Reduced DNA methylation and psychopathology following endogenous hypercortisolism – a genome-wide study

Patients with Cushing’s Syndrome (CS) in remission were used as a model to test the hypothesis that long-standing excessive cortisol exposure induces changes in DNA methylation that are associated with persisting neuropsychological consequences. Genome-wide DNA methylation was assessed in 48 women with CS in long-term remission (cases) and 16 controls matched for age, gender and education. The Fatigue impact scale and the comprehensive psychopathological rating scale were used to evaluate fatigue, depression and anxiety. Cases had lower average global DNA methylation than controls (81.2% vs 82.7%; p = 0.002). Four hundred and sixty-one differentially methylated regions, containing 3,246 probes mapping to 337 genes were identified. After adjustment for age and smoking, 731 probes in 236 genes were associated with psychopathology (fatigue, depression and/or anxiety). Twenty-four gene ontology terms were associated with psychopathology; terms related to retinoic acid receptor signalling were the most common (adjusted p = 0.0007). One gene in particular, COL11A2, was associated with fatigue following a false discovery rate correction. Our findings indicate that hypomethylation of FKBP5 and retinoic acid receptor related genes serve a potential mechanistic explanation for long-lasting GC-induced psychopathology.

Patients. The mean age of patients was 53 ± 14 years, and the mean age at diagnosis of CS was 37 ± 14 years (Table 1). Thirty-seven (77%) patients had CD and 11 (23%) had a cortisol producing adrenal adenoma. To verify that the initial diagnosis of CD and cortisol producing adrenal adenoma were correct the clinical, biochemical, radiological and histopathological data from the time of diagnosis were reviewed. In patients with CD in remission the primary treatment was transsphenoidal pituitary surgery in 25 (68%), radiotherapy in five (14%) and bilateral adrenalectomy in seven (19%). Fifteen patients needed additional treatment. In total, 29 (78%) patients with CD were treated with transsphenoidal pituitary surgery, 11 (30%) with radiotherapy and nine (24%) with bilateral adrenalectomy. All patients with cortisol producing adrenal adenoma had been treated with unilateral adrenalectomy. Eighteen (38%) patients had adrenal insufficiency and received replacement therapy with a mean daily hydrocortisone dose of 24 ± 8 mg/day. The mean urinary free cortisol (UFC) excretion was higher in patients compared to controls (Table 1). Seventeen (35%) patients had central (N = 15) or primary (N = 2) hypothyroidism and were receiving a mean L-Thyroxine dose of 104 ± 31 μ g/day. Out of 37 patients with CD, 19 (51%) had growth hormone deficiency of whom 15 were on growth hormone replacement therapy. Four out of 20 premenopausal woman (< 52 years) had hypogonadotropic hypogonadism and were receiving estrogen and progesterone, 2 of 28 postmenopausal women were receiving treatment with oral estrogen. Two women received replacement with dehydroepiandrosterone.
Controls. Controls to patients, matched for age and gender, were recruited from a random population sample obtained from the Swedish Tax Agency. Controls were approached through an invitation letter, responding subjects were interviewed per telephone and those who matched the patient's educational levels and had no previously known psychiatric or chronic diseases known to affect cognitive function, were included. In this part of the study data from one control (N = 16) per three patients (N = 48) were analysed. The mean age was 54 ± 16 years in controls ( Table 1).
Evaluation of hormone status. All patients were in remission, defined by an adequate suppression of serum cortisol concentration (≤ 50 nmol/l) following a 1 mg overnight dexamethasone suppression test. The median (interquartile range) duration of remission was 13 (5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19) years. A CRH test was performed in order to evaluate the function of the HPA-axis. Serum cortisol was measured using competitive electrochemiluminescence immunoassay (Cortisol Elecsys, Roche Diagnostics Scandinavia AB). Urinary free cortisol (UFC) was measured using radioimmunoassay (SpectRia Cortisol 125I, Orion Diagnostica Oy, Finland). Thyroid function was evaluated clinically and by measurements of free thyroxin and thyroid stimulating hormone (TSH) in serum. Gonadal function was evaluated by asking for menstruation pattern and/or age at menopause as well as measurements of estrogen and gonadotropins in serum. Growth hormone status was evaluated by review of previously performed stimulation tests and measurement of insulin-like growth factor I.
Evaluation of fatigue, depression and anxiety. Fatigue was evaluated using the fatigue impact scale, a 40 item questionnaire where different aspects of fatigue (physical, cognitive and social) are evaluated 27 . Depression and anxiety were evaluated using the comprehensive psychopathological rating scale 28 . DNA isolation and methylation assessment. DNA  was performed at the Mutation Analysis Facility (MAF) at Karolinska University Hospital. The procedure is briefly described below: Bisulfite treatment. 500ng of genomic DNA (OD260/280 > 1.8) was bisulfite treated using the EZ-96 DNA Methylation Kit (D5004; Zymo Research, Inc., Irvine, CA, USA). The CT conversion reagent was mixed with DNA and incubated in the dark at 50 °C for 16 hours. After desulfonation and washing steps, the samples were purified using spin plates, eluted in 12 μ l elution buffer and stored at − 20 °C prior to processing.
Infinium Methylation assay. The Infinium Methylation Assay was performed according to the manufacturer's instructions. Briefly, 4 μ l of denatured bisulfite-treated DNA was isothermally amplified over night at 37 °C, followed by an enzymatic fragmentation step. The fragmented DNA was precipitated, resuspended and loaded (using a Tecan EVO robot) on the 12-sample BeadChip, which was then incubated overnight at 48 °C, allowing the fragmented DNA to hybridize to locus-specific 50-mers. Non-specifically hybridized DNA was washed away, followed by a single-base extension reaction using DNP-and Biotin-labeled ddNTPs (with use of a Tecan EVO robot). Subsequently, hybridized DNA was removed from the labeled oligonucleotide and chips were dried under vacuum and imaged using an Illumina iScan scanner.

Statistical analyses. Clinical parameters.
Statistical analyses were performed with IBM SPSS statistics, version 22, or in R version 3.0.3. Data are presented as mean ± standard deviation or median (25-75 percentiles). For comparison between groups we used unpaired t-test for normally distributed data and Mann-Whitney U-test for non-normally distributed data. For proportions, Pearson Chi-square or Fishers exact test were used. Pearson's correlation was used to determine correlation between methylation and clinical parameters. Linear regression (with adjustment for age and smoking habits) was used to analyse the effect of methylation on clinical parameters. DNA methylation analyses. Data was extracted using GenomeStudio (Illumina, Methylation Module v1.9), which was also used to subtract the background and to normalize staining intensities using internal controls present on the chip. A beta-value was calculated to estimate the methylation level of each CpG locus using the ratio of intensities between methylated and unmethylated alleles (0 = unmethylated, 1 = fully methylated). The performances of the different controls used were evaluated and potential outliers identified. Data quality control and analysis was performed using the ChAMP methylation analysis package (v. 1.4.0) 29 in R. Briefly, intensity data from IDAT files were loaded, normalized using default settings (i.e. BMIQ) and corrected for batch effects using ComBat. Differentially methylated regions (DMR) were then identified using the Probe Lasso DMR Hunter function with Benjamini-Hochberg p-value adjustment. Correction for multiple testing was done using the "fdrtool" Gene ontology analyses. Gene ontology analyses were performed in DAVID Bioinformatics Resources 6.7 (NIAID/NIH) using the Functional Annotation Cluster and Functional Annotation Chart functions 30,31 . DAVID provides unadjusted p-values as well as p-values adjusted for multiple testing using both the Bonferroni and the Benjamini methods. Here we present Benjamini-adjusted p-values.

Results
Identification of differentially methylated regions and overall DNA methylation. Initial quality control (QC) analyses of the methylation raw data identified one case sample as a technical outlier due to low levels of detected CpG:s (only 38,007 CpG:s were detected with a detection p-value < 0.01), this sample was removed from further analysis. The final data set consisted of 47 cases and 16 controls. On average, 485,001 CpG:s were detected (484,979 in cases and 485,066 in controls, detection p-value < 0.01). We first performed DNA methylation analysis in ChAMP, to assess differences between patients with CS in long-term remission and matched controls (Table 2). Overall, patients had lower average percentage of DNA methylation than controls (81.2% vs 82.7%, p = 0.002; Fig. 1a). There were 3,903 probes that lay in differentially methylated regions (DMR:s; n = 461), the majority (n = 3,692; 94.6%) being hypomethylated. Of the 3,903 probes, 3,246 (83.2%) mapped to a gene (n = 337). Of the 337 genes, 278 were exclusively hypomethylated, 7 exclusively hypermethylated and 52 genes contained both hypo-and hypermethylated probes (Supplemental Table 1). Of the 3,903 probes, 55.9% (n = 2,183) had an annotated location; with the most common being gene body (33.3%), 3′ -UTR (3.9%) and TSS15 (within 1,500 base pairs upstream or downstream of the transcriptional start site, 2.8%; Fig. 1b).

Identification of probes associated with fatigue, depression and anxiety. To investigate whether
the epigenetic status of the CS subjects is associated with persistent fatigue, anxiety or depression a regression analysis adjusted for age and smoking habits was performed. We identified 731 probes in 236 genes that were associated with at least one of the three clinical traits. Of these 731 probes; 434 were associated with fatigue, 374 with depression, and 452 with anxiety. One hundred and sixty five probes in 108 genes were associated with all three traits.

Functional validation of identified probes through gene ontology analyses.
To explore the functional relevance of the identified DMR:s and clinically associated probes, we next performed gene ontology (GO) analyses using DAVID 30,31 . We initially performed GO analysis of all 337 genes with probes that lay in DMR:s and found 202 GO terms, of which 18 were significant after Benjamini correction (Fig. 2a). Terms related to retinoic acid, thyroid hormone receptor and hormone/nuclear hormone receptor binding was the most common (Figs 2b and 3a).

Genes, probes and clinical traits No
Total no of probes in DMRs 3903 Significantly associated with Anxiety 527 (75 does not match a gene) Significantly associated with Depression 436 (62 does not match a gene) Significantly associated with Fatigue 508 (76 does not match a gene)

Total no of probes in genes 3246
Total no of genes with probes 337 No of probes in genes and significantly associated with clinical traits 731 No of genes associated with Anxiety 183 No of genes associated with Depression 172 No of genes associated with Fatigue 194 No of genes associated with Anxiety, Depression and/or Fatigue 236 When focusing the analysis on the 236 genes that were associated with at least one of the clinical traits fatigue, depression, and anxiety, 184 terms were identified of which 24 terms were significant after Benjamini correction ( Table 3). As before, terms related to retinoic acid were among the most common (Fig. 3a).

DNA methylation of the GC receptor gene (NR3C1).
To explore the potential effect of hypercortisolism on DNA methylation of the NR3C1 gene, we analysed specifically DNA methylation of this gene. Fifteen out of 49 probes annotated to the NR3C1 gene were significantly differentially methylated in CS cases compared to controls. After multiple testing correction using false discovery rate (FDR, 10%), all 15 probes remained significant (qval 0.00019-0.065). The most significant difference was observed for probe cg15645634 (p = 8.31 × 10 −6 , qval: 0.00019), located in intron 8 of the NR3C1 gene. Notably, out of these 15 differentially methylated probes, 8 probes were specifically hypermethylated and 7 probes were hypomethylated (Table 4).
Correlation with markers of HPA-axis activity. To validate the functional value of DNA methylation of the NR3C1 gene and the genes involved in retinoic acid signalling (in total, n = 672 probes), we performed correlation analyses with urinary free cortisol (UFC; n = 47) and the change in serum cortisol concentration (delta cortisol; n = 24) during a CRH-stimulation test as measures of cortisol exposure and HPA-axis activity, respectively. The CRH-stimulation test was performed in a subgroup of CS patients who had not previously received  pituitary irradiation and who did not receive GC replacement therapy. Thirty-one probes were significantly correlated with UFC (Supplemental Table 2), with the strongest correlation observed for probe cg02319187 annotated to the RXRA gene (p = 0.005; Pearson's r = 0.412). Twenty-five probes were significantly correlated with change in serum cortisol in response to CRH (Supplemental Table 3). The strongest correlation was observed for probe cg00629244, located in the NR3C1 gene (p = 0.002; Pearson's r = 0.598). Three probes (cg01367322, cg03058556 and cg03825390) annotated to the ZBTB22, ZBTB9 and RGL2 genes (respectively) were significantly associated with both UFC and the change in serum cortisol. None of the correlations remained significant after correction for multiple testing (FDR, 10%).

Influence of current GC replacement therapy on DNA methylation.
To evaluate the effect of current GC replacement therapy on DNA methylation, we performed a subgroup analysis using the entire dataset, n = 468,149 probes, where patients were stratified by occurrence of current GC replacement therapy. 12,128 probes in 6,186 genes were differentially methylated between the two groups. The most significant differentially methylated probe (cg03546163, p = 2.99 × 10 −6 ; Fig. 3b) was located in the FKBP5 gene.
In total, there are 34 probes annotated to the FKBP5 gene on the Illumina 450 K methylation chip and four of these probes showed differential methylation (cg03546163, cg00058684, cg08586216 and cg25114611) in patients with, as compared to without, current GC replacement therapy (Table 5). Most probes annotated to FKBP5 were hypomethylated in cases receiving GC replacement therapy; (Fig. 3b). No probes remained significantly differentially methylated after multiple testing correction (FDR, 10%). However it is well worth noting that this correction for multiple testing takes into account a very large number of tests, and that in particular the unadjusted p-value for FKBP5 probe cg03546163 reached borderline genome-wide significance (p = 2.99 × 10 −6 ). Together, this suggests a true relevance of this finding, despite qval > 0.1.

Discussion
Hyperactivity of the HPA-axis may increase susceptibility to neuropsychiatric disorders such as depression, post-traumatic stress disorder and anxiety 2-9 . Here we provide evidence for a distinguishable pattern of genome-wide DNA methylation in patients previously treated for CS and propose a mechanism for the long-term adverse neuropsychological consequences of endogenous hypercortisolism, which is also commonly observed in a number of different psychiatric disorders.
Aberrations in DNA methylation has been associated with neurological and neuropsychiatric disorders such as autism 32 , schizophrenia 33 and Alzheimer's disease 34 , as well as early life adverse events such as child maltreatment and parental stress 17 . Childhood adverse events may influence the life-time set point of the HPA-axis 35 ; one plausible mechanism for the induction of a programmed HPA-axis is through GC-induced epigenetic changes.

GO-terms
No of genes Unadjusted P-value Benjamini-adjusted P-value  We performed a global DNA methylation analysis to assess differences between patients with CS in long-term remission and matched controls. Generally, patients with CS had lower levels of DNA methylation than controls. Four hundred and sixty-one differentially methylated regions were identified, with the majority being hypomethylated in patients. Also, we identified 731 probes in 236 genes that were associated with at least one of three psychopathological traits; fatigue, anxiety and/or depression. Gene ontology analyses revealed an enrichment of genes functioning as retinoic acid receptors, thyroid hormone receptors or hormone/nuclear hormone receptors. These receptors all belong to the nuclear receptor superfamily 36 and serve as ligand-activated regulators of gene transcription and stimulators of intracellular pathways. These genes were hypomethylated in cases as compared to controls, and associated with psychopathology in the patients.
The DNA methylation of the genes belonging to the retinoic acid receptor family was also correlated with UFC and change in cortisol concentrations during a CRH-stimulation test, suggesting a functional link between retinoic acid receptor and HPA-axis activity. The retinoic acid family includes Vitamin A and its derivatives, 13-cis retinoic acid and all-trans retinoic acid. Retinoic acid is a transcriptionally active compound that regulates gene expression via binding to specific nuclear receptors termed RARs 37 , or retinoic X receptors (RXRs) 38 . Retinoic acid is crucial during development of the central nervous system [39][40][41] and for neuronal plasticity in adult brain [42][43][44] . Previous data suggests an involvement of the retinoic acid family in the regulation of the HPA-axis. Both chronic (rats) 45 and intermittent (humans) 46 retinoic acid treatment has been shown able to induce HPA-axis hyperactivity and anxiety-like, as well as depressive, behaviour. A plausible mechanism may be that the retinoic acid interrupts the GC receptor induced negative feedback 47 by down-regulating 11β-HSD1 expression and by inhibiting GC receptor transactivation 48 . Early adverse events and poor maternal care have been linked to changes in the GC receptor (NR3C1) DNA methylation. To explore the potential effect of hypercortisolism on NR3C1 DNA methylation specifically, we analysed methylation in this gene and found that 15 out of 49 probes annotated to the NR3C1 gene were differentially methylated in cases as compared to controls. Previously, increased levels of NR3C1 methylation has been observed in post-mortem hippocampal brain tissue from suicide victims who had endured childhood abuse 15 , and in peripheral blood from subjects with a history of perinatal stress [16][17][18] and neglect or abuse during childhood 49,50 . Recently, common stressful life events were found to be associated with higher blood DNA methylation of the NR3C1 gene in adolescents 51 , suggesting that the NR3C1 DNA methylation is subject to change not only during childhood. In accordance, herein we report that adults who have endured long-term endogenous hypercortisolism have a differential pattern of NR3C1 DNA methylation than matched controls, lending further support for the importance of excess cortisol exposure as a possible cause in the programming of the HPA-axis and its psychological consequences.
To evaluate the possibly confounding effect of current GC replacement therapy on our results, we performed a subgroup analysis dividing the cases into groups of patients receiving GC replacement or not. These analyses revealed that the DNA methylation of 12,128 probes in 6,186 genes was influenced by current GC replacement therapy. One of the genes that were specifically hypomethylated in cases compared to controls, with an additional reduction in patients currently receiving GC replacement therapy, was the FK506 binding protein 5 (FKBP5). FKBP5 binds to and negatively regulates GR function, which subsequently reduces affinity of the GR to cortisol 52 . Common genetic variants in the FKBP5 gene have been associated with a relative GR resistance, and found to interact with childhood abuse to predict post-traumatic stress disorder 53 . Previously, studies in mice have reported that long-lasting exposure to GC decreases FKBP5 DNA methylation in the hippocampus, hypothalamus and blood, and that this demethylation is associated with anxiety-like behavior 54,55 and reflect previous GC load 55 . Consistent with these findings, herein we show that FKBP5 is indeed hypomethylated in CS patients as compared to controls, and that the methylation is further reduced in a sub-group of CS patients receiving GC replacement. These findings validate the suitability of CS as study model for GC exposure and further enlighten the strong effect of GC on DNA methylation.
Despite the rigorous study protocol and adequate study model this project is not without limitations. Firstly, the DNA methylation was assessed in whole blood and not in brain or any other isolated GC target tissue. A recent study, however, provided evidence that DNA methylation variation observed in the brain is in fact reflected in the blood 56 . Secondly, the findings herein remain to be validated and further explored as for whether the observed changes in DNA methylation are indeed associated with subsequent changes in mRNA and protein expression. Lastly, these findings remain to be validated in studies of patients with specific psychiatric disorders with HPA-axis hyperactivity.
In conclusion, our findings suggest that long-standing hypercortisolism reduces global DNA methylation, specifically in genes that are known to attenuate the sensitivity of the GC receptor and therefore may induce hyperactivity of the HPA-axis. Consequently, this may be of importance for the action of cortisol on the central nervous system, and by such contribute to the frequent psychopathology observed in our patients. The mechanism proposed might also apply to other disorders with transient or chronic hyperactivation of the HPA-axis that affects a considerable part of the general population; such as depression, generalised anxiety and post-traumatic stress.