Using a novel fMRI paradigm, whereby participants thought about eating food shown in images rather than passively viewing them, we have compared neural responses in women with BN and AN. These exploratory data provide the first evidence that patterns of food consumption (e.g. binges) combined with degree of restraint over eating behaviour may alter how the brain responds when thinking about eating food. However, caution must be exercised; despite stringent threshold corrections, the small sample sizes make these intriguing results somewhat tentative at this stage, and in need of further clarification with larger cohorts. Activation of the dorsolateral prefrontal cortex, visual cortex and cerebellum in response to food versus non-food images was found in women with AN and BN. Additionally, women with BN show greater somatosensory and motor responses in the right insular cortex and post-central gyrus. Women with BN showed increased activation in somatosensory, motor and appetitive regions in comparison to AN and HC. Furthermore, women with BPAN had reduced activation in comparison to women with BN in regions associated with appetitive and motor responses. Thus, although women with BPAN partake in binge eating and purging behaviour in a similar way to women with BN, their heightened control over eating (as shown by an emaciated body at less than 85 percent of normal body weight) may be achieved by a reduced drive to eat over the long term. Thus, these results suggest a clear BN-specific map of brain activation in response to thinking about eating food shown in images, which is primarily characterized by signs of increased appetitive, somatosensory and motor responses, in parallel with activation of prefrontal cortex cognitive inhibition. However, these intriguing suggestions need to be tested further.
Activation of regions associated with motor responses in women with BN, such as the caudate, supplementary motor area and the precentral gyrus suggests an increased appetitive response to food images. This pattern of activation accords with other fMRI studies of people who binge-eat 
. Increased motor responses to food images may reflect an anticipated desire for food consumption, as these regions are also activated in healthy people who have fasted and report a craving for food 
. Furthermore, activation of the caudate and motor regions (precentral gyrus, SMA) are implicated in addictive behaviour 
and thus might suggest that there are some similarities in neural systems between addiction and binge eating. Our data suggest that women with BN have an elevated neural appetitive response to food images, combined with activation of motor areas, when thinking about eating food.
We also observed an increased right dorsolateral prefrontal cortex (DLPFC) response to the food versus non-food images, but this was comparable in both eating disordered groups (there was no significant differential DLPFC activation when women with AN and BN were compared). The DLPFC forms part of the cognitive control network, and is associated with restriction of appetitive responses 
. Activation of a “top-down” cognitive control network, which interacts with “bottom-up” appetitive responses (e.g. the striatum) might be expected when women, who have a pathological desire to be thin and to control their food intake, think about eating food. This is likely to be sporadic in women with BN, who attempt to control their food intake but who are prone to yield to binge eating. In line with this notion, it is plausible that, in women with BPAN, intermittent activation of appetitive systems impinge on attempts at top-down control, but to a lesser extent than those with BN. It is currently unclear however, whether it would be greater top-down control or weaker bottom-up activation that prevents a person with BPAN from gaining weight and/or developing BN symptoms.
The balance between these behaviours is likely dependent on the level of bottom-up impingement on cognitive control, as has been shown in two studies with conflicting results: one shows greater, while the other reduced DLPFC activation (Lock et al., 2011; Marsh et al., 2009, respectively). Both studies examined response inhibition in females with BN, one study using the ‘Go/No-Go’ task 
, the other using the ‘Simon Incompatibility’ 
task, which may activate different levels of arousal (particularly during incorrect responses), and which would account for differential disruption to PFC networks. Additionally, women who have recovered from BN continue to show an aberrant pattern of activation in the striatum in response to reward
and also a reduced pattern of activation in PFC regions to the taste of glucose 
. It could be that as a woman with BN recovers, the arousal system in the brain, albeit aberrant, is tamed and causes less impingement on PFC cognitive networks. Furthermore, there could be different aetiologies for women with AN and BN, in that the former is driven by excessive activation of PFC cognitive inhibition systems, whereas the latter is driven by excessive arousal in reward regions. This could explain some of the differential neural activation observed between women with AN and BN in this study.
The insular cortex functions as an intermediary between the PFC and striatum activation, with dense functional connectivity between these regions, creating an emergent sentience of bodily state 
. Furthermore, dysregulation of the insula seems to be prominent in the pathophysiology underlying neural activation in people with eating disorders 
. We have observed increased insula activation during the food versus non-food contrast in the BN and the HC groups, but not in the AN group. However, in comparison to HC, those with BN have reduced bilateral insula activation in response to food stimuli. These data suggest that there is an intact, perhaps heightened awareness of appetitive responses in women with BN, which is comparatively reduced in women with AN.
Both BN and AN groups demonstrated an increased visual cortex response during the food versus non-food contrast. However, women with BN had a reduced visual cortex response in comparison to HC. Visual cortex activation usually reflects attention towards the stimuli. Studies of attentional bias using the Stroop task show that women with BN have a greater attentional bias to food stimuli than HC 
. However, Stroop studies do not demonstrate whether the attentional bias reflects hypervigilance towards, or avoidance of stimuli. The reduced visual cortex activation to food images (in comparison to HC) observed here suggests that the BN group may have been using strategies to cognitively avoid processing the food stimuli. It is unlikely that BN subjects were closing their eyes because there was differential visual cortex activation in the food versus non-food contrast. However, cognitive avoidance strategies are likely to be a method whereby women with BN can circumvent their appetitive drive in order to achieve self-restraint, which might be reflected in reduced visual cortex activation.
Additionally, the women with BN had significantly higher restraint scores (EDE-Q) than those with AN, and the women with BPAN had the highest restraint score, which is in line with observations that people who binge are often relapsing from attempts at restraint 
But this may also reflect and support a recent finding that the EDE-Q is less effective at identifying bulimic-type eating behaviours
. For example, a high restraint score on the EDE-Q may indicate subjective feelings of restraint, when in reality restraint of food intake is low. Despite the EDE-Q being an excellent tool, perhaps there is a now a need to re-evaluate the validity of some of the constructs tested, particularly in relation to binge eating behaviour.
It would be of benefit in future fMRI studies to compare the neural responses to high versus low calorie food images, as it has not been done in bulimic women. In healthy women it has been shown that images of high calorie foods activate brain regions associated with somatosensory responses and reward more so than compared to low calorie food images 
. This suggests that the images are cognitively evaluated for food palatability, associated with top-down PFC activation, which modulates responses in reward regions in the brain (e.g. in the striatum). Women with BN show dysregulation in a fronto-striatal neural circuit 
and so it would add to existing neural models of bulimia to compare high and low calorie food images in this group.
Additionally, future studies should ensure a totally homogeneous sample of AN women and a much larger sample size, unlike in this preliminary study, where we include 11 restricters and 7 binge-purgers in our total AN group. It could be argued that the binge-purge AN subtype bear similarities with the BN group, by the very nature that both groups partake in binge eating behaviour. However, a counter-argument is that, based on the observation that they were severely underweight, all the women with AN in this study (regardless of subtype) were more successful at restricting their food intake over the long-term than the women with BN. This was also reflected in reduced activation of appetitive and motor brain regions to food images in the women with BPAN compared to BN, and that differences on behavioural measures were comparable between BN v BPAN and between BN v RAN.However, one solution would have been to remove the binge-purgers from our total AN group. However, by including the broad range of AN pathologies we can pinpoint in this cohort the one factor they have in common: restrained eating behaviour that results in self-starvation and a dramatic loss of weight. Thus, we have compared an inhomogeneous group of AN women (who all restrict their food intake) with a group of BN women, to localise differential neural activation associated with long-term restriction of food intake.
There are some limitations to this study: the sample size in the BN group is small (n
8), and so caution must be exercised when generalising to the BN phenotype. However, in an attempt to overcome the small group numbers, we took the relatively uncommon step of using stringent False Discovery Rate (FDR) correction at both the voxel and cluster level to ensure a rate of a maximum of one false positive. In addition, we provide effect sizes for all demographic data. Furthermore, we did not examine differential neural responses to high versus low calorie food images: by doing this we could examine activation to stimuli of different motivation value. Moreover, unlike in this study, future studies should include a homogeneous and larger sample of women with AN, and also men It is entirely plausible that the small percentage of men who suffer from AN and exhibit the same symptoms also show similar patterns of neural activation to food images. Finally, we used the EDE-Q rather than the EDE to gauge eating disorder pathology, which has been shown to be a good measure overall, but weaker for bingeing behaviour 
. However, we chose to use the EDE-Q to reduce the length of the experiment for each participant, as the EDE is a longer interview version that must be completed on a one-to-one with the researcher. A longer experimental duration would have had confounding effects on neural activation (e.g. fatigue), and the EDE-Q was sufficient for our purposes to support eating disorder diagnoses derived initially from the SCID.
The data from this and other studies allow the following tentative conclusions that are in need of further testing. People with BN have increased appetitive and motor responses to food stimuli in parallel with varying (and perhaps sporadic) levels of cognitive inhibition arising from the DLPFC. An imbalanced convergence on the insular cortex between these cortical and subcortical processes alters interoceptive awareness and contributes to binge eating, purging and disrupted self-regulation. Future fMRI studies of BN should seek to confirm these findings and to extend knowledge of the interactions between the DLPFC, insular cortex and caudate, perhaps by using dynamic causal modelling (DCM).