity were not heterogeneous across experiments.

Table 3.

Separate meta-regression results for hours fasted, hunger, and body mass index as predictors of response to sweet stimuli relative to basal conditions.

	Side	x	y	z	Cluster size	SDM-Z
A
Precentral gyrus	L	−50	2	30	1661	3.102
Superior parietal lobule	R	26	−74	50	196	2.117
Middle occipital gyrus	L	−18	−98	8	660	2.800
Anterior cingulate cortex	L	−6	2	28	192	1.920
Putamen	R	28	6	4	766	2.095
Thalamus	L	−4	−4	6	592	2.174
Amygdala	L	−22	4	−20	840	3.131
Cerebellar lobule VI	R	16	−64	−22	2970	3.560
Cerebellar lobule VI	L	−30	−66	−22	2598	3.000
Supplementary motor area	L	−4	−2	58	614	−1.979
Lateral orbitofrontal cortex	L	−34	26	−14	189	−1.795
Supramarginal gyrus	L	−60	−38	24	170	−1.315
Fusiform gyrus	R	30	−34	−22	167	−1.555
B
Medial orbital frontal cortex	L	−24	44	−10	78	1.901
Medial orbital frontal cortex	R	14	38	−18	282	1.661
Anterior cingulate	L	−4	24	16	41	1.177
Insula	L	−38	2	−16	270	1.709
Anterior superior temporal gyrus	L	−54	−2	0	69	1.266
Amygdala	R	22	−6	−14	557	2.254
Caudate nucleus	L	−10	4	8	481	1.648
Supplementary motor area	L	0	−6	52	932	−1.615
Postcentral gyrus	L	−26	−42	60	254	−1.901
Lingual gyrus	R	20	−90	−14	70	−1.219
Inferior temporal gyrus	L	−48	−46	−28	1521	−2.073
Cerebellum Crus I	R	44	−54	−26	114	−1.219
Cerebellum Crus I	L	−36	−78	−30	56	−1.506
C
Postcentral gyrus	R	30	−32	62	60	1.364
Angular gyrus	L	−42	−72	36	87	1.406
Superior occipital lobe	L	−12	−98	8	100	1.274
Middle temporal gyrus	R	60	−14	−14	63	1.211
Insula	L	−40	−4	−4	549	1.530
Cerebellum lobule VI	L	−30	−56	−34	1672	2.479
Supplementary motor area	L	−4	2	58	1156	−2.934
Angular gyrus	L	−44	52	22	113	−1.695
Temporal pole	R	48	20	−20	410	−1.853
Caudate	R	12	14	2	535	−1.944
Pallidum	L	−16	4	−6	54	−1.445

A Meta-regression of hours fasted predicting response to sweet stimuli relative to a basal condition in 39 experiments, voxel threshold: p < 0.005, peak height threshold: peak SDM-Z > 1.000. B Meta-regression of hunger as a predictor of response to sweet stimuli relative to a basal condition in 20 experiments, voxel threshold: p < 0.005, peak height threshold: peak SDM-Z > 1.000. C Meta-regression with body mass index (BMI) as a predictor of response to sweet stimuli relative to an active basal condition in 23 experiments, voxel threshold: p < 0.005, peak height threshold: peak SDM-Z > 1.000.

Montreal Neurological Institute coordinates. Positive values for SDM-Z represent activity increases; negative values, activity decreases.

Fig. 3. Separate meta-regression results for hours fasted, hunger, and body mass index as predictors of response to sweet stimuli relative to basal conditions.

a Meta-regression of hours fasted as a predictor of response to sweet stimuli relative to basal conditions; b Meta-regression of hunger as a predictor of response to sweet stimuli relative to basal conditions; and c Meta-regression of body mass index as a predictor of response to sweet stimuli relative to a basal condition. All meta-regressions were conducted using an AES-SDM uncorrected threshold of p < 0.005, cluster extent of ten voxels, where SDM-Z ≤ ± 1.0. Increased activity is in RED/YELLOW, decreased activity in BLUE/GREEN. Montreal Neurological Institute coordinates.

Brain regions sensitive to hunger

Twenty of the 39 experiments recorded participants’ self-reported hunger ratings. Self-reported hunger prior to imaging was not significantly associated with hours fasted (r = 0.19, p = 0.42). Self-reported hunger prior to administration of sweet stimuli relative to a baseline condition was positively associated with activity in bilateral medial OFC, left anterior cingulate, left insula, left superior temporal gyrus, right amygdala, and left caudate nucleus (Table 3b, Fig. 3b).

Greater self-reported hunger was negatively associated with activity in the left SMA, left postcentral gyrus, right lingual gyrus, left inferior temporal gyrus, and bilateral cerebellar Crus I. Meta-regression findings did not show between-study heterogeneity.

Brain regions sensitive to BMI

BMI was not associated with hunger ratings across studies (r = −0.44, p = 0.13) nor with the number of hours fasted across studies (r = 0.26, p = 0.21). Greater BMI was positively associated with activity in the right postcentral gyrus, left angular gyrus, left superior occipital gyrus, right temporal gyrus, left insula, and left cerebellar lobule VI in response to sweet stimuli relative to baseline. Greater BMI was negatively associated with activity in the left SMA, left angular gyrus, right temporal pole, right caudate, and left pallidum (Table 3c, Fig. 3c). These results were not heterogeneous across experiments.

Comparison between studies with children/adolescents to those with adults

A sensitivity analysis using ALE with FWE cluster correction showed that adults had greater bilateral insula activity (x = 40, y = 7, z = −9, z = 2.17; x = 36, y = −8, z = 9, Z = 1.94; x = −30, y = 4, z = −11, Z = 1.81) compared to children/adolescents when comparing sweet stimuli to baseline. Nevertheless, there were not enough child/adolescent experiments to conclusively exclude the possibility of group differences, as the groups were unbalanced, with 32 experiments with adults and only seven experiments with children/adolescents.

Comparison between studies using sucrose flavor to those using chocolate/sucrose flavor

Another sensitivity analysis using ALE with FWE cluster correction compared neural responses to sucrose and chocolate/sucrose flavors. There was greater activity in the precentral gyrus (x = −54, y = −10, z = 30; where Z = 3.54; x = −56, y = −10, z = 23; where Z = 3.35) in response to sucrose/glucose relative to chocolate/sucrose in the 25 experiments with sucrose/glucose and 14 with chocolate/sucrose.

Discussion

Summary of findings

We found evidence that contextual effects are widely distributed in the later stages of a theoretical taste processing hierarchy. Longer fasts were associated with increased activity in regions related to food reward, sensorimotor processing, and homeostatic regulation and reduced activity in areas associated with sensory processing and error detection. Hunger was associated with increased activity in areas associated with taste processing in the insula, and with reward processing, including medial OFC, caudate nucleus, anterior cingulate, and amygdala. Hunger was also associated with decreased activity in areas associated with somatosensory and visual processing, including the postcentral and lingual gyri; and homeostatic regulation, such as cerebellar Crus I. Higher BMI was associated with increased activity in regions associated with reward, and visual processing. Greater BMI was also associated with higher activity in insula and cerebellar Lobule VI and reduced activity in the SMA and striatum.

Functional neuroanatomy of sweet stimulus processing

Our results are largely consistent with previous meta-analyses describing responses to sweet stimuli [2–5]. Consistent with a hierarchical theory of taste processing and previous meta-analysis results, we found that primary taste cortex—the insula—is involved in sweet stimulus processing [2–5]. Like most previous meta-analyses, we found that responses to sweet stimuli are also associated with greater postcentral gyrus, and thalamic activity [2–5]. We also replicated previous meta-analysis findings that responses to sweet stimuli were associated with greater activity in the rolandic operculum [3], claustrum [2], globus pallidus [3], and amygdala [2]. Unlike previous meta-analyses, we found that sweet stimulus responses were associated with activity in the inferior frontal gyrus pars opercularis, which is associated with inhibitory control [36–38]. Also unlike previous meta-analyses, we also found reduced activity in the right medial superior frontal gyrus (Brodmann’s area 10) in response to sweet stimuli, although this finding has to be interpreted with caution as only 5/39 experiments reported reduced activity there in response to sweet stimuli.

Furthermore, with a relaxed critical threshold, we found evidence for taste-related bilateral OFC activity [3]. The OFC has particular prominence in hierarchical models of taste processing and replication of this finding is important because OFC task-related activity is difficult to detect, as it is highly susceptible to signal dropout when using typical fMRI imaging protocols. Another notable region observed at the weaker threshold was the anterior cingulate cortex. The anterior cingulate cortex is regarded as important in food reward processing in Tier 3 of the hierarchical taste processing model [39–42], and previous meta-analyses have identified this area in sweet stimulus processing [2, 3]. With the more relaxed significance threshold, the putamen, caudate head, caudate body, and cerebellar Lobule VI also exhibited sweet stimulus responses. While caudate nucleus activity has been associated with sweet responses in previous meta-analyses [4, 5, 43], putamen and cerebellar activity have not [2–5].

Longer fasts are associated with differential activity in reward and sensorimotor regions

Longer fasts were associated with activity increases in regions associated with food reward processing (anterior cingulate cortex, amygdala) [2–5], habit learning (putamen) [44], sensorimotor integration (thalamus) [4], and error detection (cerebellar Lobule VI). Longer fasts were also associated with reduced activity in regions associated with food reward processing (supramarginal gyrus, lateral OFC, SMA) [45]. It is known that SMG functional connectivity is more associated with midbrain and limbic regions in the hungry, compared with satiated, state [45]. In contrast, fasting was not associated with greater OFC activity [39]. For the effect of hours fasted differed from a previous meta-analysis that showed that fasted conditions were associated with higher insula activity, and fed conditions, with greater superior frontal gyrus, caudolateral OFC, and superior medial prefrontal cortex activity [46]. Discrepancies may have arisen because hours fasting was treated as a categorical variable in the comparison meta-analysis.

Self-reported hunger is associated with increased activity in reward regions

Greater hunger was associated with increased activity in taste processing regions in the insula, and reward processing regions, including the medial OFC, anterior cingulate, and amygdala, as posited by a hierarchical theory of taste processing [39] (Fig. 4a). These findings complement those of a previous meta-analysis showing that greater hunger modulates the response to food pictures in the right amygdala and left lateral OFC [47]. Subjective hunger and fullness, which have substantial individual variation, may be more strongly associated with taste-related activity than fasting duration. In support of this conclusion, 5% of people do not report abdominal sensations when extremely hungry and only 50% report abdominal hunger sensations 2 h prior to a meal [48, 49]. Daily food diary studies show that subjective hunger ratings are better predictors of food intake at the next meal than stomach contents [6]. Only half of the articles (15/30) reported hunger ratings, and only one explored a relationship between hunger and neural activity [16]. Participant hunger ratings may be particularly useful in identifying hunger modulation of neural activity in fMRI sweet stimuli protocols.

Fig. 4. Hierarchical model of taste processing of Rolls, 2015, 2016a, 2016b, 2019 [39–42].

a Hierarchical model of taste processing. Tier 1 (RED) encodes gustatory, olfactory, visual, auditory, and somatosensory processing. Tier 2 (GREEN) involves reward evaluation in the OFC and amygdala. Tier 3 (BLUE) includes areas involved in decision-making, such as the medial prefrontal cortex, striatum, anterior cingulate cortex, lateral hypothalamus, and insula; b The meta-regression of positive self-reported hunger effects associated with taste processing (YELLOW) superimposed on the hierarchical model shows modulation of primarily Tier 2 and Tier 3 regions, including the OFC, and c the meta-regression of negative self-reported hunger effects (YELLOW) shows modulation of cerebellar regions, bilateral Crus I, not posited by the hierarchical model.

The effect of body mass index on responses to sweet stimuli

Greater BMI was associated with differential activity in regions associated with taste (insula) and appetite regulation (temporal pole) [50, 51], inhibitory control (SMA) [52], and goal-directed reward behavior (caudate) [53, 54]. Unexpectedly, greater BMI was associated with increased activity in cerebellar Lobule VI, a region associated with hunger and satiety regulation. Our results overlap with previous meta-analyses showing that greater BMI is associated with greater insula activity in response to sweet stimuli [2, 55–57]. Greater BMI is associated with higher activity in regions associated with reward processing, including the caudate and postcentral gyrus [2, 55–57]. Observed differences from previous reviews could be due to: (a) only including studies reporting fasting and hunger measures, or (b) the use of categorical contrasts rather than meta-regression.

Relation to theories regarding taste and reward

Our results generally support existing models of taste processing. One influential theory arranges these systems in a hierarchy (Fig. 4a). In this model, hierarchy Tier 1 is involved in sensory processing in gustatory, olfactory, visual, auditory, and somatosensory cortices. Gustatory responses are encoded in the insula taste cortex, the parvocellular portion of the ventroposteromedial nucleus of the thalamus, and the nucleus of the solitary tract. The primary taste cortex in the insula and frontal operculum processes taste intensity, temperature, and texture, but does not encode taste-specific satiety [39–42, 58–62]. Olfactory responses are encoded in the olfactory bulb and olfactory pyriform cortex. Visual responses are encoded in V1, V2, V4, and the inferior temporal visual cortex. Auditory responses are encoded in primary auditory cortex. Somatosensory responses are encoded in the ventral posterolateral nucleus, primary somatosensory cortex and SII, including the frontal and pericentral opercula. Tier 2 involves reward evaluation and affective processing, and includes the OFC and amygdala, with connections to the striatum [42, 63]. Evidence that this tier codes satiety for specific tastes comes from both monkey single-cell recording studies and human fMRI studies [64–67]. The OFC is also sensitive to fat textures [67–70]. Tier 3 in the hierarchy includes medial prefrontal cortex (Brodmann’s areas 10 and 14), which is implicated in decision-making; the striatum/basal ganglia, including the ventral pallidum, globus pallidus, and bed nucleus of the stria terminalis, associated with stimulus-response habit learning; the cingulate cortex (Brodmann’s areas 24 and 32), which is implicated in action-outcome learning, and the lateral hypothalamus and insula, both involved in autonomic and endocrine processes.

In this regionally specialized account of taste processing, our findings are consistent with the notion that hunger state moderates Tier 2 more than Tier 1 activity during sweet stimulus processing [42]. We observed support for involvement of primary taste cortex in sweet stimulus processing, and partial support for involvement of secondary taste cortex. The meta-regression showing positive effects of self-reported hunger demonstrated hunger modulation of taste activity in OFC and ventral striatum (Fig. 4b), providing support for rodent models of addiction implicating the OFC, ventral striatum and insula in hedonic responses to food [42, 54, 71–76]. Our findings also echo results showing that greater BMI is associated with increased taste sensitivity [71, 72, 74, 75] and reduced mesolimbic activity [77–79]. We did not find support for the suggestion that greater BMI is associated with lower prefrontal activity in response to sweet tastes [80, 81]. Increased BMI may be associated with reduced activity in areas assessing the incentive or motivational value (caudate nucleus) of sweet stimuli [42, 82]. It is notable that the cerebellar activity effects observed in the current study, while previously described in relation to hunger state sensitivity, have not been included in these models (Fig. 4c).

Cerebellar activity modulation by hours fasted, hunger and BMI

Perhaps the most surprising outcome of the meta-analyses was that cerebellar taste-related activity was modulated by hunger, hours fasted, and BMI. Hemispheric Lobules VI and Crus I were the only cerebellar areas sensitive to both subjective and objective measures of hunger (Fig. 4c), albeit in different directions. Bilateral cerebellar Lobule VI showed increased activity with longer fasts. In contrast, hunger was associated with decreased activity in the bilateral cerebellar lobule Crus I. Left cerebellar Lobule VI exhibited increased activity in individuals with greater BMI.

Cerebellar involvement may have been frequently over-looked in the study of hunger, satiety, and obesity, possibly because the structure is not fully covered by imaging protocols where the field of view does not capture the whole brain. Thus, the lack of published evidence for cerebellar involvement in these processes may simply reflect inadequate sampling. Nevertheless, both animal and human studies support the importance of the cerebellum in error detection and homeostatic regulation. In nonhuman primates, direct bidirectional connections have been demonstrated between the cerebellum and the hypothalamus [83–85], consistent with a modulatory role for the cerebellum on hypothalamic feeding centers. In rats, lateral hypothalamic neurons receive inhibitory monosynaptic and excitatory polysynaptic inputs from the cerebellar fastigial nucleus [86]. These connections may influence hypothalamic modulation of feeding behavior.

In humans, there is evidence that the cerebellum increases activity in response to gustatory or olfactory stimulation [59]. In addition, viewing high, compared with low, calorie foods was associated with higher cerebellar activity [87]. Resting state fMRI studies have shown that the amplitude of low-frequency fluctuations in both right and left cerebellar Lobule VI are associated with increased self-reported hunger and that cerebellar network activity is correlated with BMI [88]. In another resting state study, functional connectivity between the left posterior insula and cerebellum, and between left hypothalamus and inferior frontal gyrus were stronger during fasting [89]. Transcranial magnetic stimulation of cerebellar hemispheres (Lobules VI and VII) results in increased self-reported hunger [90].

The mechanism by which the cerebellum may alter hunger and satiety, and in so doing, alter weight, have not been fully elucidated. In rodents, lesions in cerebellar Lobule VI decrease caloric intake and weight [91]. A number of animal studies show that the cerebellar nuclei modulate lateral hypothalamic activity associated with hunger and satiety responses [92], via connections with gastric vagal nerves involved in ghrelin, glycemia, CCK, and leptin signaling [92–94]. The cerebellum is also rich in orexin and neuropeptide Y neuropeptides modulating hunger and satiety [92]. In humans, there is some evidence that leptin in the cerebellar hemispheric Lobule VI may modulate weight, with decreases in BMI due to leptin replacement therapy co-occurring with decreases in dorsal posterior cerebellar responses to high-calorie food pictures [95, 96].

Genetic evidence from both animal and human studies has shown that an Iroquois-class homeodomain protein (IRX3), expressed in the human cerebellum, may be linked to the fat mass and obesity-associated protein (FTO) gene polymorphisms associated with weight regulation in large-scale GWAS studies [97, 98]. In mouse and zebrafish, the obesity-associated FTO region has been shown to directly interact with the promoters of IRX3. Using the Genetic Investigation of Anthropometric Traits dataset, IRX3 deficient mice showed 25–30% lower weight, mainly due to browning of white adipose tissue and associated loss of fat mass with an increase in basal metabolic rate [99]. This study showed that only IRX3 brain expression showed highly significant associations with BMI. Analysis of postmortem data from the Human Brain Transcriptome also showed that the FTO allele, associated with higher BMI, is also associated with increased IRX3 cerebellar expression, suggesting that this region may be important in mediating FTO effects on BMI [99]. In mice, hypothalamic IRX3 is modulated by fasting/feeding cycles, and diet induced obesity disturbs this cycle, providing evidence for the central role of hypothalamic IRX3 in the control of whole body homeostasis [100]. These findings emphasize the potentially complex role played by hypothalamic/cerebellar interactions in energy homeostasis. Present theories of food consumption, hunger, and weight regulation may benefit from inclusion of cerebellar influences.

Study limitations and strengths

Not all studies examining sweet stimulus processing were whole-brain studies and not all report details of whether participants were fed or fasted and, if so, for how long. Meta-analysis findings are limited by the number of studies meeting these criteria. We included studies with children/adolescents as well as adults. While the hours that participants were fasted did not differ between child/adolescent and adult studies, responses to sweet stimuli compared to baseline contrasts were associated with greater insula region activity in adults compared to children/adolescents. In addition, we were unable to examine the effects of eating ad libitum as opposed to the effects of being “fed” a standard meal prior to imaging. Being “fed” is arguably not equivalent to the experience of being “sated” or “full”. Our meta-regressions are limited by assuming a linear relationship between independent and dependent variables. The results examining the effects of hunger are particularly limited by the small number of studies that recorded self-reported hunger ratings.

Unfortunately, we could not dissociate cue-only from cued and uncued sweet stimulus presentation, which may account for different findings from meta-analyses using food pictures as stimuli [56, 57]. Stimulus modality may be important in comparisons of overweight and lean individuals. Most studies presented images of food while concurrently presenting sweet tasting stimuli, which may have accounted for differential activity in visual processing regions such as the fusiform gyrus and middle occipital gyrus. Greater hours fasted were associated with greater superior occipital gyrus activity.

Finally, our findings are cross-sectional. Given this, we cannot draw conclusions about whether sweet stimuli lose their incentive value overtime with repeated consumption or if satiety and hunger systems become desensitized overtime, stimulating overeating and eventual weight gain.

Strengths of the study included use of studies using whole-brain imaging protocols and examining the neural effects of stimuli similar to the high-fat, high-sugar foods known to increase obesity risk. The use of meta-regression to examine the effects of hours fasted, hunger, and BMI allowed examination of both effect size and location of activity, with simultaneous examination of both increased and decreased activity effects.

Future directions

Future examination of self-reported hunger effects on taste processing is needed. Our findings highlight the importance of gathering similar self-report data across studies to improve the interpretability of neural findings. Investigating neural mechanisms of obesity risk will benefit from use of more controlled experiments, examination of the time-course of effects and the use of longitudinal multimodality approaches. Using larger sample sizes will facilitate detection of diffuse and small effects. Difficulties detecting OFC and cerebellar activity because of susceptibility artifacts suggest that optimization of image acquisition protocols in these areas is warranted. Further study is needed to better understand the specific role of the cerebellum in incorporating satiety and hunger influences on ingestive behavior. Although fMRI can directly capture the neural responses to food consumption non-invasively, the study heterogeneity we observed sugests the need for better control and standardization of procedures in taste processing experiments to reduce noise and enhance reproducibility.

Conclusions

Consistent with a hierarchical theory of taste processing and previous meta-analyses, we confirmed that the primary taste cortex in the insula is involved in sweet stimulus processing. Contextual effects were widely distributed in the later stages of the hierarchy. Longer fasts were associated with increased activity in regions related to food reward, sensorimotor processing, and homeostatic regulation, with reduced activity in areas associated with sensory processing and error detection. Hunger was associated with increased activity in areas associated with taste processing in the insula, and with reward processing, including medial OFC, caudate nucleus, anterior cingulate, and amygdala. Hunger was also associated with decreased activity in areas associated with somatosensory and visual processing, including the postcentral and lingual gyri; and homeostatic regulation, such as cerebellar Crus I. Higher BMI was associated with increased activity in regions associated with reward, and visual processing. Greater BMI was also associated with higher activity in insula and cerebellar Lobule VI and reduced activity in the SMA and striatum.

In summary, neural responses to sweet stimuli are modulated by both state (hours fasted and hunger), and trait (BMI) variables. Our findings motivate extensions to existing taste processing models to include cerebellar systems whose response to sweet stimuli are moderated by both state and trait variables. Control or monitoring of these effects could improve sensitivity and reproducibility in neuroimaging experiments exploring the neural mechanisms of sweet food processing.