Background

Critical illness in children, which requires treatment in a paediatric intensive care unit (PICU), represents a form of severe physical stress, hallmarked by a whole range of (neuro)endocrine abnormalities [1,2,3,4,5]. A typical response to the severe stress of critical illness is an acute activation of the hypothalamus–pituitary–adrenal (HPA) axis [1, 6, 7]. Indeed, critically ill patients show an acute rise in total and free cortisol, lasting shorter in children than in adults [1, 8,9,10,11]. Unlike long assumed, the rise in cortisol is not driven by a rise in ACTH, as ACTH levels have been shown to be normal or low during critical illness [1, 10, 11]. Instead, the elevated cortisol levels appeared explained by low levels of its binding proteins corticosteroid-binding globulin and albumin, combined with suppressed breakdown of cortisol [1, 11]. Suppressed cortisol breakdown has been revealed by reduced expression and/or activity of the cortisol metabolising enzymes 11β-hydroxysteroid dehydrogenase (11βHSD) 2 and the A-ring reductases, and by strongly reduced cortisol plasma clearance during tracer infusion and after administration of a hydrocortisone bolus [1, 11].

Follow-up studies of children who needed PICU admission for a wide variety of surgical or medical reasons have shown that these children reveal high risk for important impairments in physical and neurocognitive development and behavioural problems years after the acute illness, also in the absence of pre-existing conditions known to affect or possibly affect development [12,13,14,15,16,17]. Other severe early-life adverse events, such as abuse or neglect occurring during several sensitive neurodevelopmental windows, have been associated with similar neurocognitive impairments and behavioural problems and with risk of psychiatric and metabolic diseases later in life [18,19,20]. Some of these associations appear to be sex- and/or age-specific [20].

Research has suggested that stress affects development of brain regions crucial for many aspects of cognition and behaviour through increased glucocorticoid signalling that, when prolonged or excessive, can induce abnormal DNA methylation within different levels of the HPA-axis and glucocorticoid signalling [21,22,23,24]. More specifically, such DNA-methylation changes have been reported for genes encoding corticotropin-releasing hormone (CRH) and its receptor CRHR1, arginine vasopressin (AVP) and its receptor AVPR1b, pro-opiomelanocortin (POMC), corticotropin receptor (MC2R), glucocorticoid receptor GRα (NR3C1) and for the GRα-regulated chaperone protein FK506 binding protein 51 (FKBP5) [23, 25,26,27,28,58,59] and exposure to endocrine-disrupting plasticisers leaching from indwelling medical devices [60,61,62]. Also, as critically ill patients are unable to eat, artificial feeding is often provided [63]. Feeding management of critically ill children has been shown to affect the DNA methylome up until PICU discharge [41, 42]. Since prolonged or excessive stress can induce abnormal DNA methylation within different levels of the HPA-axis and glucocorticoid signalling in the context of other adverse early-life exposures [21,22,23,24, 59], the development of such abnormalities after paediatric critical illness seemed plausible. Our present findings are indeed in line with those from studies on the impact of other forms of early-life stress which revealed mostly hypermethylation in NR3C1 and hypomethylation in FKBP5 within the gene body (intron 7) and the promotor [64, 65]. Via research performed in humans in combination with experiments in cells, it has previously been shown that stress reduces methylation in the FKBP5 gene, interestingly at a CpG site which we also found to be hypomethylated, whereby FKBP5 expression was found to be upregulated [66]. Of importance in the context of critical illness, a study in mice revealed that lipopolysaccharide injection increased FKBP5 expression in the hippocampus, driving increased neuroinflammation [67]. Whether the stress-induced rise in endogenous glucocorticoids mediated this effect is currently unclear. However, in patients with Cushing syndrome, it has been shown that long standing excessive endogenous hypercortisolism induces hypomethylation in FKBP5, explaining their long-lasting psychopathological sequelae [24]. In experimental models, also exogenous hypercortisolism has shown to reduce DNA methylation in FKBP5 in neuronal cells, and in vivo this was associated with increased FKBP5 expression across several brain regions [68]. These data provided a plausible explanation for our finding that hypomethylation in the FKBP5 gene was aggravated in patients who 2 years earlier had been treated with glucocorticoids in the PICU, taking into account potential confounders affecting the need of glucocorticoid treatment. It currently remains unclear what brings about the hypomethylation, though downregulated DNA methyltransferase-1 and cleavage of the DNA backbone directly by the GRα have been suggested to play a role [39, 69,70,71]. Some of the differentially methylated CpG sites that we found within NR3C1 and FKBP5 have also been described with other forms of stress. For example, cg15910486 within the promotor/5’UTR of NR3C1 was reported to be hypermethylated in children who experienced early-life adverse events; cg23416081 and cg15929276 located within the 5’UTR region of FKBP5 have previously been shown to alter cortisol reactivity and behaviour in children; cg20813374 located in the promotor/5’UTR of FKBP5 is a known stress-related epigenetic signature previously associated with myocardial infarction and inflammation; and cg03546163 in the 5’UTR of FKBP5 was shown to be differentially methylated in the context of Cushing’s syndrome [24, 65, 66, 72].

Also glucocorticoid metabolism can be altered by early-life stress [33]. We here found that genes encoding cortisol metabolising enzymes were differently methylated in former PICU patients, more specifically for 5α- and 5β-reductase and 11β-HSD1. These changes could again be partly explained by glucocorticoid treatment during PICU stay years before. The methylation differences within HSD11B1, SRD5A1 and AKR1D1 may result in altered systemic cortisol availability and hereby affect development as was suggested by our functional outcome analysis. The HSD11B1 hypomethylation we observed in former patients appeared to some extent protective against rather than contributing to some of the developmental impairments. However, a strong harmful association was found between most functional outcome measures and the abnormal DNA methylation in AKRD1, which encodes the 5β-reductase enzyme, an association that has not been reported before.

We also observed altered DNA methylation in former patients within the genes encoding CRH receptor, ACTH and its receptor MC2R, and three GR-regulated proteins, DUSP1, GILZ and TNFα, which has also been reported for other conditions of early-life stress [26, 28,20, 73]. Higher vulnerability to illness-induced DNA-methylation changes in older than in younger children, in particular from age of adrenarche onwards and into puberty, has previously been shown by our group [74].

This study has some strengths and weaknesses to highlight. The large sample size and the multicentre, prospective study design with predefined long-term assessments of former PICU patients and healthy children were strengths. In addition, our methodology applying tenfold cross-validation over 100 iterations reduced the odds of findings by chance and reduced the impact of outliers. Our study also has some limitations. First, we have studied DNA methylation in buccal mucosa, whereas the HPA-axis links well-defined brain structures with the adrenal cortex via hormonal regulation. Studying the DNA-methylation markers of stress is obviously not possible in the physiological tissues and organs where it exerts its effects through the HPA-axis. It was thus for pragmatic reasons that we analysed buccal mucosa, as these epithelial cells are accessible to clinical research. These cells are also exposed to stress hormones, but we do not know whether the methylation changes observed in these cells reflect those supposed to occur in the HPA-axis itself. This difficulty has been encountered in other studies, where blood cells or buccal cells have been used as proxies to physiological tissues [23,24,25, 29, 31, 40,41,42,43, 66]. Second, due to lack of a sample before PICU admission, we cannot discriminate between pre-existing abnormal methylation and abnormal methylation induced by the critical illness. Third, although we preferentially recruited siblings and relatives of the patients to the control group, extended with unrelated children from the same geographical area and adjusted as much as possible for baseline risk factors, we cannot exclude residual confounding by genetic background and environment. Fourth, the treatment with glucocorticoids during PICU stay had not been randomised, arguing for caution when interpreting these results. Indeed, specific conditions triggering the need for glucocorticoid treatment theoretically may confound these results. However, with the extensive adjustment for risk factors which also included type, severity and duration of illness we aimed to reduce as much as possible the impact of this limitation. Nevertheless, we cannot exclude that there may be some residual unmeasured confounding in these and the other multivariable analyses. Finally, we were not able to assess what effect the abnormal DNA methylation might have on gene transcription, nor on cortisol or ACTH levels, as we did not have the samples for these analyses. However, differential methylation in the order of magnitude as observed in the present study was associated with differential gene expression in our earlier study on DNA methylation in muscle of adult critically ill patients and controls [75].

Conclusions

Two years after critical illness in children, buccal mucosa DNA revealed abnormal methylation of CpG sites within genes of the HPA-axis, most extensively within the FKBP5 and AKR1D1 genes, which occurred largely independent of sex and age. In addition, glucocorticoid treatment while in the PICU was found to be associated with an aggravation of the methylation changes in FKBP5 and AKR1D1 detected 2 years later. The observed abnormal DNA methylation within these genes in former PICU patients statistically explained part of the long-term physical and neurocognitive developmental impairments. These findings call for attention regarding safety of a liberal glucocorticoid use in the PICU in the absence of strong underlying evidence of benefit.