To compare levels of methylation of the glucocorticoid receptor (GR) gene (NR3C1) promoter between women with bulimia nervosa (BN) and women with no eating disorder (ED), and also to explore, in women with BN, the extent to which methylation of the GR gene promoter corresponds to childhood abuse, suicidality, or borderline personality disorder (BPD).
We measured methylation levels in selected NR3C1 promoter regions using DNA obtained from lymphocytes in 64 women with BN (32 selected as having a history of severe childhood abuse and 32 selected as having no such history) and 32 comparison women with no ED or history of childhood abuse.
Compared to noneating disordered women, women with BN and comorbid BPD (or BN with a history of suicidality) showed significantly more methylation of specific exon 1C sites. There was also a (nonsignificant) result indicative of greater methylation in some 1C sites among women with BN, when compared (as a group) to women with no ED. No parallel effects owing to childhood abuse were observed.
Bulimia nervosa (BN) often co-aggregates with problems of impulse control, affect regulation, and anxiousness, but there are striking individual differences as to comorbid characteristics1, 2: About 30% of those affected are frankly “dysregulated” (emotionally labile and impulsive), a third “over-regulated” (emotionally constricted and inhibited), and another third quite free of apparent psychopathology. Available evidence links such phenomenological variations to different constitutional risks, environmental exposures, and treatment needs.1, 2
Gene-Environment Interactions in BN
Most contemporary etiological theories attribute BN to the activation, by environmental pressures, of hereditary susceptibilities.2, 3 In keeping with this view, our group has documented several gene-environment interactions (involving selected neuroregulatory genes and childhood-abuse exposures) that appear to be relevant to risk for BN or its common comorbid symtpoms.2, 4 Forming a direct rationale for the current study, we have observed the combination of low-function variants of the glucocorticoid receptor (GR) polymorphism, Bcll, and exposure to childhood sexual or physical abuse to be significantly more common in women with BN than in women who eat normally and, furthermore, to predict more-pronounced affective symptoms in women with BN.6, 7 The GR plays a critical role in modulating individual stress reactivity. Based on the preceding, we infer that traumatic stress may contribute particularly strongly to risk of BN (and related symptoms) in individuals who are genetically disposed towards lower GR modulation of stress responses. The preceding would be consistent with evidence associating BN with glucocorticoid-system anomalies—a modal tendency in the population seeming to be nonsuppression after dexamethasone and/or heightened basal cortisol—implying reduced inhibition of hypothalamic-pituitary-adrenal stress responses.8, 9
GR Gene and Selected Promoter Region Sites
The human GR gene, located on chromosome 5 at locus q31-q32, is composed of eight coding exons preceded by a number of untranslated first exon variants, each preceded by a promoter containing a transcription start site.10–12 Seven of these first exons (D, J, E, B, F C, and H) are found within a CpG island in a region encompassing 4.5 kb upstream of exon 2 while the remainder (A and I) are located in region 34.5 kb upstream of exon 2. First exon variants are thought to influence GR expression in various tissues. 1D variant expression seems to be exclusive to the hippocampus, whereas 1B and 1C variants are reportedly expressed in brain, various blood cell types and other tissues. The 1C and 1B variants are most strongly expressed in the hippocampus followed by 1F and 1H, and finally by 1D, 1J, and 1E.12
Gene expression appears to be strongly influenced by epigenetic mechanisms.13–15 One process involved is DNA methylation—the addition of methyl groups to the cytosines of CpG islands (DNA sequences in which a cytosine is followed by a guanine) that are common in the regulatory regions of most genes. Methylated CpGs in genes' promoters reduce access of the transcriptional machinery to the DNA and, in turn, block gene expression. Some theorists postulate that epigenetic programming effects, because they produce stable alterations in the DNA transcriptional profile, allow for the principled environmental regulation of gene expression.15, 16 In evidence, variation in early maternal care has been shown to regulate hippocampal GR promoter methylation and GR expression in animals, in a fashion linking neglect to increased methylation, reduced GR gene expression and increased stress reactivity.17 In parallel, human studies have associated childhood abuse with increased methylation of different GR gene exon-1 promoter variants.18, 19 Such processes could explain associations between adverse environmental exposures, during early development, and later problems of adjustment. Whether or not the preceding is true, evidence associates hypermethylation of various neuroregulatory genes with diverse mental-illness phenotypes, including suicidality,18 borderline personality disorder (BPD),20, 21 and eating disorders (EDs).22–24
The Present Study
To date, studies on epigenetic processes in eating-disordered individuals are rare, and none have yet examined the possibility that bulimic eating syndromes coincide with alterations in GR promoter methylation. This study was concerned with the latter association. Based on findings indicating generally altered glucocorticoid functioning in BN,8, 9 and a role of a main GR gene polymorphism in bulimic and associated phenomenology,6, 7 we explored the possibility that women with BN, when compared to noneating-disordered (NED) women, might display propensities toward hypermethylation of the GR gene promoter. We also explored two alternative hypotheses: The first, based on evidence linking developmental stress to altered GR methylation in animals and humans,15–18 was that GR gene methylation might be more characteristic of individuals with BN who reported exposure to childhood abuse than of people with BN alone; the second, based on studies associating suicidality18 and BPD20, 21 with increased GR gene methylation, was that GR methylation might be more pronounced in individuals with BN who also displayed comorbid “borderline” or “suicidal” manifestations, than of people with BN alone. Guided by previous findings, we targeted CpG sites in GR gene first exon sites 1B, 1C, 1F, and 1H,16, 18 measuring percent methylation at each CpG. Given studies suggesting that dietary factors can affect DNA methylation,25, 26 we made efforts to control for alterations in DNA methylation that might correspond to eating-disorder linked dietary anomalies (such as bingeing, purging, or abnormal body weight).
We drew the sample addressed in this paper from a pre-existing database, developed during a large-scale, institutional ethics board approved study on the relevance of genetic factors and childhood abuse to eating and comorbid-psychiatric symptoms. Our database included data from 247 women displaying a BN-spectrum disorder (i.e., either full-blown BN as defined in DSM-IV-TR27 or subthreshold variants) recruited through a specialized EDs program for adults, and 138 NED women, recruited through public and university/college-based announcements. To be eligible for the NED group, participants had to be free of current or past clinical ED symptoms according to the Eating Disorders Examination (EDE) interview (described below). Normal eaters were screened for presence of psychiatric problems or past treatments during the recruitment phase.
We selected a subsample of participants so as to enable specific tests of hypotheses concerning variations in GR promoter methylation owing to presence/absence of BN or childhood abuse, as follows: From among the women in our database who reported at least weekly binge episodes, we selected the 32 who represented the most extreme forms of childhood abuse reported (criteria described below), and the first 32 who denied experiencing any abuse. All cases in the resulting sample met DSM-IV-TR criteria for BN, except for one who was binge-eating once (rather than twice) weekly at the time of testing, and another who had been diagnosed as having BN upon entry into the study, but whose BMI had dropped to 17.3 by testing day, so that she would have qualified for Anorexia Nervosa, Binge-Purge subtype at that moment. On average, the selected women had been ill for 10.27 (±7.08) years. Likewise, we selected 32 NED women based on the requirement that, aside from having been drawn from the same age group as were the women with BN, these individuals showed no history of childhood abuse or BPD. To avoid creating a “supernormal” comparison group, we accepted some individuals who (upon the assessment described below) displayed diagnosable (but relatively mild) mental-health problems—one of our NED participants having, during the preceding year, met criteria for a specific phobia, and one each, a milder form of alcohol abuse or cannabis dependence. Including the preceding participants was conservative, as it (if anything) biased against clinical-nonclinical differences. We note that excluding the three participants in question did not alter main results we report here.
Participants were between the ages of 17 and 48, with mean = 26.05 (±6.59) for BN and 23.67 (±5.70) for NED individuals. Body Mass Index (BMI: Kg/m2) fell between 17.32 and 33.73 for BN participants (mean = 22.42 ± 3.85) and between 18.02 and 29.38 for NED participants (mean = 21.79 ± 2.68). BN and NED groups did not differ on either dimension according to t tests. Data reflecting educational attainment and household income are shown in Table1. Chi-squared tests detected no differences on education or income variables. Limiting recruitment to unmedicated individuals was impractical (and undesirable on grounds of representativeness), and we therefore included 33 BN women (52.4% of the sample) who were using a psychoactive medication when tested. None of the NED women were using psychoactives.
Table 1. Frequency (and percent) of members in each group reporting different levels of education and income
Note: Because of isolated missing values, data on “education” are available for 60 of 64 BN and 31 of 32 NED participants. Data on “income” are available on 52 BN and 31 NED participants.
ED diagnoses and symptoms were assessed using the EDE,28 a semistructured interview assessing severity of core ED symptoms. The current “Gold Standard” for ED diagnosis, the EDE has established reliability and validity28 Screening for comorbid (past 12 months) DSM-IV Axis-I disorders was accomplished using the Structured Clinical Interview for DSM-IV Axis I disorders29 (SCID-I), a computer-guided, interview-based version of the Diagnostic Interview Schedule, Version IV30 (DIS4), and/or the Clinician-Administered Post-Traumatic Stress Disorder Scale31—all “industry standard” measures, exhibiting excellent reliability, and convergent and discriminant validity. (Variations in interviews applied reflected shifts in study protocols occurring during the recruitment of the full sample of participants reported here). Elsewhere, we have evaluated agreement between DIS4 and SCID-I diagnoses, and have obtained solid Kappas (and percent agreements) for past 12-month presence of Axis-I disorders.32 BPD was assessed using the relevant module from the Structured Clinical Interview for DSM-IV Axis-II Disorders,33 excluding the criterion referring to overeating. Audits on interrater reliability for a BPD/non-BPD distinction, conducted in sets of 12, 14, and 33 interviews selected from three different study periods in our laboratory, produced kappas (and percent agreements) of 0.80 (91.7%), 0.81 (92.86%), and 0.77 (90.91%). We produced a single dichotomous (present/absent) score reflecting history of suicidality, using the relevant item score for the BPD ascertainment.
Childhood abuse was assessed using the Childhood Trauma Interview,34 a roughly 30-minute structured interview addressing experiences of abuse occurring up to age 18. Interrater reliability for indices reflecting the nature, severity, frequency and duration of trauma are very good. Subjects were classified as having experienced childhood abuse when they received a severity score of 3 or greater for sexual abuse or physical abuse with a minimum frequency score of 3 (meaning at least once every 4 months). Examples of criterion-level sexual abuse were: Fondling of genitals or breasts, or having to watch someone else being sexually abused. Examples of criterion instances of physical maltreatment were: Being whipped on bare legs, slapped in the face, or pushed to the ground. We felt that lower scores (reflecting concepts such as “being looked at in a sexualized way” or “pushed, but not pushed down”) indicated abuse too ambiguously. Prior studies in our lab, using thresholds comparable to those applied here, have led to interpretable effects of childhood abuse upon diverse clinical and neurobiological indices.2
DNA samples, obtained from whole blood, were amplified by polymerase chain reaction, and then treated with sodium bisulfite using the EpiTect Bisulfite Kit (Qiagen) according to the manufacturer's instructions to assure complete conversion of unmethylated cytosine into uracil. DNA methylation was then assessed using the EpiTYPER platform (Sequenom, CA), which employs base-dependent cleavage contingent on methylated cytosine followed by MALDI-TOF mass spectrometry quantification of methylation content in the targeted region. Bisulfite-treated DNA was sent to the Innovation Center of Genome Quebec where EpiTYPER assays were designed and processed. The regions selected for DNA methylation experiments were the promoters of GR exon 1B, 1C, 1F, and 1H, which are sequences that have been extensively studied in depression and suicide (Fig.1). We retained data for only those CpG sites for which at least 70% of values were present. Where individual data points were missing on retained sites, we imputed missing values using site means.
Percent methylation at CpGs in each of the four selected promoter regions (1B, 1C, 1F, and 1H) served as outcome variables. For each region, we ran a series of ANOVAs testing effects of group, promoter-region site (a repeated measure), and then the group × promoter-region site interaction. A first ANOVA defined the grouping factor to compare “BN with BPD” (BN-BPD: n = 14), “BN without BPD” (BN-noBPD: n = 47) and “NED controls” (n = 32). A second ANOVA assigned participants to the groups “BN with suicidal history” (BN-suicide; n = 14), “BN with no suicidal history” (BN-no suicide n = 47) and NED. (We note that BN-BPD and BN-suicide samples displayed some degree of overlap, eight of the 14 BPD and the 14 “suicidal history” participants being the same individuals). A third analysis reassigned participants to the groups “BN with history of abuse” (BN-abused: n = 32), “BN with no history of abuse (BN-not abused: n = 32)” and NED controls. A fourth analysis compared BN and NED groups globally. Because of observed inconsistency with sphericity assumptions, effects were tested using relatively conservative multivariate test statistics.35 Additional analyses of covariance (ANCOVAs) were performed to evaluate the extent to which observed group × promoter-region site interactions (BN-BPD vs. BN-no BPD vs. NED; BN-suicidal vs. BN-nonsuicidal vs. NED) on Site 1C might have been confounded by effects of binge frequency, vomit frequency, BMI or medication use. Parallel analyses were performed on the BPD group effect on the 1H site. As variations on eating symptoms and medication usage were fully confounded with a BN/NED distinction, we performed the analyses in question among BN women alone. To avoid overparameterization, we conducted a separate ANCOVA for each group classification and for each covariate of interest. As binge and vomit frequencies conformed (as is usual) to a non-normal (Poisson) distribution, we recoded continuous frequency measures into values reflecting the lowest, middle and highest tertiles on each continuum. Medication was coded as a dichotomous (present/ absent) variable.
Among the ANOVAs testing BN-BPD vs. BN-no BPD vs. NE effects, we observed a significant group × promoter-region site interaction in the 1C region [F (34,146) = 1.75, p < 0.02] (Fig.2). Group comparisons at individual CpG sites indicated that BN-BPD individuals showed elevated methylation on 1C positions 10 and 21 when compared to levels obtained in NED controls (p < 0.03 in both cases; p < 0.07 in both cases after Bonferroni correction). Methylation levels in BN-no BPD individuals were not elevated relative to those in controls. A second analysis involving the BPD/non-BPD distinction revealed a main effect of group in the 1H region [F (2, 88) = 4.86, p < 0.01]. Here, group comparisons indicated an opposite effect, suggesting significantly lower methylation (p < 0.005; p < 0.01 after Bonferroni correction) in BN-BPD individuals (7.78 ± 0.71) than in NED individuals (8.89 ± 1.18). Methylation levels in BN-no BPD individuals (8.63 ± 1.16) did not differ from those of controls. Medication usage in BN-BPD participants (n = 8, or 61.5%) did not differ from that in BN-no BPD participants (n = 23, or 48.9%), according to a Fisher Exact test. Likewise, BN-BPD and BN-no BPD groups did not differ on income or education variables (data not shown). A second set of ANOVAs tested an alternative grouping (BN-suicidal/BN-non suicidal/NED), and again revealed a significant group × promoter-region site interaction in the 1C region [F (34,150) = 1.84, p < 0.01], illustrated in Figure3. Group comparisons at individual CpG sites indicated that BN-suicidality individuals showed significantly elevated methylation on the 1C positions 10 (p < 0.04), 22 (p < 0.02), and 29 (p < 0.005), compared to those obtained in NED controls (ps < 0.10, 0.05, and 0.01, respectively, after Bonferroni correction). Methylation levels in BN-no suicidality individuals never differed significantly from those in NED participants (see Table2 for precise locations of 1C CpG sites). Not surprisingly, medication usage in BN-suicidal participants (n = 11, or 84.6%) differed significantly from that in BN-nonsuicidal participants (n = 20, or 42.6%), on a fisher Exact test (p < 0.01). Likewise, the BN-suicidal and BN-nonsuicidal groups did not differ on income or education variables (data not shown). ANOVAs designed to test for BN-abused vs. BN-not abused vs. NED effects revealed no significant main or interaction effects on any of the sites tested. Finally, ANOVAs testing global BN vs. NED effects on each of the four selected promoter regions revealed no significant main effects, but did yield a nonsignificant trend toward a group × promoter-region site interaction [F (17,78) = 1.541, p < 0.10], consistent with hypermethylation of some 1C sites in individuals with BN when compared to the pattern obtained in NEDs.
Table 2. Localizing information for specific exon 1C CpGs retained for statistical analyses
chr5:142763289 to 142763295
chr5:142763314 to 142763322
chr5:142763355 to 142763361
chr5:142763375 to 142763377
chr5:142763398 to 142763407
chr5:142763442 to 142763445
chr5:142763450 to 142763455
chr5:142763473 to 142763475
chr5:142763620 to 142763626
chr5:142763629 to 142763632
Controls for Eating Symptoms and Medication Use
“Binge frequency” and “medication usage” were significant covariates for effects obtained in the 1C region. However, we found no tendency towards attenuation of group × promoter-region site effects when these covariates were controlled. None of the covariates were significant in the 1H region, and we observed no tendency towards attenuation of the group effect when covariates were included. In other words, findings did not seem to be confounded with eating-symptom or medication effects. For brevity's sake, we do not report results of analyses involving covariates in detail here.
This study provides evidence of altered DNA methylation at certain CpG sites on the promoter of a regulatory exon-1 variant of the glucocorticoid gene (noncoding exon 1C) in association with comorbid BPD or history of suicidality in bulimic individuals (Figs. 2 and 3). Results also show individuals with concurrent BN and BPD to display significantly lower methylation on GR region 1H, when compared to BN-no BPD or NED individuals. In contrast, we find no reliable tendency for bulimic individuals reporting childhood abuse to show altered methylation of the 1C promoter (or of any of three other regions studied). In other words, our results imply that individuals with bulimic syndromes, when they display marked impulse-dysregulation and mood instability (as is the case for those with comorbid BPD or a history of suicidality) show site-specific propensities towards altered GR promoter methylation. Our findings indicate no parallel effects owing to a history of childhood abuse.
Why should hypermethylation of 1C covary with psychopathology in BN, when comparable effects are not observed on exons 1B, 1F, or 1H? We note that accumulating evidence links variation in methylation patterns of GR exon 1C to such psychopathological expressions as depression and suicidality.18 Furthermore, because it is strongly expressed in brain, the 1C variant may constitute a good candidate to mediate such behavioral manifestations of epigenetic programming as impulsivity or suicidality.11 Finally, on theoretical grounds, because the 1C region is associated with the GR alpha receptor (thought to be excitatory), silencing of this region would be consistent with reduced inhibitory feedback within the HPA axis, and correspondingly increase basal cortisol levels—which is a modal effect observed in individuals with BN.8, 9
An apparent tendency in our data for epigenetic effects associated with 1C methylation to be more specific to comorbid psychopathology (i.e., BPD or suicidality) than to BN warrants comment. The literature on putative vulnerability factors acting in BN often shows selected “risk” factors (whether they be developmental, familial, or neurobiological) to be more pertinent to the expression of comorbid clinical traits in BN sufferers than they are to bulimia-specific symptoms.2,4 Elsewhere, we have interpreted such tendencies as meaning that BN may often represent the activation, by excessive caloric restraint, of susceptibilities towards nonspecific tendencies, such as affective instability or impulse dysregulation, rather than the action of etiological agents that are truly bulimia-specific.2 In keeping with the preceding, the epigenetic effects studied here seem to have greater influence over generalized psychopathological expressions (manifested as BPD or suicidality) than over the likelihood of having (or not having) BN. In other words, epigenetic processes of the type explored here seem to be of greater value in explaining comorbid expressions (e.g., borderline vs. nonborderline tendencies) occurring in bulimic individuals, than they are of “being” or “not being” bulimic itself. We would argue that we are observing the epigenetic programming of a generalized vulnerability to psychiatric instability—that may then be transformed into risk of an ED by effects of excessive dieting (for a full treatment of such processes see Refs. 2 and 4). Whether triggering effects of dietary factors are enacted via epigenetic processes that amplify pre-existing vulnerabilities, or indirect (nonepigenetic) effects of malnutrition on epigenetically prepared vulnerabilities (enacted through such effects as altered neurotransmitter availability) remains a question for future research. What is suggested at present, however, is that epigenetic factors may underlie clinical variations that occur within the bulimic population.
Previous studies have supported the theory that severe environmental stressors can cause epigenetically regulated alterations in gene expression, especially when such stressors occur early in the life cycle.16,19 Given the preceding, we had anticipated that we might observe an association in our study between DNA methylation and childhood abuse. One possible account for our failure to detect epigenetic markings associated with childhood abuse could be that our “abused” group, although selected using unambiguously severe instances of abuse, may not have consisted of people who reported incidents that were consistently severe enough to produce epigenetic imprinting. Another possibility is that our study lacked power to detect abuse-linked effects, even though it was adequately powered to detect (apparently stronger) effects attributable to comorbid psychopathology. Indeed, it is likely that DNA methylation, measured at any moment in any individual, may reflect diverse influences (related to heredity, perinatal insults, developmental experiences, current stressors, nutritional and mental status, and other factors), so that DNA methylation may be in constant flux in function of changing environmental exposures. If so, methylation levels and manifest psychopathology might be brought into closer synchrony than are methylation levels and temporally more distal events, such as childhood abuse. Regardless, we note that a comparable finding (i.e., hypermethylation of GR and other neuroregulatory gene promoters in individuals with BPD, regardless of presence of reported childhood abuse) has been reported elsewhere.20
It remains to discuss an effect suggesting 1H hypomethylation in participants with concurrent BN and BPD—which, we note, corroborates a similar result obtained by one of the authors on GR region 1H in the hippocampus of suicide completers with a history of childhood abuse.18 An implication of the preceding may be that the GR regulatory region may be poised for epigenetic regulations that run in different directions across different regions of promoter. Although we are uncertain how to interpret any such effect, we are intrigued by the fact that parallel patterns of alteration (hypermethylation of region 1C and hypomethylation of 1H) are obtained across PTSD and BPD samples. Clearly, such effects constitute a target for future research efforts.
Our findings suggest that a highly dysregulated bulimic presentation, characterized by affective and behavioral instability, may often be associated with greater levels of GR methylation (in at least the 1C region). The preceding is not surprising, given that GR hypermethylation should, in theory, interfere with the optimal regulation of stress-system responses mediated by the GR. Such effects may help inform the understanding of heterogeneous presentations of BN—or bulimic subphenotypes.2
We would argue that epigenetically informed models also offer several immediate benefits in clinical practice. First, although they pertain to biological processes, such models help clinicians and patients alike understand that disorders such as BN occur in individuals who are susceptible, not because of “weakness,” “capriciousness,” or “superficiality,” but because they carry biological vulnerabilities that have been activated by specific environmental triggers—that act beyond their willful control. In other words, epigenetically informed perspectives inform more and blame less.
Use of peripheral biomaterials to obtain measures of DNA methylation raises legitimate concerns about the pertinence of findings to central brain processes. However, we note that the use of blood-derived tissues to assess psychiatric disorders is increasingly being recognized and a growing body of research is documenting correspondence between brain- and blood-derived gene expression signals15,38—highlighting that patterns of epigenetic variation observed in blood are likely to provide a proxy to patterns acting in brain. We also acknowledge that, in the absence of a “neutral” control gene, we cannot ascertain that effects we have observed are associated with gene-specific epigenetic effects, rather than genomewide influences. However, given presence of effects in only some of the regulatory regions of the gene of interest we have tested, we have a basis for inferring that effects we have observed do display a degree of site-specificity. Finally, we note that our treatment of medication use as a unitary effect (regardless of different molecules applied) provides a rather “blunt” index of possible medication effects. Available sample size, unfortunately, rendered any more finely grained analysis impossible. In a related vein, given that psychoactive medications can influence methylation levels, it is important to consider that greater usage of medication in our BN-suicide group could have confounded our observation associating a history of suicidality with elevated GR gene methylation. Arguing against the idea of a pure medication confound, however, we also observed relative hypermethylation of the GR gene in BN-BPD participants, who did not tend to be more frequently medicated than were BN-non-BPD individuals. Furthermore, the covariance analyses we reported earlier help rule out medication as a confound in observed effects on methylation levels.
Summary and Conclusions
In keeping with previous findings linking various psychopathological entities to epigenetic alterations in the expression of disorder-relevant genes, our findings associate BN, when accompanied by BPD or suicidality, with significant hypermethylation of specific sites on the GR gene promoter. Further work will be required to ascertain the extent to which alterations in DNA methylation of the type observed here represent markers of an active psychopathological state, of exposure to early-life adversity, of processes related to nutritional status, or some combination of all three.