Introduction

In the central nervous system, myelin is a cholesterol rich extension of oligodendrocytes that coats axons, i.e. the long processes of neurons1,2. Axon myelination in the nervous system enables fast, saltatory electrical impulse propagation through action potentials modulation. Myelin can be thought of as an insulator, which boosts the conduction speed of the action potential along the axonal body, increasing the neural signal transmissibility from neuron to neuron, ultimately sustaining adequate cerebral intra-and-inter regional communication2. Motor, sensory, and cognitive functions of the nervous system all require rapid impulse propagation relying on an adequate axonal myelination2,3. Not only does myelin surround white matter bundles, it also plays a key role in cortical gray matter. Intracortical myelin is most prevalent in the deeper layers of the cortex, and it seems to enhance cortical function by fine-tuning timing and synchrony of neural networks4. Being particularly sensitive to experience3 and in continuous development throughout life5, intracortical myelin participates in brain plasticity and remodeling. This is evidenced by the fact that myelin content is associated with cognitive function, and that higher levels of intracortical myelin are associated with better performance on cognitive tasks5. Therefore, it is safe to assume that an abnormal concentration of intracortical myelin may play a role in psychiatric disorders. This is the case with schizophrenia, where a loss of intracortical myelin has been shown3,6,7.

Animal studies have provided consistent support for an adverse impact of undernutrition on myelination, given the potentially detrimental effects of long-lasting starvation on white matter (WM) composition8,9. Despite the lack of equivalent investigations in humans, myelin seems to be sensitive to micro-and-macro nutrient deficiencies: in fact, evidence in the literature suggests that iron is directly involved in myelin production as a required co-factor for cholesterol and lipid biosynthesis10. Authors have suggested that inflammation, oxidative stress, and loss of oligodendrocytes might all play a role11,12,13.

Anorexia nervosa (AN) is a severe psychiatric disorder with one of the highest long-term mortality rates of all psychiatric disorders14,15,16. It affects predominantly women, emerging typically during adolescence and early adulthood17. The core features of the disease are an intense fear of gaining weight, a body-image distortion leading to weight loss, behaviors and strategies aimed at weight loss and resulting in a state of severe malnutrition18. There is growing evidence of the impact of AN on the brain, both on structure and function19,20,21,22,23,24,25. In particular, WM microstructure alterations, gray and WM volume abnormalities have been shown in patients with AN in both acute and chronic phases20,21,22,26,27,28,29.

Given that animal models suggest a myelin decrease following malnutrition, the evidence of intracortical myelin abnormalities in other psychiatric disorders, and its maturational pattern (that extends into late adolescence and adulthood)30,31,32, it is reasonable to hypothesize a myelin deficit in patients with AN. However, since in AN WM investigations have been almost-exclusively carried out with standard Magnetic Resonance Imaging (MRI) acquisition sequences, a direct quantitative detection of myelin concentration in-vivo has been challenging33.

New techniques based on ultra-high field MRI imaging34,35,36,37,38 may address this gap in the literature, allowing visualization of the brain at submillimeter resolution39 and examination of myelin in a more fine-grained fashion. In view of the fact that the human cortex is approximately 2–4 mm thick, standard spatial resolution (1 mm) does not allow for proper detection of intracortical myelin40. Moreover, evidence shows that depth-dependent intracortical myelin properties obtained by ultra-high field MRI are reliable and consistent with ex-vivo data34,39,40. Based on this premise, in our preliminary study we aimed to measure intracortical myelin in patients with AN and controls, capitalizing on a newly validated method able to ascertain depth-dependent intracortical myelin organization with an ultra-high field 7 T scanner39,40. Specifically, using an ad-hoc acquisition sequence and data analysis procedure, in-vivo T1 values (in milliseconds) as inverse measure of intracortical myelin39 were extracted and compared in a sample of women with AN and healthy controls. To our knowledge, this type of investigation of cortical myelo-architecture has never been carried out in AN. Moreover, the use of quantitative MRI to measure actual tissue parameters (such as T1 values) prevents the risk of tissue or hardware-related bias that may impact generalizability of results41. We hypothesized that individuals with AN would show higher T1 values (i.e., lower intracortical myelin concentration) compared to controls. We also explored whether the differences in intracortical myelin concentration would be particularly visible in the outermost cortical vs. innermost depth-levels in AN. We further investigated any relationships between disorder-related variables and intracortical myelin.

Results

Socio-demographic and illness-related data

Individuals with AN did not differ from controls in terms of age. As expected, individuals with AN had a significantly lower Body Mass Index (BMI) compared to controls (Welch t = 6.057, p < 0.001) (Table 1). Osmolality levels were within the normal range for 4/7 (57.1%) individuals with AN, while three patients had slightly higher values than the normal range42. Duration of illness in individuals with AN ranged between 1 and 21 years. No current psychiatric comorbidity, or past psychiatric diagnosis was identified. Only one participant reported psychotropic medication use (venlafaxine).

Table 1 Descriptive statistics:
Full size table
Table 2 Global T1 values across individuals with AN and controls: Analysis of Covariance.
Full size table

Global T1 values

Participants with AN and controls had equal variances distribution for total T1 values (in milliseconds) (Levene’s Test p > 0.05). A statistically significant difference emerged between groups for T1 values (F(1,15) = 6.95, p = 0.03, η2 = 0.29). In detail, patients with AN showed higher T1 values than controls (Mean Difference MD = 48.55 ms, Standard Error SE = 19.80, 95% Confidence Intervals (C.I.) = 6.34, 90.77, Cohen’s d = 1.26). These differences also emerged in the bootstrap post-hoc comparison (MD = 48.28 ms, Bias = −0.406, SE = 17.29, 95% bias corrected accelerated C.I. = 15.28, 85.14). See Table 2 and Fig. 1b for all details. A graphical representation of total T1 values maps is available in Fig. 1a. Images were rendered with the ggsegDesterieux (https://github.com/LCBC-UiO/ggsegDesterieux) and ggseg43 (https://lcbc-uio.github.io/ggseg/) packages via RStudio v. 1.2.50. working on R v.4.0.4.

Figure 1
figure 1

(a) Standardized total T1 values map for AN patients (top) and controls (bottom). Maps are based on the Destrieux’s Atlas and show T1 values’ average across 10 depth-levels per each Region-of-Interest (ROI) in AN patients and healthy controls. For illustrative purposes, the distribution of T1 values was rescaled to a distribution with mean = 0. (b) Boxplot with violin and jitter elements displaying distribution of T1 values (in ms) as intracortical myelin concentration index averaged across all 10 cortical depth-levels for AN patients and controls; (c) boxplot with violin and jitter elements displaying distribution of inner cluster’s T1 values (in ms) for AN patients and controls; (d) boxplot with violin and jitter elements displaying distribution of outer cluster’s T1 values (in ms) for AN patients and controls; (e) scatterplots between BMI and T1 values (in ms) in all depth-levels, inner and outer clusters in AN patients and controls. For illustrative purposes groups are shown separately, although the statistical significance symbol refers to the analysis on overall population. T1 values distributions between AN patients and controls and relationship between T1 values and BMI. *p < 0.05; +p = 0.05.

Full size image

Based on the structural indications provided by the Principal Component Analysis (PCA), we averaged the T1 values from depth levels 1 to 5 independent of ROIs to form a composite measure of intracortical concentration in an “inner cluster” (i.e., cluster extending from the middle of the cortical ribbon to the gray matter/white matter boundary). The corresponding measures from depth levels 6 to 10 were also averaged independent of ROIs to generate a composite measure of intracortical concentration in an “outer cluster” (i.e., cluster extending from the middle of the cortical ribbon to the gray matter/CSF boundary). The mean cortical concentration of the inner and outer clusters were included in two separate set of analyses. See Table 3 and Fig. 1c,d for details about inner and outer cluster distributions.

Table 3 T1 values in the inner and outer cluster across individuals with AN and controls: Analysis of Covariance.
Full size table

Inner cluster

Patients with AN and controls had an equal variance distribution in T1 values within the inner cluster (Levene’s Test p > 0.05). The ANCOVA reported a difference approaching statistical significance between groups (F(1,15) = 4.32, p = 0.05, η2 = 0.22). In the inner cluster, AN patients showed higher T1 values in respect to controls (MD = 44.07 ms, SE = 21.20, 95% C.I. −1.11, 89.26, Cohen’s d = 1.07). This result was confirmed by a bootstrapped post-hoc comparison based on 1000 replications (MD = 43.33 ms, Bias = −2.08, SE = 19.26, 95% bias corrected accelerated C.I. = 8.93, 86.58).

Outer cluster

Both groups reported equal variance distribution in T1 values within the outer cluster (Levene’s Test p > 0.05). The ANCOVA showed a significant effect of group on T1 values (F(1,15) = 7.61, p = 0.01, η2 = 0.34). In outer cluster, individuals with AN showed significantly higher T1 values than controls (MD = 53.03 ms, SE = 19.29, 95% CI = 12.05, 95.01, Cohen’s d = 1.42), confirmed by a bootstrapped post-hoc test (MD = 51.29, Bias = −1.46, SE = 15.93, 95% bias corrected accelerated C.I. = 26.66, 91.62).

Correlations between illness-related variables and intracortical myelin

At the general level BMI was negatively correlated with global T1 values (rho = −0.68, p = 0.003, C.I. = −0.87, −0.31), T1 values in the inner (rho = −0.57, p = 0.017, C.I. = −0.82, −0.14) and outer cluster (rho = −0.75, p < 0.001, C.I. = −0.90, −0.44). Visually, such correlation seems more prominent in AN patients, although within group analyses did not reveal any significant correlations (p > 0.05) (Fig. 1e for scatter-plots). In addition, we explored the relationship between intracortical myelin concentration, osmolality and duration of illness in AN individuals. Osmolality marginally correlated with total myelin concentration (rho = 0.79, p = 0.05, C.I. = 0.08, 0.97) and with myelin in the inner cluster (rho = 0.89, p = 0.01, C.I. 0.43, 0.98), while it was less correlated to myelin concentration in outer cluster (rho = 0.50, p = 0.27, C.I. −0.41, 0.91). Duration of illness did not correlate with total intracortical myelin concentration, inner and outer cluster myelin concentration (all p > 0.05).

Discussion

In this preliminary study, we used an innovative method based on 7 T ultra-high field MRI to examine depth-dependent intracortical myelin in a population of women with AN. In accordance with our hypothesis, women with AN showed higher T1 values across the cortical ribbon. These differences were stronger in the five outermost intracortical depth-levels, whilst they were less evident in the five inner intracortical layers. Such results may index an intracortical myelin deficit in women with AN compared to healthy controls.

The use of a high-resolution MRI scanner allowed for the first time to use T1 values as a proxy to investigate depth-dependent intracortical myelin organization in-vivo. Our results are consistent with previous findings of reduced myelin concentration in AN27,44, and go beyond these, showing that such a deficit is widespread and mainly affects the outer depth-levels dependent intracortical myelin organization. Travis and colleagues27 reported a relationship between decreased fractional anisotropy (FA, defined as a marker of WM integrity) and decreased relaxation time R1 (considered as an index of myelin content) in AN adolescents. This suggests that observed WM differences in AN might be due to reductions in myelination10,45. However, as stated recently by Barona and colleagues in their meta-analysis46, differences in WM could be due to several factors, including a larger axon diameter, lower packing density of fibers, altered membrane permeability as well as a myelin deficit47. In fact, Travis and colleagues27 also underlined that the most significant WM disruptions occurred where the axons showed the largest diameter and a thicker myelin shell (i.e. the corticospinal tract and the corpus callosum)21.

Whereas in our study higher T1 values were found in patients with AN, in a recent study44 performed with a 3 T MRI scanner an opposite pattern emerged, i.e., patients with AN showed lower T1 relaxation times (i.e. time constant indicating how long the protons need to return to the equilibrium state41) than controls in both gray matter and WM. Nonetheless, the authors suggested that these differences could be related to starvation-driven myelin loss44. Although the two interpretations seem counterintuitive, it is worth highlighting that T1 relaxation time is strictly dependent on the physical properties of the underlying tissue41, including free water content48,49 and concentration and types of macromolecules50 such as iron content49 and myelin51. Thus, to date, such methodologies based on in-vivo quantitative MRI do not allow confirmation that between group differences may be due to various factors other than myelin content. Nevertheless, it is known that while water content is directly proportional to T1 time, iron content and myelination are inversely related to T141. This is confirmed by the fact that T1 relaxation time is generally longer in WM, which is myelin dense41. While Boto and colleagues44 did not exclude the impact of water content in the measurement of relaxation times, in our study hydration level was not a confounder. Moreover, the relationship between T1 and myelin has also been confirmed by ex-vivo data34,39,40,51, even in populations characterized by strong myelin loss such as multiple sclerosis52,53.

It is known that myelin assembling persists during development and in adult life54,55,56,57. Animal studies highlight that the process of myelin lipid remodeling is continuous, and that the rate of change of such lipids is different throughout life58,59. Myelin membranes are known to have a very high lipid-to-protein ratio, with lipids accounting for at least 70% of the dry weight2. Myelin construction therefore requires a large amount of lipids, including cholesterol, galactolipids, plasmalogen, and fatty acids60. In consideration of the reduced caloric intake, decrement in intracortical myelination in AN may mirror oligodendrocytes’ dysfunctions due to lack of micronutrients necessary for the synthesis of specific lipids. Nevertheless, further studies targeted on AN are needed to confirm such a claim.

A myelin deficit may also play a fundamental role in the symptomatic manifestations of AN. Animals studies suggest intracortical myelin is related to cortical functions and behavior61, and changes in myelin preceded changes in behavior54,55. Therefore several authors have suggested that intracortical myelin might be a marker of plasticity in the cortex40,61. Myelin damage, demyelination or oligodendrocyte loss results in a global malfunction in many neurological conditions such as multiple sclerosis, leukodvstrophies, Alzheimer’s disease, stroke, cerebral palsy, traumatic brain-spinal injury and cognitive decline62,63. Therefore intracortical myelin loss might contribute to the neurocognitive deficits of patients with AN such as cognitive flexibility, weak central coherence, emotional processing difficulties64.

Further investigations will need to study closely the relationship between neurocognitive differences and myelin concentration in AN.

Strengths and limitations

This preliminary study is the first to investigate depth-dependent intracortical myelin organization in individuals with AN. Using the high spatial resolution afforded by 7 T imaging in conjunction with a previously validated method for intracortical myelin estimation40 we obtained information about T1 values used as a proxy for intracortical myelin concentration across the cortical ribbon. Given the small sample size and the lack of specific a-priori assumptions, we chose not to focus on specific ROIs and this limitation should be addressed in future studies. The small sample size represents the main limitation but the current findings albeit preliminary encourage further studies in larger and more fully characterized samples. Osmolality as an index of hydration was collected in AN patients only, precluding us from including it as a covariate in our main analysis. However, correlation between osmolality and myelin concentration in the outermost depth-levels was weaker than within the inner cluster in AN patients, and only two out of seven patients with AN had abnormal osmolality levels indicative of dehydration, hence it is unlikely that dehydration fully confounds our findings. At the time of the study, the patients were all in the acute phase of illness, but duration of illness varied across our sample. Although no relationship emerged between illness duration and variables of interest, there might have been heterogeneities in the clinical sample. Although we are aware of the aforementioned limitations, we believe that our study provides promising evidence of altered intracortical myelin organization in AN.

Conclusions

In this preliminary study, we leverage a novel method for the in-vivo examination of T1 values as an inverse index of intracortical myelin organization in AN. We found generally higher T1 values in women with AN compared to healthy women, mirroring a widespread lower myelin concentration. These differences might be due to malnutrition and starvation, and general brain atrophy, or might be trait markers of the illness. However, the specific mechanisms (nutritional or structural) that might lead to myelin loss, and most importantly if this loss is reversible, need to be elucidated further. Moreover given the newly hypothesized role of metabolic factors in the etiology of AN65, whether a premorbid abnormality in myelin structure exists needs specific investigation. In a more speculative fashion, our study also raises the possibility that cognitive, visuo-spatial (and other) difficulties seen in active AN might be the result of altered intracortical myelin. Despite limitations, our findings provide an important foundation on which to build future studies. If confirmed, these results could shed new light on the neural basis of AN and on brain effects of the illness.

Methods

Participants

The study was carried out at the Icahn School of Medicine at Mount Sinai, New York. Participants were recruited amongst patients from the Mount Sinai Eating and Weight Disorders Program, the Mount Sinai Psychiatry service and from the community via flyers and adverts. Initial screen for eligibility was conducted over the phone by a research coordinator. We included female participants, with an age range from 18 to 45, who were able to provide informed consent and who spoke and understood English fluently. Patients had to have: (a) a diagnostic statistical manual of mental disorders, 5th edn (DSM-5)66 diagnosis of AN; (b) a body mass index (BMI) between 15.5 and 18.5; (c) a history of food restriction of more than 1 year; (d) to be medically stable. Healthy controls were screened for eligibility and had to have no history of mental health disorders (including eating disorders) and no chronic medical conditions.

Exclusion criteria for all participants included any contraindication to the MRI examination (i.e., embedded or implanted metallic bodies and claustrophobia) and pregnancy.

Participants who met initial criteria for the study were invited for an in-person assessment. After providing informed consent, participants completed initial study measures and completed a urine pregnancy test. Seven female adults with current AN and 11 healthy female adults were included.

The study was approved by the institutional review committee of the Icahn School of Medicine at Mount Sinai (ISMMS). All research was performed in accordance with relevant guidelines and regulations. Patients and controls provided informed consent to participate.

Measures

Socio-demographic data were collected using a short questionnaire. Weight and height were measured objectively to calculate BMI (kg/m2) (Table 1). Data were collected on socioeconomic status, duration of illness, and current medications in individuals with AN.

The Eating Disorder Examination interview version16.0D (EDE67) was used for diagnostic purposes. The EDE is a semi-structured interview considered the “gold standard” for assessing eating disorder symptoms. The EDE assesses a range of eating disorder features and yields a global score of eating disorder symptoms (Table 1).

The SCID Screening Module68 was administered to screen participants for other psychiatric disorders conditions.

AN patients underwent a blood draw in order to establish their level of osmolality (i.e. an hydration index based on molecular weight/1 L water presence ratio expressed in milliosmoles per kilogram).

MRI acquisition

Patients and controls were scanned at the ISMMS using a 7 T MR scanner (Magnetom, Siemens Healthcare) with a 32-channel with a Nova head coil (Nova Medical, Wilmington MA). An ultra-high resolution MP2RAGE sequence was used to acquire whole brain T1-weighted images. Acquisition parameters consisted of 0.5 mm isotropic resolution, repetition time (TR) = 5000 ms, echo time (TE) = 5.75 ms, inversion time (TI) TI1/TI2 = 900 ms/2780 ms, 224 axial slices with slab thickness 11.5 cm, field-of-view 224.5 × 203 × 112 mm3, and slab selective excitation and flow suppression36. Bias-field corrected quantitative T1 maps and T1 images were obtained from images at the two inversion times (TI1/TI2).

High resolution intracortical myelin profiling using T1 maps

Image preprocessing followed the procedures developed and validated previously39,40. Skull-stripping, background-marking and alignment (using rigid 6 degrees-of-freedom registration, normalized mutual information) to a 0.4 mm MNI template using the CBS Tools (https://www.nitrc.org/projects/cbs-tools) were applied on the T1 maps, T1-weighted and T2 images. No data were discarded due to poor quality data. Then, images were corrected for field inhomogeneities and matched to a template image39 using 3D Slicer (https://www.slicer.org). Images were filtered with arteries and dura segmentation, corrected for partial volume effects and segmented into gray matter, white matter and cerebrospinal fluid (CSF) using the Topology-preserving, Anatomy-Driven Segmentation (TOADS)69 and Multi-object Geometric Deformable Model (MGDM)70 segmentation algorithms as integrated in the CBS Tools. The CRUISE algorithm71 was applied in order to extract the cortex. Such an algorithm is robust to white matter lesions72. Capitalizing on the CBS Tools’ Volumetric Layering module73, the cortex was then dissected into 10 cortical depth-levels using a volume-preserving approach. This level-set approach generates one 2-dimensional surface per layer containing T1 values at each surface location. Next, the 2-dimensional level-sets were transformed into 3-dimensional representations of T1 values using the CBS Tools’ Profile Sampling Module. This transformation represents each vertex T-values of each level-set surface as a column of identical T1 values perpendicular to the surface39,40. The final result of this pipeline is 10 volumes of cortical profiles, one for each depth-level, for each of 7 patients with AN and 11 healthy controls. The whole process was completed within each individual's native space.

The cortex was then parcellated in the native space using Freesurfer (https://surfer.nmr.mgh.harvard.edu/). This parcellation was based on the pre-processed T1-weighted images alone (independent of the cortex extraction described above) and was used to obtain 148 cortical regions-of-interest (ROIs) based on the Destrieux atlas, that is particularly suitable for this type of investigation by separating gyri from sulci73,74. In each participant and for each cortical ROI, we estimated the absolute T1 value as an inverse index of myelin concentration at each depth-level: so, we obtained a matrix consisting of 148 total rows (ROIs) and 10 columns (i.e. each intracortical myelin depth-level) per participant.

Statistical analyses

All demographic variables were individually checked for missing data or outliers. Normality and equality of variances were verified using Shapiro–Wilk and Levene’s tests. Group comparisons for age and BMI were carried out using two-sided Independent-samples t-tests.

With the aim of getting a single index of total intra-cortical myelin concentration per participant, we averaged the values of the 10 myelin depth-measures across all the ROIs. The mean total intra-cortical myelin concentration was compared between patients and controls using an analysis of covariance (ANCOVA) with the group as main factor and age as covariate. Further, a bootstrap post-hoc comparison based on 1000 replicates was carried out in order to get a more stable group comparison result.

A Principal Component Analysis (PCA) with an oblique rotation (Promax algorithm) was applied on the original T1 values matrix across all participants. This matrix contained of 2664 rows (148 myelin concentration values per participant, one value per ROI) and 10 columns (10 depth-levels). PCA was applied in order to reduce the number of computations and to deal with correlated predictors (i.e. T1 values between multiple cortical depth-levels). Bartlett’s Test of Sphericity has been used to check PCA assumptions. The number of components was established through a parallel analysis75 that compared the real eigenvalues obtained from the matrix to the eigenvalues obtained from a Monte-Carlo simulated matrix based on random data with the same size76. As output of parallel analysis, a scree plot of eigenvalues was used to set up a threshold of significance and reliability of components using the Kaiser criterion (i.e. eigenvalues > 1). Based on the number of components extracted, we averaged the T1 values in different analyses. Following PCA, independent comparisons between groups were carried out by means of ANCOVA with averaged T1 values as dependent variable, the group as the main factor, age as a covariate and a bootstrap post-hoc comparison based on 1000 replicates.

Finally, for exploratory analyses, we investigated relationships between T1 values, BMI, osmolality and duration of illness by means of non-parametric correlations (Spearman’s rho rank coefficient) with significant level set at p < 0.05 uncorrected.

All statistical analyses were carried out using JASP (https://jasp-stats.org/, v. 0.11).

Intracortical myelin clusters by principal component analysis

Bartlett’s Test of Sphericity resulted significant (X2 (26) = 19,657.73, p < 0.001), suggesting a redundancy in the T1 values’ matrix. Two components (RC1 and RC2) were detected by parallel analysis using an eigenvalue threshold = 1 (Kaiser criterion) (see scree plot, Fig. 1SM). Table 4 shows the contribution of each depth-measure in each component.

Table 4 PCA results. Loading coefficients for RC1 and RC2 components.
Full size table

Specifically, T1 values at depth levels 1 to 5 contributed to RC2, whereas those in depth levels 6–10 contributed to RC1 (a path diagram is available in Fig. 1SM). In case of participation in both components (as in the case of measures in depth levels 5 and 6), the layer was associated with the component whose expression was greater. The path diagram and scree plot available in Fig. 1SM visually confirm the contribution of each depth measure to RC1 and RC2 and the reliability of the number of components extracted from the PCA.