The Circulating Methylome in Childhood-Onset Inflammatory Bowel Disease

Abstract

Background

The genetic contribution to inflammatory bowel disease (IBD), encompassing both Crohn’s disease (CD) and ulcerative colitis (UC), accounts for around 20% of disease variance, highlighting the need to characterize environmental and epigenetic influences. Recently, considerable progress has been made in characterizing the adult methylome in epigenome-wide association studies.

Methods

We report detailed analysis of the circulating methylome in 86 patients with childhood-onset CD and UC and 30 controls using the Illumina Infinium Human MethylationEPIC platform.

Results

We derived and validated a 4-probe methylation biomarker (RPS6KA2, VMP1, CFI, and ARHGEF3), with specificity and high diagnostic accuracy for pediatric IBD in UK and North American cohorts (area under the curve: 0.90-0.94). Significant epigenetic age acceleration is present at diagnosis, with the greatest observed in CD patients. Cis-methylation quantitative trait loci (meQTL) analysis identifies genetic determinants underlying epigenetic alterations notably within the HLA 6p22.1-p21.33 region. Passive smoking exposure is associated with the development of UC rather than CD, contrary to previous findings.

Conclusions

These data provide new insights into epigenetic alterations in IBD and illustrate the reproducibility and translational potential of epigenome-wide association studies in complex diseases.

Graphical Abstract

Open in new tabDownload slide

1. Introduction

The inflammatory bowel diseases (IBDs), including Crohn’s disease (CD) and ulcerative colitis (UC), are common causes of chronic illness in both children and adults worldwide. Disease prevalence in the Western world is projected to reach 1% by 2030.^1–3 Childhood-onset disease is typically characterized by extensive intestinal involvement and long-standing effects on health, educational attainment, and well-being,^4–6 along with increased mortality.⁷

Both genetic and environmental factors are unequivocally implicated in disease pathogenesis. Linkage studies followed by genome-wide association studies (GWAS) have been successfully applied in IBD and, to date, have allowed the characterization of over 300 susceptibility loci,⁸ accounting for up to 20% of disease variance. More recently, the application of -omics technologies has allowed the exploration of the epigenetic contributions to complex diseases, with parallel aims of understanding gene-environmental mechanisms in pathogenesis and informing biomarker discovery. Epigenome-wide association studies (EWAS) and GWAS are complementary in defining disease biology,^9,10 with a consideration of cell specificity and reverse causation.¹¹ Successful translation of biomarkers derived from the epigenome has been applied in immune-mediated, malignant, neurodegenerative, and metabolic diseases.^12,13

In previous work, we reported the profile of genome-wide DNA methylation in peripheral blood leukocytes at diagnosis from a cohort of newly diagnosed children with CD.¹⁴ This characteristic pattern of genome-wide alterations was replicated in adult inception patient cohorts from the United Kingdom and Scandinavia,^15,16 allowing the definition of the circulating IBD methylome (differential methylation changes highly associated with the diagnosis of IBD) in adult-onset disease. Recently, this profile was also replicated in adult patients with medically refractory CD who required resection surgery.¹⁷ In these studies, site-specific DNA methylation changes found in peripheral blood in IBD patients are underpinned by genotypes and cell-specific alterations in gene expression. Meta-analysis has provided additional validation of robust methylation changes associated with diagnosis of IBD from whole blood.¹⁸

On the basis of evidence currently available, it appears that methylation alterations associated with IBD are influenced by genetic variations (methylation quantitative trait loci [meQTL]) and modulated by environmental exposures as well as by the presence or absence of active inflammation. Lifestyle factors, particularly nutrition, tobacco smoking, and alterations of the microbiome, are strongly implicated in IBD and known to exert significant effects on key epigenetics processes.^19–21 The interplay between the exposome (totality of environmental exposures an individual is exposed to from conception throughout life) and genome in shaping epigenetic alterations in complex disease remains inadequately understood and necessitates further investigation.

We hypothesize that the environmental and genetic influences on epigenetic mechanisms may be most tractable in childhood-onset disease. In the present study, we, therefore, extend the previous research on the circulating methylome in children with IBD. We utilize these previously published data to define and validate a simple accurate diagnostic biomarker panel for IBD, which outperforms other blood-based markers, including C-reactive protein (CRP), and shows disease specificity compared with other immune diseases. Furthermore, we report epigenetic evidence for accelerated biological aging in affected children and define the contributions of germline variation, parental imprinting, and passive smoke exposure to the disease methylome using parent-child trios.

2. Methods

2.1. UK inception cohorts (UK1 and UK2 cohorts)

Children with pediatric IBD in this study were defined as those under 18 years of age, diagnosed in line with the modified Porto criteria.²²

The UK1 inception cohort consisted of 36 children with CD and 36 controls, as described previously in Adams et al’s study.¹⁴

In the present study, a new inception cohort (UK2) totaling 86 patients with a diagnosis of IBD (33 CD, 31 UC, 22 IBD-unclassified [IBDU]; median age of 12 years [range: 0.9-17.5] [Supplementary Table]) was recruited from the pediatric clinics at John Radcliffe Hospital, Oxford, and Addenbrooke’s Hospital, Cambridge, United Kingdom. All patients were sampled within a year of diagnosis. Early disease progression was defined as the need for treatment escalation within 18 months of diagnosis and sample collection—specifically need for surgery or for biological therapy and/or more than 2 courses of steroids while on thiopurine therapy. A total of 30 non-IBD controls (children who were investigated for gastrointestinal [GI] symptoms without IBD or healthy siblings from other disease cohorts) with a median age of 9 years (range: 2-17) were also recruited from the John Radcliffe Hospital, Oxford.

2.2. Parental trios cohort (UK3)

A total of 90 pediatric IBD patients (60 CD and 30 UC/IBDU) were recruited alongside their trio families (child with IBD, mother, and father), totaling 270 participants, at Southampton Children’s Hospital, United Kingdom. The median age of the affected children was 12.5 years (range: 5-17) for CD and 11.9 (range: 2-17) for UC, with corresponding mean disease duration of 1.3 years for CD and 1.7 years for UC, as detailed in Supplementary Table.

2.3. Ethical approval

Ethical approval for these studies was obtained from the Southampton & South West Hampshire Research Ethics Committee (09/H0504/125), the Yorkshire & The Humber Sheffield Research Ethics Committee for the Oxford Gastrointestinal (GI) Cohort (21/YH/0206), and local research committees (REC 12/EE/0482 and REC 17/EE/0265).

2.4. Sample size estimation

There were 2 primary outcomes of this study: (1) further characterization and validation of the methylome in pediatric IBD using epigenome-wide analysis and (2) the definition of a methylation biomarker panel, which has the potential for clinical translation. Group sizes ranging from 30 to 60 would yield a statistical power of 0.8 at an effect size of 10%²³ to replicate previous EWAS findings. For our targeted biomarker analysis, we need group sizes of 15 to achieve a statistical power of 0.8 at an effect size of 20% for p-value significance. We additionally performed post hoc analyses, which are more exploratory, including EWAS for IBD subtypes.

2.5. DNA extraction and genome-wide methylation analysis

Genomic DNA was extracted from peripheral blood samples collected in EDTA by the salting-out method for the Southampton samples, Qiagen Puregene Blood Core Kit C for Oxford samples, and DNA Blood Mini Kit for Cambridge samples. For methylation arrays, 500 ng of DNA was randomized onto a 96-well plate for bisulfite conversion and analyzed at UCL Genomics (London, UK) using the HumanMethylationEPIC platform (Illumina, San Diego, CA). A further 750 ng of genomic DNA was analyzed using Global Screening Arrays v3.0 (Illumina) at UCL Genomics.

2.6. Data processing

Idat files were processed using pipelines from minfi²⁴ in R (R Foundation for Statistical Computing, Vienna, Austria). Two samples were removed due to incomplete bisulfite conversions. Failed probes (detection p value >0.01 in at least 50% of the samples) or probes found on the sex chromosomes were removed from the analysis. The raw data were background-adjusted and corrected for dye color bias, and quantile-normalized. Batch effects were adjusted for both slide and position using ComBat.²⁵ Estimated cell proportions (CD4⁺, CD8⁺, T cells, natural killer [NK] cells, B cells, monocytes, and granulocytes) were computed using the Houseman algorithm.²⁶ EWAS analysis of differential DNA methylation analysis was performed using limma,²⁷ and models were adjusted for the first principal component of the cell type proportions, sex, age, and center. Unless stated otherwise, all multiple correction testing was adjusted using the Benjamini-Hochberg false discovery rate (FDR).

2.7. Gene ontology enrichment analysis and identification of cell type of origin

The function gometh within the R package missMethyl²⁸ was used to test gene ontology (GO) enrichment for CpG sites with an adjusted p value of <0.001. The function corrects for the bias between the differing number of probes per gene present on the Illumina EPIC array and CpGs associated with multiple genes. The top 1000 differentially methylated probes (DMPs) were then used to identify the cell type of origin using eFORGE v2.0 (https://eforge.altiusinstitute.org/).²⁹

2.8. MeQTL analysis

A sex check was performed using PLINK (v1.07) to identify and remove any sex mismatches. MeQTLs were estimated using Matrix-eQTL (v2.3), with a cis distance threshold of 1 Mb. We used a genome-wide approach of all single nucleotide polymorphisms (SNPs; n = 654 027) and all methylation sites (n = 773 449) with covariates of age, sex, and cell count proportions included in the model.

2.9. Epigenetic clock analysis

DNA methylation age was determined using 328 of the 353 probes in the Horvath epigenetic clock,³⁰ implemented in the R package watermelon.³¹ Age acceleration was calculated as the difference between DNA methylation age and chronological age. Statistical differences between IBD and controls were computed using a general linear model. We determined the median age acceleration between disease groups from residuals obtained after regressing DNA methylation age from controls.

2.10. Correlation of IBD-associated methylation findings

We analyzed the correlation of alterations in methylation seen in IBD between the UK1 and UK2 inception cohorts for the methylation sites most strongly associated with pediatric IBD. We carried out a correlation analysis of the top 100 overlapping methylation sites’ beta values reported in the UK1 with the new UK2 inception cohort.

2.11. Biomarker validation and 4-probe-based diagnostic model

Differentially methylated sites in the discovery cohort with the highest statistical significance were selected for analysis as potential biomarkers.¹⁴ The top 4 nonredundant sites were taken forward as our 4-probe diagnostic logistic regression model, which was initially trained in the normalized beta values in the UK1 cohort¹⁴ and then tested in the UK2 inception cohort. A 2-dimensional t-distributed stochastic neighbor embedding (t-SNE) plot was created for a visual analogue of the data using the UK2 inception cohort. Publicly available and previous datasets were utilized for model validation (Supplementary Methods). Performance of the models was defined using the area under the curve (AUC), receiver operating characteristic (ROC) curve, accuracy, specificity, sensitivity, positive predictive value (PPV), and negative predictive value (NPV).

2.12. Assessment of intergeneration inheritance of IBD-associated methylation changes and imprinted genes

Correlations in methylation between father and child, mother and child, and mother and father were assessed for a panel of IBD-associated methylation sites, by deriving Spearman correlation coefficients.¹⁴ The package DMRcate³² was used to compute differentially methylated regions (DMRs), with significance determined by an FDR <0.05. We then further selected out DMRs within imprinted region using 2 databases: the geneimprint (http://www.geneimprint.com/) and Otago imprinting (http://igc.otago.ac.nz).³³ We firstly compared DMRs in the UK2 inception cohort. We then carried out formal EWAS and DMR analyses in the parental trios cohort, by comparing levels of methylation in child-father and child-mother pairings.

2.13. Passive smoking exposure

Fisher’s exact test was used to assess differences between IBD subtypes and passive smoking exposure in both the parental trios cohort and UK2 inception subcohort. Parental smoking data were collected by parental and patient interviews at the time of recruitment into the UK3 cohort. Information on parental past and current smoking, including pack-year history, was obtained (Supplementary method).

3. Results

3.1. The circulating methylome in childhood-onset UC and CD

Epigenome-wide analysis within the UK2 inception cohort identified 384 CpG sites with significant evidence for altered methylation levels when comparing IBD with controls (Holm-corrected p < 0.05) (Figure 1A, Supplementary Table). The site showing the greatest evidence for differential methylation implicates the gene ZBTB16 with 3 CpG sites with >5% hypomethylation in IBD patients. Other genes showing strong evidence for differential methylation include PP4R3B (12% hypermethylation), Interferon Alpha and Beta Receptor Subunit 1 (IFNAR1; 9% hypomethylation), and Suppressor of Cytokine Signaling 2 (SOCS2; 6% hypermethylation). Overall, of the 100 most differentially methylated sites, there were 51 CpG sites with hypermethylation and 49 with hypomethylation. GO analysis of the 384 DMPs revealed 3 significantly enriched GO terms involving regulation of alpha-beta T cells (Supplementary Figure 1).

We next performed an EWAS for each of the IBD subtypes (CD and UC) compared with controls. For CD vs controls, a total of 157 CpG sites were differentially methylated (FDR < 0.05) (Figure 2A and Supplementary Table). For UC vs controls, a total of 107 CpG sites were differentially methylated (FDR p < 0.05) (Figure 2B and Supplementary Table). One CpG site found within LOC100131347 was found to be differentially methylated (FDR = 0.034) when comparing CD and UC directly (Supplementary Table).

There were no statistically significant DMPs associated with a more severe disease course/progression (Supplementary Table and Supplementary Figure 2 and 4).

3.2. Monocytes and NK cells are involved in the observed methylation changes in pediatric IBD patients

Active inflammation is known to affect the proportion of immune cell types in IBD. We assessed the estimated cell proportions derived from methylation profiling of the UK2 inception cohort. An increase in estimated cell type proportions was found in both granulocytes (p = 0.00137) and monocytes (p = 0.000261) between IBD patients and controls (Figure 1B), whereas a decrease in the estimated cell type proportions was observed in B cells (p = 0.0411), CD4 T cells (p = 0.0411), CD8 T cells (p = 0.0411), and NK cells (p = 0.00406) between patients and controls. The top 1000 significantly differentiated methylation CpG sites between IBD and controls were assessed for identification of cell type origin using eFORGEv2.0. This analysis revealed differentially methylated sites that colocalize with regulatory elements of specific tissue enrichments from primary monocytes (binomial p = 1.82 × 10⁻⁶) and NK cells (binomial p = 4.60 × 10⁻⁸) (Supplementary Figure 3; Supplementary Table).

3.3. Derivation and validation of a pediatric IBD diagnostic model

We compared the evidence for altered methylation for the 100 sites most strongly implicated by EWAS analysis in the 2 UK inception cohorts: UK1 and UK2 (Supplementary Table). A strong correlation was seen across the cohorts (R = 0.93, p < 2.2e-16) (Figure 1C).

The top 4 methylation sites within the genes RPS6KA2, VMP1, CFI, and ARHGEF3 were selected from the EWAS analysis in the UK1 cohort (Table 1).¹⁴

Of these 4 sites, the methylation sites within the RPS6KA2 and VMP1 loci (adjacent to RPS6KB1) contributed the most weight in the model, and the individual marker with the highest diagnostic accuracy was found to be RPS6KA2, with an accuracy of 0.85. Incorporating 4 probes into the model improved its performance. A logistic regression model consisting of methylation of these 4 probes was then tested in the UK2 cohort for a first validation; following ROC/AUC analyses, the model demonstrated a diagnostic accuracy with an AUC of 0.912 (95% confidence interval [CI], 0.86-0.96) (accuracy: 0.74; PPV: 0.98) (Figure 3A and B and Table 2).

To further validate our model, we separately compared children in the UK2 cohort who were CRP-positive (>5 mg/L; n = 32) and children with no CRP rise at presentation (denoted CRP-negative [<5 mg/L; n = 49]) against non-IBD controls. In the CRP-positive group, we found an AUC of 0.99 (95% CI, 0.99-1) (accuracy: 0.88; PPV: 0.89) (Figure 3D and Table 2). Within the CRP-negative group, an AUC of 0.90 was observed against controls (95% CI, 0.83-0.96) (accuracy: 0.82; PPV: 0.85) (Figure 3E and Table 2).

We then sought to externally validate our prediction model using the available methylation data at the baseline time point for the North American inception RISK cohort (164 children with CD and 74 controls³⁴). In this cohort, we demonstrated an AUC of 0.93 (95% CI, 0.90-0.96) (accuracy: 0.87; PPV: 0.81) (Figure 3C and Table 2). At the second time point (1-3 years after diagnosis), accuracy was markedly reduced, with an AUC of 0.66 (95% CI, 0.59-0.74) (accuracy: 0.69; PPV: 0.63), demonstrating a rapid reduction in utility over a relatively short time.

To investigate the utility in adults at presentation, we analyzed a large adult IBD inception cohort¹⁵ for accuracy of diagnosis with our model. In this cohort, a more modest accuracy for diagnosis was noted—AUC 0.74 (95% CI, 0.68-0.81) (accuracy: 0.71; PPV: 0.73) (Figure 3G).

To confirm diagnostic specificity for IBD against other GI and inflammatory diseases, our prediction model was tested in cohorts of pediatric patients with celiac disease and non-celiac controls. The observed AUC in celiac disease was 0.55 (95% CI, 0.44-0.71) (accuracy: 0.55; PPV: 0.55) (Figure 3F and Table 2). Additionally, the model exhibited no discrimination between adult rheumatoid arthritis patients and controls, as evidenced by an AUC of 0.64 (95% CI, 0.60-0.68) (accuracy: 0.622; PPV: 0.63) (Figure 3H).

3.4. Epigenetic age acceleration at diagnosis in pediatric IBD cohorts

We assessed the correlation between chronological and epigenetic age using the Horvath clock in IBD vs controls in the UK2 inception cohort (Figure 4A). A strong positive correlation of chronological and epigenetic age was observed in both groups, with a higher degree exhibited in the control group (R = 0.78) compared to IBD patients (R = 0.53) (Figure 4A). Increased age acceleration was found in children with IBD compared to non-IBD children (p = 0.0011). Following regression analysis, the median age acceleration was found to be 3.53 years in IBD patients (Figure 4B). Subtype analysis showed an increased median age acceleration of 4.52 years in the CD group. A total of 20 children with IBD were identified as extreme outliers, with >10 years age acceleration at diagnosis. One child was noted to have derived age acceleration of 39 years; this individual escalated to biologics within 1 month of sampling. Advanced age acceleration was not correlated with clinical parameters, including location and behavior for CD and extent and severity for UC. Escalation to the need of biologic or surgery was also not associated with increased age acceleration.

3.5. Genetic variant-driven methylation differences defined by meQTL analysis

We compared the contribution of all cis-meQTLs (SNPs n = 654 027, methylation n = 773 449) between IBD and controls in the UK2 cohort. A total of 800 515 cis-meQTLs were found to be significant following FDR analysis. Across the genome, the majority of meQTLs reside within intergenic regions. Chromosome 6 contains a disproportionately high number of meQTLs (n = 120 309, 15.2%) compared to any other chromosomes (Figure 5A). On further analysis by subchromosomal location, a nonrandom distribution of meQTLs was observed with a disproportionate number residing between 6p22.1 and 6p21.33, which encompasses HLA-associated genes. Implicated loci include RNF39, HLA-A/B/E/F/G, LTA, and HLA-DQB1-DQB2-DQA1 (Figure 5B).

3.6. Investigation into intergeneration inheritance of IBD-associated methylation changes and imprinted genes

To address the issue as to whether methylation signatures associated with childhood-onset IBD are inherited, we examined statistical evidence of correlation of methylation levels in children with IBD and their parents. The 100 methylation sites, which were most strongly associated with IBD in both the UK1 and UK2 inception cohorts, were assessed for evidence of correlation in the parental trios cohort (UK3). None of the CpG sites tested were correlated between parents and children after multiple correction testing (Supplementary Table).

We extended our analysis by carrying out a 2-step process for investigating DMRs within regions of the genome with known a priori scientific evidence for intergenerational imprinting. We firstly assessed DMRs using the UK2 cohort as the discovery cohort; 11 DMRs were found to reside within imprinted regions of the genome (Supplementary Table). Within Guanine Nucleotide-Binding Subunit (GNAS), which encodes highly isoform-dependent products, 3 different DMRs were observed. The same analysis was then performed in data generated from UK2; 6 DMRs within imprinted regions were implicated in both cohorts (Table 3). Of the overlapping DMRs found between both cohorts, 2 are from paternal alleles, 2 from maternal alleles, and 2 regions are from isoform-dependent regions.

We further performed DMR analysis by comparing either father and child or mother and child pairs in the parental trios cohort. In total, evidence for imprinting of 3 DMRs was found between child and father; these had been previously implicated in the UK2 cohort (Supplementary Table) but were not found in the UK1 cohort. These regions were within the gene KCNQ1 (11p15.5), close to CD81, PLAGL1 (6q24-q25), and HOXA2-A3 (7p15-p14). No DMRs were observed between mother and child.

3.7. Parental smoking habit associated with UC but not CD

We assessed the influence of passive smoking exposure on IBD subtypes in the UK3 cohort of parent-child trios. We observed a difference in the proportion of children with UC who had been exposed to passive smoking compared to children with CD in the parental trio cohort. In total there were, two nonexposed children with UC vs 11 definitely exposed; 30 nonexposed CD children vs 15 definitely exposed; Fisher’s exact test between CD and UC (p = 0.001; 95% CI, 0.009-0.51; OR = 0.09) (Supplementary Table).

We sought to replicate this finding. A total of 44 of the patients or parents of the child provided data on smoking habit in the UK2 cohort (involving 17 children with CD and 14 UC). In children developing CD, 15 had no exposure to passive smoking vs 2 who had definite exposure; whereas for UC, 7 children had no exposure vs 7 who had definite exposure (p = 0.04387; 95% CI, 0.01-1.01; OR = 0.14) (Supplementary Table).

Fisher’s exact test for the combined group of children studied in the UK2 and UK3 cohorts confirmed that children developing UC were more likely to have had passive smoke exposure in early life compared to children with CD (p = 0.0008; 95% CI, 0.06-0.55; OR = 0.19) (Supplementary Table).

4. Discussion

This study provides further insights into the reproducibility, biological importance, and translational potential of epigenetic analyses in complex diseases. We provide evidence confirming the characteristic pattern of DNA methylation present at diagnosis in children with CD and UC, complementing and supporting previous studies in children with CD, and in North European adults with CD and UC. Together, these studies demonstrate a remarkable degree of replication of the circulating methylome in both adult-onset and childhood-onset disease, providing a catalyst to explore other immune-mediated diseases. Using these data, we have been able to derive and validate a simple 4-probe biomarker panel with high accuracy and specificity for both CD and UC in children. Additionally, the findings provide insights into pathogenic mechanisms complementary to those discovered by GWAS studies. Thus, we define the genetic contribution to shaping the epigenome, highlighting the contribution of the major histocompatibility complex (MHC); provide first evidence for paternal imprinting in IBD; and demonstrate the influence of aging, inflammation, and passive smoke exposure on the epigenome in disease.

The consistency observed between the UK inception cohorts for the most strongly associated sites demonstrating altered methylation is striking.^14,34 The present dataset demonstrates more regions of altered methylation in CD than in UC, although only 1 marker is implicated in direct comparison of CD vs UC; and the top markers are strongly correlated in both diseases. Our findings in children are also highly consistent with the results from reported EWAS in adult IBD populations in Northern Europe, and the associated meta-analysis.^15–17 In the present EWAS, the differential methylation at 3 CpG sites within the gene zinc finger and BTB domain containing 16 (ZBTB16) is of particular note. ZBTB16 has been linked to autophagy through the degradation of Atg14L³⁶; zinc-binding groups also play a crucial role in interacting with histone deacetylase.³⁷ Additional sites resided in SOCS2 and IFNAR1, both playing a role in modulating JAK-STAT signaling pathways. We confirm the dysregulation of methylation at CpG sites associated with the VMP1 and ZBTB16 loci, the most consistent sites in previous studies.

The predictive accuracy (0.91-0.93) of the 4-site-based diagnostic model derived in this study for children offers promising implications for clinical practice, with similar predictive power in both CD and UC. Notably, the methylation sites within RPS6KA2 and VMP1 (mir-21) carry substantial weight in the model. These sites are situated in genomic regions previously associated with IBD pathogenesis.^15,38 The diagnostic accuracy derived is greater than that for other blood-based biomarkers and is similar to that reported for fecal calprotectin, currently the most commonly used biomarker in this context. The model appears to be specific for IBD in contrast to  celiac disease, which also involves intestinal inflammation, and rheumatoid disease, a systemic inflammatory condition. The high accuracy in children with no CRP elevation at diagnosis is particularly noteworthy in clinical practice and suggests that described biomarker model is not secondary to inflammation alone. Interestingly, our model performed less effectively in adult IBD populations than in children. This is perhaps explained by the well-accepted stochastic methylation changes that occur as a result of lifetime environmental exposures and the aging process per se.³⁹ Further work is needed to explore these methylation patterns in other ethnicities and geographical populations as an early diagnostic and prognostic markers, to explore the predictive ability in other immune-mediated conditions and to look at the impact of IBD disease course and therapy over time on the biomarker, particularly with regards to exposure to both immunosuppressant therapies and cumulative inflammation. The model performs with high accuracy as a diagnostic tool at presentation; however, performance drops following a period of treatment after diagnosis in the RISK cohort. However, of particular note is the fact that the model performs accurately as a diagnostic test in children with low CRP levels at presentation and that the data demonstrate specificity for IBD when compared to other inflammatory diseases.

DNA methylation panels have recently emerged as molecular biomarkers for biological aging; several epigenetic clocks, which utilize methylation across cell types, predict a spectrum of ages from early in life to centenarians.⁴⁰ Positive age acceleration is associated with adverse outcomes, such as cancer⁴¹ and cardiovascular disease,⁴² and with all-cause mortality.⁴³ Furthermore, acute age acceleration has been found during pregnancy and been associated with heightened circulating markers of inflammation. In our cohorts, we found significant age acceleration in IBD patients compared to controls, most notably in CD patients. In some instances, extreme age acceleration of greater than 30 years was observed. These data are consistent with findings in adult IBD populations, where CD patients appear to show the greatest age acceleration.¹⁷ Recent data suggest reversibility can occur in some disease states.^44,45 Further studies are now needed to identify whether either acute or chronic inflammation is the biggest driver of increased age acceleration in patients; to assess the reversibility of the observed aging with therapy; and, most importantly, to characterize the consequences of observed premature epigenetic aging in children with IBD.

Our targeted approach of cis-meQTL analysis confirms the importance of germline variation in the shaping of the methylome.^14–16,46 Thus, we identified a high proportion of methylation differences in IBD patients independent of sex, age, and cell count proportions. Of particular interest, our genome-wide analysis highlights the nonrandom clustering of meQTLs residing within chromosome 6 (6p22.1-p21.33), strongly implicating the functional relevance of HLA-related genes in shaping the methylome in immune diseases, as previously reported in rheumatoid arthritis and multiple sclerosis.⁴⁷ These findings support the hypothesis that the observed methylation-associated differences in children with IBD largely reflect the combined effects of the exposome, and active inflammation and germline variants acting as meQTLs.

Regions within the gene GNAS (20q13.2-3) were repeatedly found to be shared between IBD and control. GNAS is a complex gene locus, which gives rise to multiple gene products including NESP55 (maternally derived isoform) and XLas (paternally derived isoform). Genetic mutations in GNAS have previously been associated with both colorectal cancer⁴⁸ and colitis-associated cancers.⁴⁹ Downstream from GNAS at position 20q13.3.2 is the zinc finger protein 831 (ZNF831), which is also a well-described IBD genetic susceptibility locus.⁵⁰ We found several methylation sites implicated at rs259964 within ZNF831 following our targeted meQTL analysis. Further investigation is required into understanding this locus to understand functional implications of isoform expression in IBD.

Exploring passive smoking exposure in pediatric IBD, we observed a strong bias towards the development of UC rather than CD in children exposed to parental smoking (p = 0.0008). Although these observations are discordant with some historical reports in the literature,^51,52 our study has the strength not only that we involved 2 independent cohorts but also that we were able to confirm smoking status in the parents of the Southampton cohort (in which the effect size was greatest).

While our study contributes valuable insights into the circulating methylome of pediatric IBD, it is essential to acknowledge areas for further work. We have studied North European populations and extended the research using available data from North America, confirming the accuracy of the diagnostic biomarker we derived; however, these findings need to be explored in other diverse populations, especially in ethnic groups where the incidence has increased rapidly over the last generation. We have limited our studies to blood samples, rather than gut samples^14–17; and direct comparison of circulating and gut-associated methylomes is valuable.^53–56 The impact of these methylation findings in blood on changes at the tissue level needs to be evaluated. A level of consistency has been found between meQTL findings in blood and gut,⁴⁶ and our whole blood findings also replicate 21 of these loci found in ileum samples. The cohort recruited for our diagnostic biomarker study did not allow direct comparison of the accuracy of the 4-probe panel against fecal calprotectin. Notwithstanding this, the accuracy we report here compares favorably against published literature for fecal calprotectin⁵⁷ and suggests comparability to other blood-based markers. Further prospective investigation is needed to assess the accuracy formally before clinical translation in the context of a blood-based screening or diagnostic tool. Our findings highlight the need for key data to be collected rigorously regarding environmental exposures in the development of IBD.

In conclusion, these data provide a strong impetus for studies of the epigenome in IBD, and more widely in other complex immune-mediated diseases. The studies provide confidence that the methylation alterations may be detected in cohorts of well-characterized patients, using available technologies. The findings have implications for early clinical translation in biomarker discovery as well as for understanding pathogenesis, specifically the role of the epigenome in mediating gene-environmental interactions. Further work is needed to explore these findings in diverse populations and to define the pathophysiological effects of these alterations in disease.

Supplementary Data

Supplementary data are available online at ECCO-JCC online.

Funding

This study was funded by the Action Medical Research (AMR) (GN2880). HUU was supported by the Biomedical Research Centre, Oxford, The Leona M. and Harry B. Helmsley Charitable Trust, Medical Research Council (MRC) grant, and National Health Service (NHS). JS was funded by the National Institute for Health and Care Research (NIHR), The Leona M. and Harry B. Helmsley Charitable Trust, Crohn’s Colitis United Kingdom (CCUK), European Crohn’s Colitis Organisation (ECCO), and European Council (EC). JJA was funded by an NIHR advanced fellowship. RK was funded by CCUK, National Research Scotland Chief Scientist Office (NRS CSO), and Medical Research Council Clinical Academic Research Partnerships (MRC CARP) fellowships. The views expressed are those of the authors and do not necessarily represent those of the NHS, the NIHR, or the Department of Health.

Conflict of Interest

HHU has received research support or consultancy fees from Janssen, UCB Pharma, Falk, Eli Lilly, GSK, Celgene/BMS, and AbbVie. A patent application (N42837GB) has been filed for the 4-probe diagnostic biomarker discussed.

Acknowledgments

We thank the GI Biobanking Team at TGU, Oxford, for their help in collecting patient samples and DNA extractions. We also thank Shana-Lee Bownes and Jennifer Hollis for collecting patient data. Lastly, we thank all patients and their families for their participation in this study.

Author Contributions

AA, SE, and JS conceived the study. AN, KN, MK, and GC conducted the experimental work. GC, KM, AQ, FG, CJ, RK, NTV, MZ, HHU, RH, GLH, EE-O, DW, JJA, RMB, NMC, and SE collected the clinical samples and medical data. AN, AA, JN, RK, HHU, and JS performed the formal analysis. AN, AA, JN, GC, KM, AQ, FG, CJ, RK, NTV, NMC, RH, GLH, EE-O, MK, JJA, RMB, DW, MZ, SE, HHU, and JS contributed to the writing of the manuscript. AA and JS acquired funding for the study.

Data Availability

Data are available upon reasonable request.

References