Introduction

Anorexia nervosa (AN) is a serious psychiatric and physical health disorder, characterized by severe food restriction, perception of being overweight despite emaciation, and an intense fear of weight gain (Amercian Psychiatric Association, 2013). AN affects approximately 2.9% of females and has the highest mortality rate among all psychiatric disorders [2]. However, despite a recent increase in studies investigating the neurobiology of AN, its underlying pathophysiology is still poorly understood, as are the contributions of psychological factors, such as cognitions relating to body image, versus physiological factors, such as nutrition.

Several studies have reported widespread decreases in gray and white matter volumes during underweight phases of AN [3, 4], with a meta-analysis revealing global reductions of 5.6% and 3.8% in gray and white matter, respectively [5]. Some have found associations between gray matter volume, AN symptoms and predisposing factors, such as a rigid eating pattern and body image distortion [6, 7], and state-based physiological factors such as body mass index (BMI) [3]. While alterations have been seen to partially reverse with weight gain, there are conflicting data on whether they resolve completely [7,8,9].

Diffusion tensor imaging (DTI) is a method that allows us to move beyond basic volumetrics and to assess white matter integrity within the brain [10]. It allows for the assessment of microstructural properties of white matter based on water diffusion within the brain, using two key measures of interest: fractional anisotropy (FA) and mean diffusivity (MD). FA is a scalar index of the amount of diffusion at every voxel, reflecting the direction and degree of water diffusion. MD is an index of overall diffusivity, irrespective of direction. MD is usually higher in damaged tissue due to increased free diffusion, while FA decreases due to the loss of coherence in the preferred diffusion direction [11]. These measures are aggregates of radial diffusivity (RD) and axial diffusivity (AD), which reflect perpendicular and parallel diffusivity directions within white matter tracts, respectively. Increased RD is interpreted as dysmyelination or demyelination [12], while decreased AD is thought to indicate axonal injury [13].

Several studies have used DTI to examine potential white matter pathology associated with AN. A recent meta-analysis identified decreased FA in the posterior areas of the corpus callosum, the left superior longitudinal fasciculus, and the left precentral gyrus, as well as increased FA in the right cortico-spinal projections, and lingual gyrus in AN relative to controls [14]. A systematic review by our research group identified these regions and also lower FA in the fornix, cingulum and several thalamic regions, and higher MD in the fornix and superior longitudinal fasciculus in AN [15]. These studies also found positive correlations between body mass index during underweight AN and FA, which may indicate an association between white matter integrity and acute malnutrition. The majority of studies have examined FA alone, or only looked at MD within regions with significant FA differences. Although FA and MD are calculated on the same eigenvariates and are typically inverse, they do measure different aspects of white matter microstructure and are most informative when both utilized [16]. Those who did examine MD mostly found increased MD in underweight AN relative to HC [17,18,19,20], although a few studies have found no differences [21,22,23] or reduced MD [24].

To shed light on whether white matter differences observed in underweight AN are due to disorder-related pathology or are secondary to acute undernutrition, three studies conducted longitudinal studies to examine alterations after partial weight restoration. At a whole-brain level, Cha et al. [25] found no differences between HC and the AN group in either state, nor any changes in FA from underweight AN to weight-restored AN. Vogel et al. [24] reported widespread increased FA and decreased MD in underweight AN relative to HC, which was normalized after weight restoration treatment, while von Schwanenflug et al. [26] found reduced FA in the body of the corpus callosum in underweight AN which normalized after weight restoration. Despite the inconsistencies reported during underweight AN, these longitudinal studies suggest that white matter alterations may be state-dependent and reversible with weight restoration treatment.

The primary aim of this study was to further our understanding of trait-based and state-dependent variations in white matter (patho-)physiology in adolescents with AN. Adolescence is a critical period of neurodevelopment and is also the peak time of anorexia nervosa onset [27]. By examining individuals who are closer to their disorder onset, we can expect fewer confounding factors relating to chronicity, and potentially identify pathophysiology that is pertinent to early intervention. We used DTI to examine FA and MD in female adolescents with AN who presented to a hospital setting with acute starvation, relative to age-matched healthy controls (HC). We also examined the relationship between altered white matter microstructural parameters, BMI, and cognitive and behavioral features as measured by the Eating Disorders Examination Questionnaire (EDE-Q) [28]. Further, in a small exploratory longitudinal analysis, we examined change in white matter microstructural properties after short-term weight restoration. We hypothesized that patients would exhibit reduced FA and increased MD in underweight AN relative to HC, and that individual differences in these measures would be associated with both psychological and physiological measures. We also anticipated partial normalization of white matter microstructure upon weight rehabilitation.

Methods

Participants

Twenty-six underweight adolescent girls hospitalized with AN and 19 HC, all aged 14–19 years, underwent a magnetic resonance imaging (MRI) scan. Two AN participants were excluded due to incomplete scan data. Three HC participants were excluded due to an abnormal neurology finding, a history of eating disorder, and to aid in matching groups on their mean age. This left 24 AN and 17 HC participants for analyses. AN participants were recruited at the Children’s Hospital at Westmead (Sydney, Australia) and HC participants were recruited through public advertisements in the same catchment area (Greater Western Sydney). Participant characteristics are presented in Table 1.

Table 1  Demographic characteristics of the participants
Full size table

AN participants were diagnosed according to the diagnostic and statistical manual of mental disorders criteria (DSM-5) [1], and had AN for less than 3 years. They were recruited in the first week of their admission to a specialist inpatient pediatric eating disorder program at the hospital. Underweight AN participants underwent MRI scans after being medically stabilized with normal biochemistry levels [29]. Psychiatric comorbidities, other than psychotic conditions, were not excluded as this would not have represented a typical AN sample [30]. The following comorbid conditions were identified in the AN group: major depressive disorder (n = 12), dysthymia (n = 2), hypomania (n = 1), panic disorder (n = 1), social phobia (n = 7), obsessive compulsive disorder (n = 3), general anxiety disorder (n = 5), alcohol dependence (n = 1).

HCs were of a normal weight range for age, height and gender and did not have a current or past history of an eating disorder or any major psychiatric disorder, as assessed with the mini structured clinical interview for the DSM [31]. All participants were 14–19 years of age, English speaking, had no history of significant brain injury or neurological conditions, and no contraindications for MRI (e.g. claustrophobia, metallic implants).

Informed written consent was obtained from all participants and from parents of those under the age of 16. The study protocol (HREC/13/SCHN/385) received approval by the Human Research Ethics Committees of Children’s Hospital at Westmead, Western Sydney University and Westmead Hospital, and was carried out in accordance with the Declaration of Helsinki.

Questionnaires

AN and HC participants completed the Eating Disorder Examination Questionnaire (EDE-Q) [28] to assess general eating disorder symptomatology. It contains subscales of restraint, eating concern, shape concern and weight concern. It provides subscale scores reflecting the severity of eating disorder behaviors. Subscale scores and a global score are derived from 22 items addressing attitudinal aspects of eating-disorder psychopathology. Australian community normative data for the EDE-Q have been published [32]. The scale has acceptable internal consistency, test–retest reliability and temporal stability.

Longitudinal cohort

Treatment included refeeding, individual psychological support and preparation for the outpatient family-based therapy [33], with family meetings and psychoeducation. Girls with AN were refed initially with 24–72 h of continuous nasogastric feeds which ceased with day-time medical stability (heart rate > 50 bpm, blood pressure > 80/40 mmHg, temperature > 36°C, no cardiac arrhythmias and normal biochemistry). This was followed by a combination of nocturnal feeds and an oral meal plan for a total intake of 2400–3000 kcal/day. The nasogastric feeding duration was determined by markers of medical instability [29]. They were managed on a specialist eating disorder unit with a lenient behavioral program which included hospital-based school attendance, a daily adolescent group program and a second daily physiotherapy program. There were daily medical and psychiatric reviews and three supportive psychotherapy sessions each week [29]. When AN participants improved to at least 85% of their Expected Body Weight (%EBW) [33], they were invited to complete a second MRI scan. Ten AN participants completed the follow-up research scan; however, only data from eight individuals could be included due to two return participants having incomplete baseline scan data.

MRI acquisition

Diffusion-weighted images (DWI) were collected on a 3 T MRI (GE HDxt Twinspeed magnet system) using an eight-channel head coil. DWI acquisition parameters included 55 axial contiguous 2.5 mm slices; 1.5 mm × 1.5 mm resolution; 128 × 128 matrix; repetition time = 1700 ms; echo time = 95 ms; frequency direction = right/left. For each slice, diffusion gradients were applied along 32 directions (b = 1000 s/mm2) and 4 b0 images were collected.

Pre-processing

DWI data were processed and analyzed using FMRIB Software Library (FSL) v6.0 [34]. Non-brain tissue was removed using FSL’s brain extraction tool (BET) and diffusion-weighted images were corrected for subject motion and eddy current-induced distortions using ‘eddy’ as implemented in FSL [35]. Data were corrected for B-matrix rotation following subject motion. Diffusion tensors were fitted to the data by linear regression using a least-square minimization approach (DTIFIT; FMRIB Diffusion Toolbox). FA, MD, AD (λ1) and RD((λ2 + λ3)/2) maps were estimated using the eigenvalues associated with the fitted tensor model.

Tract-based spatial statistics

Tract-based spatial statistics (TBSS) was used to test for between-group differences in white matter microstructural measures. Each participant’s FA images were aligned to a common target in MNI 152 standard space (FMRIB58_FA template; https://fsl.fmrib.ox.ac.uk/fsl/fsl6.0/tbss/FMRIB58_FA.html) using nonlinear registration, and resampled to a 1 × 1 × 1 mm3 voxel size. These normalization parameters were subsequently applied to the other diffusivity parameters maps (MD, RD, AD). FA images were averaged to create a mean FA map and a mean FA skeleton, which represents the center of all tracts common to the group. The mean skeleton map was thresholded to FA > 0.2 to exclude any peripheral areas with high variability and/or partial volume effects. FA maps of each individual were then projected onto the mean FA skeleton. This same procedure was also applied to the MD, RD and AD maps.

Statistical analyses

Group comparisons were carried out using non-parametric permutation-based statistics using “randomize” in FSL [36], with 5,000 permutations run for each contrast. Threshold-free cluster enhancement (TFCE) [37] was used to obtain clusters and the threshold for significance was set at family-wise error-corrected p value (pFWE) of < 0.05.

Group differences in FA and MD were tested between underweight AN and HC using two-sample t tests, with age as a covariate of no interest. Paired t tests were used to compare weight-restored AN with underweight AN, with time between scans as a covariate of no interest. Finally, two-sample t tests were used to compared weight-restored AN with HC.

For any analyses with a significant group difference in FA or MD, values from significant voxels within the region were averaged as a summary measure of the diffusion properties in each participant. To better characterize microstructural changes underlying the between‐group differences in FA and MD, AD and RD were also examined within significant clusters. Post hoc two‐sample t tests compared the groups on each component measure within these regions‐of‐interest. The John Hopkins University (JHU) White Matter Labels Atlas were used for anatomical localization of results.

Correlational analyses

Correlation analyses were performed separately for AN and HC groups using randomize in FSL to examine associations between FA and MD, and BMI and EDE-Q subscale scores, while covarying for age.

Post hoc comparisons

To explore effects of common comorbidities, two-sample t tests we performed between individuals with AN with comorbid depression versus those without comorbid depression for MD and FA.

Results

Underweight AN v HC

Group comparisons revealed a significant increase in MD in underweight AN relative to HC in a single white matter cluster of 8797 voxels, peak MNI coordinates: − 16, 12, 36 (L superior corona radiata). It was comprised of 18 white matter regions identified with the John Hopkins university atlas (Table 2). These regions included the corpus callosum, association fibers, namely the left sagittal stratum and external capsule, projection fibers such as the bilateral corona radiata (anterior, superior and posterior portions), left internal capsule (anterior, posterior and retrolenticular) and limbic fibers such as the bilateral cingulum and left fornix (Fig. 1). Post hoc examination of the component measures revealed that within this significant MD cluster, there was significantly increased RD and AD in underweight AN relative to HC (Fig. 2).

Table 2  Summary of average mean diffusivity values for significant tracts for underweight AN versus HC comparisons
Full size table
Fig. 1 
figure 1

Global fractional anisotropy and mean diffusivity values for underweight anorexia nervosa (undAN) and healthy control (HC) participants. Bars represent group mean

Full size image
Fig. 2 
figure 2

TBSS results comparing MD, AD and RD in underweight adolescents with anorexia nervosa (AN) versus age-matched healthy controls (HC). Significant tracts (p < .05) with underweight AN > HC are shown in orange. MNI coordinates, x = − 1, y = − 25. Underlying white matter skeleton shown in green. MD mean diffusivity, AD axial diffusivity, RD radial diffusivity

Full size image

There were no significant differences between groups in FA. Mean FA and MD were extracted from the whole-brain white matter skeleton for each individual and are presented in Fig. 1 to provide an overview of individual differences.

The AN sample consisted of 23 with restrictive subtype (ANr) and 1 with purging subtype (ANbp). Results did not change when the individual with ANbp was excluded from analyses.

Correlational analyses with symptom measures

There were no significant correlations between any MD or FA cluster and BMI or EDE-Q subscale score, for either the AN or HC group after controlling for age (uncorrected for multiple comparisons).

Longitudinal analyses (underweight AN vs weight-restored AN)

Eight clinical participants completed a follow up assessment after they had completed short-term weight restoration. The second MRI scan was conducted on average 71 days (range 13–294 days; ± 92 SD; 41 median) after the initial MRI scan.

After weight restoration, this subsample had a significant increase in BMI (mean change 3.14 ± 1.04, p = 0.003) and decrease in EDE-Global scores (mean change 1.36 ± 0.91, p = 0.004) relative to at their scan whilst underweight. On average, patients gained 1.63 kg per week of treatment (range 0.25–4.2 kg per week).

There was a significant reduction in MD after weight restoration in a single white matter cluster of 104,171 voxels. These widespread changes occurred in all 46 JHU-atlas regions of interest (see Table 3 for mean values and percent change). All results remained significant when covarying for differing lengths of treatment period between individuals and rate of weight gain. There were, however, no significant changes in FA after weight restoration. Figure 3 shows the average whole-brain MD and FA for each individual pre- and post-treatment (underweight AN vs weight-restored AN), relative to the treatment duration. Change in BMI and EDE scores after weight restoration and rate of weight gain did not correlate with the magnitude of change in MD or FA either globally or in any of the regions of interest.

Table 3  Summary of average mean diffusivity values for significant tracts for the underweight AN versus weight-restored AN
Full size table
Fig. 3 
figure 3

Global mean diffusivity values for each individual at baseline (underweight AN; undAN) and after weight restoration (w-resAN), relative to the treatment period (i.e. number of days between scans)

Full size image

Weight-restored AN v HC

There was significantly lower MD in the weight-restored AN group relative to HC in a single white matter cluster of 102,447 voxels. See supplementary Table 1 for a full list of regions with significant difference in MD. There were no significance differences in FA between the weight-restored AN group and HC.

Post hoc exploration of the effects of comorbid depression

There were no significant differences between individuals with AN and comorbid depression (n = 12) versus those without comorbid depression (n = 12) in any WM microstructural measure.

Discussion

Widespread increases in MD of up to 9% were found in underweight AN relative to HC, which was associated with both increased AD and RD, suggestive of dys- or de-myelination. However, there were no significant differences between the underweight AN and HC groups in FA, and no significant correlations between DTI measures and BMI or EDE-Q subscales. In a small, preliminary analysis, weight restoration therapy reduced MD, to levels significantly lower than age-matched HC, but did not consistently alter FA across individuals. These data show that white matter microstructure is significantly altered in female adolescents with AN, with preliminary longitudinal data suggesting that it may be reversible with short-term weight restoration.

Pattern of microstructural alterations

Similar to previous studies, we found increases in MD, AD and RD in underweight AN relative to HC [17, 19, 20]. While there were no significant group differences in FA between underweight AN and HC, Fig. 1 highlights a trend for lower FA in underweight AN relative to HC. This pattern of alterations in DTI measures is thought to reflect higher extracellular volume fraction and lower membrane density due to chronic myelin and axon loss, potentially after widespread inflammation [38]. Demyelination of white matter fibers disrupts the restricted diffusion in the perpendicular direction, causing an increase in RD, and cellular debris is subsequently cleared by microglia, causing an increase in AD [39]. In a neurodevelopmental context, however, it is possible that this pattern of findings reflects a critical disruption to the rapid myelination process that occurs throughout adolescence via a lack of glial cell proliferation and subsequent insufficient production of myelin [40].

Regional findings

Consistent with the recent meta-analysis by Barona et al. [14], the most significant difference in underweight AN relative to HC was found in the corpus callosum. The corpus callosum is critical for the interhemispheric integration of cognitive, emotional and perceptual functions, all of which are implicated in AN symptomology [41, 42]. Other key alterations were found within the internal and external capsule and corona radiata, which are critical for information flow between cortical and subcortical regions responsible for reward processing, cognitive flexibility and goal-directed behavior [43]. The uncinate fasciculus also contains reciprocal connections between the ventrolateral prefrontal cortex and amygdala, regions that are also integral to decision-making and reinforcement learning [43, 44]. Impairment of these processes could contribute to the impaired response inhibition and behavioral avoidance often seen in AN [4].

The bilateral cingulum bundles were also found to have significantly reduced white matter integrity in underweight AN. These diffuse tracts interconnect core components of the default mode network (DMN), which is activated during introspection and suppressed during cognitive functions [45]. Cingulum integrity is positively associated with individual differences in cognitive abilities and functional activation of the DMN [46]. While there are mixed findings regarding the role of the DMN in the neuropathology of AN [47, 48], it is possible that DMN dysregulation plays a role in self- and body-focused ruminations.

Multiple association fibers, and in particular, the left superior longitudinal fasciculus, exhibited lower white matter integrity in underweight AN relative to HC in the current study. This long-range fiber connects frontal to parietotemporal and occipital areas and allows for the integration of information from non-adjacent regions [44]. Previous studies have implicated poor connectivity between these regions as being a contributing factor in distorted perception of body size, a putative endophenotype of AN [47]. Impairment within long-range tracts more generally may also contribute to deficits in global processing while local processing remains intact. This may be expressed as weak central coherence, which some have proposed as an endophenotype of AN [49].

Collectively, altered white matter diffusivity across such widespread regions likely contributes to the altered functional connectivity reported in the cognitive control network and visual and homeostatic integration [50]. Despite this, we were unable to find any significant correlations between measures of white matter microstructural integrity and symptom severity (as measured by the EDE-Q), or BMI.

Longitudinal findings

In our exploratory longitudinal sample, short-term weight restoration significantly reduced MD in almost all of the white matter skeleton. Figure 3 demonstrates that this was a consistent direction of findings across all eight individuals, the degree of which was not related to the duration of treatment (range 13–289 days). Alternatively, changes in FA after weight restoration were inconsistent, with five individuals showing the expected increase in FA, while three showed no change or decreases. Previous longitudinal studies of short-term weight restoration have reported no change in FA [25], reduced FA [24] and increased FA [26]; therefore, this inconsistency is not unprecedented and highlights that there are likely multiple factors contributing to individual changes in microstructural properties after short-term weight restoration.

The magnitude of change in MD observed was striking, with some regions seeing reductions of up to 120%, compared to the initial group difference from HC of 3–9%. It is possible that this ‘overshoot’ in MD reduction may relate to the relatively rapid realimentation that occurred with this specific treatment protocol (e.g. 1.63 kg/week versus 0.4 kg/week [25]). Adolescence is a critical period for neurodevelopment, with increasing myelination allowing for better neural integration and the subsequent improvement in higher level cognitive skills [51]. It is possible that reintroduction of dietary substrates that are precursors to myelin production may provide short-term compensatory neuroplasticity.

It is important to note that although differences in age were accounted for statistically in all analyses, this is not reflected in the magnitude of change percentages. The HC and AN post-treatment groups were marginally older than the AN pre-treatment group (~ 10 months); therefore, the reported magnitude of change for MD may be slightly inflated due to normal developmental changes.

Our post hoc analyses did not find any significant WM microstructural differences between those with and without comorbid depression. It is likely, however, that the analysis was inadequately powered to detect small to moderate differences. Replication of this analysis in a larger sample is necessary to test this meaningfully and to allow for adequate statistical power to investigate the factors modulating microstructural plasticity.

On the basis of our preliminary results and evidence from previous studies, we tentatively suggest that weight-recovery treatment of young people with early onset of AN (less than 3 years of illness) and the neuroplasticity in adolescence could be key factors relevant to brain recovery. This is important for providing hope to young people with AN and their families that the effects of acute starvation on the adolescent brain may be reversible. While we did not find linear correlations between change in EDE-Q measures and microstructural properties, it is still possible that changes in white matter microstructure contribute to changes in symptoms, even if they are not the key driver. Improved neural transmission and communication through core tracts such as the corpus callosum likely facilitate improved general cognition, which may enhance the efficacy of cognitive therapies. This complex relationship requires further exploration.

Limitations

Our findings should be considered in light of the following limitations. First, our sample size was relatively modest and replication is required. In particular, our longitudinal sample was small due to many participants being unwilling to continue engaging in research after hospitalization, and it was limited to short-term weight restoration. We did not collect comparative longitudinal data in HC to account for normative aging data, however accounted for time between scans statistically. Finally, although including psychiatric comorbidities represents a more typical AN sample [30], the range of included comorbidities has the potential to influence the results. Despite these limitations, our results were consistent with some previous literature and provided additional information to build upon in future research.

Conclusion

In conclusion, widespread alterations in white matter diffusivity are found in underweight female adolescents with AN, which preliminary longitudinal data suggests may be predominantly due to the effects of malnutrition. Our exploratory data found significant reversal, and in fact, overcompensation of white matter deficits was observed after short-term weight restoration. Replication in a larger sample is required before any firm conclusions can be made and future research on longer term outcomes in longitudinal samples is required to provide critical information regarding the effects of acute starvation on neurodevelopmental trajectories.

What is already known on this subject?

Inconsistent findings of altered FA and MD are reported in adolescents with AN. Most studies examine FA alone or MD in limited regions, thus providing an incomplete view of white matter microstructure.

What your study adds?

There is widespread dys- or de-myelination in underweight AN, identified via increased MD, AD and RD. Realimentation in a small subsample reduced MD but did not consistently alter FA.