Is Serum BDNF Altered in Acute, Short- and Long-Term Recovered Restrictive Type Anorexia Nervosa?

Brain-derived neurotrophic factor (BDNF), a neurotrophin involved in the regulation of food intake and body weight, has been implicated in the development and maintenance of Anorexia nervosa (AN). The majority of previous studies reported lower BDNF levels in acutely underweight AN patients (acAN) and increasing levels after weight rehabilitation. Here, we investigated serum BDNF concentrations in the largest known AN sample to date, both before and after weight restoration therapy. Serum BDNF was measured in 259 female volunteers: 77 in-patient acAN participants of the restrictive type (47 reassessed after short-term weight rehabilitation), 62 individuals long-term recovered from AN, and 120 healthy controls. We validated our findings in a post-hoc mega-analysis in which we reanalyzed combined data from the current sample and those from our previous study on BDNF in AN (combined sample: 389 participants). All analyses carefully accounted for known determinants of BDNF (age, sex, storage time of blood samples). We further assessed relationships with relevant clinical variables (body-mass-index, physical activity, symptoms). Contrary to our hypotheses, we found zero significant differences in either cross-sectional or longitudinal comparisons and no significant relationships with clinical variables. Together, our study suggests that BDNF may not be a reliable state- or trait-marker in AN after all.

To overcome this limitation, we carried out Bayesian analyses using JASP [6], using an uniformed Cauchy distribution centered on a zero effect size (δ) and a width of 0.7070 as a prior in all Bayesian analyses. This allowed us to investigate which hypothesis our data supported the best and assess evidence for the null-hypothesis that the cross-sectional and longitudinal study groups do not differ in serum BDNF concentrations [4].

Results from the post-hoc mega-analysis
To verify and extend the results obtained from the sample at the focus of the current study, a supplementary post-hoc mega-analysis was conducted pooling all participants from the current study together with participants from Site A (Berlin) of our previous multicenter study on serum BDNF concentrations in AN [7]. To test for differences in the cross-sectional mega-sample, a multiple regression model with planned contrasts (i.e. acAN vs HC, recAN vs HC) was employed. In the sample examined in this current study, the influence of age on serum BDNF concentrations [8] was addressed by pair-wise matching of acAN and recAN to HC by age. However, this was unfeasible in the combined sample and therefore this effect was controlled for by adding age as an explanatory variable to the multiple regression model. Further, a dummy variable was included in the model, indicating if subjects belonged to the sample of our current or our previous study. Results showed that the cross-sectional groups did not differ significantly on serum BDNF concentrations in the posthoc mega-analysis (Table S4). Longitudinally, differences in serum BDNF concentrations in acAN before and after partial/short-term weight rehabilitation were assessed using a mixed-effects model allowing to control for the effect of the aforementioned dummy variable. Results indicated no significant differences in serum BDNF concentrations after short-term weight increase in the combined sample (Table S5).
In contrast to the sample at the focus of the current study, we did not limit participants to acAN of the restrictive type in our previous study [7]. Therefore, we repeated our combined-samples analyses after excluding acAN and recAN of the binge-purge type. Nonetheless, results did not change and the differences in serum BDNF concentrations between acAN and HC, recAN and HC and acAN before and after partial/short-term weight rehabilitation remained non-significant (Table  S6, Table S7).

Results from confirmatory analyses, excluding participants receiving pharmaceutical treatment
Previous work has suggested that serum BDNF concentrations are influenced by administration of selective serotonin reuptake inhibitors (SSRI) [9]. In our current study, three participants in the acAN group and two participants in the recAN group in the cross-sectional, as well as one participant in the longitudinal comparison received treatment with SSRI within 4 weeks before blood sample collection. To ensure the observed absence of differences in serum BDNF concentrations in all groups was not mediated by receiving treatment with SSRI, we repeated the analyses after excluding all SSRI-medicated participants as well as, for the cross-sectional comparisons, their age-matched HCs.

Controlling for BMI-SDS and age in the longitudinal comparison
To further verify results from the longitudinal comparison of serum BDNF concentrations before and after partial weight-recovery, a linear mixed effects model was administered to both the data of our current sample and the combined sample from our previous investigation of serum BDNF levels in AN [7]. Administration of the linear mixed effects model allowed us to investigate whether controlling for differences of both age and BMI-SDS of participants between the two study timepoints proved to mediate a significant increase of serum BDNF after partial weight-rehabilitation. The differences of both age and BMI-SDS before and after weight-increase were consequently added as fixed effects to the multilevel model. However, previous unsignificant results remained unaffected by this exploratory analysis (Table S8, Table S9).

Quantifying the type II error in the longitudinal comparison
The test result in the longitudinal arm of our main analyses (paired t-test t(46) = −1.59, p = 0.118) quantifies a probability of observing the test result in our sample, or an even more extreme result, given that the null-hypothesis of no group differences in serum BDNF concentrations across partial weight-rehabilitation is true. Since this result did not meet the widely accepted criterion of a type I error below 0.05 we don't have sufficient evidence to reject the null-hypothesis and therefore the change in serum BDNF across partial weight-recovery cannot be considered statistically significant. However, the aforementioned probability was lower in the longitudinal comparison than in our cross-sectional comparisons and we therefore further quantified the type II error, the probability of incorrectly not rejecting a false null-hypothesis, to validate our results. Since statistical power is the probability of correctly rejecting a false null-hypothesis, the type II error is defined as β = 1 -Power.
We carried out power-analysis using the "pwr"-package [10] in R [11], assessing the power to detect a medium effect size, as defined by Cohen [12], resulting in a power of 0.919. Therefore, the type II error of incorrectly not-rejecting a false null-hypothesis is 0.081.

Analysis of serum BDNF concentrations between acAN and recAN
Since recAN were significantly older than acAN in our sample (Welch Two Sample t-test t(127.03) = −9.87, p < 0.001), our approach of pair-wise age-matching participants was not possible for analyzing group differences of serum BDNF concentrations between recAN and acAN. We therefore controlled for possible undue influence of age on serum BDNF concentrations [13] in a multiple linear regression model. To assess differences in serum BDNF levels between recAN and acAN in the posthoc mega-analysis an additional dummy variable was added to the model indicating if subjects belonged to our current sample or that of our previous study. However, no significant group difference in serum BDNF levels between recAN and acAN could be observed either in the current sample or in the combined sample (Table S10, Table S11). All values are presented as means ± standard deviation. Analysis of variance and post-hoc pairwise comparisons (corrected for multiple comparisons using the Benjamini-Hochberg procedure) were used to assess between-group differences in all variables; F-values and p-values are reported as test results. a post-hoc comparison between acAN and recAN b post-hoc comparison between recAN and HC c post-hoc comparison between acAN and HC * p < 0.05 in post-hoc test ** p < 0.01 in post-hoc test Abbreviations: acAN acute anorexia nervosa participants, HC healthy control participants, recAN long-term recovered anorexia nervosa participants, BMI body mass index, BMI-SDS body mass index standard deviation score, EDI-2 Eating Disorder Inventory-2.    Pearson (BDNF and BMI-SDS, EDI-2, BDI-2) and Spearman (BDNF and physical activity) correlation coefficients are reported as test statistics along with their corresponding p-values before correction for multiple comparisons. Abbreviations: acAN-T1 acute anorexia nervosa participants at study timepoint 1 (admission), acAN-T2 acute anorexia nervosa participants at study timepoint 2 (after short-term weight rehabilitation), recAN long-term recovered anorexia nervosa participants, HC healthy control participants, BMI-SDS body mass index standard deviation score, EDI-2 Eating Disorder Inventory-2, BDI-II Beck Depression Inventory.   Estimates and their respective standard errors, as well as t-values and p-values for each coefficient are presented. The bottom part of the table shows the number of observations, R-squared, adjusted R-squared, the residual standard error and the F statistic (i.e. parameters to assess the fit of the model).

BDNF -BMI-SDS BDNF -EDI-2 BNDF -BDI-II
No recAN in this sample was of the binge-purge type. Abbreviations: acAN acute anorexia nervosa participants, HC healthy control participants, recAN long-term recovered anorexia nervosa participants. Observations 118 Estimates and their respective standard errors, as well as t-values and p-values for each coefficient are presented, with the total number of observations in the bottom part of the table. Abbreviations: acAN-T2 acute anorexia nervosa participants at study timepoint 2 (after short-term weight rehabilitation). Observations 94 Estimates and their respective standard errors, as well as t-values and p-values for each coefficient are presented, with the total number of observations in the bottom part of the table. Abbreviations: acAN-T2 acute anorexia nervosa participants at study timepoint 2 (after short-term weight rehabilitation), BMI-SDS body mass index standard deviation score.    Abbreviations: BDNF brain-derived neurotrophic factor, acANT1 acute anorexia nervosa participants at study timepoint T1, acANT2 acute anorexia nervosa participants at study timepoint T2, HCacAN healthy control participants, age-matched to the acAN group, recAN long-term recovered anorexia nervosa participants, HCrecAN healthy control participants, age-matched to the recAN group Figure S2: Serum BDNF concentrations in ng/ml in all study participants plotted over squared storage time at -80°C before analyzing. Batches are represented by different shapes. Study groups are indicated by different colors. The regression line follows the formula of y ~ x and the model is highly significant (p = 0.001). Grey background around the regression line indicates the 95% confidence interval. Abbreviations: BDNF brain-derived neurotrophic factor, acANT1 acute anorexia nervosa participants at study timepoint T1, acANT2 acute anorexia nervosa participants at study timepoint T2, HCacAN healthy control participants, age-matched to the acAN group, recAN long-term recovered anorexia nervosa participants, HCrecAN healthy control participants, age-matched to the recAN group