Comparison of Food Cue–Evoked and Resting-State Functional Connectivity in Obesity

Abstract

Objective:

Obesity is associated with differences in task-evoked and resting-state functional brain connectivity (FC). However, no studies have compared obesity-related differences in FC evoked by high-calorie food cues from that observed at rest. Such a comparison could improve our understanding of the neural mechanisms of reward valuation and decision making in the context of obesity.

Methods:

The sample included 122 adults (78% female; mean age = 44.43 [8.67] years) with body mass index (BMI) in the overweight or obese range (mean = 31.28 [3.92] kg/m²). Participants completed a functional magnetic resonance imaging scan that included a resting period followed by a visual food cue task. Whole-brain FC analyses examined seed-to-voxel signal covariation during the presentation of high-calorie food and at rest using seeds located in the left and right orbitofrontal cortex, left hippocampus, and left dorsomedial prefrontal cortex.

Results:

For all seeds examined, BMI was associated with stronger FC during the presentation of high-calorie food, but weaker FC at rest. Regions exhibiting BMI-related modulation of signal coherence in the presence of palatable food cues were largely located within the default mode network (z range = 2.34–4.91), whereas regions exhibiting BMI-related modulation of signal coherence at rest were located within the frontostriatal and default mode networks (z range = 3.05–4.11). All FC results exceeded a voxelwise threshold of p < .01 and cluster-defining familywise error threshold of p < .05.

Conclusions:

These dissociable patterns of FC may suggest separate neural mechanisms contributing to variation in distinct cognitive, psychological, or behavioral domains that may be related to individual differences in risk for obesity.

INTRODUCTION

Components of executive function may be impaired in obesity (1). For instance, obesity is associated with impaired performance on tasks that require inhibition of primed responses (e.g., flanker tasks) as well as cognitive flexibility (e.g., Wisconsin Card Sorting Task (2–5)). Adults with obesity also show a preference for smaller, immediate rewards over larger, delayed rewards relative to normal-weight individuals (6). Importantly, steeper discounting of delayed future rewards has been associated with increased purchasing (7) and consumption (7,8) of highly palatable, calorie-dense foods, as well as binge eating disorder (4). Furthermore, greater capacity for self-regulation has been associated with more frequent consumption of healthy low-calorie foods and regular engagement in physical activity, both in healthy weight adults and adults with obesity (9–11). These findings indicate that impaired self-regulation and reward processing may predispose some individuals to obesity by influencing decisions regarding diet and exercise.

Impaired self-regulation and reward-related decision making in obesity may be indicative of altered signaling in the frontostriatal neural circuit linking prefrontal cortical (PFC) regions involved in inhibitory control to midbrain regions such as the ventral striatum and nucleus accumbens (NAc) that are involved in reward valuation (12,13). Indeed, body mass is inversely associated with activation of regions of the PFC involved in inhibitory control in response to palatable food cues (14) and during attempts to regulate craving for unhealthy foods (15). This is accompanied by greater activation in frontostriatal regions involved in reward processing during reward valuation tasks including high-calorie food cues (16–19). Altered signaling in frontostriatal regions is also prospectively associated with weight gain and response to weight loss interventions (16,19,20). These findings are consistent with the hypothesis that vulnerability to obesity is linked to underrecruitment of PFC-supported inhibitory control coupled with overvaluation of rewards as reflected by overrecruitment of medial PFC and ventral striatum. However, further work is necessary to evaluate whether the connectivity between these regions is also altered in obesity.

There are also abnormal brain connectivity patterns among adults with obesity, both in the context of specific task demands and at rest. For instance, obesity is associated with altered signal coherence between regions in the frontostriatal network during the processing of monetary (21,22) and food rewards (22–24). Interestingly, successful regulation of craving, as well as successful response inhibition, are associated with the opposite pattern of cortical-subcortical connectivity among adults with obesity, with stronger coupling predictive of more successful attenuation of craving and response inhibition (25–27).

Resting-state brain networks also exhibit obesity-related differences in connectivity (28–30). Obesity has been associated with decreased global and local efficiency, as well as modularity of functional networks throughout the brain (23,31), including the default mode network (DMN (32)). Furthermore, individuals with obesity show enhanced resting-state connectivity between regions involved in reward valuation and decision making such as the NAc, the anterior cingulate cortex (ACC), and ventromedial PFC (vmPFC (33)), suggesting that altered reward network function in obesity is not specific to valuation of food cues but may reflect more fundamental context-independent disruptions. Importantly, several studies have demonstrated that diet, exercise, and weight loss surgery modify resting-state functional connectivity patterns (34–36), indicating that differences in resting-state functional connectivity may be mechanistically linked to obesity and are modifiable with weight loss.

Despite this prior work, several questions remain. First, studies have not analyzed task-evoked and resting-state connectivity in the same individuals or examined whether differences in functional connectivity at rest correspond to variation in connectivity of these same regions during the processing of food cues. Furthermore, most studies have compared functional connectivity among adults with obesity with that of healthy-weight individuals. Implicit in this approach is the assumption that individuals characterized as obese are a homogenous group, which may obscure potentially meaningful individual differences among those who meet the clinical criteria for overweight or obesity. Finally, few studies have determined whether weight-related differences in connectivity predict performance on tasks that evaluate the cognitive processes thought to be supported by the affected networks. To address these gaps, we used a hypothesis-driven seed-based approach to determine whether body mass was associated with variation in frontostriatal connectivity with other brain areas during the processing of high-calorie food cues and at rest. Furthermore, we tested whether variation in functional connectivity of these networks mediated the relationship between body mass and performance on a reward task.

METHODS

Participants

Participants were recruited from a behavioral weight loss intervention focused on cardiovascular outcomes. Of the 383 individuals enrolled in the intervention, 125 (32.6%) participated in this neuroimaging study. Participants who completed the neuroimaging arm of the intervention did not differ significantly from the parent sample on any clinical or demographic indicator, with the exception of systolic blood pressure; however, this difference was not clinically meaningful (parent sample, 120.2 [11.7] mm Hg; neuroimaging sample, 118.4 [11.7] mm Hg). Comparisons of the two samples are provided in Table 1. The results from this study are from the data collected at baseline, within 1 month of the start of the intervention. Data for the neuroimaging arm of the study were collected between November 2013 and July 2016. Participants for the parent study were primarily recruited through direct mailings and research registries at the University of Pittsburgh. The study was approved by the University of Pittsburgh Institutional Review Board, and all participants provided informed consent before initiation of any procedures.

TABLE 1.

Comparison of Participants in the Neuroimaging Arm of the Study to Total Intervention Sample

	Total Sample (n = 383)	MRI Ancillary Study (n = 125)	p
Age, M (SD), y	45.0 (7.9)	44.3 (8.6)	.274a
Weight, M (SD), kg	90.9 (13.7)	91.3 (14.1)	.687a
Body mass index, M (SD), kg/m2	32.4 (3.8)	32.4 (3.9)	.801a
Resting systolic blood pressure, M (SD), mm Hg	120.2 (11.7)	118.5 (11.63)	.044a
Resting diastolic blood pressure, M (SD), M (SD), mm Hg	72.2 (8.9)	71.1 (8.9)	.077a
Waist circumference, M (SD), cm	106.4 (9.9)	106.1 (9.8)	.680a
Cardiorespiratory fitness, M (SD), ml kg−1 min−1	22.6 (4.4)	22.9 (4.4)	.381a
% Female	79.4	78.4	.743b
% White	72.8	76.8	.226b

MRI = magnetic resonance imaging; M = mean.

^ap Value based on independent t test.

^bp Value based on χ² test.

Procedures

Exclusion criteria were as follows: women who were currently pregnant, breastfeeding, or reported planning a pregnancy in the next 12 months; history of bariatric surgery; report of current medical condition or treatment of a medical condition that could affect body weight (e.g., diabetes mellitus, hyperthyroidism or hypothyroidism, etc.); report of a current cardiovascular condition, myocardial infarction, or cardiac surgery in the previous 12 months; resting systolic blood pressure ≥160 mm Hg or resting diastolic blood pressure ≥90 mm Hg or taking medication that would influence blood pressure; eating disorders; substance abuse; current psychological treatment; taking psychotropic medication in the previous 12 months; hospitalization for depression within the previous 5 years; and contraindications to having a magnetic resonance imaging (MRI) assessment (e.g., metallic implants, report of claustrophobia, etc.). Additional exclusion criteria for this ancillary study included left-handed individuals or individuals with any form of traumatic brain injury or neurological illness. Given these exclusion criteria, the sample was in generally good health apart from having overweight or obesity.

Assessments

Body Mass Index

Participants’ height and weight were obtained to calculate body mass index (BMI) according to the standard formula: kg/m².

MRI Visual Food Cue Stimuli and Design

MRI was performed using a Siemens Verio 3-Tesla MRI scanner. For the resting-state scan, participants viewed a fixation point for 5 minutes and 28 seconds. Participants then completed a visual food cue paradigm (37,38) that was adapted from Killgore et al. (38). Briefly, participants passively viewed 24 second blocks of high-calorie food images (e.g., pizza), low-calorie food images (e.g., carrots), and images of neutral nonfood items of similar visual complexity, texture, and color (e.g., houses), with 20-second rest periods between blocks. During each block, 12 images were displayed for 2000 milliseconds each. Each block type was presented three times. Participants also rated their level of hunger on a scale ranging from 0 (not at all hungry) to 10 (very hungry) between blocks. This was done to ensure that participants were awake and attending to the task.

Iowa Gambling Task

The Iowa Gambling Task (IGT) is a monetary decision-making task that quantifies the degree to which individuals learn the decision-making strategy leading to the highest earnings, as well as disadvantageous selections (39). Briefly, participants completed a computerized version of the IGT in which they selected 100 cards one at a time from four possible decks, with the goal of maximizing the amount of money won at the end of the task. Card selections were divided into five blocks to permit evaluation of decision making trajectories over the course of the task. Two metrics were derived based from this. First, a net payoff score was calculated by subtracting the total number of cards selected from the disadvantageous decks from the total number of cards selected from the advantageous decks ((C + D) − (A + B)). Higher payoff scores are indicative of a greater proportion of selections from advantageous decks. Second, a difference score was calculated by subtracting the net payoff score in block one from the net payoff score in block five to quantify change in decision making over the task. Higher difference scores suggest a positive learning curve, such that feedback on early trials informs the adoption of a more successful decision-making strategy in later trials.

MRI Data Acquisition and Preprocessing

For the visual food cue task, 204 T2*-weighted volumes were obtained using a fast echo-planar imaging sequence with blood oxygen level–dependent (BOLD) contrast (time repetition, 2000 milliseconds; echo time, 28 milliseconds; flip angle, 90 degrees). Thirty-four slices were collected at 3.2-mm thickness in the posterior to anterior direction. For the resting-state scan, 210 T2*-weighted volumes were obtained using an echo-planar imaging sequence with BOLD contrast (time repetition, 1540 milliseconds; echo time, 25 milliseconds; flip angle, 90 degrees). Thirty slices were collected at 3.5-mm thickness in the posterior to anterior direction. High-resolution T1-weighted anatomical volumes were also collected in the sagittal plane using a magnetization-prepared rapid gradient-echo sequence for each participant (256 slices, voxel dimensions 1 × 0.976 × 0.976 mm).

Task-related and resting-state functional data were preprocessed using FEAT version 5.98, part of FSL (FMRIB’s Software Library). Motion correction was conducted using MCFLIRT (40). Functional volumes from the visual food cue task were temporally filtered with a Gaussian high-pass cutoff of 100 seconds. Conversely, a band-pass temporal filter between 0.01 and 0.1 Hz was applied to resting-state volumes to remove noise attributable to physiological processes (e.g., respiration) and noise due to scanner drift. Images from both scans were spatially smoothed with a 5-mm full-width half-maximum three-dimensional Gaussian kernel. Non–brain matter (i.e., skull) was removed using the robust brain extraction technique (41). Functional images were registered to Montreal Neurological Institute space using 12-parameter affine transformations using FMRIB’s linear image registration tool (FLIRT (40,42)). No errors were observed in image registration.

Analysis of Visual Food Cue Data

Regional Activation and Region-of-Interest Selection

Subject- and group-level analyses of neuroimaging data were conducted using general linear models (GLMs) in FSL (FMRIB, Oxford, United Kingdom). Data from the visual food cue task were deconvolved using a gamma function. A GLM using multiple regression examined signal change for each condition (i.e., neutral images, high-calorie images, and low-calorie images). Contrast parameters compared high- and low-calorie conditions relative to neutral conditions and relative to one another. Individuals with motion displacement exceeding 1 mm were included as nuisance covariates of no interest in first-level models (n = 32).

Results from these first-level comparisons were then forwarded to higher-level mixed-effects group analyses. Age, sex, and framewise displacement (FD) were included as covariates. To be more inclusive in the generation of seeds (described hereinafter), the statistical parametric maps from the visual food cue contrasts were thresholded using a voxelwise threshold of p < .01 and a cluster extent threshold of 25 contiguous voxels. Seed regions were chosen based on regional activation patterns observed during the task according to the following criteria: a) a region exhibited differential activation in response to food images (high or low calorie) relative to neutral images, b) activation in response to food images was significantly correlated with BMI, and c) the region had been previously linked with obesity or to processes known to be disrupted in obesity. This approach circumvents the potential problems of “double dipping” for several reasons. First, the selection of seeds based on activity during the visual food cue paradigm does not bias or inflate BMI associations with measures of connectivity during the task or at rest. That is, the demonstration of a relationship between BMI and regional activation in response to food cues does not reveal information or bias the likelihood that BMI relates to the manner in which these regions communicate with the rest of the brain. Indeed, examination of task activation and task-evoked connectivity has been shown to provide complementary rather than redundant information on the function of a given region (43–45) and therefore should be regarded as independent approaches.

Analysis of Task-Evoked Functional Connectivity

Functional connectivity during the visual food cue task was assessed using a psychophysiological interaction (PPI) approach. A PPI analysis quantifies BOLD signal covariation between a seed region and all voxels across the brain, and determines the degree to which seed-to-voxel covariation is modulated by contextual factors like task condition (46).

Seed masks were created for regions meeting the criteria specified previously by placing a 10-mm sphere around detected maxima coordinates. The BOLD signal time series from each seed was then extracted for each individual. These time series were then entered into a GLM along with the task contrast of interest and a PPI term representing the interaction between the seed region time series and the task contrast. In this case, the contrast of interest was BOLD responses to high-calorie food images versus BOLD responses during the baseline fixation. The GLM including these terms was then applied, which generated correlation maps for each individual. The β values for the PPI term represented the degree to which task-evoked variation during high-calorie foods corresponded to task-evoked variation in the seed.

The β values were then forwarded to higher level GLMs to examine whether BMI was associated with variation in seed-to-voxel correlations during the presentation of high-calorie images. Age, sex, and FD were also included as covariates of no interest. The statistical parametric maps generated for each contrast were thresholded using a voxelwise threshold of p < .01 and a cluster familywise error threshold of p < .05 (47).

Resting-State Analyses

We examined whether the connectivity patterns during the visual food cue paradigm were also different during rest. Thus, we used a seed-based approach to assess resting-state functional connectivity, with the same seeds used in the task-evoked connectivity analysis described previously.

To reduce the influence of physiological confounds, functional volumes were segmented into gray matter, white matter, and cerebrospinal fluid. Using a nine-parameter denoising model (48), physiological noise as represented by the first two principal components from white matter and cerebrospinal fluid voxels were regressed out of the whole-brain time series. The global signal and motion parameters were also regressed out of the whole-brain time series. Signal time series were then extracted from each seed and entered into a GLM using multiple regression to model BOLD signal covariation between the seed region and all other voxels controlling for the six motion parameters. Contrast parameters examined which voxels positively and negatively covaried with the seed region time series. The β values derived from these analyses were then forwarded to higher level GLMs to examine whether BMI was associated with variation in seed-to-voxel correlations at rest. Age, sex, and FD were used as covariates. The statistical parametric maps generated for each contrast were thresholded using a voxelwise threshold of p < .01 and a cluster familywise error threshold of p< .05.

Mediation Analyses

To assess the relationship between IGT performance and functional connectivity, regions of interest (ROIs) exhibiting significant BMI-related signal covariation were selected for further analysis with IGT performance. The mean parameter estimates for the identified ROIs were extracted and independent multiple linear regression analyses were conducted using the PROCESS macro with bias-corrected bootstrapped confidence intervals ((49); model 4) in SPSS version 25.0 (IBM Corp., Armonk, NY). This analysis examined whether the relationship between BMI and IGT performance (total score and change score) was mediated by task-evoked or resting-state functional connectivity. Hunger was included as a covariate in these models.

RESULTS

Participant Characteristics

Of the sample of 125, 3 did not complete the visual food cue task. Therefore, the final sample size was 122 (mean [SD] age = 44.43 [8.67] years; 78.7% female) with BMI in the overweight and obese range (mean [SD] BMI = 32.41 [3.95] kg/m², range = 25.10–40.29 kg/m²). The sample was predominantly white (74.6%) and non-Hispanic (97.5%). BMI was not significantly associated with race (β = −0.02, p = .86), ethnicity (β = −0.04, p = .68), sex (β = 0.04, p = .69), age (β = −0.12, p = .19), average hunger rating during the task (β = 0.14, p = .13), likelihood of having functional MRI (fMRI) data corrected for excess motion (odds ratio = 0.334, p = .52), or IGT net payoff score (β = −0.03, p = .76). BMI was marginally associated with change in net payoff score from block one to block five (β = −0.16, p = .081), with lower BMI predicting greater improvements in decision making over the course of the task.

Seed Selection: Regional Activation During the Visual Food Cue Task

Consistent with prior work using the visual food cue paradigm, a diverse set of regions were active during the task. Specifically, BMI was inversely associated with activation in regions located in the left orbitofrontal cortex (OFC), right OFC, and left dorsomedial PFC (dmPFC) during the presentation of food compared with neutral images. These regions were selected as seeds to be used for the task-evoked and resting-state connectivity analyses and are depicted in Figure 1 and listed in Table 2.

FIGURE 1.

Regional activation patterns during the visual food cue task from which seeds were selected for connectivity analyses. A, Contrast map for activation that was negatively correlated with BMI during high-calorie + low-calorie food > neutral blocks, with the seeds derived from that contrast. The left dmPFC seed is depicted in red, the right OFC seed depicted in green, and the left lateral OFC seed depicted in light blue. B, Contrast map for activation that was positively correlated with BMI during high-calorie > low-calorie food blocks, with the left medial OFC seed derived from that contrast depicted in dark blue. C, Contrast map for activation that was negatively correlated with BMI during high-calorie > low-calorie food blocks, with the left hippocampus seed derived from that contrast depicted in yellow. BMI = body mass index; dmPFC = dorsomedial prefrontal cortex; OFC = orbitofrontal cortex; MNI = Montreal Neurological Institute; ROI = region of interest.

TABLE 2.

MNI Coordinates (in mm) of Local Maxima in Regions Used as Seeds Based on Regional Activation During the Visual Food Cues Task

	Maximum z Score	x	y	z
Left lateral orbitofrontal cortex	3.6	−36	24	−14
Right orbitofrontal cortex	3.5	18	20	−18
Left dorsomedial prefrontal cortex	3.47	−16	58	22
Left medial orbitofrontal cortex	3.35	−30	20	−16
Left hippocampus	2.99	−32	−14	−20

MNI = Montreal Neurological Institute.

Regions met a voxelwise threshold of z > 2.3 and cluster extent threshold of 25 contiguous voxels.

Aim 1: BMI-Related Variation in Functional Connectivity Evoked by High-Calorie Food

It was hypothesized that, during the presentation of high-calorie food cues, higher BMI would be associated with reduced connectivity between regions involved in reward valuation (e.g., amygdala, medial PFC [mPFC] and OFC) and lateral regions of PFC, but with increased connectivity between regions involved in motor planning and execution (e.g., premotor cortex, inferior parietal cortex, and cerebellum) and those involved in reward valuation. These hypotheses were partially supported. Specifically, BMI was associated with stronger task-evoked functional connectivity between regions involved in reward valuation and motor planning (e.g., OFC, dmPFC, and basal ganglia). However, contrary to our hypotheses, there were no negative associations between BMI and task-evoked connectivity between prefrontal cognitive control regions and reward valuation regions. Given that the direction of the relationship between BMI and task-evoked connectivity was consistent across seeds, results obtained using the left lateral OFC seed will serve as an exemplar of these effects (Figure 2). However, details regarding the regions that exhibited significant BMI-related increases in connectivity with each seed during the presentation of high-calorie food are described in Table 3.

FIGURE 2.

Networks exhibiting BMI-related increases in engagement with the left lateral OFC seed (MNI coordinates −36, 24, −14) during the presentation of high-calorie food cues. Map was cluster thresholded at z > 2.3 and p < .05. B, Scatterplots of the relationship between BMI and seed-to-ROI signal covariation during the presentation of high-calorie food cues. BMI = body mass index; OFC = orbitofrontal cortex; MNI = Montreal Neurological Institute; ROI = region of interest.

TABLE 3.

MNI Coordinates (in mm) of Local Maxima in Regions Showing BMI-Related Increases in Functional Connectivity With Each Seed During the Presentation of High-Calorie Food

Seed	Region	Maximum z Score	x	y	z
Left lateral orbitofrontal cortex	Left middle frontal gyrus	3.15	−40	12	28
	Right hippocampus	3.32	24	−18	−22
	Right thalamus	3.66	18	−26	4
	Posterior cingulate cortex	3.26	−4	−36	32
	Precuneus	3.49	0	−54	12
	Right medial temporal gyrus	2.79	60	−20	−10
Right orbitofrontal cortex	Right medial temporal gyrus	4.45	62	−16	−16
	Right hippocampus	3.15	28	−22	−14
Left dorsomedial prefrontal cortex	Left caudate	3.30	−12	−2	18
	Left hippocampus	3.36	−30	−12	20
	Left putamen	3.15	−28	0	0
	Right putamen	3.31	18	10	−2
	Right pallidum	4.22	18	2	0
	Medial subcallosal cortex	3.44	−2	12	−10
Left medial orbitofrontal cortex	Right superior temporal sulcus	2.34	60	−20	−10
	Right postcentral gyrus	3.20	60	−14	34
	Left supramarginal gyrus	3.51	−62	−38	34
Left hippocampus	Left dorsomedial prefrontal cortex	4.91	−4	66	16
	Right inferior temporal gyrus	3.74	62	−42	−16
	Left dorsolateral prefrontal cortex	3.30	−38	50	12

MNI = Montreal Neurological Institute; BMI = body mass index.

Regions met a cluster threshold of z > 2.3 and p < .05.

For the left lateral OFC seed, BMI was associated with stronger left lateral OFC functional connectivity with regions involved in sensory processing, memory formation and recall, reward valuation, and self-directed thinking during the presentation of high-calorie food cues. These regions included the left medial frontal gyrus right hippocampus, right medial temporal gyrus, right thalamus, medial posterior cingulate cortex, and medial precuneus. Figure 2 depicts the networks exhibiting BMI-related increases in engagement with the left lateral OFC seed during the processing of high-calorie food.

Aim 2: BMI-Related Variation in Resting-State Functional Connectivity

Similar to the hypotheses regarding the relationship between BMI and task-evoked connectivity, it was predicted that, at rest, higher BMI would be associated with reduced connectivity between regions involved in reward valuation (e.g., amygdala, mPFC, and OFC) and lateral regions of PFC, but with increased connectivity between regions involved in motor planning and execution (e.g., premotor cortex, inferior parietal cortex, and cerebellum) and those involved in reward valuation. Across each seed selected, BMI was associated with weaker seed-to-ROI connectivity, but not in a manner consistent with our hypotheses. BMI was consistently associated with weaker connectivity between regions involved in sensory processing, reward valuation, and conflict monitoring, including regions of the basal ganglia, vmPFC, NAc, and ACC. Furthermore, BMI was not associated with resting connectivity between lateral prefrontal regions and regions involved in reward valuation, as had been predicted. It was also hypothesized that the relationship between BMI and resting-state functional connectivity would be similar to patterns observed during the presentation of food cues. However, there was very limited overlap in the regions exhibiting BMI-related variation in connectivity during the visual food cues task as compared with rest, and BMI predicted opposing patterns of connectivity in each of these contexts. Again, because the direction of the relationship between BMI and resting connectivity did not vary by seed, the patterns of connectivity observed using the left lateral OFC are presented in more detail as an exemplar of the effects observed in each seed. Networks exhibiting BMI-related differences in intrinsic connectivity with the left lateral OFC are illustrated in Figure 3. The relationship between BMI and regional resting-state connectivity for all seeds examined is described in more detail in Table 4.

FIGURE 3.

A, Networks exhibiting BMI-related decreases in intrinsic connectivity with the left lateral OFC seed (MNI coordinates −36, 24, −14). Map was cluster thresholded at z > 2.3 and p < .05. B, Scatterplots of the relationship between BMI and seed-to-ROI signal covariation at rest. BMI = body mass index; OFC = orbitofrontal cortex; MNI = Montreal Neurological Institute; ROI = region of interest.

TABLE 4.

MNI Coordinates (in mm) of Local Maxima in Regions Showing BMI-Related Decreases in Intrinsic Functional Connectivity With Each Seed at Rest

Seed	Region	Maximum z Score	x	y	z
Left lateral orbitofrontal cortex	Left thalamus	3.64	−10	−10	14
	Right ventromedial prefrontal cortex	3.91	32	56	4
	Right nucleus accumbens	3.21	14	18	−8
	Medial ACC	3.16	0	34	22
Right orbitofrontal cortex	Right caudate	4.11	16	18	12
	Left paracingulate gyrus	3.91	−4	50	0
	Left amygdala	3.86	−28	−6	−16
	Right ACC	3.63	4	40	8
	Left ventromedial prefrontal cortex	3.56	−10	50	−6
	Right nucleus accumbens	3.41	14	18	−8
	Medial posterior cingulate cortex	3.24	−4	−36	38
	Medial precuneus	3.05	−4	−58	18
Left dorsomedial prefrontal cortex	Right putamena	3.44	30	6	−2
	Left dorsomedial prefrontal cortex	3.43	−10	66	14
	Right dorsomedial prefrontal cortex	3.20	22	44	34
	Right insula	3.14	40	−6	8
	Medial posterior cingulate	3.10	0	−42	26
	Right palliduma	2.96	16	6	0
Left medial orbitofrontal cortex	Right ACC	3.04	10	42	18
	Right putamen	2.77	22	8	−10
	Right superior frontal gyrus	3.03	24	40	32
	Right nucleus accumbens	3.07	12	20	−6
Left hippocampus	Medial ACC	3.64	−4	32	22
	Medial paracingulate gyrus	3.17	10	46	12
	Ventromedial prefrontal cortex	3.14	−4	48	−10

MNI = Montreal Neurological Institute; BMI = body mass index; ACC = anterior cingulate cortex.

Regions met a cluster threshold of z > 2.3 and p < .05.

^aDenotes regions that showed BMI-related differences in functional connectivity with a given seed both at rest and during the visual food cue task.

Mediation Analyses

In an effort to contextualize the observed associations of BMI with functional connectivity, additional analyses were conducted to determine whether task-evoked or resting-state functional connectivity differences were associated with IGT performance. Although the direct pathway (path c) between BMI and IGT performance was not significant, this is not considered a necessary condition for proceeding to test for mediation (50).

No indices of seed-to-ROI connectivity during the visual food cue task were associated with IGT performance or emerged as significant mediators of the relationship between BMI and IGT performance. In contrast, resting-state functional connectivity with several OFC seeds was associated with IGT performance, with stronger connectivity consistently being associated with more advantageous decision making. Interestingly, the strength of resting connectivity for several seed-to-ROI pathways significantly mediated the relationship between BMI and IGT performance, with BMI being associated with lower net payoff and lesser improvements in performance over the course of the task (Table 5). These results suggest that weaker resting connectivity between regions involved in reward processing and decision making is associated with impaired monetary decision making overall, as well as the adoption of less advantageous monetary decision making strategies in response to performance feedback, with the effects being more pronounced with increasing BMI.

TABLE 5.

Indirect Effect of BMI on Overall IGT Performance and on Change in IGT Performance From the First to the Final Block of the Task Operating Through Resting-State Seed-to-ROI Connectivity Values

	B	SE	95% ULCI	95% LLCI
Mediating effects on overall igt performance
Right orbitofrontal cortex
Left paracingulate gyrus	−0.191	0.086	−0.402	−0.050
Left ventromedial prefrontal cortex	−0.178	0.090	−0.397	−0.037
Medial posterior cingulate cortex	−0.170	0.086	−0.377	−0.034
Medial anterior cingulate cortex	−0.108	0.064	−0.282	−0.017
Right nucleus accumbens	−0.232	0.092	−0.449	−0.075
Left dorsomedial prefrontal cortex
Right pallidum	−0.178	0.099	−0.411	−0.021
Mediating effects on change in IGT performance
Left lateral orbitofrontal cortex
Medial anterior cingulate cortex	−0.345	0.181	−0.820	−0.081
Left thalamus	−0.022	0.010	−0.045	−0.005
Right nucleus accumbens	−0.283	0.168	−0.731	−0.036
Right ventromedial prefrontal cortex	−0.442	0.185	−0.861	−0.134
Right orbitofrontal cortex
Left ventromedial prefrontal cortex	−0.266	0.160	−0.673	−0.030
Left medial orbitofrontal cortex
Medial anterior cingulate cortex	−0.212	0.145	−0.631	−0.021
Right ventral striatum	−0.407	0.187	−0.886	−0.130

BMI = body mass index; IGT = Iowa Gambling Task; LLCI = lower limit of confidence interval; ULCI = upper limit of confidence interval.

Confidence intervals were computed using bias-corrected bootstrapping with 5000 samples. Seeds with which ROI connectivity values were not associated with IGT performance are not included in the table.

DISCUSSION

We examined the association between BMI and functional connectivity during the processing of high-calorie food cues, as well as at rest, to determine whether the association between body mass and patterns of neural connectivity is only apparent in specific contexts (i.e., in the presence of high-calorie food cues) or is evident even in the absence of external demands or exogenous cues. Comparison of task-evoked and intrinsic network organization revealed interesting though somewhat unexpected patterns of network connectivity. For each seed, higher BMI was associated with greater regional connectivity in the presence of high-calorie food but weaker connectivity at rest. This suggests that regions that participate in reward valuation, learning, and memory are organized into dissociable context-dependent functional networks among individuals who are overweight or obese. Furthermore, these networks may mediate distinct cognitive, affective, and behavioral processes known to be disrupted in obesity. In support of this hypothesis, weakened resting-state functional connectivity between subdivisions of the mPFC and regions involved in reward valuation (NAc, vmPFC) and conflict monitoring (ACC) was associated with impaired monetary decision making, whereas there was no relationship between functional connectivity during the presentation of high-calorie food and monetary decision making. There is evidence that adults with obesity engage discrete functional networks when evaluating food cues as compared with monetary rewards (22), a finding that is consistent with the patterns of functional connectivity observed in the present study.

Functional Connectivity During the Processing of High-Calorie Food Cues

Modulation of DMN Connectivity

Many of the regions that exhibited BMI-related increases in connectivity in the presence of high-calorie food cues are located within the DMN, including the precuneus, posterior cingulate gyrus, medial temporal gyrus, and hippocampus. This was particularly the case when examining regional connectivity with the OFC, which is also a major hub of the DMN. The DMN supports internally guided and self-referential cognitive processes (51,52). Weight-related associations with task-induced connectivity to regions of the DMN may reflect more thorough engagement with and enhanced processing of high-calorie food cues that extends beyond the experience of reward or pleasure, but may also include recall of personal memories about food (e.g., sharing meals with friends) and the affective states associated with these memories (see more hereinafter). Furthermore, elevated DMN activation during moderately demanding cognitive tasks has been associated with poor task performance and attentional difficulties (53,54). It is possible that increased engagement of the DMN in the presence of food cues in overweight and adults with obesity may come at the expense of optimal performance with higher cognitive demands. Additional research will be necessary to characterize the cognitive and behavioral correlates of altered functional connectivity, including whether these patterns predict problematic eating behavior.

Involvement of the Hippocampus in Processing of Food Cues

In addition to increased coupling of the hippocampus with the OFC described previously, unique functional pathways linking the hippocampus to other regions of the brain were revealed when using this region as a seed. Body mass was associated with greater hippocampal connectivity to the dmPFC, inferior frontal gyrus, and dorsolateral PFC in the presence of high-calorie food, an association not previously reported. In the context of exposure to high-calorie food, increased signal coherence between the hippocampus and dorsolateral PFC may suggest that adults with obesity are not only recalling food-related memories with strong affective content (as indicated by increased OFC-hippocampal connectivity) but also more thoroughly processing the details of these memories. Hippocampal connectivity with the inferior frontal gyrus and dmPFC, two regions supporting working memory (55,56), may similarly reflect enhanced processing of memories involving palatable food at higher body mass. These findings suggest that it may be important to address the way in which an individual engages with cues associated with unhealthy eating during weight loss treatment, perhaps by incorporating elements of cognitive behavioral therapy for binge eating disorder (57).

Functional Connectivity at Rest

Disruption of Reward Networks

Many of the regions exhibiting BMI-related reductions in signal coherence have been previously linked to reward valuation, decision making, and reward-guided action selection, including the NAc, vmPFC, caudate, pallidum, putamen, amygdala, ACC, and paracingulate gyrus. Weakened functional integration of regions involved in reward processing and reward-related decision making in the absence of externally directed processing has been observed in multiple populations with impairments in reward-related decision making (58–60). It is possible that reduced intrinsic functional connectivity between regions that support reward valuation, decision making, and action selection may be indicative of impaired reward processing and contingency learning in overweight and obesity. Furthermore, given that weaker resting connectivity between several of these regions was associated with impaired monetary decision making, and that these patterns were not observed in the presence of high-calorie food, it is possible that the observed disruptions in functional connectivity reflect a generalized reward processing deficit with increasing body mass.

Disruption of the DMN

Signal coherence between regions of the DMN at rest was also found to be inversely associated with body mass, replicating findings reported by several previous studies (28,30,32). Reduced functional cohesion of the DMN at rest may lead to less effective integration of signals from spatially distributed regions, and subsequent dysregulation of processes dependent on such integration (32,51,52). It is possible that reduced functional connectivity in this network leads to a diminished capacity to effectively plan and execute new behaviors that would promote weight loss, and a consequent reliance on habitual yet unhealthy behaviors.

Intrinsic and Evoked Modulation of Basal Ganglia Connectivity

Higher body mass was associated with differences in connectivity between the dmPFC and several regions of the basal ganglia, including the dorsal caudate nucleus, putamen, and pallidum, both at rest and during the presentation of high-calorie food cues. The dmPFC is involved in planning and selecting context-appropriate behavioral responses that maximize the probability of goal attainment, including the receipt of a valued reward such as food (61–63). As such, it is possible that increased signal coherence between the dmPFC and regions of the basal ganglia during the processing of high-calorie food cues may be indicative of a stronger tendency to imagine physically interacting with the food as it is presented. This may be one mechanism through which food cues in the environment influence eating behavior, and may contribute to biasing of attention toward food cues in obesity (64,65).

In contrast, BMI was associated with weaker connectivity between these regions at rest. Previous research has suggested that functional interactions between the dmPFC and basal ganglia support performance monitoring and flexible adjustment of behavior to prevent future errors (66), and is also thought to promote the acquisition of action-outcome contingencies (67) necessary for the execution of behaviors that lead to optimal outcomes. In the context of overweight and obesity, weaker signal coherence between the dmPFC and the basal ganglia at rest may contribute to difficulty adopting healthy life-style behaviors and the failure to devalue high-calorie food despite the negative medical and social consequences associated with overconsumption of such foods. It will be important to examine whether dmPFC–basal ganglia connectivity predicts weight loss, as well as adherence to physical activity and dietary prescriptions.

SUMMARY AND CONCLUSIONS

There are several strengths of the present study, including a comparatively large sample and a multimodal approach to examine the associations between body mass and functional connectivity. Nevertheless, results should be interpreted in the context of several limitations. One limitation is the use of a cross-sectional design, which precludes the ability to draw inferences about causality. To determine the temporal relationship between body mass and functional network organization, it will be necessary to use longitudinal approaches in future research. One such approach is to examine how network-level signaling dynamics change after successful weight loss (68–71). This experimental approach will provide some indication as to whether abnormal functional network architecture is a modifiable state marker of excess weight, or whether it represents a stable characteristic of individuals who are prone to weight gain. Importantly, each of these possible outcomes has different mechanistic implications and may yield alternative approaches to treatment and prevention.

Another important limitation of the current study is that, by design, analyses were constrained to those networks exhibiting significant signal covariation with a relatively small number of seed regions, with the selection of seeds further being constrained by regional activation during a single task, albeit a highly disease relevant one. This approach was adopted to test specific mechanistic hypotheses regarding the pathways underlying weight gain and weight maintenance. However, it is likely that obesity exerted an effect on functional connectivity in other networks that were not assessed in the present study. Future studies may consider using a combination of analytic approaches to simultaneously assess multiple facets of network integrity. Relatedly, it is worth noting that there are alternative approaches to computing task-evoked connectivity (e.g., gPPI (72)) as well as for comparing task-evoked and resting-state connectivity (e.g., (73)). It is possible that the pattern of results observed in the present study may have differed should another approach have been used to compute task-evoked connectivity metrics, or to compare connectivity across task and resting states. As such, it unclear to what extent the differences between patterns of connectivity observed during the task and those observed during rest could be attributed to the analytical procedures adopted in the present study. It has yet to be resolved which analytic approach is most valid, highlighting the need for additional research directly comparing each.

The present study was also limited by the exclusion of healthy-weight individuals. Although the results suggest that there are important individual differences among adults with obesity that are obscured by the use of weight categories, it would be informative to compare patterns of functional connectivity across the full range of BMI. Doing so may reveal important weight-related differences in functional network organization that could have important mechanistic implications regarding the transition from healthy weight to overweight or obese. Furthermore, we used traditional approaches for statistical thresholding of fMRI images, although recent criticism of these approaches (74) suggests that it may be important in future work to apply other (e.g., nonparametric) approaches for determining thresholds of fMRI data in the context of obesity and weight loss. In addition, we did not standardize the menstrual cycle phase in which we collected data from female participants, although there is evidence that circulating sex hormones influence indices of functional connectivity (75). Finally, the time elapsed between a participant’s last meal and their MRI visit was not standardized across participants. Given that the inclusion of hunger ratings in statistical models did not modify the relationship between BMI and indices of functional connectivity, it is unlikely that subjective reports of hunger account for the patterns of BMI-related functional connectivity. However, lack of information on meal time leaves open the possibility that variation in circulating glucose and insulin levels may have influenced reactivity to food cues.

To conclude, our results indicate that higher body mass is associated with altered functional connectivity of several networks. During the presentation of high-calorie food cues, higher body mass was associated with greater functional connectivity between regions involved in reward valuation, contingency learning, memory formation and recall, and self-referential cognitive processes. In contrast, higher body mass was associated with weaker resting connectivity of the same seeds with regions that support reward valuation, decision making, and reward-guided action selection. This suggests that neural mechanisms underlying obesity may be modulated by contextual factors. It is possible that the weight-related disruptions in connectivity observed in one context (e.g., processing of high-calorie food cues) may contribute to deficits in some cognitive domains (e.g., attentional control) but not others. These findings may be leveraged to develop treatments that directly target the processes supported by these networks, a strategy that will ideally maximize treatment efficacy by modifying the underlying neurobiological systems. For example, there is growing interest in using indices of brain function as a form of biofeedback to teach individuals how to develop volitional control over the function of regions and networks that may be responding in suboptimal ways (76). Identification of aberrant patterns of functional connectivity in obesity will inform neurofeedback approaches to modifying network connectivity to support weight loss efforts. This may represent a novel treatment approach to specifically target disruptions in self-regulation, reward-related decision making, and attentional biases for food that could be applied in conjunction with more traditional behavioral weight loss interventions. Furthermore, should future research find evidence that differences in functional network architecture precede the development of obesity, it would suggest that these differences reflect a biomarker of risk for obesity that may be useful for identifying high risk groups that would benefit from early intervention. Additional research exploring the effect of obesity and weight loss on functional network organization will help further the development of innovative approaches to addressing obesity and related disease.

Glossary

ACC

anterior cingulate cortex

BMI

body mass index

BOLD

blood oxygen level dependent

DMN

default mode network

dmPFC

dorsomedial prefrontal cortex

FD

framewise displacement

GLM

general linear model

IGT

Iowa Gambling Task

mPFC

medial prefrontal cortex

NAc

nucleus accumbens

OFC

orbitofrontal cortex

PFC

prefrontal cortex

PPI

psychophysiological interaction

ROI

region of interest

vmPFC

ventromedial prefrontal cortex