Background
Abnormal eating behaviours are commonplace in modern societies; from psychiatric restrained eating and yo-yo dieting to epidemic overeating. An abundance of hyper stimulating advertising and easily-obtainable high energy palatable food likely encourages dysfunctional cognitive biases that prompt pathological eating behaviours [
1]. Imbalances in energy intake caused by restrained or overeating are known to influence cognitions over many domains [
2], and long-term alterations in energy intake are also associated with brain volume changes [
3]. For example, acute adolescent onset and prolonged lifetime starvation, as in those with anorexia nervosa (AN) is associated with global brain volume reductions [
4,
5], which can be reversed following long-term regular normal eating [
6]. However, gray matter reduction has also been shown to persist in those recovered from, but with a history of AN [
7]. Additional to gray matter changes, reduced white matter and increased cerebrospinal fluid has been observed consistently in many studies of those with AN [
2,
8‐
16]. Regional brain differences in AN are less consistent, although gray matter alterations have been shown in the anterior cingulate cortex, hippocampus and amygdala [
17‐
19]. Against this background, a meta-analysis of previous studies examining brain volume alterations in those with abnormal eating behaviour would strengthen our understanding of the neural underpinnings of AN.
Magnetic Resonance Imaging (MRI) is a technique that can be used to measure brain volume differences, often with a voxel-by-voxel statistically automated technique known as voxel-based morphometry (VBM) [
20]. The VBM approach has advantages over other methods, such as manual tracing, a time-consuming technique susceptible to human error, because VBM uses standardized anatomical measures of voxel-wise estimations and parametric mapping to determine tissue-type probabilities, and is quicker to conduct statistical estimations between groups on a whole brain or region of interest basis than some other software approaches.
In a previously published qualitative review of VBM studies in eating disorders [
21], it was advised that an important task for future studies is to address whether starvation or irregular energy intake results in the loss of brain tissue. In this previous review, the authors focused on a limited number of heterogeneous studies of eating disorders (e.g. anorexia nervosa, recovered anorexia nervosa and bulimia nervosa) and therefore wisely did not attempt to conduct a meta-analysis. However, unlike the previous meta-analysis, in our novel meta-analysis, we focus on a quantitative meta-analysis of global and regional differences between patients with AN and healthy controls. Furthermore, in contrast to the previous review, we do not include studies that measure those who have recovered from AN, because we aim to conduct an analysis of the effects of current low food intake on brain volume. Additionally, in contrast to the previous review, we conduct extensive meta-analyses on global and regional brain volume differences between AN versus healthy controls. Thus, here we conduct for the first time an extensive meta-analyses of global and regional brain volume differences in AN vs. healthy controls.
Discussion
We demonstrate extensive global and regional meta-analyses comparing brain volume differences in patients currently ill with AN versus healthy controls in a total of 228 people with AN and 240 healthy controls. We show statistically significant data on global gray matter and white matter reductions; with concomitant increases in cerebrospinal fluid in those with AN. Broadly, the temporal and occipital lobes showed the most susceptibility to gray matter decrease in those currently ill with AN, and more studies report reduced volume in temporal and parietal cortices and mid-brain regions than others. Regionally, in those with AN, there were reductions in the hypothalamus, left inferior parietal lobe, right lentiform nucleus and right caudate. These regions are linked to appetite and somatosensory perception, functions which are often observed to be aberrant in those with AN. Hemispheric lateralisation did not seem to occur in those with AN, suggesting that starvation effects on the brain do not target one particular side of the brain.
The previous qualitative review [
21] advised that future meta-analyses should include data on standard global analyses, taking into account BMI measures (e.g. comparing restrained vs. binge eaters), advice we heed here and extend by also conducting regional and frequency analyses. As was the case in the previous qualitative review, for our meta-analysis the sample heterogeneity prevented consideration of life-time history of eating disorders, psychiatric and medical comorbidity, handedness, eating disorder measures and medication use. Additionally, the ALE method does not enable covarying for other confounding factors, such as varying ages, sample sizes, different software packages used or other comorbidies, which is therefore a limitation of the ALE we present here. Perhaps future ALEs in the field, when more VBM experiments have been conducted, will be able to apply stricter selection of included studies. Our preliminary and novel meta-analyses concur with the previous qualitative review of VBM studies examining those with eating disorders, which cautiously concluded that their preliminary findings “…hint at gray matter reductions in people with anorexia nervosa, whereas the opposite (gray matter increase) may occur in people with binge eating [behaviour]”.
The authors of the previous review also conclude that “…efforts across the diagnostic eating disorder spectrum are likely to be most informative and such studies would allow for the investigation of the potential role of clinical and participant characteristics” [
21]. This view is also in line with recent reviews discussing how updates to the DSM-IV will likely involve relaxing rigid diagnostic criteria in favour of a spectrum approach to classifying people with eating disorders [
34,
35]. Understanding which brain regions and their functions are more susceptible to over- or under eating might enable specific cognitive tests to gauge severity of eating pathology in order to rectify eating behaviour at an individual level. As part of this meta-analysis we also ran ALE on those with binge eating disorders, but given the heterogeneity of current VBM studies our data could not be included here (data available on request). However, it would be useful for future VBM studies to explicitly examine how varying levels of food consumption influence brain volume across the lifespan.
From the standpoint of under- versus normal/increased food consumption and the effects on brain volume, it is important to mention in the context of our meta-analyses that a recent VBM study comparing currently ill female adolescents with AN and following them through to recovery, showed results in a similar direction to ours [
36]. Specifically, Mainz and colleagues showed GM reductions along the cortical midline that were reversible following recovery, with most significant increases in GM in the cerebellum, thalamus, hippocampus and amygdala, which also correlated with weight and hormone normalisation.
Considering volume differences in the context of functional MRI studies of those with AN may help to elucidate brain circuits most susceptible to acute malnutrition and the development of AN. In a recent review of neurobiological findings in AN [
35] a summary of recent fMRI findings was given. Those with AN have reduced 3activation in
bottom-up regions (e.g. mid-brain), such as the striatum, hippocampus, amygdala, hypothalamus and cerebellum, often in conjunction with increased
top-down activation in prefrontal cortex regions such as the DLPFC, MPFC, ACC and OFC. Bottom-up activations are largely consistent with reward, motivation and general arousal, whereas top-down activations are linked to cognitive inhibition of appetite, self-referential goals and evaluation of salience. Additionally, the review highlights that brain imaging studies of AN also report aberrant activation in the insula, a temporal lobe structure associated with interoceptive awareness and cognitive/emotional perceptions of the body. Thus, our meta-analyses of reduced brain volume in bottom-up regions in those with AN are consistent with functional studies. During our review we found only one study demonstrating increased activation in the DLPFC in those with AN [
33], suggesting that more VBM studies of AN need to closely examine PFC regions. Broadly, these structural and functional data in AN suggest that neural circuitry in the fronto-striatal pathway (linked to impulse control), which also involve connections with the insular cortex, is most susceptible to cognitively-maintained restraint of appetite in AN.
Our meta-analyses have comparable limitations to the recent qualitative review of VBM studies, such as differences in age, different analysis versions (although all studies used MRI), comorbidities with other disorders. Duration of illness, which was not taken into account may have been another factor in the differences observed in brain structure, one recent study of anorexia nervosa found no difference on brain structure between ‘short’ and ‘long’ duration of illness [
4], whereas another recent VBM study found normalisation of brain regions on recovery [
36]. An important strength of our work is that all the studies we included were corrected for total brain volume in the original publications. A further strength was that we conducted extensive meta-analyses from global, regional and frequency (T-scores and number of studies) perspectives. It would be beneficial for future VBM studies that examine the effects of food intake on the central nervous system to study otherwise healthy individuals (e.g. restrained eaters vs. overweight or food cravers), so that psychiatric and medical confounds can be avoided.
One seemingly conflicting finding in our analyses is that the ALE reports a significant reduction in the left inferior parietal lobe in patients with AN (Figure
4), whereas metric data of average difference scores (t-values) and number of reported studies per region (Figure
5. And Figure
6 respectively), does not. However, this merely highlights the significance of the ALE of reduced volume in the parietal lobe, which counts the number of foci reported across all studies, and weighs them up against other reported foci across the whole brain. It could be that there were fewer foci reported in other regions illustrated in the metric data, and also that the difference scores across all studies reporting reduced volume in the parietal lobe between AN patients and controls was consistently high. Given that the finding of reduced left inferior parietal lobe volume survived threshold-corrected whole brain ALE analysis, this should be taken as robust.
Acknowledgements
This work was supported by the Swedish Research Council, Swedish Brain Foundation, Novo Nordisk, A Karlssons foundation, Tore Nilsons foundation, Gunvor och Josef Anérs foundation, Åhlens foundation, LR Åkerhams foundation. The funding sources had no input in the design and conduct, interpretation or preparation of this review, or in the approval of the manuscript.
We would like to thank Santino Gaudio (Department of Neuroscience, Bambino Gesù Children’s Hospital, Rome, Italy), Sonja Yokum (Oregon Research Institute, Eugene, OR, USA), Jason Warren (Dementia Research Centre, Institute of Neurology, London, UK), and Valtteri Kaasinen (Departments of Neurology, University of Turku, Turku, Finland) for providing us with additional information on global brain tissue volumes and/or activation foci coordinates and/or T scores.
Competing interests
All authors declare no competing interests, financial or otherwise. All authors approve submission of the manuscript.
Authors’ contributions
SJB devised the study and supervised OET, SJB, OCH and OET conducted the searches, analysed the data and illustrated the results, SJB, OET, OCH and HBS contributed to the writing of the manuscript. All authors read and approved the final manuscript.