The Role of Circadian Rhythms and Sleep in Anorexia Nervosa

Key Points Question What is the association between anorexia nervosa, circadian rhythms, and sleep traits? Findings In this genetic association study, a bidirectional association between anorexia nervosa and morning chronotype was found using mendelian randomization. In addition, an association between anorexia nervosa and insomnia was found using mendelian randomization and a polygenic risk score for anorexia nervosa in a clinical biobank. Meaning These findings suggest that anorexia nervosa is a morningness eating disorder, in contrast to most other evening-based psychiatric diseases, and corroborate the association between anorexia nervosa and insomnia, concordant with earlier observational studies.


Introduction
The circadian clock is the intrinsic biological system that produces 24-hour cyclical processes called circadian rhythms. 1,2Circadian rhythms affect a wide spectrum of human physiology, including cycles in body temperature and sleep-wake patterns.Chronotype, the preference for morningness or eveningness, is a correlate of circadian rhythms and an individual's propensity for earlier or later timing for sleep and other activities and is associated with psychological, neurological, gastrointestinal, and respiratory disorders. 3,4ere is a bidirectional association between circadian rhythms and psychiatric disorders.For example, tendency for an evening chronotype is associated with depression and schizophrenia. 5,61][12] Characteristics of generally healthy adults with an evening chronotype, such as emotional eating, later food timing, and impulsivity, further support the role of eveningness tendency in binge eating-like behaviors. 13,14However, the relevance of the circadian system in anorexia nervosa remains largely unknown.Anorexia nervosa is a major eating disorder characterized by severe food intake restriction, weight loss, and intense fear of weight gain 15 and has among the highest mortality rates of all psychiatric diseases. 16Some evidence suggests the role of the circadian clock: for example, morningness tendency, with lack of appetite, of which the most prominent is the well-recognized morning anorexia and evening hyperphagia dichotomy that characterizes patients with night eating syndrome, 17 but also with evening preference in small observational studies. 18Additional evidence includes early morning awakening insomnia among people with anorexia nervosa, indicative of perturbations in sleep-wake cycles. 19Directionality of the association and whether any effects are causal or mediated by mood and anxiety pathophysiology that often cluster with anorexia nervosa remain unknown. 20netic analyses can further interrogate the link between anorexia nervosa and circadian and sleep traits.Genome-wide association studies (GWAS) for anorexia nervosa revealed several loci. 21,22ndelian randomization (MR) uses genetic variants that are robustly associated with a trait to examine potential causal effects on outcomes of interest. 23Due to the properties of genetic variants, MR is less likely to be affected by unmeasured confounding factors or reverse causation than observational studies.The application of MR has provided additional evidence supporting observational data on hypoleptinemia among patients with anorexia nervosa. 24,25It is also speculated that there is a bidirectional association between sleep and eating processes, such that eating pathology disrupts sleep, and conversely, dysregulation in sleep influences eating behaviors. 26e aim of this study was to investigate the bidirectional association between anorexia nervosa and chronotype and other sleep traits through MR and a polygenic risk score (PRS) for anorexia nervosa.

GWAS for Anorexia Nervosa
This study followed the Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization (STROBE-MR) reporting guidelines. 31The largest GWAS meta-analysis for anorexia nervosa from the Eating Disorders Working Group of the Psychiatric Genomics Consortium aggregated data from participants of European ancestry across 33 data sets and included 16 992 cases and 55 525 controls, of which 87.7% to 97.4% of cases were female, 49.6% to 63.4% of controls were female, and 5.3% of the samples were from the UK Biobank. 22The case definition of self-reported diagnosis.The GWAS identified 8 independent genome-wide significant (P < 5 × 10 −8 ) single-nucleotide variants (SNVs).SNV-based heritability was estimated at 11% to 17%, reflecting the polygenicity of anorexia nervosa.

GWAS for Self-Reported Chronotype and Sleep Traits
The primary outcome was chronotype.Summary statistics for the largest GWAS of European ancestry predominantly from the UK Biobank for chronotype (morning preference) was retrieved from the Sleep Disorder Knowledge Portal 27 (eMethods in Supplement 1).Four sleep traits were also considered secondary outcomes, including daytime napping, daytime sleepiness, and sleep duration from the Sleep Disorder Knowledge Portal 27 and insomnia from the Center for Neurogenomics and Cognitive Research in Amsterdam. 28Traits were determined using primarily self-reported data (eMethods in Supplement 1).

MR Primary Analyses
Bidirectional 2-sample MR between anorexia nervosa and chronotype and sleep traits was conducted using the largest publicly accessible GWAS studies in participants of European ancestry (eTable 1 in Supplement 2) using the "TwoSampleMR" R package.All data sources obtained participant informed consent and relevant ethical approval; as we used publicly available data which does not constitute human participant research, per the US code of federal regulations (45 CFR   46.102), no additional institutional review board approvals were necessary.The primary analysis considered chronotype (morning preference) and 4 secondary sleep traits: daytime napping, daytime sleepiness, insomnia, and sleep duration.For each trait, a genetic instrument was generated using lead GWAS variants (eTables 2-15 in Supplement 2), and standard data harmonization was conducted (eMethods in Supplement 1).The inverse variance weighted (IVW) regression method was used as the primary analysis.The IVW method yields an unbiased estimate in the absence of horizontal pleiotropy. 23,29A range of sensitivity analyses robust to the presence of horizontal pleiotropy were tested, including MR-Egger and weighted median estimator (WME) methods.
Consistency in effect sizes and overlap of 95% CIs were examined. 30A threshold of .05 was used for statistical significance in the MR for our primary analysis with chronotype and secondary analyses with sleep traits.All analyses were performed between February and August 2023.

MR Sensitivity and Secondary Analyses
For any significant finding from the IVW analyses, additional sensitivity analyses were conducted including MR pleiotropy residual sum and outlier (PRESSO), 32 leave-one-out analysis, 29 MR-Egger intercept, 29 I 2 quantification, 33 and a Steiger test 34 (eMethods in Supplement 1).Secondary analyses were conducted to examine the robustness of the findings including MR analyses for a binary selfreport chronotype variable and 3 objective measures of chronotype, 35 and MR using female-only summary statistics from the insomnia GWAS. 28

PRS for Anorexia Nervosa in the Mass General Brigham Biobank
The Mass General Brigham (MGB) Biobank is the health care enterprise clinical biobank from the MGB health care network in Boston, Massachusetts, and links electronic health records to genetics and lifestyle data.The present protocol was approved by the MGB institutional review board.Patients provided blood samples for genotyping with written consent (eMethods in Supplement 1).All genetic analyses were restricted to participants of European ancestry.A PRS for anorexia nervosa based on effect estimates from the recent meta-analysis GWAS for anorexia nervosa of European ancestry was calculated for each patient. 22The PRS was generated using PRS-continuous shrinkage (PRS-CS) based on bayesian regression that places a continuous shrinkage prior on SNV effect sizes. 36The UKBiobank EUR-ancestry linkage disequilibrium (LD) panel was set.A total of 1 734 712 SNVs were retained in the PRS following clumping.The PRS was standardized with a mean of 0 and an SD of 1.
Disease designation for anorexia nervosa and sleep disorders was based on PheCode classifications, which aggregate clinically relevant ICD-9 and ICD-10 billing codes into disease groups as described in the PheCode catalog 37 (eMethods in Supplement 1).Anorexia nervosa cases were based on the anorexia nervosa-specific code (305.21), which uses ICD-9 billing code 307.1 and ICD-10 billing code F50.0.Association analyses for the anorexia nervosa PRS were conducted for anorexia nervosa and sleep disorders with at least 1% prevalent cases, using the PheWAS R package. 38sociations were tested using logistic regression models with adjustments for age, sex, genotyping array, batch, and 10 principal components of ancestry to account for genetic substructure (minimal model), then further adjusted for body mass index, employment status, smoking status, exercise, alcohol intake, and education attainment (fully adjusted model).Significance was considered at the Bonferroni threshold for 9 tested sleep disease outcomes (P < .05/ 9 tests).
Enrolled patients were invited to complete an optional Health Information Questionnaire, which included questions on sleep habits (eMethods in Supplement 1).The following circadian and sleep variables from self-report were examined for associations with the PRS: time in bed, sleep debt, sleep midpoint, and social jetlag.Associations were tested using adjusted linear regression models as described previously, and in sensitivity analyses, further adjusted for PheCodes for depression (296.2) and anxiety (300.1), 2 prevalent comorbidities. 39Significance was considered at the Bonferroni threshold for 4 circadian and sleep outcome traits (P < .05/ 4 tests).All analyses were conducted using R software version 4.2.3 (R Project for Statistical Computing).
We performed a series of sensitivity analyses to further assess the presence of horizontal pleiotropy.Results from the MR-Egger intercept were not significant for horizontal pleiotropy with anorexia nervosa as the exposure or with chronotype as the exposure.Findings from MR PRESSO suggested possible pleiotropy with chronotype as the exposure (P < .001),however this was mitigated by removing 3 outlier variants, resulting in estimates consistent with the main analysis (β = 0.197; 95% CI, 0.067-0.327)(Figure 1; eTable 16 in Supplement 2).Assessment of horizontal pleiotropy using MR PRESSO was not significant for pleiotropy with anorexia nervosa as the exposure.Leave-one-out analysis did not suggest undue influence by single outliers or systemic bias (eTables 17-18 in Supplement 2).The calculated I 2 quantity was 49.6% with anorexia nervosa as the exposure and 48.2% with chronotype as the exposure.The Steiger test held true in both directions.
The observed direction of the association between chronotype and anorexia nervosa was consistent in secondary analyses using summary statistics for binary chronotype (morning vs evening preference) as the outcome (β = 0.014; 95% CI, 0.002-0.026)(Figure 1 and Forest plot of MR primary, sensitivity, and secondary results for the associations between anorexia nervosa and chronotype (morning), including accelerometer-derived objective chronotype measures of midpoint of least active 5 hours, midpoint of most active 10 hours, and sleep midpoint.β Values reflect the difference in the outcome for each 1-unit increase in the exposure, with blue boxes denoting statistically significant results.Error bars represent 95% CIs.Primary analyses included inverse variance weighted (IVW) MR, weighted median estimator (WME), and MR-Egger, and sensitivity analyses included MR pleiotropy residual sum and outlier (PRESSO).
We found no evidence for the association between anorexia nervosa and chronotype using accelerometer-derived sleep timing (Figure 1 and Table 2).However, MR direction of associations for risk of anorexia nervosa were consistent.

MR Analyses for Anorexia Nervosa and Sleep Traits
In MR, we found an association between insomnia and increased risk of anorexia nervosa, and no associations between anorexia nervosa and daytime napping, daytime sleepiness, or sleep duration (Table 1).Genetic liability for insomnia was associated with increased risk of anorexia nervosa using IVW MR (β = 0.339; 95% CI, 0.052-0.627);however, no association was observed for the genetic liability of anorexia nervosa with increased risk of insomnia (Table 1 and Figure 2).In sensitivity analyses, effect sizes were consistent using WME but not MR-Egger models.Results from the MR-Egger intercept were not significant for horizontal pleiotropy.However, results from MR PRESSO were significant (P < .001),and subsequently mitigated through removing 6 outlier variants resulting in estimates consistent with the main analysis (β = 0.449; 95% CI, 0.195-0.703)(Figure 2; eTable 16 in Supplement 2).Leave-one-out analysis did not suggest influence by single outliers or systemic bias (eTable 20 in Supplement 2).The I 2 quantity was calculated as 51.5%, and the Steiger test proved false, suggesting the direction of causality of insomnia on anorexia nervosa cannot be ascertained.
When using effect sizes from the female-only insomnia GWAS, we also found that genetic liability for insomnia was associated with increased risk of anorexia nervosa using IVW MR (β = 0.324; 95% CI, 0.015-0.632)(Table 2 and Figure 2).Results from the MR-Egger intercept were not significant for horizontal pleiotropy.However, results from MR PRESSO were significant (P < .001)and subsequently mitigated through removing 4 outlier variants, resulting in estimates consistent with the main analyses (β = 0.312; 95% CI, 0.021-0.603)(Figure 2; eTable 16 in Supplement 2).
Leave-one-out analysis did not suggest influence by single outliers or systemic bias (eTable 21 in Supplement 2).The I 2 quantity was calculated as 54.0%, and the Steiger test was also false.

Anorexia Nervosa in the MGB Biobank
To further examine associations between anorexia nervosa and sleep disorders in a disease-enriched and independent health enterprise clinical biobank, we conducted a series of analyses in the MGB Each SD increase in the PRS was associated with 36% higher odds of anorexia nervosa (odds ratio [OR], 1.36; 95% CI, 1.14-1.63)(Table 3).In association analyses with 9 prevalent sleep disorders, we found associations with higher organic or persistent insomnia.Each additional SD in the anorexia nervosa PRS was associated with 10% higher odds for organic or persistent insomnia (OR, 1.10; 95% CI, 1.03-1.17),which remained unchanged when accounting for lifestyle factors.Associations with other sleep disorders including sleep apnea and restless legs syndrome were not significant.In addition, among patients with self-reported sleep habits (n = 16 109), there were no associations evident between the anorexia nervosa PRS and time in bed, sleep midpoint, sleep debt, or social jetlag (eTable 22 in Supplement 2).

Discussion
Using MR, we provide evidence for bidirectional associations between anorexia nervosa and morning chronotype, which were consistent in a series of sensitivity and secondary analyses.We also found associations between insomnia and higher risk for anorexia nervosa concordant with earlier observational studies; however, these effect sizes were possibly confounded by psychosocial comorbidities.We did not find associations between anorexia nervosa and other sleep traits, including daytime napping, daytime sleepiness, and sleep duration.The bidirectional association between anorexia nervosa and morning chronotype proved robust to sensitivity analyses using alternative approaches to ascertain chronotype and to major bias due to pleiotropy.Results using accelerometer-derived measures of chronotype were consistent with findings from self-report implicating morningness with anorexia nervosa but were ultimately not significant likely due to limited power (85 670 participants in the accelerometer GWAS in comparison with 449 734 participants in the self-report GWAS).In a clinical biobank, we did not observe associations between the anorexia nervosa PRS and self-reported sleep midpoint, a measure of chronotype, possibly due to limited power or confounding comorbidities.As many as 62% and 54% of people diagnosed with an eating disorder also have an anxiety disorder and a mood disorder such as depression, respectively. 39Anxiety and depression, and most other psychiatric diseases, have consistently been associated with eveningness, potentially confounding previous findings using observational data. 40,41Furthermore, anorexia nervosa is most common at a younger age, peaking in onset around ages 15 to 16 years, when people experience a delay in chronotype toward eveningness. 42,43MR mitigates these confounding effects through the use of random genetic variation, allowing for robust analysis of the connection between anorexia nervosa and chronotype independent of psychosocial comorbidities and resolution of seemingly conflicting evidence from observational studies.
Our findings suggest that anorexia nervosa is another eating disorder with likely a circadian feature.First, in observational data, anorexia nervosa was associated with breakfast and lunch skipping, reinforcing the connection between anorexia and food temporality. 44Second, morning bright light therapy was deemed less effective in anorexia nervosa treatment compared with binge eating disorders and bulimia nervosa, 2 eating disorders presumed to be evening based. 7,9Third, compared with healthy controls, malnourished patients with anorexia nervosa have an enhanced and advanced cortisol awakening response often exhibited with earlier chronotype. 45,46Lastly, patients with anorexia nervosa often experience early morning awakening insomnia indicative of perturbations in sleep-wake cycles. 19Contradictory evidence from observational data linking anorexia nervosa with evening chronotype 18 or showing no associations with sleep timing 47 are small and may be confounded by psychosocial comorbidities.The precise role of the circadian system in the etiology of anorexia nervosa remains unclear.
In addition, we found a potential association between insomnia and increased risk of anorexia nervosa.Although findings from the IVW method were inconsistent in sensitivity analyses using MR-Egger, a low I 2 test statistic 48 indicated that the MR-Egger result was likely biased. 49Our findings provide evidence for a robust association between insomnia and anorexia nervosa without bias due to horizontal pleiotropy; however, the direction of causality cannot be proven.The results from MR were further corroborated in a clinical biobank using a PRS for anorexia nervosa, in which we observed positive associations between anorexia nervosa and organic or persistent insomnia.
Directionality cannot be inferred from this cross-sectional analysis.
anorexia nervosa was based on a lifetime diagnosis of anorexia nervosa from hospital or register records, structured clinical interviews, online questionnaires based on standardized criteria (ie, JAMA Network Open | Psychiatry Diagnostic and Statistical Manual of Mental Disorders [Third Edition Revised]; Diagnostic and Statistical Manual of Mental Disorders [Fourth Edition]; International Classification of Diseases, Eighth Revision; International Classification of Diseases, Ninth Revision [ICD-9]; or International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10]), or a Biobank.A total of 47 082 adult patients of European ancestry with genetic data were included in the MGB Biobank analyses (mean [SD] age, 60.4 [17.0] years; 25 318 [53.8%] female).We first examined associations between the anorexia nervosa PRS and anorexia nervosa diagnosis based on PheCodes.

Table 1 .
MR Primary Analyses Results for the Associations Between Anorexia Nervosa and Chronotype and Sleep Traits Abbreviations: IVW, inverse variance weighted; MR, mendelian randomization; SNV, single-nucleotide variant; WME, weighted median estimator.a β Values reflect the difference in the outcome for each 1-unit increase in the exposure.

Table 2 )
. In sensitivity analyses, there was no evidence of horizontal pleiotropy using the MR-Egger intercept test or MR PRESSO.Leave-one-out analysis did not suggest undue influence by single outliers or systemic bias (eTable 19 in Supplement 2).The I 2 quantity was calculated as 41.0%, and the Steiger test held true.Figure 1.Forest Plot of Mendelian Randomization (MR) Primary, Sensitivity, and Secondary Results for the Associations Between Anorexia Nervosa and Morning Chronotype

Table 2 .
MR Secondary Analyses Results for the Associations Between Anorexia Nervosa and Chronotype and Sleep Traits Abbreviations: IVW, inverse variance weighted; MR, mendelian randomization; SNV, single-nucleotide variant; WME, weighted median estimator.a β Values reflect the difference in the outcome for each 1-unit increase in the exposure.

Table 3 .
Associations Between Polygenic Risk Score for Anorexia Nervosa and Anorexia Nervosa and Sleep Disorders in the Mass General Brigham Biobank (n = 47 082) a Effect sizes are per SD increase in the polygenic risk score for anorexia nervosa.Associations were tested using adjusted logistic regression models.Model 1 was adjusted for age, sex, genotyping array, batch, and 10 principal components of ancestry (minimal model).Model 2 was further adjusted for body mass index, employment status, smoking status, exercise, alcohol intake, and education attainment (fully adjusted model).Findings for anorexia nervosa are missing for model 2 due to insufficient number of cases.Significance was considered at the Bonferroni threshold for 9 tested sleep disease outcomes (P < .05/ 9 tests).
Abbreviations: NA, not applicable; OR, odds ratio.a The clinical implications of our novel bidirectional chronotype findings are unclear; however, our study provides several opportunities for future research into the prevention and treatment of anorexia nervosa.Our findings suggest the role of morningness as an underrecognized risk factor for anorexia nervosa whose importance should be further investigated in the context of other Genome-wide association studies for anorexia nervosa and circadian and sleep traits eTable 2. SNVs and proxies included in genetic instrument for anorexia nervosa with chronotype eTable 3. SNVs and proxies included in genetic instrument for anorexia nervosa with binary chronotype eTable 4. SNVs and proxies included in genetic instrument for anorexia nervosa with daytime napping eTable 5. SNVs and proxies included in genetic instrument for anorexia nervosa with daytime sleepiness eTable 6. SNVs and proxies included in genetic instrument for anorexia nervosa with insomnia eTable 7. SNVs and proxies included in genetic instrument for anorexia nervosa with sleep duration eTable 8. SNVs and proxies included in genetic instrument for anorexia nervosa with accelerometer derived midpoint of least active 5 hours, midpoint of most active 10 hours, and sleep midpoint eTable 9. SNVs and proxies included in genetic instrument for chronotype with anorexia nervosa eTable 10.SNVs and proxies included in genetic instrument for daytime napping with anorexia nervosa eTable 11.SNVs and proxies included in genetic instrument for daytime sleepiness with anorexia nervosa eTable 12. SNVs and proxies included in genetic instrument for insomnia with anorexia nervosa eTable 13.SNVs and proxies included in genetic instrument for female only insomnia with anorexia nervosa eTable 14.SNVs and proxies included in genetic instrument for sleep duration with anorexia nervosa eTable 15.SNVs and proxies included in genetic instrument for accelerometer derived midpoint of least active 5 hours with anorexia nervosa eTable 16.Variants identified as outliers by MR-PRESSO eTable 17.Leave-one-out analysis for results of anorexia nervosa with chronotype.IVW MR conducted with listed SNV removed in each respective analysis eTable 18. Leave-one-out analysis for results of chronotype with anorexia nervosa.IVW MR conducted with listed SNV removed in each respective analysis eTable 19.Leave-one-out analysis for results of anorexia nervosa with binary chronotype.IVW MR conducted with listed SNV removed in each respective analysis eTable 20.Leave-one-out analysis for results of insomnia with anorexia nervosa.IVW MR conducted with listed SNV removed in each respective analysis eTable 21.Leave-one-out analysis for results of female only insomnia with anorexia nervosa.IVW MR conducted with listed SNV removed in each respective analysis eTable 22. Association between polygenic risk score for anorexia nervosa and self-reported sleep traits in the Mass General Brigham Biobank (n =16,109) 50Whether this would impact transdiagnostic treatment of more evening-based psychiatric comorbidities, eg, depression, remains unknown.Replication analyses and mechanistic studies are crucial to warrant any circadian-based therapies suggested by our findings.