Memory impairment is associated with the loss of regular oestrous cycle and plasma oestradiol levels in an activity-based anorexia animal model.

OBJECTIVES
Patients with anorexia nervosa (AN) suffer from neuropsychological deficits including memory impairments. Memory partially depends on 17β-oestradiol (E2), which is reduced in patients with AN. We assessed whether memory functions correlate with E2 plasma levels in the activity-based anorexia (ABA) rat model.


METHODS
Nine 4-week-old female Wistar rats were sacrificed directly after weight loss of 20-25% (acute starvation), whereas 17 animals had additional 2-week weight-holding (chronic starvation). E2 serum levels and novel object recognition tasks were tested before and after starvation and compared with 21 normally fed controls.


RESULTS
Starvation disrupted menstrual cycle and impaired memory function, which became statistically significant in the chronic state (oestrous cycle (P < 0.001), E2 levels (P = 0.011) and object recognition memory (P = 0.042) compared to controls). E2 reduction also correlated with the loss of memory in the chronic condition (r = 0.633, P = 0.020).


CONCLUSIONS
Our results demonstrate that starvation reduces the E2 levels which are associated with memory deficits in ABA rats. These effects might explain reduced memory capacity in patients with AN as a consequence of E2 deficiency and the potentially limited effectiveness of psychotherapeutic interventions in the starved state. Future studies should examine whether E2 substitution could prevent cognitive deficits and aid in earlier readiness for therapy.


Introduction
Anorexia nervosa (AN) is a severe psychiatric disorder characterised by a restriction of energy intake, weight loss, fear of gaining weight and body image disturbance affecting primarily adolescent girls and young women (Herpertz-Dahlmann 2015). The majority of patients with AN suffer from endocrine deficits and a reduced brain volume that is potentially associated with reduced cognitive function (Castro-Fornieles et al. 2009). Amenorrhoea is nearly ubiquitous in patients with AN (Dempfle et al. 2013), although it has been eliminated as a prerequisite diagnostic criterion in the DSM-5 in order to include prepubertal and male patients as well as patients on contraceptives (American Psychiatric Association 2013). The loss of menstrual cycle or primary amenorrhoea involves a reduction of the steroidal sexual hormone 17b-oestradiol (E2) in patients with AN compared to normally developing adolescents (Ramoz et al. 2013;Singhal et al. 2014). The (re-)appearance of menses is an important criterion for the recovery of patients with anorexia (Mitan 2004;Dempfle et al. 2013).
In addition to acting as a reproductive hormone, E2 has diverse trophic and regulatory functions in the brain, such as influencing the development of distinct non-reproductive networks including amygdala and hippocampus growth, the modulation of central nervous system activity in adolescence, and a neuroprotective role under neuropathological conditions (Beyer 1999;Kipp & Beyer 2009;Neufang et al. 2009). E2 has been associated with improved cognitive performance in recognition and spatial memory tasks linked to hippocampus and prefrontal cortex, and has been shown to enhance spine density in the hippocampus (Luine & Frankfurt 2012;Sundstr€ om Poromaa & Gingnell 2014). The effects of AN on brain structure and function are diverse, and include reduction of grey and white matter volume (Seitz et al. 2014) and involve cognitive impairments in non-verbal intelligence functions, verbal memory and motor speed (Lawrence et al. 2003;Chui et al. 2008;Kjaersdam Tell eus et al. 2015). Moreover, first studies indicate an interaction between E2 and memory function in AN because patients with AN with amenorrhoea or irregular menses exhibit significant cognitive impairments in verbal ability and cognitive efficiency (Chui et al. 2008), and E2 levels correlate positively with verbal learning after weight recovery (Buehren et al. 2011).
Currently, it is unclear how acute and chronic E2 deficiency differs with regard to cognitive function and how these deficiencies differentially affect patients with AN. An analysis of discount habits involving delayed gratification tasks indicated increased self-control in adults but not in adolescent patients with AN as differing neuropsychological characteristics (Ritschel et al. 2015). Similarly, visuospatial impairments seem to increase with age in patients with AN (Stedal et al. 2012), which could be an indicator of chronic AN causing more deficits, as adolescent patients are more acutely ill than adult patients, who tend to be ill for a longer time. Although maturational effects or selection could also play a role, it seems important to differentiate between the cognitive effects of E2 depletion after different lengths of food restriction.
Intervention studies have successfully attempted to remediate E2 deficiency in patients with AN and have countered another consequence of amenorrhoea, i.e., osteoporosis (Misra et al. 2011). Thus, if E2 deficiency is indeed linked to learning and memory impairment in AN, this finding could likely lead the way to future E2 replacement studies and would potentially improve cognitive deficits as well. We decided to investigate this in an acute and chronic starvation animal model using activity-based anorexia (ABA), which is the most widely utilised rodent model of AN simulating weight loss by food restriction and hyperactivity by access to a running wheel (Routtenberg & Kuznesof 1967;Epling et al. 1983;Adan et al. 2003;Gutierrez 2013). Structural and neurobiological changes in the brain of rodents with ABA include cerebral glucose hypometabolism, changes in dendritic branching and more GABA receptors in the hippocampus; the latter change is linked to more fearful behaviour after remittance (Barbarich-Marsteller et al. 2005;Chowdhury et al. 2014;Aoki et al. 2014). However, to the best of our knowledge, neither the expected E2 deficit nor the potential cognitive function impairments have been shown in the ABA rodent model. Thus, we aimed to test memory functions by applying the novel object recognition (NOR) task introduced by Ennaceur & Delacour (1988). It has been extensively validated as a working memory task measuring recognition memory relying on the prefrontal cortex function as well as the hippocampus and parahippocampal functions. These regions are affected by volumetric reduction in AN concomitantly (Antunes & Biala 2012;Luine & Frankfurt 2012;Seitz et al. 2014). The NOR task has already been used for testing negative effects of E2 deficiency and its reversal using E2 therapy on learning in female rodents (Jacome et al. 2010;Luine 2015). These observations suggest that the NOR is an appropriate task for testing the potential interaction between E2 deficiency and memory function in ABA.
The aim of the present study was to investigate whether E2 levels are reduced in female adolescent ABA rats, and whether learning ability is decreased. Furthermore, we correlated E2 deficits with impaired memory functions and the length of starvation. We analysed bodyweight, running wheel activity (RWA), cyclicity, E2 levels and object recognition memory in acutely and chronically food-restricted ABA rats to differentiate between newly occurring AN and chronic illness.

Subjects
Forty-seven adolescent female Wistar rats (Charles River, Sulzfeld, Germany) served as subjects. The 4week-old rats arrived in the laboratory, with an average weight of 88.32 ± 14.35 g, and were individually housed in Type IV, 1820-cm 2 cages (Polysulfone, Tecniplast GmbH) under 12/12-h light/dark cycle (lights on at 07:00 h) with ad libitum access to water. All of the rats had 24 h/day running wheel access and showed a regular 4-day oestrous cycle at the end of the 10-day acclimatisation period.

Study design
The current study was part of a larger ongoing study design (LANUV registration number 84-02.04.2013. A221). All of the animal experiments were approved by the Animal Experimentation Committee of the Ministry for Nature, Environment and Consumer Protection of Northrhine-Westfalia and were conducted in agreement with the German Law and European regulations (Guideline 86/609/EEC).
Adolescent rats were selected to imitate the disorder in humans as exactly as possible as most patients with AN have an illness onset during adolescence (Herpertz-Dahlmann 2015). The original ABA paradigm of self-starvation (Routtenberg & Kuznesof 1967;Gutierrez 2013) was modified in so far that we included a standardised reduced amount of food instead of reduced feeding times. This allowed us to also test a chronic starvation condition to enhance the potential translational value of these data, as many patients with AN are chronically ill. Classical selfstarvation ABA cannot be used here as it leads to high animal mortality when applied chronically (Routtenberg & Kuznesof 1967;Exner et al. 2000). Standardised reduced feeding furthermore helped to maintain comparable body weight conditions in different animals for testing recognition memory and hormonal status. A schematic summary of the testing protocols pertaining to this study is shown in Figure 1. The animals were randomly assigned to ABA and control groups. Two different ABA groups varied according to the length of starvation as follows: ABA_ acute finalised directly after acute weight loss (n ¼ 9); and ABA_ chronic with a continued 2-week weight-holding phase after weight loss (n ¼ 17). Two groups of separate control animals adjusted for the different study length in acute and chronic were kept under the same housing conditions but were fed ad libitum (Control_ chronic, n ¼ 12; and Control_ acute , n ¼ 9).
The rodents had 10 days to acclimatise to their running wheel cages with food ad libitum. The ABA group received 40% of their baseline daily food intake until 25% weight reduction was achieved, except for six animals in the chronic starvation group, where the target weight reduction was 20%. In the chronic group, body weight was maintained by adjusting daily food intake for 2 weeks according to the needs of the individual animals to maintain a stable weight. Body weight, food consumption and RWA were measured daily at 12:00 h, menstrual cycle was measured daily beginning the latest four days prior to starvation. Behavioural tests were performed at two points in time, at the end of acclimatisation and directly before completion. Blood samples were collected at the end of acclimatisation and after completion. Both, behavioural tests and blood sampling took place between 09:00 and 12:00 h directly before feeding to measure the strongest effects of starvation. The cycle phases were compared between ABA and control animals for days of behavioural tests and blood sampling to ensure a similar distribution of cycle phases in both groups.

RWA
RWA were measured once daily using tachometers (BC 5.12, Sigma Germany). RWA during the 10-day habituation phase was averaged. The relative RWA increases during acute starvation and the weightholding phase were calculated by normalising the daily average RWA with the RWA during the habituation phase. Figure 1. Experimental protocol. Schematic experimental protocol for acute and chronic starvation activity-based anorexia (ABA) and control animals and respective controls. Daily measurements of body weight, running wheel activity and oestrous cycle were obtained. NOR, novel object recognition task; E2, blood withdrawal to determine the oestradiol level.

Determinations of oestrous cycle and E2 plasma levels
Vaginal smears were stained with Giemsa solution, and the cycle was determined by the characteristic cornification of vaginal epithelium in metoestrous, diestrous, proestrous and oestrous phases (Dos Santos et al. 2011). Vaginal opening was ensured before the beginning of the ABA model. The percentage of animals with oestrous phases was calculated in the last 4 days before completion as this is the average duration of one whole cycle in rats. For E2 determinations, individual blood samples were collected by retro-orbital blood-draw under anaesthesia with 5% isoflurane (Forene, 100%, v/v, B506, Abbott, TX, USA) before starvation and from the right atrium of the heart after completion. The blood samples were centrifuged at 12,000 rpm for 15 min at 4 C and the serum specimen was extracted. Serum samples were frozen at -40 C until E2 assays were performed. E2 concentrations were determined using the Rat E2 ELISA (kit # ES180S-100 -Calbiotech Inc., Austin, TX, USA) according to manufacturer's instructions.

Determination of leptin levels
To further validate the model, a leptin serum assay was performed after starvation using the Rat Leptin, LEP ELISA KIT (# CSB-E07433r -Cusabio Biotech Co., Wuhan, Hubei Province, China) according to manufacturer's instructions.

NOR task
The NOR task analyses the rat's behaviour when it is exposed to a familiar object and a novel object. Two similar objects are presented and, after a delay, one sample object is exchanged. Because a novel stimulus provokes approach behaviour, the animal spends more time with the novel object if it remembers the old sample. Thus, we analysed the recognition memory of rats by comparing the exploration rates of the new object divided by the old sample.
The open field arena (70 Â 70 Â 40 cm) consisted of wood coated with white plastic. For subsequent analyses, a digital video camera was placed above the field to monitor every trial. The tests were performed in a dimmed room with white noise of approximately 70 dB. Before each test, the animal home cages were placed in the test room for at least 30 min to avoid stress effects due to transport. The apparatus and objects were disinfected after each trial using Antifect N liquid (propan-1-ol 35%, ethanol 23.5%). Distinct objects made of ceramic, metal and plastic, varying in height between 25 and 28 cm, were used. The positions and the objects were counterbalanced within and across animal groups. Exploration was defined as a snout orientation toward the object with a distance 2 cm or sniffing or touching with the snout. The animals exploring less than 10 s were excluded. Thus, the complete data set from a reduced number of animals could be used in the acute starvation group (n ¼ 5 of 9) and the corresponding control group (n ¼ 6 of 9), as well as in the chronic starvation group (n ¼ 13 of 17) and corresponding control animals (n ¼ 10 of 12).
The test consisted of a habituation period without objects (5 min), a sample trial (3 min), a 1-h delay and a recognition trial (3 min). The test was modified from previously described paradigms (Ennaceur & Delacour 1988;Frye & Lacey 2001;Antunes & Biala 2012;Luine & Frankfurt 2012). In the habituation period, the rats were placed in the centre of the field and were allowed to freely roam. Subsequently, when the rats were removed, two identical objects were placed 20 cm from the opposing corners. The animals were placed in an empty corner of the apparatus facing away from the objects. The time that the rats spent exploring each object was measured. After the 1-h delay, a third copy of the sample object and a new object were presented in the same positions. The exploration ratio was defined as:

Statistical analysis
The data of the continuous outcomes E2, NOR and RWA were described using the mean and corresponding standard deviation (SD) values in each subgroup (acute/chronic) and in the combined groups (ABA/controls). A two-way ANOVA, measuring the differences of E2 or NOR before and after starvation as outcome and the factors group (with characteristic ABA or control) and type of starvation (with characteristic acute or chronic), was fitted to the data. Additionally, the model included the two-way interaction between the factors group and type of starvation to allow the comparison of ABA versus controls in the acute or chronic phase of starvation by means of linear contrasts.
The comparisons of the average RWA measurement during the 7 days of acute starvation between ABA and control groups and the last 7 days of the weightholding phase for the chronic ABA and chronic control groups and the comparisons of leptin levels between ABA and controls were performed using unpaired t-test. To rule out an effect of the oestrous cycle phase, we compared the number of acute ABA with acute control animals and chronic ABA with chronic control animals that were in each of the four cycle phases on the days of the memory test and blood withdrawal before starvation using chi-squared tests. To check for a potential relationship between memory and reduction of E2 levels, we calculated Pearson correlations between the two difference scores of E2 levels and object recognition exploration ratio.
The comparisons between ABA and controls within each phase of starvation were evaluated by one-sided tests. For all other comparisons in which no direction of the treatment effect were presumed, two-sided P values 0.05 were considered to be statistically significant. Because of the explorative nature of the present study, no adjustment to the significance level was made. Results were reported as P values, value of the test statistic with corresponding degrees of freedom (df) and effect size (Cohen's d or f for continuous data, odds ratio, OR, in the case of binary data). All of the statistical analyses were conducted using SPSS version 20 for Windows (IBM, Chicago, IL, USA).

Results
The summary measures of the data in all of the subgroups are displayed in Table 1. Daily body weight, RWA and oestrous cycle are presented in Figure 2. RWA was significantly higher during acute weight loss in both groups exposed to ABA compared to the controls (P ¼ 0.0073, t(df) ¼ 16, t ¼ 3.07, Cohen's d ¼ 1.45 (acute) and P ¼ 0.0015, t(df) ¼ 27, t ¼ 3.54, Cohen's d ¼ 1.34 (chronic), respectively). However, RWA was not significantly differental between controls and chronic starvation ABA animals during the weight-holding phase (P ¼ 0.5018, t(df) ¼ 27, t ¼ 0.68, Cohen's d ¼ 0.26). Most rats showed a normal oestrous cycle before the onset of the ABA paradigm and all rats showed at least one complete cycle during the course of the experiment. The oestrous cycle showed a trend towards disruption already during acute starvation (chi-squared tests, P ¼ 0.114, v(df) ¼ 1, v 2 ¼ 2.4923, OR ¼6.4) and was significantly reduced during the weightholding phase of the chronic restriction group (chisquared tests, P < 0.001, v(df) ¼ 1, v 2 ¼ 25.1946, OR ¼384). Except for one, all of the rats in the chronic condition were without a regular oestrous cycle phase within the last 4 days. The E2 levels decreased in the ABA rats, whereas there was a physiological increase due to maturation in the control group. The results of the overall F-tests were presented in Supplementary Table 1, available online. Evaluation of the linear contrasts showed that the serum concentration differences comparing E2 levels before and after starvation were not significantly reduced after acute starvation (E2_standard_delta, P ¼ 0.2147, t(df) ¼ 43, t ¼ -0.80, Cohen's d ¼ 0.24) but were significantly reduced after chronic starvation compared to the control group (E2_standard_delta, P ¼ 0.0413, t(df) ¼ 43, t ¼ 1.78, Cohen's d ¼ 0.54) (Figure 3). Regarding object recognition, all of the groups showed mean exploration ratios above 0.50, indicating learning. As expected, leptin was significantly reduced in both ABA groups after starvation (n ¼ 9 ABA acute: 0.39 ± 0.47 ng/l, n ¼ 9 control acute: 1.87 ± 0.73 ng/l, Student's t-test P < 0.001; n ¼ 16 ABA chronic: 0.14 ± 0.22 ng/l, n ¼ 12 Control chronic: 0.59 ± 0.70 ng/l, Students t-test P ¼ 0.023). Leptin levels after starvation showed a trend towards correlation with oestrogen levels after starvation in the whole sample (Pearson correlation r ¼ 0.267, P ¼ 0.073) and a highly significant correlation with oestrous cyclicity (Spearman rho ¼0.427, P ¼ 0.003).
Evaluation of the linear contrasts showed that no significant decrease in learning in ABA rats in the acute starvation group compared to the control group (NOR_Delta, P ¼ 0.1926, t(df) ¼ 30, t ¼ 0.88, Cohen's d ¼ 0.32) could be observed; however, the chronic starvation group showed significantly impaired learning after starvation (NOR_Delta, P ¼ 0.0316, t(df) ¼ 30, Learning impairments significantly correlated with the decrease in E2 levels in the chronic starvation group (r ¼ 0.633, P ¼ 0.020), whereas the acute starvation group only revealed a similar tendency (r ¼ 0.870, P ¼ 0.055).
When combining both ABA groups, the decrease in E2 levels was significant compared to the combined control group (P ¼ 0.0451), as was the decrease in object recognition (P ¼ 0.0391, t(df) ¼ 43, t ¼ 1.73, Cohen's d ¼ 0.53) and their correlation (r ¼ 0.506, (Figure 4).

Discussion
To the best of our knowledge, this study is the first to show E2 deficits and memory function impairment in ABA rats. These effects were already, in part, present after acute starvation and became significant after chronic starvation. We have shown that more pronounced reductions in E2 plasma levels are associated with more severe learning impairment. The ABA paradigm was successfully implemented in our modified version with adolescent animals without self-starvation. Both ABA groups demonstrated, as expected, a reduction in weight and leptin levels, an increase in RWA, and a disruption in the oestrous cycle.

Oestrous cycle disruption
Approximately 50% of our acutely starved ABA animals lost their oestrous cycle towards the end of the weight loss period, whereas the chronically starved animals showed a near-complete loss of oestrous phases. This observation is consistent with previous studies showing that the incidence of animals with an oestrous phase decreases under the ABA paradigm (Dixon et al. 2003;Dos Santos et al. 2011) and after simple food or caloric restriction (Riddle et al. 2013). Dixon et al. (2003) reported that the majority of animals lose their oestrous cycle within 8 days in adult Long-Evans rats, Figure 2. Weight, running wheel activity and oestrous cycle in ABA. Effects of acute (left) and chronic (right) activity-based anorexia (ABA) on weight, running wheel activity (RWA) and oestrous cycle: (A) standardised body weight with SD; (B) standardised RWA with SD, the average of the first 10 days is set to 100%; (C) incidence of animals with the oestrous cycle phase within 4-day blocks, which coincides with the length of the normal rat oestrous cycle. ***P < 0.001 (A and B) two-way ANOVA, the comparisons within the subgroups (acute/chronic) were evaluated using linear contrasts. (C) Chi-square test comparing ABA and controls.
this breed being potentially more susceptible to cycle disruption. In our animal population, complete disruption of the ovarian cycle was only achieved in chronic starvation, which is generally perceived to reflect the exhaustion of metabolic resources such that a pregnancy cannot be realised. Hence, anovulation involving E2 deficiency is regarded as the body's natural protective mechanism.

E2 levels
The reduction of E2 plasma levels was significant in the chronic ABA group compared to the control animals and showed similar absolute but non-significant changes in the same direction in the acute ABA group. This finding corresponds well with the partial or complete disruption of the oestrous cycles mentioned above in the acute or chronic ABA groups, respectively (see Table 1). The only other report that studied E2 plasma levels in a model comparable to ABA was by Dos Santos et al. (2011). The authors analysed E2 levels in chronic ABA after 4 weeks in five animals and showed a decrease in absolute terms. However, the differences were not significant in that study, which could be because Dos Santos et al. did not provide free access to an RWA that maximally simulates AN and induces hyperactivity to a higher extent. Instead of furnishing a running wheel, Dos Santos et al. had the rodents run for 40-60 min on a motor treadmill.
Amenorrhoea (Vyver et al. 2008;Dempfle et al. 2013) and E2 deficiency (Singhal et al. 2014) are consistent findings in patients with AN, thus validating the findings in our ABA animals as a naturalistic model for human patients with respect to hormonal changes after starvation. We showed that a longer starvation period (>7 days as used in the acute starvation group) and continuous RWA result in the menstrual cycle being completely eclipsed and cause significant E2 deficiency. We also have to discuss that a serious disturbance during the maturation phase of the oestrous cycle such as starvation might also lead to persisting amenorrhoea. A similar phenomenon is observed in humans as very young girls not regularly menstruating before developing AN suffer from a much longer period of amenorrhoea than those with an established cycle (Dempfle et al. 2013).

Memory impairment
All of the groups showed an exploration ratio above 0.50 indicating learning. Group means of 0.58-0.67 before starvation are consistent with previous studies (Hopkins et al. 2011). Regarding the potential Figure 4. Correlation of oestradiol and memory. Correlation of oestradiol reduction and novel object recognition (NOR) memory deficit in all activity-based anorexia (ABA) animals. Delta NOR, NOR after starvation minus before starvation; Delta oestradiol, oestradiol after starvation minus before starvation; r, Pearson's correlation coefficient. implications of amenorrhoea and E2 reduction, our study revealed a significant loss of object recognition memory in ABA after chronic starvation. Acute ABA showed similar but not significant absolute reduction rates. This finding is consistent with certain cognitive deficits in AN. Stedal et al. (2012) demonstrated spatial recognition deficits, Lawrence et al. (2003) found impaired visual discrimination learning, and Oltra-Cucarella et al. (2012) showed impaired working memory in patients with AN. These observations are consistent with the dysfunction of object recognition in our model. Simple pattern recognition memory has not been shown to be decreased in patients with AN (Lawrence et al. 2003;Kjaersdam Tell eus et al. 2015). Thus, the task of object recognition in our ABA animals potentially extends beyond simple pattern recognition and is more linked to spatial recognition and visual discrimination. First studies in patients with AN seem to corroborate this findings, as Lozano-Serra showed reduced visuo-spatial organisation and recall during starvation (Andr es- Perpiña et al. 2011). Several of the other typical findings in AN such as increased perseverance and decreased flexibility are likely character traits that already existed before the acute onset of illness (and might even be a predisposition) (Oltra Cucarella et al. 2012). However, learning and memory deficits may add to this already inflexible character type and further impair the adaptation of new suggestions and make behavioural changes more difficult in the commencement of therapy. This process is consistent with our own clinical experience of patients feeling ''walled in'' and not being responsive to guidance from their therapists, especially in the beginning of therapy when these patients present in the starved state. Thus, the impairments in memory and learning documented in this animal study could, if similarly appearing in patients with AN, contribute to the patient's incapacity for learning new strategies and properly engaging in the psychotherapeutic process in a state of starvation.

Association of E2 and memory
We aimed to analyse the interaction between cognition and E2 levels. We described a correlation of E2 reduction and object recognition memory deficit significant in chronically starved ABA animals. This observation suggests that cognitive impairment could depend on E2 deficiency. A strong causal association between greater E2 deficits and more severe recognition memory deficits have previously been documented in ovariectomised and E2 supplemented rodents (Jacome et al. 2010;Luine 2015) suggesting a potential mechanistic link between the two components. These results also fit well with at least partially effective E2 supplementation effects on memory and learning in post-menopausal women (Luine & Frankfurt 2012). However, a causal relation in AN patients can only be definitively proven by an intervention study. The first correlative study of this association in patients with AN was conducted by Buehren et al. (2011). They found an association between verbal learning and E2 increase after recovery. Chui et al. (2008) furnished indirect evidence by showing an association of amenorrhoea or irregular menses with reduced cognitive functions as diverse as verbal ability, cognitive efficiency, oral language and broad mathematics in former patients with AN who were partially recovered. Lozano-Serra et al. (2014) extended these findings showing reduced visuo-spatial abilities in weight recovered patients without menses compared to those with menstruation. Regarding morphological brain changes, Mainz et al. (2012) demonstrated an interdependency between follicle-stimulating hormone (FSH) and the increase of structural brain volume in the hippocampus and thalamus after weight recovery in AN. FSH is known to increase before full resumption of menses is reached in weight-recovering patients with AN and, thus, is perceived to be a sensitive precursor of an ensuing increase in E2. The hippocampus is presumed to play an important functional role in object recognition memory besides the prefrontal cortex (Luine & Frankfurt 2012). Thus, our findings are consistent with previous studies indicating an association between hormone status and cognition in AN. As has been shown in numerous prior studies, the hippocampus is a unique source for classical and non-classical types of oestrogen receptors (Ivanova & Beyer 2000;Foster 2012). E2 can quickly modulate both neuronal function and structure in the hippocampus, which enables this steroid hormone rapidly and persistently, depending on the type of receptor interaction, modulating cognition (Prange-Kiel & Rune 2006;Ogiue-Ikeda et al. 2008;Sellers et al. 2015). This observation is consistent with findings that established a role for oestrogen receptor 2 gene haplotypes in genetic susceptibility to AN (Eastwood et al. 2002) and indicated that mutations in the E2-related receptor alpha increase the risk for developing eating disorders (Cui et al. 2013). It is concluded that abnormalities in the biochemical and developmental pathways related to classical and nonclassical E2 signalling, including E2 co-activators and selective oestrogen receptor modulators, converge to create a physiological condition that facilitates or triggers the onset of AN.

RWA
RWA was only significantly increased during active weight reduction but not during the weight-holding phase. This observation suggests that chronically starved animals adapted to the new condition and reduced hyperactivity as a survival strategy. This finding is consistent with previous findings illustrating that RWA increases for only 3-4 days, subsequently plateaus, and in the long-term converges with the activity of the controls (Holtkamp et al. 2006;Barbarich-Marsteller et al. 2013). A detailed analysis of RWA in ABA (Wu et al. 2014) confirmed that the increase of RWA has a limited time frame.

Comparing acute and chronic starvation
Although chronically starved animals did not differ from control animals in the final phase of the experiment concerning RWA, they showed continued weight reduction, amenorrhoea, E2 and memory deficiency. Thus, we illustrated that long-term starvation combined with running wheel access is an adequate model for chronic AN modelling the clinical characteristics of chronically ill patients. Starvation considerably affected menstrual cycle, E2 secretion, memory and their correlations; however, this became statistically significant only in the chronic group with a 100% of amenorrhoea in the animals.
We started the experiment during mid-adolescence of rats at post-natal day 38. Thus, the chronically starved animals did not only experience adolescence but also young adulthood during the experiment. This corresponds well to patients with AN as the main age of onset is between 14 and 19 years in adolescence while chronically ill patients are often affected well into adulthood (Herpertz-Dahlmann 2015).

Limitations
Because we wanted to analyse acute and chronic starvation in ABA, we decided not to include self-starvation in our model, as this experimental set-up often proves to be lethal to the animals when continued for a longer time frame (Routtenberg & Kuznesof 1967;Exner et al. 2000). Instead, we prescribed a weight reduction of 20-25% by adjusting the daily food intake plus an optional weight-holding phase for the chronic starvation group. In that manner, chronic starvation could be tested and the results could be analysed, thereby reducing the influence of variable body weights in ABA rats. However, this difference in the protocol must be considered when we compare our data with those of other ABA studies using self-starvation by reducing the time of food availability. The variable weight reduction of 20-25% might have slightly increased the variability of outcomes. Lastly, ABA cannot imitate some of the key features of AN like fear of gaining weight and body image distortion or resulting behaviour, thus our results can only be applied to patients with AN with caution. However, ABA appears to be a good model for AN regarding body weight, oestrous cycle as well as hormonal changes and even some starvation-induced behaviours like hyperactivity and learning deficits.

Conclusion
The primary findings of our study were that E2 deficiency and loss of memory were correlated in the chronic starvation ABA animals. This finding is consistent with a potential link between amenorrhoea, E2 plasma level reduction and impaired memory function in patients with AN. This connection might help to explain cognitive deficits and probably limited effectiveness of psychotherapy in patients with AN at the commencement of treatment; e.g., cognitive behavioural therapy would involve repeated processes of learning and relearning. Memory function should be included in the early diagnostic tests of patients with AN after admission to the hospital, potentially to test their readiness for new input during psychotherapeutic interventions. Future animal studies should provide information regarding whether starvation causes structural deficits (dendritic plasticity, synaptic connectivity or cell loss) in the hippocampus and other affected brain areas and whether E2 substitution restores memory function and structural changes. Such positive E2 effects might offer the potential to correct learning deficits and memory impairment and to facilitate psychotherapeutical interventions for patients with AN.