Investigating the DNA methylation profiles of children with oligoarticular juvenile idiopathic arthritis (JIA)

Abstract

Juvenile idiopathic arthritis (JIA) is a complex autoimmune disease affecting children aged between 6 months and 16 years. JIA represents a group of 7 subtypes of disease, with the most common being oligoarticular JIA (oJIA). Despite a prevalence of up to 1 in 400, rates similar to those in T1D, JIA research is relatively sparse. Research into disease pathogenesis has largely focussed on genetic risk factors, and has also identified CD4+ T-cells as likely to mediate the autoimmune process. However, research is particularly needed regarding diagnosis and prognosis of disease and its outcomes. Currently, diagnosis is almost entirely dependent on clinical observation and history, with little in the way of biomarkers to classify patients or to guide clinical management.

Epigenetics represent biological modifications to DNA and chromatin that control gene expression and chromatin structure. DNA methylation is perhaps the most accessible modification available for study, and is known to modulate immune cell function particularly amongst CD4+ T-cell subsets. A number of autoimmune diseases have reported significant DNAm associations, and have also provided intriguing data on the potential of DNAm to predict clinical outcomes. This study hypothesised that DNAm is important in oJIA pathogenesis, and potentially provides a biological basis for the diagnosis and prognosis of disease. This study utilised CD4+ T-cells and a case-control study design to analyse the associations between DNAm and oJIA, with data generated from the Illumina Infinium HumanMethylation450 BeadChip array. Cases were matched with controls according to age and sex. Further, cases were subtyped according to current diagnostic criteria and had active disease, both of which attempted to ensure all cases were clinically homogeneous.

The first aim was to profile DNAm in oJIA cases compared to controls. Processing of data through analysis pipelines resulted in high quality data. Differential methylation analysis suggested that oJIA cases and controls could be segregated in cluster analysis using DNAm data, despite no genome wide significant hits being produced. Immune system pathways analysis suggested the top hits were relevant to disease, being enriched for receptor binding of cytokines such as IL6, IL17 as well as MHC class II. In addition, a number of top ranking probes were enriched within cell death and survival functions. Indeed, gene expression data suggested genes within those pathways were also correlated with DNAm. Technical validation of a selection of probes was highly successful, with all probes validating. A small replication study, however, was not able to reproduce these findings. Of particular note, a wide distribution of DNAm values was observed for many of the validated probes. Since technical validation was so successful, this DNAm heterogeneity potentially derived from sample group heterogeneity, which may well have played a part in difficulties replicating data.

Therefore, biological sources of heterogeneity were explored in chapter 5, focussing primarily on the genetic associations with DNAm. Probes utilised for technical validation were analysed for genetic associations associating with either mean or variable DNAm. Both analyses suggested that the most robust associations were for known mQTLs and enhancer SNPs. Indeed, DNAm differences according to genotype were up to 13% and 27% for 2 probes analysed, representing a many-fold difference over case-control differences (typically approximately 5%). Combined with an intermediate level of minor allele frequency for many of these robustly associated SNPs, these mQTLs represent a likely source of biological variation contributing to oJIA DNAm variation. These minor allele frequencies increase the likelihood of inadvertent sampling bias, potentially resulting in difficulties in replicating DNAm data. Deeper analysis provided some initial indication that these mQTLs may also be potential oJIA risk loci, with the most significant associations again coming from known mQTL or enhancer SNPs. This also suggested DNAm data may well identify regions of interest for genetic risk loci discovery.

The final chapter hypothesised that sources of potential clinical heterogeneity not captured within current classification criteria may well lead to DNAm heterogeneity, as could recognised subgroups within oJIA. Of primary focus, age of disease diagnosis was assessed for associations with DNAm. This study found that case-control analyses of older diagnosed samples (greater than or equal to 6 years) resulted in case-control clustering using far fewer probes. Indeed, the reduction of probes required for clustering was more pronounced in the analysis of younger diagnosed samples (less than 6 years of age), and also resulted in a genome wide significant hit. These subgroups represented 2 highly divergent populations, since top ranking probes from each subgroup had virtually no overlapping probes. This data suggested that age subgroups in oJIA represent sources of sample heterogeneity, leading to DNAm heterogeneity. Technical validation for a large majority of the select probes from the younger-diagnosed analysis was also successful. However, a small replication study could not reproduce these initial findings. In light of the potential for mQTLs to have pronounced effects on DNAm, as explored in chapter 5, larger replication groups (or, indeed, discovery groups) will likely be needed to mitigate the risk of sampling error to enable reproduction of findings.

OJIA heterogeneity was also explored by looking at known subgroups, Persistent vs Extended disease. A number of oJIA cases would go on to develop extended disease, and the possibility existed for DNAm signatures to identify these cases prior to disease extension. This was indeed the case, with an exploratory analysis suggesting a number of probes can cluster persistent cases from extended-to-be cases. Further, these probes were able to produce a highly sensitive and specific test to predict disease extension, thereby providing a proof of principle for a prediction test using DNAm data.

This study is the largest case-control analysis of JIA DNAm to date, and provided insights into the potential for DNAm to identify pathogenic pathways, identify sources of oJIA heterogeneity, and opened the possibility for biological markers of disease to be used in clinical management. The findings regarding the pronounced effect of mQTLs on DNAm also suggest that genetics is a large source of DNAm variability, far larger than group differences typically found in a complex diseases (such as oJIA). The identification of subgroup specific differences, even with a clinically homogeneous subtype, warrants further investigation to explore potential differences in pathogenesis between age groups and the use of DNAm as biomarkers for classification or disease management.