Restrained eating (RE) or dietary restraint (DR), refers to the intentional, sustained restriction of food intake for purposes of weight loss or maintenance (Herman and Polivy, 1980). RE and DR have been assessed with several self-report measures including the Restraint Scale (RS; Herman and Polivy, 1980), Three Factor Eating Questionnaire (TFEQ; Stunkard and Messick, 1985) and Dutch Eating Behavior Questionnaire (DEBQ; Van Strien et al., 1986). Consistent with predictions of RE theories (Herman and Polivy, 1980; Heatherton et al., 1988), some evidence indicates women who score higher on these measures have significantly higher risk for future onset of binge eating, weight gain, and bulimic symptomatology compared to lower-scoring peers (Laessle et al., 1989; Johnson and Wardle, 2005; Stice et al., 2005). For example, Laessle (Laessle et al., 1989) found high RS scores were closely related to consequences of unsuccessful dieting including disinhibited eating and weight fluctuations, but were not associated with successful overall caloric restriction in everyday life. Other recent studies have also indicated REs are more prone to impulsivity, difficulties with response inhibition, and overeating (Jansen et al., 2009; Jasinska et al., 2012). Taken together, these studies indicate RE is typically difficult to maintain over extended intervals, especially in environments wherein cues for easily accessible, high-calorie foods abound.

Paralleling behavior differences, neuroimaging studies have corroborated differences between REs and unrestrained eaters (UREs) in the processing of food stimuli. For example, a positron emission tomography (PET) study indicated DEBQ DR subscale scores were positively correlated with dorsal striatum dopamine release during exposure to the smell and taste of food (Volkow et al., 2003). Coletta et al. (2009) used the RS to investigate neural activation differences between REs and UREs via functional magnetic resonance imaging (fMRI); they found REs showed more activation than UREs did in the left orbitofrontal cortex (OFC) and dorsolateral prefrontal cortex (DLPFC) in response to food pictures. Burger and Stice found higher RE scores based on the Dutch Restrained Eating Scale were related to hyper-responsivity in motivational and reward regions (right OFC, bilateral DLPFC) during food intake (Burger and Stice, 2011). Dong et al. (2014) used regional homogeneity (ReHo) analysis to assess resting-state activity among REs and UREs identified on the basis of RS scores. They found REs showed comparatively more ReHo in brain regions associated with food reward (i.e., OFC, DLPFC), attention (i.e., lingual gyrus, cuneus, inferior parietal lobule), and somatosensory functioning (i.e., paracentral lobule, anterior insula). DLPFC is known to be a high level cortex with multiple functions. One of the major function is the cognitive control (Greene et al., 2004; Egner and Hirsch, 2005). Nevertheless, as mentioned above, DLPFC has also shown to be a reward related region (Coletta et al., 2009; Burger and Stice, 2011; Dong et al., 2014).

Functional neuroimaging studies have shown food and food-related visual or olfactory cues can activate brain circuits implicated in reward such as the OFC and insula among REs. Whereas, the OFC is thought to encode information related to the reward value of food, the insula may contribute to processing information related to the taste of food and its hedonic value. Collectively, research implicating activation of these structures underscores how RE may strengthen the incentive salience of food cues, hence increasing risk for eating behavior in food-rich environments.

Behavior and fMRI correlates of RE have been documented, yet there is a relative paucity of research on associations between individual differences in RE and regional gray matter volume (rGMV), particularly within non-clinical samples. Voxel-based morphology (VBM) is an unbiased, whole-brain objective technique used to identify differences in the brain's gray matter (Ashburner and Friston, 2000). Gray matter comprises unmyelinated neurons and other cells distributed throughout the cerebral cortex, brainstem, cerebellum, and spinal cord. GMV is a product of cortical thickness and variations in surface area due to patterns of associated folds. Studies of brain anatomy can provide unique insights into the neural basis of dieting behavior. However, data are scarce in this regard. To our knowledge, only one study has investigated the association between eating behaviors and brain structural changes. Yao et al. (2016) employed VBM to examine the relationship between eating behavior traits, measured by TFEQ, and brain structural changes. They found cognitive restraint of eating was positively correlated with the GMV in the DLPFC and negatively correlated with the GMV in the putamen. As well, disinhibition scores were negatively associated with the GMV in the left middle frontal gyrus (Yao et al., 2016).

Yao et al.'s study is potentially instructive in establishing links between different aspects of eating behavior and alterations of various brain structures. However, it is not clear whether findings identified in that sample (age 20–60 years) apply to specific at risk groups such as young adults whose brains have not matured fully. Furthermore, despite considerable consistency in fMRI studies of RE, one possible source of variability in results may be in relation to the measurement of RE. Although popular RE scales share a common motivational component reflecting desire for thinness and concerns with shape and weight (Laessle et al., 1989), they diverge in other respects. For example, while high scores on salient TFEQ-2 and DEBQ subscales are said to tap successful dieting, the RS is a well-validated alternative on which high scores are more reflective of unsuccessful dieting (Laessle et al., 1989; Allison et al., 1992; Lowe, 1993; van Strien, 1999; Strien, 2002; van Strien et al., 2007; Stroebe et al., 2013). Finally, evaluating individual differences in RE and rGMV within non-clinical groups may clarify why people who score higher on RE measures have significantly higher risk for future onset of binge eating, weight gain, and bulimic symptomatology compared to lower-scoring peers. To address these issues, we assessed associations between RE and rGMV within a non-clinical sample of young women and men.

In addition, because RE may be multidimensional in nature, it is not clear whether some facets have stronger associations than others with rGMV. For example, the revised RS (Herman and Polivy, 1980) includes subscales that reflect concern for dieting (RS-CD) and weight fluctuations (RS-WF) that may be used in addition to total RS scores. Some authors have argued that RS-WF subscale items should be disregarded because they are potentially confounded with obesity and overweight (Stroebe, 2008). However, to explore these concerns empirically, links of each RS subscale and total RS scores with GMV can be assessed, after first statistically controlling for the impact of body mass index (BMI). Presumably, significant associations between RS-WF scores and GMV, independent of BMI, would suggest this subscale has utility beyond BMI.

Extending previous structural MRI research (Volkow et al., 2003), the objective of the current study was to examine structural changes associated with dietary restraint. In this study, we first used VBM to quantify GMV. Then, we conducted regression analyses to assess relations between GMV and total RS scores as well as e brain structural alterations associated with (I) concern for dieting (RS-CD) and (II) weight fluctuation dimensions of RS.

To the best of our knowledge, this study is the first to use VBM analyses to investigate associations between RE and rGMV in a large, non-clinical sample of women and men. Results were in line with the general hypothesis that individual differences in overall RS scores would correspond to significant structural differences in brain regions associated with reward and inhibition.

First, within the entire sample, high total RS scores were associated with larger GMV in the left insula (BA 13), an area implicated in processing interoceptive information reflecting hunger, thirst, and visceral sensations (Craig, 2011). The insula, striatum and midbrain structures encode the subjective value of rewards regardless of their type (Sescousse et al., 2010). Moreover, increased insula GMV in eating disorder groups relative to non-disordered controls is associated with taste pleasantness and reward value. Smucny and colleagues found GMV was lower in the insula, medial OFC, and cerebellum in obese-prone compared to obese-resistant individuals, suggestive of structural differences in brain regions important in energy regulation among persons at risk for weight gain (Smucny et al., 2012). Functional neuroimaging studies have linked stronger insula activation to encoding food liking (Volkow et al., 2011) and exposure to appetizing food pictures among REs (Coletta et al., 2009). Other research has found REs are more likely than UREs to show heightened reward circuitry in response to food cues that increase risk for overeating (Jasinska et al., 2012). However, Yao et al. (2016) found cognitive restraint of eating was positively correlated with DLPFC GMV and negatively correlated with GMV in the putamen. We may conclude that the inconsistent results are probably due to differences between the various restraint scales (Yao et al., 2016).

In the context of most other studies, our finding of an increased GMV in the insula implies REs may showy an enhanced drive for compulsive eating that interferes with the control of eating, in turn, increasing risk for obesity. However, because there is unlikely to be a one-to-one correspondence between fMRI activation and GMV in specific regions, future studies that evaluate associations between RE, insular rGMV and BOLD signal changes in response to food stimuli are needed to elucidate these associations further.

Total RS scores were also inversely correlated with GMV in the posterior cingulum gyrus (CGp). Several MRI studies have suggested the CGp has a more fundamental role in the allocation of neural resources to cognitive control akin to that of attention in the selective processing of sensory stimuli (Maddock et al., 2001, 2003; Greicius et al., 2004; Luhmann et al., 2008). In food-rich environments, the long-term goal of weight control as a self-regulation process requires more cognitive resources than the proximal goal of food enjoyment does. There is evidence that both (i) proactive self-control efforts in response to immediate temptations and (ii) externally-imposed controls can help to protect long-term interests such as losing or maintaining weight (Takeuchi et al., 2014). However, it is plausible that weak self-control efforts and/or lack of responsiveness to external controls in response to immediate temptations impair one's ability to monitor or inhibit inappropriate behavior, consequently disrupting long-term weight control goals. An associated hypothesis that warrants attention within future longitudinal extensions is the possibility that more rGMV in food reward regions and/or less rGMV in the CGp undermines self-control capacities of REs in response to enticements, in turn, increasing vulnerability to future overeating and weight gain.

In disentangling features of RE further, higher RS-CD subscale scores were correlated with increased GMV in a cluster that included left insula areas extending to the orbitofrontal cortex (OFC). The insula (BA 13) and OFC (BA 12) comprise areas in a network of regions involved in processing food stimuli, emotion, awareness of internal state, and behavioral inhibition, and are believed to integrate sensory information that affects eating and reward-related behavior (Craig, 2010; Burger and Stice, 2011; Weise et al., 2013). Also, Cristofori et al. (2015) pointed out that connections from the insula to the OFC could allow the insula to integrate the value of rewards in modulating internal states (Cristofori et al., 2015). In another VBM study, Schäfer et al. (2010) found that, compared to non-eating disordered controls, both bulimia nervosa and binge-eating disorder groups were characterized by increased medial OFC GMV reflecting possible group differences in food reward processing. Neuroimaging studies have also supported medial OFC involvement in subjective evaluations of reward (Kringelbach et al., 2003) and activation during the anticipation and experience of pleasant tastes (O'Doherty et al., 2002; Kringelbach, 2005). Another recent study found that high restraint scores of adolescent girls were related to more right OFC activation in response to the taste of small amounts of milkshake (Burger and Stice, 2011). In the context of other VBM and fMRI studies, the present findings suggested high levels of concern with dieting correspond to larger GMV in food reward regions.

RS-WF subscale scores in the sample were negatively correlated with GMV in the posterior cingulum gyrus (CGp), partially in line with Brooks et al. (2011) who observed reduced right posterior cingulate GMV among women with AN, a group preoccupied with weight changes, relative to non-disordered controls. Past fMRI research has implicated CGp involvement in allocating neural resources toward cognitive control (Maddock et al., 2003). Presumably, when attempts to control food intake among REs are accompanied by reduced cognitive control or impaired self-regulation, vulnerability to overeating and weight changes increase (Vohs and Heatherton, 2000). In light of the negative relation of RS-WF scores with GMV in this sample, prospective studies that evaluate (i) relations between reported weight fluctuations and changes in CGp GMV volumes as well as (ii) complementary associations between CGp GMV and changes in BMI or perceived weight fluctuations can clarify implications of this correlation further.

Finally, within each gender, associations between RE and GMV in the left insula and posterior cingulum gyrus replicated those observed for the entire sample. However, among the women, RE also had positive and negative correlations with GMV, respectively, in the parahippocampal gyrus and middle frontal gyrus (MFG). The parahippocompal gyrus may have an key role in memory encoding and retrieval. However, of more possible relevance here, inhibition of inappropriate responses, control of goal-directed behaviors, and ability of error detection/correction are key MFG functions (Menon et al., 2001). Yokum et al. found trend-level reductions in MFG GMV were related to BMI increases over a 1-year follow-up (Yokum et al., 2012). In concert with that study, a plausible hypothesis that follows from the present finding is that lower GMV in the MFG among women with higher RE levels may correspond to reduced inhibitory control that increases vulnerability to overeating and weight gain in food-rich environments. Among the men, RE scores had a unique significant positive correlation with middle temporal gyrus GMV, consistent with hypothesized links between RE and GMV in food reward areas. Specifically, these data converge with Dong et al. (2014) who observed positive correlations between RE and resting state activity in the MTG and other regions involved in visual perception and food reward.

This study was carried out in accordance with the recommendations of the guidelines for human research, the human research ethics committee at SWU with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the the human research ethics committee at SWU.

JQ and HC design the study; YS and DW collect the data; YS analyzes and interprets the data; YS and TJ write the manuscript.