Differential alterations of amygdala nuclei volumes in acutely ill patients with anorexia nervosa and their associations with leptin levels

Background The amygdala is a subcortical limbic structure consisting of histologically and functionally distinct subregions. New automated structural magnetic resonance imaging (MRI) segmentation tools facilitate the in vivo study of individual amygdala nuclei in clinical populations such as patients with anorexia nervosa (AN) who show symptoms indicative of limbic dysregulation. This study is the first to investigate amygdala nuclei volumes in AN, their relationships with leptin, a key indicator of AN-related neuroendocrine alterations, and further clinical measures. Methods T1-weighted MRI scans were subsegmented and multi-stage quality controlled using FreeSurfer. Left/right hemispheric amygdala nuclei volumes were cross-sectionally compared between females with AN (n = 168, 12–29 years) and age-matched healthy females (n = 168) applying general linear models. Associations with plasma leptin, body mass index (BMI), illness duration, and psychiatric symptoms were analyzed via robust linear regression. Results Globally, most amygdala nuclei volumes in both hemispheres were reduced in AN v. healthy control participants. Importantly, four specific nuclei (accessory basal, cortical, medial nuclei, corticoamygdaloid transition in the rostral-medial amygdala) showed greater volumetric reduction even relative to reductions of whole amygdala and total subcortical gray matter volumes, whereas basal, lateral, and paralaminar nuclei were less reduced. All rostral-medially clustered nuclei were positively associated with leptin in AN independent of BMI. Amygdala nuclei volumes were not associated with illness duration or psychiatric symptom severity in AN. Conclusions In AN, amygdala nuclei are altered to different degrees. Severe volume loss in rostral-medially clustered nuclei, collectively involved in olfactory/food-related reward processing, may represent a structural correlate of AN-related symptoms. Hypoleptinemia might be linked to rostral-medial amygdala alterations.

Here we investigated for the first time the volumes of individual nuclei and transition areas of the amygdaloid complex in a large sample of acutely underweight females with AN in comparison with age-matched healthy females.We applied FreeSurfer-based amygdala subsegmentation (Fischl, 2012;Saygin et al., 2017) and multi-stage sMRI quality assessment.Given evidence for widespread (sub-)cortical neuroanatomical alterations in AN (King et al., 2018), we expected that amygdala nuclei volumes would also be reduced in the AN group.Importantly, however, we strived to elucidate differences in the relative severity of amygdala nuclei alterations to unveil potential significant effects beyond whole amygdala and other subcortical GM volume alterations.Specifically, motivated by anxiety-/ reward-based models of AN (Murray et al., 2018;O'Hara et al., 2015) and the above-noted functional amygdala subdivisions, we hypothesized that volumetric reductions would be most prominent in basolateral and baso-/corticomedial amygdala subdivisions predominantly involved in fear and reward processing (Baxter & Murray, 2002;Davis & Whalen, 2001;Janak & Tye, 2015;LeDoux, 2007;Sah et al., 2003).Finally, we explored the clinical relevance of amygdala nuclei alterations in AN by investigating associations with leptin levels serving as an indicator of AN-related neuroendocrine alterations (Hebebrand et al., 2007(Hebebrand et al., 2022)), degree of underweight, illness duration, and ED-specific, depressive, anxious, and general psychiatric symptom levels.

Participants
Female patients with acute AN were admitted to ED treatment programs at a child and adolescent psychiatry or psychosomatic medicine department of a tertiary care university hospital and underwent MRI within 96 h after beginning nutritional rehabilitation.Current AN, according to DSM-5 criteria, was diagnosed using a modified version of the expert form of the Structured Interview for Anorexia and Bulimia Nervosa (SIAB-EX) (Fichter & Quadflieg, 1999)  The initial study sample was subjected to quality control (QC) of sMRI scans; participants with misapplied general (sub-)cortical segmentation and/or amygdala subsegmentation were discarded (online SM 1.4-1.5,Table S1).To optimize group comparisons, HC were age-matched to AN via optimal pair matching pursuing a minimized sum of absolute pairwise distances in the matched sample (Hansen & Klopfer, 2006).The difference in age means between matched groups was 0.2 years (maximum age distance among AN-HC pairs was 0.6 years).The final study sample consisted of 336 female volunteers: 168 AN (aged 12-29 years) and 168 age-matched HC (aged 12-29 years).
For leptin measurements, fasting venous blood was collected into EDTA vacutainer tubes at 7-9 a.m., for AN within 96 h after treatment initiation.Plasma leptin was measured using a commercially available enzyme-linked immunosorbent assay (BioVendor, Brno, Czech Republic) with intra-/inter-assay variation coefficients <6%.Leptin values were logarithmically transformed (log 10 -leptin) to achieve normality (Haas et al., 2005).Plasma samples were available from 142 AN and 156 HC.Non-detectable leptin concentrations below the lower limit of detection of the leptin assay (LOD = 0.20 μg/L) occurred in 39 of 142 AN (27.46%) and were subsequently imputed using censored likelihood multiple imputation to preserve their natural variability [CLMI (Boss et al., 2019), online SM 1.3; see online SM 2.4 for confirmatory analysis including only leptin levels ⩾LOD in AN].Leptin levels <LOD did not occur in HC.Missing/unavailable leptin values were not imputed.
The automated amygdala subsegmentation into nuclei was performed using FreeSurfer functionality for combined amygdala/hippocampus subsegmentation, based on a Bayesian probabilistic atlas to assign an anatomical label to each voxel (Saygin Fig. 1.Amygdala nuclei and amygdaloid complex.(a) Two-dimensional (2D) illustration of FreeSurfer-based amygdala subsegmentation.(b) Bar graph visualization of age-, age 2 -, and eTIV-adjusted GLM0 for individual amygdala nuclei volumes in AN v. HC.AN, patients with acute anorexia nervosa; HC, healthy control participants; lh, left brain hemisphere; rh, right brain hemisphere; eTIV, estimated total intracranial volume; GLM, general linear model; HPA, hypothalamic-pituitaryadrenal axis; CRH, corticotropin-releasing hormone; GABA, gamma-aminobutyric acid; -R, receptor.(a) Amygdala nuclei location in axial/coronal/sagittal planes, groupings/subdivisions, functions, and connectivity (abstracted and simplified illustration).Amygdala subdivisions (A-D) are highlighted with gray background (LeDoux, 2007(LeDoux, , 2008;;McDonald, 2020;Sah et al., 2003;Saygin et al., 2011Saygin et al., , 2017;;Watson et al., 2012).Brain regions in boxes with dashed outline refer to important input or output regions.Arrows indicate predominantly unidirectional or reciprocal connectivity (gray arrows, joined projections to/from all nuclei within an amygdala subdivision; arrows colored according to nuclei labels, projections to/from individual amygdala nuclei).2D FreeView-snapshots of a T1-weighted sMRI scan after FreeSurfer version 7.1.1-basedamygdala subsegmentation, mapped onto a preprocessed and normalized T1-weighted brain image of a typical patient with acute AN in the study sample.(b) Bar graphs with error bars for study groups AN (n = 168) and age-matched HC (n = 168) displaying adjusted means (EMM, mm 3 ) ± standard error of the mean (S.E.M.) of individual whole amygdala and amygdala nuclei volumes in separate brain hemispheres (color of bars matches color of nuclei labels in panel a).Model estimates were obtained with GLM0 [performed separately for each amygdala (sub-)region, computed as F test: dfs = 1, 331] covarying for age at date of research (linear and quadratic orthogonal polynomials) and eTIV (covariates were grand mean-centered).FDR-q, p values were multiple testing adjusted using false-discovery rate (Benjamini & Hochberg, 1995) across all amygdala nuclei (whole amygdala adjusted separately using FDR).Significance levels for volume differences between study groups are stated as: ***q < 0.001; *q < 0.05; ns, non-significant.Effect size statistics are provided as partial η 2 (Cohen, 1988).et al., 2017).The atlas was created from high-resolution ex-vivo MRI data (≈0.1 mm isotropic at 7 T, manually segmented postmortem human brain samples) and in-vivo training MRI data (Saygin et al., 2017).Ten amygdala (sub-)region volumes were generated and analyzed separately for left (lh) and right (rh) brain hemispheres: whole amygdala, accessory basal, basal, central, cortical, lateral, medial, and paralaminar nuclei, anterior amygdaloid area, and corticoamygdaloid transition.

Quality control
There are no established QC standards for amygdala subregions to date (Sämann et al., 2020).Therefore, we developed a visual and partly outlier-guided multi-stage QC procedure for combined amygdala/hippocampus subsegmentations in line with recently published recommendations for hippocampal subfield QC by the ENIGMA consortium (Sämann et al., 2020) and under expert consultation with two independent ENIGMA representatives (http://enigma.ini.usc.edu/).Briefly, our QC involved: (1) a-priori exclusion of participants with low scan quality (contrast-/ signal-to-noise ratios) or insufficient ratings in general (sub-)cortical QC (online SM 1.4), (2) snapshot-based visual QC of amygdala/hippocampus subsegmentations of all participants after a-priori exclusions, and (3) dynamic visual inspection of amygdala/hippocampus subsegmentations with statistical outliers using FreeSurfer's FreeView tool (based on group-wise outlier detection via combined volume and bilateral symmetry criteria).Visual QC was manually conducted by two trained raters with  (Cohen, 1988) are stated.Censored likelihood multiple imputation (Boss et al., 2019) was applied for left-censored leptin values in AN below the lower limit of detection of the leptin assay (LOD = 0.20 μg/L).
General brain segmentation measures eTIV and total subcortical GM volume were extracted from FreeSurfer's segmentation statistics (Fischl, 2012).In AN, the mean ± S.D. age at first onset of AN was 14.5 ± 2.8 years (assessed in n = 163) and the mean ± S.D. duration of the current AN episode (DOI) was 14.1 ± 18.3 months (n = 164).AN subtype was determined via SIAB-EX: 142 AN (84.52%) were restrictive and 22 (13.10%) were binge-purge [subtype not assessed in n = 4 (2.38%)].Of the patients with AN, 26 had one or more co-existing psychiatric conditions: 12 had a depressive disorder, 11 an anxiety disorder, 6 an obsessive-compulsive disorder, 1 a post-traumatic stress disorder, 1 an adjustment disorder, 2 a personality disorder, 1 a developmental disorder, 1 Tourette syndrome, and 1 a somatization disorder.Selective serotonin reuptake inhibitors were taken by 4 AN and mirtazapine by 1 AN within the last 6 months before study participation.None of the HC participants had any psychiatric diagnosis currently or in the past or any psychoactive medication.All study participants were female and identified as 'European'.S1].

Statistical analyses
Statistical analyses were conducted in R v4.1.1 (online SM 1.5) (R Core Team, 2022).Raw volumetric measures of amygdala (sub-) regions were approximately normally distributed in the study sample according to visual inspection and Shapiro-Wilk test (online SM 2.1, Fig. S2).All amygdala (sub-)region volumes were modeled using general linear models (GLMs) with study groups as the predictor, and a selection of covariates based on our research questions.To assess general/global AN-driven alterations in amygdala nuclei volumes, GLM0 (Fig. 1b) was adjusted for linear and quadratic orthogonal polynomials of participant age in line with recent ENIGMA studies due to evidence for nonlinear age effects on amygdala volumes (Chen et al., 2016;Han et al., 2020;Sämann et al., 2020;Vinke et al., 2018;Zugman et al., 2022).GLM0 was also adjusted for estimated total intracranial volume (eTIV), which is an established correction method of brain volumes for head size variation and recommended prior to any volumetric brain analysis (https://surfer.nmr.mgh.harvard.edu/fswiki/eTIV) (Malone et al., 2015;Sanfilipo, Benedict, Zivadinov, & Bakshi, 2004;Sargolzaei et al., 2015;Voevodskaya et al., 2014).In order to uncover specific effects of AN on amygdala nuclei against the background of AN-related (1) whole amygdala alterations, and (2) total subcortical GM alterations (King et al., 2018), we considered two relative GLMs: (1) GLM1 (Fig. 2a) included whole amygdala volume in addition to covariates from GLM0 to investigate subregional (i.e.within-amygdala) effects of AN; and (2) GLM2 (Fig. 2b) additionally covaried for total subcortical GM volume to examine amygdala nuclei alterations beyond AN-related generalized subcortical GM reductions.Multiple testing adjustment of p values using false-discovery rate (FDR) (Benjamini & Hochberg, 1995) was applied across all amygdala nuclei in GLM0, and across both relative GLMs (separately from GLM0  1) were implemented to confirm group differences from GLM0 (Frank, Favaro, Marsh, Ehrlich, & Lawson, 2018;King et al. 2018).We further investigated potential effects of AN subtype (restrictive/binge-purge), hydration status [urine-specific gravity (Biller et al., 2015;Streitbürger et al., 2012)], and oncotic pressure [serum albumin concentration (Wagner et al., 2006)] on amygdala nuclei volumes in AN (online SM 2.3).
To follow-up on amygdala nuclei significantly decreased in AN according to all GLM approaches, robust multiple linear regression analysis (RLM) (Huber, 1981) was performed within the AN group to test for associations between individual FDR-q, p values were multiple testing adjusted using false-discovery rate (Benjamini & Hochberg, 1995) across all amygdala nuclei and both GLM1 and GLM2 (whole amygdala adjusted separately using FDR but also across both GLMs).Significance levels for volume differences between study groups are stated as: ***q < 0.001; **q < 0.01; *q < 0.05; ns, non-significant.Effect size statistics are provided as partial η 2 (Cohen, 1988).

Sample characteristics
Study groups AN and HC did not differ in age, IQ, eTIV (Table 1), and handedness (online SM Table S2).Parental SES was higher in HC than AN (Table 1).Cigarette smoking prevalence was lower in AN than HC (online SM Table S2).As expected, AN had significantly lower BMI-SDS and, correspondingly, lower log 10 -leptin levels than HC, whereas ED-specific, depressive, trait anxiety, and general psychiatric symptom measures were markedly higher in AN (Table 1).Moreover, AN presented with reduced total subcortical GM volume [t(331) = 8.89, p < 0.001).

Amygdala nuclei
When controlling for age and eTIV (GLM0, Fig. 1b), model estimates of whole amygdala (lh = −4.34%,rh = −4.22%)and most amygdala nuclei volumes were significantly smaller in AN than HC with maximum volumetric reductions in medial nucleus (lh = −10.70%,rh = −8.59%),cortical nucleus (lh = −9.33%,rh = −8.54%),corticoamygdaloid transition (lh = −7.98%,rh = −7.69%),and accessory basal nucleus (lh = −6.81%,rh = −6.00%;central and paralaminar nuclei rh were only nominally smaller in AN).Reported group differences from GLM0 remained robust (online SM Fig. S3B) after excluding AN with co-existing depressive, anxiety, obsessive-compulsive, and post-traumatic stress disorder diagnoses, and/or antidepressant pharmacotherapy (see Table 1).Likewise, findings were mostly unchanged when controlling for parental SES, IQ, Fig. 3. 3D visualization of FreeSurfer-based amygdala subsegmentation into amygdala nuclei displaying group differences in AN v. HC.AN, patients with acute anorexia nervosa; HC, healthy control participants; lh, left brain hemisphere; rh, right brain hemisphere.Amygdala nuclei where significantly smaller volumes in AN v. HC were detected as overlapping findings according to GLM0 (eTIV-adjusted), GLM1 (whole amygdala volume-adjusted), and GLM2 (total subcortical GM volume-adjusted) are displayed with 100% opacity (namely, accessory basal nucleus lh and rh, cortical nucleus lh and rh, corticoamygdaloid transition lh and rh, and medial nucleus lh).Amygdala nuclei where group differences (i.e.smaller volumes in AN v. HC) were not significant at threshold FDR-q < 0.05 (according to at least one GLM approach) are displayed with 50% opacity.The figure shows a single-subject 3D model of amygdala nuclei (n = 9 lh, n = 9 rh), obtained from a preprocessed and normalized T1-weighted brain image of a typical patient with acute AN in the study sample, segmented via the FreeSurfer v7.1.1 automated amygdala subsegmentation tool, and rendered using the 'Blender' software (Blender Online Community, 2018).Orientation in the brain is given for amygdala lh (in gray/italics).
handedness, and cigarette smoking in the main sample (except for the no longer significant volumetric reductions in lateral, central, and paralaminar nuclei lh, online SM Fig. S3A).Amygdala nuclei volumes were not related to AN subtype (online SM Table S3) and hydration status in AN (mostly within normal range, de-/ hyperhydration in n = 2/6 AN, online SM Tables S4 and S5).Hypoalbuminemia did not occur in AN (online SM Table S4).
After adjusting for whole amygdala volume, significant within-amygdala differences between groups emerged for the following nuclei (GLM1, Fig. 2a): accessory basal and cortical nuclei lh and rh, corticoamygdaloid transition lh and rh, and medial nucleus lh were smaller in AN than HC.In contrast, basal nucleus rh, lateral and paralaminar nuclei lh and rh were larger in AN than HC relative to whole amygdala volume.The GLM2 approach, adjusting for total subcortical GM reductions in AN v. HC, yielded no group differences in whole amygdala volumes but smaller accessory basal and cortical nuclei lh and rh, corticoamygdaloid transition lh and rh, medial nucleus lh, and anterior amygdaloid area rh volumes in AN than HC.However, AN showed larger paralaminar nucleus rh than HC relative to total subcortical GM volume (Fig. 2b).
Notably, accessory basal, cortical, and medial nuclei, and corticoamygdaloid transition showed absolutely (GLM0) as well as relatively [i.e. compared to whole amygdala (GLM1) and total subcortical GM (GLM2) volume alterations] reduced volumes in AN v. HC, with almost perfect symmetry across brain hemispheres (except for medial nucleus: only lh affected).These subregions share an anatomically aggregated location in the rostral-medial amygdala (Fig. 3).

Associations with clinical measures
RLMs in the AN study group examining all amygdala nuclei within the bilateral rostral-medial cluster [accessory basal, cortical, medial (lh) nuclei, and corticoamygdaloid transition, online SM Table S6] yielded significant positive associations of accessory basal nucleus lh and corticoamygdaloid transition lh volumes with BMI-SDS.Importantly, all rostral-medially clustered amygdala nuclei volumes, comprising accessory basal, cortical, medial nuclei, and corticoamygdaloid transition, were significantly and positively associated with log 10 -leptin levels in AN at medium strength (η p 2 = 0.044-0.126,Fig. 4).To test for the unique effect of leptin above and beyond BMI-SDS effects in follow-up RLMs in AN, we covaried for BMI-SDS while orthogonalizing log 10- -leptin: after FDR-adjustment, log 10 -leptin explained additional variance in all amygdala nuclei volumes within the rostral-medial cluster (i.e.accessory basal, cortical, medial nuclei, and corticoamygdaloid transition).No significant associations between amygdala nuclei volumes and DOI or psychiatric symptom levels emerged in AN (online SM Table S6).Exploratory RLMs in HC did not reveal significant relationships between amygdala nuclei volumes and BMI-SDS or log 10 -leptin levels (online SM 2.5, Table S7).

Discussion
In this first study to investigate amygdala substructure volumes in AN using sMRI, we found significant volumetric reductions of most amygdala nuclei in acutely underweight AN compared to HC.More importantly and going beyond previous findings of generally reduced whole amygdala volumes in AN (Burkert et al., 2019;Friederich et al., 2012;Giordano et al., 2001;Fig. 4. Scatter plots for associations between significantly reduced amygdala nuclei volumes and leptin concentrations in AN.Log 10 -leptin, logarithmically transformed (base 10) leptin concentration; lh, left brain hemisphere; rh, right brain hemisphere.Scatter plots with individual data points, linear regression lines, and 95% confidence intervals around the regression line (gray band) in the AN study group (plasma leptin measurement available in n = 142 of 168 AN) displaying associations between individual amygdala nuclei volumes that were significantly reduced in AN v. HC according to all GLM approaches (GLM0/1/2, Figs 1b-3) and log 10 -transformed plasma leptin concentrations [non-detectable leptin values <LOD of the leptin assay (n = 39 of 142) were multiple imputed using CLMI (Boss et al., 2019), associations were examined via RLMs applying M-estimation and Huber weighting for fitting via iterated re-weighted least squares].Standardized residuals of amygdala nuclei volumes are plotted after adjustment of raw volume measures for age at date of research (linear and quadratic orthogonal polynomials) and eTIV using robust multiple linear regression.RLM statistics are provided as t value (unstandardized β divided by its standard error), unadjusted p value (computed via robust Wald F test), and effect size estimate partial η 2 (Cohen, 1988) for each RLM for log 10 -leptin as the predictor [RLM formula: individual amygdala nuclei volumes (lh, rh) ∼ log 10 -leptin + poly(age 1 , age 2 ) + eTIV].Log 10 -leptin remained a significant predictor of the plotted amygdala nuclei volumes after multiple testing FDR-adjustment, additional adjustment for BMI-SDS in follow-up RLMs, and after excluding all AN with leptin concentrations <LOD in a confirmatory analysis (online SM 2.4, Table S6).
Psychological Medicine Kappou et al., 2021;King et al., 2015;Zhang et al., 2018), a bilateral cluster located in the rostral-medial amygdala was particularly affected in AN as indicated by large effect sizes of group differences.This anatomical cluster comprised bilateral accessory basal and cortical nuclei, corticoamygdaloid transition, and left medial nucleus.Critically, the magnitudes of volumetric reductions in these subregions were more extensive relative to those observed for the whole amygdala and other subcortical GM volumes.In contrast, despite absolute volumetric reduction, basal nucleus rh as well as bilateral lateral and paralaminar nuclei were less affected in AN relative to whole amygdala and total subcortical GM volume reductions.These findings demonstrate differential alterations of individual nuclei within the amygdaloid complex suggestive of locally differing vulnerability to the effects of AN.Of note, lower leptin levels in patients with AN independently predicted greater volumetric reduction of amygdala nuclei within the rostral-medial cluster.This suggests an underlying or modulating role of hypoleptinemia, resulting from severe underweight, in relation to amygdala substructure alterations in AN, which might have relevant clinical implications.Amygdala nuclei alterations observed in our study seem robust: they were largely independent of potential confounders including variation in demographic variables SES, IQ, handedness, and cigarette smoking (main results confirmed except for three nuclei outside of the rostral-medial cluster), AN subtype, co-existing psychiatric diagnoses, antidepressant use, and hydration status (Frank et al., 2018;King et al., 2018).Although generic amygdala substructure reductions may be expected informed by previous research on subcortical GM and whole amygdala alterations in AN (Bernardoni et al., 2016;Burkert et al., 2019;Eynde et al., 2012;Friederich et al., 2012;King et al., 2015;Monzon et al., 2017;Seitz et al., 2014Seitz et al., , 2016;;Su et al., 2021;Titova et al., 2013;Walton et al., 2022), degrees of absolute reduction of nuclei volumes within the amygdaloid complex appear rather heterogenous than uniform in AN as revealed by our study.In fact, they ranged from almost 11% (medial nucleus lh) to less than 2% (paralaminar nucleus rh) and showed focal maxima in subregions anatomically aggregated in the bilateral rostral-medial amygdala.Recent studies in major depressive disorder (Yao et al., 2020), obsessive-compulsive disorder (Zhang et al., 2020), posttraumatic stress disorder (Morey et al., 2020), and schizophrenia (Barth et al., 2021;Tesli et al., 2020;Zheng et al., 2019) have also documented differential alterations of amygdala subregions but spatial patterns differed from our findings.Importantly, despite limited comparability due to differing etiology, the medium-to-large effect sizes of amygdala nuclei reductions in acute AN found here substantially exceed the small effects previously reported under other neuropsychiatric conditions where widespread GM alterations are well-established, such as schizophrenia (van Erp et al., 2016).However, AN-related alterations in amygdala substructures do not reach the magnitude of chronic amygdala nuclei atrophy in neurodegenerative disorders such as Alzheimer's disease (20-30% tissue loss) (Cavedo et al., 2011) or frontotemporal dementia (35-50% volume loss) (Bocchetta, Iglesias, Cash, Warren, & Rohrer, 2019).
As anticipated based on evidence indicating that global GM reduction in acute AN rapidly returns to normal levels during short-term weight restoration (Bahnsen et al., 2022;King et al., 2018;Walton et al., 2022), we found that lower BMI-SDS in AN predicted smaller volumes of left hemispheric accessory basal nucleus and corticoamygdaloid transition (small effect size).Remarkably, the volumes of all rostral-medially clustered amygdala nuclei were more clearly associated with leptin levels than BMI-SDS in AN, hinting at specific leptin effects above and beyond the degree of underweight.In other words, leptin explained additional/other variance components of aforementioned amygdala nuclei volumes that were not attributable to BMI-SDS variation.This finding indicates predictive relevance of AN-related hypoleptinemia (Föcker et al., 2011;Hebebrand et al., 2007Hebebrand et al., , 2022;;Lawson & Klibanski, 2008) for the severity of rostral-medial amygdala nuclei reductions.Hence, we speculate that hypoleptinemia, in consequence of severe underweight, might be a possible (patho-)mechanism causally underlying and/or modulating the degree of amygdala substructure alterations in AN.This could, in a broader sense, offer novel insight into the neurobiology of striking and widespread, yet poorly understood, GM alterations in AN (King et al., 2018;Scharner & Stengel, 2019;Treasure et al., 2015;Zatorre et al., 2012).Basic research lends some support to our speculation by indicating leptin receptor expression in accessory basal and other amygdala nuclei and brain regions connected with baso-/corticomedial amygdala subdivisions (Fig. 1a) (Wada et al., 2014).Leptin signaling in the amygdaloid complex itself and innervating midbrain dopaminergic neurons interacts with mesolimbic reward pathways (Coccurello & Maccarrone, 2018;Fernandes et al., 2015), modulates anxiety-related behaviors (Liu, Perez, Zhang, Lodge, & Lu, 2011) and conditioned taste aversion (Han, Yan, Luo, Liu, & Wang, 2003) in rodents.Leptin also affects neural activity in the olfactory bulb (Sun et al., 2019) closely communicating with corticomedial amygdala nuclei (LeDoux, 2007;Sah et al., 2003).Furthermore, leptin promotes neurogenesis and synaptic plasticity in the amygdala-hippocampus formation in rodents (Bouret, 2010;Ge et al., 2018;Lu et al., 2006;Schepers et al., 2020).Translating to human research, leptin levels correlate with whole amygdala volume in female/male adults independent of BMI (Zonneveld et al., 2021) as well as with olfactory sensitivity (Fernandez-Garcia et al., 2017) and mood (Lawson et al., 2012) in females across the weight spectrum.According to case reports, leptin administration in leptin-deficient patients increased regional GM tissue concentration (Matochik et al., 2005) and, further, affected functional amygdala activation (Frank et al., 2013(Frank et al., , 2011)).In AN specifically, where the impact of hypoleptinemia on diverse neuroendocrine systems and AN-related symptoms is wellestablished (Ehrlich et al., 2009a(Ehrlich et al., , 2009b;;Hebebrand et al., 2007Hebebrand et al., , 2022;;Lawson & Klibanski, 2008;Schneider et al., 2009), a greater leptin increase during weight restoration predicted less rumination about food at discharge independent of BMI (Fürtjes et al., 2018).Leptin might modulate food-related cognition in AN via its interactions with accessory basal and medial amygdala nuclei (Petrovich, 2011).
Clinical trials on recombinant human leptin administration in patients with AN have recently been encouraged (Hebebrand et al., 2022(Hebebrand et al., , 2019)), given tolerability and potential beneficial effects of metreleptin on cognitive, emotional, and behavioral functions in AN as reported in case studies (Antel et al., 2021;Milos et al., 2020).Hyperactivity, mood, rumination about food, weight phobia, and even appetite/hunger seem to improve under metreleptin treatment applied in combination with inpatient nutritional rehabilitation (Antel et al., 2021;Milos et al., 2020).The associations between suppressed leptin levels and anatomically specific, severe GM reductions in the rostralmedial amygdala in AN may provide a novel mechanistic explanation for proposed metreleptin effects on core symptoms of AN considered to be linked to (dys-)functions of the severely affected amygdala nuclei unveiled by our study (Janak & Tye, 2015;LeDoux, 2007LeDoux, , 2008;;Sah et al., 2003).However, given hypoleptinemia forms a protective adaptation to chronic starvation signaling the need to reduce energy expenditure and increase caloric intake (Hebebrand et al., 2007), possible negative effects of metreleptin treatment on body weight (Welt, Smith, & Mantzoros, 2004) may limit therapeutic use in AN and, therefore, need to be closely monitored in studies.
Less severe volumetric reductions (i.e.relative to whole amygdala and total subcortical GM volumes) of basal, lateral, central, and paralaminar amygdala nuclei might represent preserved or at least residual structural (and potentially functional) integrity.Human amygdala volume and neuron number normally increase by 40% and 11%, respectively, during adolescence (Avino et al., 2018) when AN episodes usually interfere.However, immature excitatory neurons in paralaminar nucleus migrating to basolateral nuclei well into adulthood have been revealed in post-mortem human amygdala specimens (Sorrells et al., 2019).Thus, our finding that paralaminar nucleus seems more resistant to the effects of AN might correspond to mostly unaffected neuronal plasticity within the amygdaloid complex in AN.
This is a sufficiently powered study in a large and homogeneous (regarding age, biological sex, ethnicity, IQ, eTIV) sample of AN and HC participants with sMRI acquisition at the same scanner following a standardized protocol throughout the entire study.However, our findings should be considered in the light of the following limitations: first, despite overall good test-retest reliability of FreeSurfer-based amygdala subsegmentation in within-and across-scanner comparisons (Kahhale, Buser, Madan, & Hanson, 2020;Morey et al., 2009;Quattrini et al., 2020), the applied FreeSurfer tool was designed using adult postmortem and in-vivo brain samples (Saygin et al., 2017).Its performance is less well-established for adolescent brains (Schoemaker et al., 2016).We addressed this issue by developing a thorough QC procedure with extensive visual inspections and a-priori scan quality checks (Backhausen et al., 2016;Gilmore, Buser, & Hanson, 2021).Second, good numeric but lower spatial reliability has been proposed for single, smaller amygdala nuclei like medial and paralaminar nuclei (Kahhale et al., 2020;Quattrini et al., 2020).Thus, either cautious interpretation of corresponding FreeSurfer outputs, higher-resolution (>3 T) T1-weighted and additional T2-weighted scans, or more macrolevel amygdala nuclei groupings (Oshri et al., 2019) have been suggested (Kahhale et al., 2020).Nonetheless, we analyzed amygdala nuclei individually and still recognized prominent bilateral groupings of severely reduced nuclei in AN bearing in mind that replication by future studies is needed to substantiate our results.Third, more sensitive measurement methods for leptin are available and could reduce left-censoring in AN [e.g.singlemolecule array technology (Quanterix, 2016)].We conducted CLMI, yielding efficient unbiased parameter estimates (Boss et al., 2019), to keep valuable data of AN with leptin concentrations <LOD of the leptin assay in our sample and preserve their natural variability.Finally, to clearly identify whether amygdala nuclei alterations form state-related reversible phenomena of AN or constitute trait markers, longitudinal investigations over the course of weight restoration and after long-term weight recovery are essential.
In conclusion, this study provides evidence for heterogeneous subregional alterations within the amygdaloid complex in acute AN showing local specificity in rostral-medially clustered amygdala nuclei of both brain hemispheres.Volumetric reductions in rostral-medial amygdala nuclei exceed other psychiatric disorders (e.g.schizophrenia) and even AN-related whole amygdala and total subcortical GM volume reductions.These anatomically Psychological Medicine

a
SES was determined according to the parental (household) educational level/occupation group [range: 0 (lowest), leaving school without graduation -5 (highest), graduating from university](Patrick et al., 2004), given most study participants were adolescent, current students at school, university, or professional training institutions (AN: 83.33%, HC: 77.38%) and still lived with their parents or guardians (AN: 85.71%, HC: 68.45%).See online SM 1.2 for further details.b Median ± interquartile range (IQR) are shown for SES (ordinal scale), and group differences in SES were tested using Wilcoxon rank-sum test with continuity correction (W, p value, and effect size estimate Wilcoxon's r are stated as test statistics).6292Marie-LouisWronski et al.substantial interrater reliability (κ = 0.76, online SM 1.4).Participants with misapplied amygdala/hippocampus subsegmentation were excluded [3.35% overall which is similar to previous hippocampal subfield studies (3.50%)(Sämann et al., 2020), online SM 1.4, Fig.S1, Table

Fig. 2 .
Fig. 2. Bar graph visualization of GLM1 (a, whole amygdala volume-adjusted) and GLM2 (b, total subcortical GM volume-adjusted) for individual amygdala nuclei volumes in AN v. HC.AN, patients with acute anorexia nervosa; HC, healthy control participants; lh, left brain hemisphere; rh, right brain hemisphere; subcortical GM volume, total subcortical gray matter volume; GLM, general linear model.Bar graphs with error bars for study groups AN (n = 168) and age-matched HC (n = 168) displaying adjusted means (EMM, mm 3 ) ± standard error of the mean (S.E.M.) of individual whole amygdala and amygdala nuclei volumes in separate brain hemispheres.Model estimates were obtained with either main GLM [performed separately for each amygdala (sub-)region, computed as F test: dfs = 1, 330].(a) GLM1 covarying for age at date of research (linear and quadratic orthogonal polynomials), eTIV, and whole amygdala volume (lh, rh).(b) GLM2 covarying for age at date of research (linear and quadratic orthogonal polynomials), eTIV, and total subcortical GM volume [covariates in GLM1/2 (panels a/b) were grand mean-centered].FDR-q, p values were multiple testing adjusted using false-discovery rate(Benjamini & Hochberg, 1995) across all amygdala nuclei and both GLM1 and GLM2 (whole amygdala adjusted separately using FDR but also across both GLMs).Significance levels for volume differences between study groups are stated as: ***q < 0.001; **q < 0.01; *q < 0.05; ns, non-significant.Effect size statistics are provided as partial η 2(Cohen, 1988).

Table 1 .
Demographic variables and clinical measures AN, patients with acute anorexia nervosa; HC, healthy control participants; SES, socioeconomic status; IQ, intelligence quotient; BMI, body mass index; BMI-SDS, body mass index standard deviation score; BMI min , minimum lifetime BMI; log 10 -leptin, logarithmically transformed (base 10) leptin concentration; eTIV, estimated total intracranial volume; subcortical GM volume, total subcortical gray matter volume; EDI-2 core, averaged score comprising the core subscales 'drive for thinness', 'body dissatisfaction', and 'bulimia' of Eating Disorder Inventory-2; BDI-II, Beck Depression Inventory-II; STAI, State-Trait Anxiety Inventory (for participants aged ⩾15 years); STAI(K), State-Trait Anxiety Inventory for Children (for participants aged <15 years); SCL-90-R GSI, Global Severity Index of the Symptom Checklist-90-Revised. Number of participants and mean value ± standard deviation (S.D.) for each variable and study group (AN, HC) are shown.Group differences were tested using two-sample t tests (age-, age 2 -, and eTIV-adjusted GLM for total subcortical GM volume).As test statistics, t value (absolute value), degrees of freedom (df), p value, and effect size estimate Cohen's d