DNA methylation in people with anorexia nervosa: Epigenome-wide patterns in actively ill, long-term remitted, and healthy-eater women

Abstract

Objectives

Recent studies have reported altered methylation levels at disorder-relevant DNA sites in people who are ill with Anorexia Nervosa (AN) compared to findings in people with no eating disorder (ED) or in whom AN has remitted. The preceding implies state-related influences upon gene expression in people with AN. This study further examined this notion.

Methods

We measured genome-wide DNA methylation in 145 women with active AN, 49 showing stable one-year remission of AN, and 64 with no ED.

Results

Comparisons revealed 205 differentially methylated sites between active and no ED groups, and 162 differentially methylated sites between active and remitted groups (Q < 0.01). Probes tended to map onto genes relevant to psychiatric, metabolic and immune functions. Notably, several of the genes identified here as being differentially methylated in people with AN (e.g. SYNJ2, PRKAG2, STAT3, CSGALNACT1, NEGR1, NR1H3) have figured in previous studies on AN. Effects also associated illness chronicity and lower BMI with more pronounced DNA methylation alterations, and remission of AN with normalisation of DNA methylation.

Conclusions

Findings corroborate earlier results suggesting reversible DNA methylation alterations in AN, and point to particular genes at which epigenetic mechanisms may act to shape AN phenomenology.

Keywords:

• Anorexia nervosa
• epigenetics
• genetics
• eating disorders
• methylation

1. Introduction

Epigenetic mechanisms are believed to modulate gene expression, and in so doing, to influence phenotypic manifestations in the absence of changes in the DNA code (Szyf 2015). In the context of physical- and mental-illness phenotypes, epigenetic mechanisms may therefore have the potential to influence the progression from genetic susceptibility to full-blown illness, and back. Data implicate epigenetic processes in the aetiology of several mental illnesses, including eating disorders (EDs: (Booij and Steiger 2020; Steiger and Booij 2020)).

The best-studied of epigenetic processes, DNA methylation, involves the addition of methyl groups to DNA regions at which cytosine is followed by guanine (Razin 1998). Methylation generally blocks or suppresses gene expression (Razin 1998; Szyf 2015). Three available studies have examined epigenome-wide DNA methylation in people with AN. The first, by our group, measured methylation of DNA from leukocytes drawn from 29 women with AN and 15 women with no ED (Booij et al. 2015). Findings suggested altered methylation levels at genes pertinent to known psychological, metabolic and physical concomitants of AN—including HDAC4, involved in histone deacetylation (Sild and Booij 2019), TNXB and DSE, associated with Ehlers-Danlos syndrome, a connective tissue disorder that has been linked to EDs (Baeza-Velasco et al. 2021), and NR1H3, involved in lipid metabolism and inflammation (Zhao et al. 2021). Using the same methylation profiling technique, Kesselmeier et al. (2018) examined DNA methylation levels in whole blood DNA from 47 women with AN, 47 lean women without AN, and 100 population-based women. A probe near the CSGALNACT1 gene, linked to bone and cartilage development (Watanabe et al. 2010), was nominally associated with AN and, intriguingly, results corroborated hypermethylation at TNXB and altered methylation at NR1H3 in AN (although the direction of methylation change was opposite to that we reported). The third study, again by our group, measured methylation levels in leukocyte DNA from 75 women with AN, 31 showing stable remission of AN, and 41 being normal eaters (Steiger et al. 2019). Comparisons between the active and no ED groups showed 58 differentially methylated sites (Q < 0.01). Chronicity of illness correlated (usually inversely) with DNA methylation levels at 64 sites that mapped onto genes regulating neurotransmitters, insulin function and ageing. Intriguingly, findings in remitted participants suggested a normalisation of changes seen in those who were actively ill. TNXB and NR1H3 again figured among identified probes. This study extended our previous report, this time with expanded samples, including roughly double the number of individuals with active AN.

2. Materials and methods

2.1. Study design

Full methods, clinical evaluations, and laboratory procedures have been described in detail by Steiger et al. (2019). Participants included 145 women with an active DSM-5 AN diagnosis (AN-Active), 70 with AN restrictive subtype (AN-R) and 75 AN binge-purge (AN-BP) subtype. We also recruited 49 women who previously fulfilled DSM-5 criteria for AN but no longer met ED criteria, and who had maintained a self-reported body mass index (BMI) above 18.4 kg/m² for at least 12 months (AN-remitted). Finally, we recruited 64 women with no ED (NED) history. Diagnoses were based on interviews with the Eating Disorders Examination (EDE: Fairburn et al. 2008) or in 10 cases, clinical interviews conducted by specialist clinicians, complemented by results from the EDE Questionnaire (EDE-Q: Fairburn and Beglin 2008). The samples described here included 75 AN-Active, 31 AN-remitted, and 41 NED women present in our earlier report (Steiger et al. 2019).

Descriptive data on participants are provided in Table 1. Groups differed predictably on mean BMI and EDE-Q scores, with the AN-active group having significantly lower BMI and more ED symptoms than did either other group (p<.01). Although the remitted group had normal-range EDE-Q scores on average, its members had higher scores than did the NED group. Incidentally, AN-remitted participants were slightly older, on average, than were members of the other two groups. We controlled statistically for effects attributable to age, smoking and psychotropic medication use.

Table 1. Sample characteristics.

Variable	AN-active (n = 145)	AN-remitted (n = 49)	NED (n = 64)
Age (M ± SD)^†	24.73 (7.4)^a	27.39 (6.2)^b	24.50 (5.4)^a
BMI (M ± SD)^‡	15.05 (1.8)^a	21.86 (2.7)^b	22.59 (2.6)^b
Cigarette smoker, n (%)	30 (20.7%)	5 (10.2%)	7 (10.9%)
Psychotropic medication use, n (%)^‡	94 (64.8%)^a	17 (34.7%)^b	0 (0%)^c
AN subtype:
AN-R, n (%)	70 (48.3%)	N/A	N/A
AN-BP, n (%)	75 (51.7%)
Chronicity in months (SD)	95.63 (87.6)	98.69 (61.2)	N/A
EDE-Q
Global EDE-Q^‡	3.93 (1.2)^a	1.04 (0.8)^b	0.42 (0.5)^c
Restraint^‡	3.93 (1.7)^a	0.59 (0.8)^b	0.28 (0.6)^b
Eating Concern^‡	3.03 (1.4)^a	0.47 (0.8)^b	0.14 (0.4)^b
Shape Concern^‡	4.58 (1.2)^a	1.65 (1.3)^b	0.75 (0.8)^c
Weight concern^‡	4.16 (1.5)^a	1.41 (1.1)^b	0.51 (0.7)^c

Note: Means with different letters in their superscripts differed at p < 0.05.

^†p < 0.05; ^‡p = 0.001

AN: anorexia nervosa; NED: non-eating-disordered; BMI: body mass index; AN-R: anorexia nervosa, restricting type; AN-BP: anorexia nervosa, binge-eating/purging type; EDE-Q: Eating Disorder Examination Questionnaire; AN-active, n = 123; AN-remitted, n = 40; NED, n = 60. Chronicity, n = 144.

For genome-wide DNA methylation analyses, we collected whole blood in EDTA tubes and extracted DNA from leukocytes using a DNA extraction kit (Qiagen). Epigenome-wide analyses were conducted using the Infinium Human Methylation 450 BeadChip Kit (Illumina Inc.) available at the time of collection of our first 173 samples and later, the Infinium Methylation EPIC BeadChip Kit (Illumina Inc.) for a subsequent 96 samples. Although Illumina discontinued production of the 450 BeadChip kit during the early phase of our data collection, manufacturer information and an independent study (Moran et al. 2016) suggest that samples can be combined effectively across 450 and EPIC kits. We entered only those probes that are common to both the 450 BeadChip Kit and the Infinium Methylation EPIC BeadChip Kit into our analyses. Further technical details are provided by Steiger et al. (2019).

2.2. Statistical analyses

We compared groups on descriptive and clinical variables with chi-squared tests or one-way analysis of variance (ANOVA) as appropriate, using SPSS 27 (SPSS Inc.). Raw methylation data underwent preprocessing for quality control and functional normalisation. Probes showing random technical variation, as described in Steiger et al. (2019), were removed. We performed analyses on methylation levels with Matlab, using linear mixed models to compare groups at individual probes, with methylation levels as the dependent variable and group as the independent variable, and estimated cell proportions, batch, age, smoking, and use of psychotropic medication as covariates. We also tested the linear effects of chronicity and BMI within the AN group alone, and differences between AN subtypes. False Discovery Rate (FDR) correction (Q < 0.01) was applied throughout.

2.3. Data screening

The final dataset included 152,169 probes. Comparison of log2 median intensity of the methylated and unmethylated channels with Minfi PlotQC detected no outliers. Based on intensities of X and Y chromosomes, all participants were determined to be female. Probes that were SNP associated, cross hybridised, non-specifically bound, or affected by issues in detection p-values (>20% with p>.01) were not removed from our analysis—and 26,104 of the 152,169 analysed probes (17.15%) met at least one of the criteria listed. Although 455 probes had above-criterion detection p-values, none of these probes figured among our significant results. Likewise, as noted in footnotes in Tables S1, S2 and S3, a negligible number of probes were affected by the other issues noted. We also tested for possible genomic inflation effects, and corresponding QQ plots and lambda values are shown in Supplementary Figures S1 and S2.

3. Results

A preliminary analysis revealed no differences on methylation indices between AN-R and AN-BP subgroups [t>|4.84|, Q > 0.22]. We therefore limited our group comparisons to overall differences among AN-active vs. AN-remitted vs. NED groups. Across these three groups, linear mixed models showed differential methylation on 273 probes (representing 195 genes). Comparisons between the AN-active and NED groups revealed 205 differentially methylated sites (at Q < 0.01), and isolated genes relevant to metabolism and nutritional status (leptin sensitivity, insulin resistance, lipid metabolism), psychiatric status (HPA-axis regulation, glutamate) and immune function (immunoregulation and autoimmune diseases) (see Supplementary Figure S3, a). Notably, several of the genes identified – ZNF608, SYNJ2 (2 probes), GATA2, NOD1, PRKAG2 (2 probes), STAT3 (3 probes), CSGALNACT1, RPTOR (4 probes), NEGR1, DSE, and SP6 – have been associated with AN or other EDs in previous studies. Other relevant genes include IRS2 (2 probes) and JAZF1 (both involved in diabetes), ADD1 (involved in fat cell development and insulin metabolism), PPM1H (linked to Attention Deficit Hyperactivity Disorder), CSK (involved in bone health and immunity), AKAP13 (implicated in body weight regulation, autism and compulsivity) and CNIH3 (linked to schizophrenia).

Comparisons between AN-active and AN-remitted groups showed differential methylation on 162 probes (Figure S3, b). Ninety-four of the identified probes matched those found in the AN-active vs. NED comparisons (Q < 0.01), and again included ZNF608, SYNJ2, NOD1, PRKAG2, STAT3, RPTOR, NEGR1, SP6, IRS2, JAZF1, ADD1, PPM1H, AKAP13, AUTS2 and CNIH3. Consistent with our previous work, we also noted hypermethylation at the NR1H3 gene.

In striking contrast, AN-remitted vs. NED comparisons showed no differences on any of the probes identified in comparisons involving the AN-active group (Q > 0.74, see supplementary information Figure S3, c). In other words, remission appeared to coincide with restored methylation levels. A full list of significant findings that overlapped between active versus NED and active versus remitted comparisons at Q < 0.01 is provided as supplementary information in Table S1. Means and standard deviations of the probes that overlapped between AN-active versus NED and AN-active versus AN-remitted comparisons are listed in Table S2.

A final pair of analyses was performed in the AN-active group alone to assess associations between methylation levels and variables representing extent of exposure to the active eating disorder (Table S3 and Figure S4a). We identified 19 probes at which methylation was significantly associated with chronicity of illness (Q < 0.01), that mapped onto genes related to serotonin function (HTR2A), fertility (ZP2, NWD1), and immunity (TGFBR2, IGSF11, CASS4, FKBP5). On 95% of the probes (18 of 19), the direction of effect suggested lower methylation levels in people with more long-standing AN. A separate analysis, examining the association between BMI and methylation status, associated BMI with methylation levels at 20 probes—on 16 (80%) of which lower BMI corresponded to lower methylation (Table S3 and Figure S4b). Identified genes included modulators of stress responses (FKBP5), leptin receptor function (LEPR), oestrogen (PRMT1), and lipid metabolism (SCARB1, 2 probes). Several of the probes isolated here (FKBP5, ZNF608, KIAA0146, SCARB1, NCRNA00114) were also identified in the AN-active vs. NED or AN-active vs. AN-remitted comparisons described above.

4. Discussion

The present findings corroborate earlier results, but with expanded samples, indicating differential methylation in actively ill individuals when compared to people in remission from, or who have never had, AN (Steiger et al. 2019). Furthermore, findings in our ‘AN-remitted’ group are similar enough to those in our ‘no ED’ group to suggest a return to normal levels after remission. In other words, results imply the existence of state-related alterations in DNA methylation levels that can, encouragingly, be ‘reset’ with remission of illness.

Findings involving measures of illness chronicity and body mass index are compatible with the notion that many of the observed methylation-level changes may be illness sequelae—in the sense that severity of malnutrition (as reflected by low BMI) or duration of exposure to malnutrition (as reflected by chronicity of illness) both generally corresponded to lower methylation levels. The preceding would be consistent with the known role of dietary nutrients in supporting DNA methylation (Stevens et al. 2018).

Aspects of our findings bearing upon probe-wise differences offer intriguing indications concerning genes that may be particularly susceptible to epigenetic regulation in AN. First, we replicate a tendency seen in previous work suggesting that DNA methylation changes affect genes acting upon the mental status (e.g. AUTS2, GATA2, C1RL, VWF, FKBP5, CNIH3, PPM1H), metabolic functions (e.g. IRS2, RPTOR, TCAP, NEGR1, SYNJ2, ZNF608) and immunity (e.g. NOD1, FKBP5, CSK, DSE). We note that GATA2 reportedly influences serotonin function (Vadodaria et al. 2016), AUTS2 regulates neurodevelopment (Hori et al. 2022), C1RL stress reactivity (Föcking et al. 2021), and VWF post-traumatic reactions (Robicsek et al. 2011), whereas FKBP5 methylation is implicated in Hypothalamic–Pituitary–Adrenal-axis regulation (Park et al. 2019). Other identified genes have been implicated in metabolic regulation, including insulin responses (IRS2 and RPTOR), lipid metabolism (TCAP) and body mass (NEGR1, PRKAG2). Yet others have been linked to inflammatory processes (NOD1) and auto-immune disease (CSK, DSE). It is also noteworthy that several of our probe-wise findings corroborate reports from previous methylation studies in AN (Booij et al. 2015; Kesselmeier et al. 2018; Steiger et al. 2019). These include STAT3, associated previously with AN and with leptin signalling in ‘activity-based-anorexia’, a murine AN model (Hebebrand et al. 2019), CSGALNACT1, nominally linked to AN in a previous methylation study (Kesselmeier et al. 2018), PRKAG2, a gene linked elsewhere to binge-eating and obesity (Rodríguez-López et al. 2021), NEGR1, linked to weight regulation and obesity (Breton et al. 2020), and NR1H3 and SYNJ2, both involved in lipid metabolism and inflammation (Zhao et al. 2021; Iranzo-Tatay et al. 2022). We note, in contrast, that we do not replicate previous findings linking AN to altered TNXB methylation.

We add a comment on strengths and limitations of this study. Our ‘actively ill’ sample is larger than that in any other epigenome-wide methylation study in Anorexia Nervosa to date, and our design allows for a unique comparison between actively ill and remitted groups. Judicious interpretation of results is nonetheless warranted. Given tissue specificity of DNA methylation, the relevance of findings obtained using peripheral tissues to brain function may be dubious (Edgar et al. 2017). Furthermore, without measures of gene expression, we cannot be certain of the functional significance of altered methylation levels we observed. Finally, analyses examining genomic inflation factors suggest the possibility of some inflation of p-values in our data.

Cognisant of the limitations noted, we find the set of genes identified here to have an intuitive relevance to AN and its associated features. The patterning of findings lends itself to the potentially important interpretation that diverse aspects of phenomenology in AN may be epigenetically regulated. Furthermore, noted tendencies for methylation changes to coincide with the actively ill state, to be reversible with remission of symptoms, and to increase with longer chronicity of illness, all suggest that methylation changes have potential as markers of illness staging, entrenchment and recovery. Further research may, in addition, help determine whether methylation studies have a contribution to make in the effort to identify targets for therapeutic intervention (e.g. via pharmaceuticals or nutraceuticals) in AN.

Acknowledgements

None.

Statement of interest

No potential conflict of interest was reported by the author(s).

Data availability statement

Data upon which this study were based are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Additional information

Funding

Research described in this paper was supported by a grant from the Canadian Institutes of Health Research [CIHR; MOP-142717], on which Howard Steiger and Linda Booij are co-principal investigators. Linda Booij is supported by a salary award from the Fonds de Recherche du Québec. The authors are grateful for the assistance of Amy Côté-Croteau.