This fMRI study uniquely demonstrated how thinking about eating food presented in images activates differential neural responses between healthy women and those with a current diagnosis of anorexia nervosa (AN). Additionally, we conducted a preliminary analysis of the differential neural responses between the subtypes of AN who have varying levels of appetite restraint. This study progresses our recent fMRI findings that compared women with AN with those who had bulimia nervosa (BN) 
, by focusing only on the differential effects of thinking about eating food between healthy women and those with AN (and the subtypes). Our main finding was that in the total AN group, thinking about eating food corresponded to a reduced activation in the cerebellar vermis and increased activation in the visual cortex. FMRI studies of those with AN showing food stimuli also report reduced activation in the cerebellum 
and this also accords with reports that the cerebellum, particularly the vermis, has a prominent role in feeding behaviour 
. It is likely that asking females with AN to think about eating food does not activate appetitive regions of the brain in the same way as healthy females, and it is likely that there will be a reduced appetitive response in the brain, given that one of the core symptoms of this eating disorder is restraint of appetite.
Our strongest result in terms of cluster size, was observed in preliminary analyses comparing the subtypes of AN separately with healthy controls, yielding a reduced activation in the right insular cortex in AN. The insular cortex, particularly the anterior insular cortex (AIC) is associated with subjective feelings of the body, or interoceptive awareness
and generally with all types of emotional feelings 
, and a recent fMRI study has also shown reduced AIC activation in response to rewarding stimuli in people with AN 
. Activation of the insular cortex in response to food images in healthy controls might reflect an interoceptive awareness of appetitive responses. However, in women with AN, appetitive responses to food images are likely to be in conflict with the desire to be thin, and to activate anxiogenic concerns about shape, weight and eating. It has been suggested that disruption to insular cortex activation in those with AN might reflect a rate-limiting defect, that is, a failure to efficiently process information from appetitive brain regions and to effectively orchestrate with higher-order cognitions 
. A rate-limiting defect in the insular cortex likely underlies a proneness for anxiety 
, which often precedes and is highly comorbid with eating disorders 
. Thus, in females with AN, activation of brain regions involved in the Cognitive Control Network (CCN), such as the DLPFC and the ACC, may be associated with attempts to cognitively control and/or disrupt somatic responses 
that is the basis of an anxiogenic interoceptive awareness of appetitive stimulation, via interaction with the insular cortex 
It is clear from the literature that using body image stimuli, as opposed to food images during fMRI activates different patterns of neural activation in those with AN. Recent studies showing images of bodies (e.g. own, others, drawn figures) report reduced activation in the superior temporal lobe, visual cortex, ACC and frontal regions, as well as increased insula and premotor cortex activation 
. Therefore, it is likely that body image distortions arise from a different neural pattern to the neural mechanisms underlying restriction of appetite, given that we and others using food stimuli observe increased activation in frontal, ACC and visual regions, but reduced activation in the striatum, cerebellum, insular cortex e.g. 
, although we failed to observe increased amygdala and ACC activation as in some studies e.g. 
. Furthermore, some fMRI studies demonstrate that hunger and satiety can alter the response to food stimuli in those with AN, for example, reduced visual cortex activation when hungry but greater visual and lateral prefrontal activation when satiated 
, which compares to our results and may demonstrate that females with AN were satiated after eating lunch.
Greater visual cortex activation in the women with AN may be associated with cognitive biases 
linked with thoughts about consuming the food shown in the images and strategies to cognitively restrain appetite. There is some evidence that the visual cortex is modulated by the dorsolateral prefrontal cortex (DLPFC), via increased attention resources, and the combined activation of these regions are strongly linked to object recognition 
. Given that we found, in comparison to healthy controls, increased DLPFC and visual cortex activation in just over half the group of AN participants who were sub-classified as restricting
AN (and not when combining with, or separately analysing the binge-purging
AN group), it seems a plausible preliminary conclusion to suggest that excessive appetite suppression and cognitive biases for food stimuli in those with RAN are linked to DLPFC function. It could be that combining the subtypes of AN during fMRI analysis prevents DLPFC activation from being observed, given their varying levels of control over appetite. This is a particularly attractive explanation in line with a recent review of fMRI studies in AN that reported a strong association between RAN and DLPFC function 
. Furthermore, mounting evidence, in both healthy and disordered eating behaviour, implicates the DLPFC in the cognitive control of appetite 
, particularly in females 
, and artificial stimulation of the DLPFC reduces the experience of craving for food in those who are prone to binge eat 
. Finally, there is evidence for lateralization of DLPFC-driven cognitive control: the left DLPFC activating to anticipation of conflict, the right to immediate, impulsive conflict 
, and here we found increased right DLPFC activation in RAN, suggesting that thinking about eating while looking at pictures of food causes immediate conflict with desires to remain thin.
Preliminary findings when comparing restricting
with binge purging
AN also revealed greater activation of the anterior cingulate cortex (ACC) and the parahippocampal gyrus. The ACC lies caudal-medial to the DLPFC, and is associated with orchestrating top-down cognitive and bottom-up arousal (e.g. appetitive) responses, in line with current predictions about desired and likely future rewards 
, in order to make adjustments for effective cognitive control via interaction with the DLPFC 
. It is plausible that there is more conflict between cognitive strategies and appetitive responses in those with RAN, given that they are perhaps ruminating on restraint cognitions to a greater degree than those who temporarily relinquish their cognitive control during sporadic binge eating, and that appetitive responses interfere with these cognitions. However, it could also be that predictions are computed via ACC activation in those with BPAN, reflecting a greater impingement of appetitive neural circuitry on cognitive restraint, as seen in some fMRI studies of those who are prone to binge eating 
. Observing a greater parahippocampal response in RAN compared to BPAN might reflect that the former group are engaging in more ruminations about previous experiences of eating food, e.g. 
. Conversely, it might be that a general lack of appetitive behaviour in those with RAN leads to a downregulation of parahippocampal gray matter volume, e.g 
, causing this region to be hyper-stimulated when explicitly thinking about eating food. However, in our current small preliminary samples of women with RAN and BPAN, it is difficult to ascertain the true nature of the relationship between appetitive processes and cognitive restraint, and so a comparison with larger subgroup groups of AN is needed.
According to a recent clinical review of neuroimaging studies in AN, summarising findings from fMRI and other methods (e.g. using Positron Emission Tomography, Single Photon Emission Tomography) 
, three main regions in those with restricting
AN show the most consistent dysfunctional activation: the frontal cortex (particularly the DLPFC) 
, hypo-responsiveness in the left inferior parietal lobule 
and overall dysfunction in the bilateral anterior cingulate cortex 
. In those with binge purging
AN there appears to be a similar pattern of neural dysfunction, in the frontal lobe 
, parietal cortex 
and bilateral anterior cingulate dysfunction 
. However, when examining neural dysfunction in AN as a whole group (without separately analysing the subtypes) a different neural profile emerges. Instead, it appears that core neural pathology in those with AN in general, includes reduced function/metabolism in the temporal lobe 
seemingly improving following recovery 
. Moreover, in response to food stimuli, a left-dominant reduction in the temporal and parietal lobes is prevalent in people with AN 
. Increased activity in the fusiform gyrus and amygdala are also commonly observed 
. Furthermore, dorsal striatum hypo-activation in AN regardless of subtype has also been reported by functional neuroimaging studies 
. Thus, in conjunction with the main findings of previous neuroimaging studies, we find that thinking about eating food is associated with neural dysfunction in the visual cortex, DLPFC and ACC, and that a different neural signature is observed when separating the subtypes of AN, prompting the need for further research in to the subtypes.
In the present study, we used a symptom-provocation fMRI paradigm that has been used in investigations of Obsessive Compulsive Disorder (OCD) 
, an anxiety disorder that is highly comorbid with AN 
. In the previous study of those with OCD, participants were cognitively engaged when they were shown anxiogenic pictures related to their disorder. They reported being anxious and showed an increased medial and dorsal PFC activation. In comparison, during our scanning period, we found that cognitively engaging participants while presenting food images (by asking them to think about eating the food) induced significantly higher levels of anxiety in women with AN compared to healthy controls. In addition, the RAN group showed a prefrontal response to the food images. Using instructions to cognitively engage the participants is a methodological variation from other fMRI studies that merely present appetitive stimuli (images of food, drink, and taste in the mouth) for passive engagement (e.g. not explicitly instructing to cognitively engage) to women with AN 
. It is of note that these previous studies did not observe an increased DLPFC response to the appetitive stimuli. Thus, activation of the DLPFC may reflect a specific recruitment of anxiety-related cognitions in women with AN in relation to food (e.g. concerns about shape, weight and eating, striving for thinness), that may drive an inhibition of the insular cortex and other brain regions linked to appetitive responses (e.g. the cerebellum) that arise when thinking about eating food shown in images.
There are some caveats to the explanations above, and some limitations in this study. We did not find amygdala or OFC activation to food images in women with AN as in other studies 
. However, the size of the amygdala, combined with the reduced power of the study due to small sample sizes, and the heterogeneity of the AN group as a whole may account for this. However, again these differences could be due to a greater cognitive component during this paradigm in comparison to other studies. It is a fact that our subgroup numbers are small. However, we emphasise the preliminary nature of this unique fMRI study utilising cognitive engagement in the images in line with our previously published fMRI data 
. In an attempt to counteract some of the limitations of small group numbers we used stringent voxel- and cluster-wise False Discovery Rate (FDR) threshold correction. Additionally, some women with AN were taking SSRI medication, but independent t-tests showed no significant differences in neural activation between those who were and were not taking SSRI. Also, it must be considered that since the females with AN were currently ill, the neural activation we observe could be due to malnutrition effects, and future fMRI studies should add covariates for gray matter volume 
. Furthermore, we only used high calorie food images, and did not compare neural and cognitive responses to low calorie images, nor were food preferences assessed, which could have caused differential activation. Also, we did not collect data to ensure that all healthy participants had, as instructed, eaten lunch, although all females with AN were collected from the hospital after eating lunch; nor did we collect data on IQ or duration of impatient care (although we did collect duration of illness data). Lastly, we did not explicitly test cognitive engagement with the images: however, increased visual cortex activation (in the food versus non-food individual group contrasts) suggests visual engagement in the images during the experiment.
The data from this novel preliminary study progresses the field and can provide the following tentative conclusions that need further testing. Women with AN have reduced appetitive and somatosensory neural responses to food images when explicitly thinking about eating food shown in images, but the subtypes are differentiated by increased DLPFC-ACC and reduced cerebellar vermis activation. These regions are associated with a cognitive control network (CCN), specifically controlling basic appetitive drives with cognitive restraint mechanisms. Conflict between appetitive responses to food and anxiety cognitions about shape, weight, eating and desire for thinness are likely to activate neural mechanisms associated with the DLPFC. An imbalanced convergence on the insular cortex by the CCN is likely to lead to a rate-limiting defect and disruption to appetitive neural responses and difficulties in interoceptive awareness in females with AN. Activation of the CCN is perhaps more prominent in women with RAN.