Cindy Boer

48 | Chapter 1.2 possible interaction partners. Thus, chromatin conformation maps are a valuable re- source to identify potential causal variants and possible causal genes in GWAS. TAD boundaries can be used to limit possible causal genes, and reference epigenome maps can help to identify active genes and variants in cell‐specific active regulatory elements. Interpretation of Epigenetic Studies: Problems and Tools Cause or effect: confounding and reverse causation Interpretation of epigenetic studies is not trivial because many factors can influence the association between the phenotype and epigenetic features. In contrast to DNA‐ sequence variants, epigenomic features are dynamic, meaning that they can be highly tissue specific (especially enhancers), dependent on environmental stimuli and devel- opmental stage. This makes epigenetic analysis vulnerable for classical epidemiological pitfalls such as confounding and reverse causation. Besides the previously mentioned cellular heterogeneity as a potential confounder in case of multicellular tissues, such as bone or synovial tissue, a major issue is reverse causation. In cross‐sectional studies, one can never be sure whether the epigenomic features cause the phenotype or vice ver- sa. With respect to the epigenetic studies performed in cartilage and bone, this problem is also realistic. The massive epigenomic deregulation apparent in degraded cartilage can be a consequence of a process initiated by an entirely different cause of the disease. It is possible to use known genetic association and the concept of mendelian random- ization to investigate the potential causal relationships between DNA methylation and the phenotype of interest.[115] In these kinds of studies, single nucleotide polymor- phisms (SNPs) known to influence the CpG site are used as genetic instruments to test for causation. Similarly, SNPs associated with the trait/disease of interest can then be used to test whether they are associated with methylation levels at the same CpG site. In this way, the direction of cause can be disentangled ( Figure. 2 ). This approach has recently been used to show that DNA methylation differences associated with BMI are predominantly a consequence of adiposity, rather than a cause[116]. In addition, this methodology has also been used to suggest that several known risk factors do not show a causal effect on bone fracture, except for BMD[117]. Similarly, high BMI was shown to cause OA, but other known clinical risk factors did not have a causal relationship with OA[118]. Typically, these mendelian randomization studies need genetic data with large sample sizes, which is increasingly available for both osteoporosis (GEFOS con- sortium) and osteoarthritis (Genetics of OA‐consortium). However, sample sizes for the methylation studies in target tissue are typically small ( Table 1 ), and therefore collab- oration and meta‐analysis across the different data sets are needed. This is increasingly recognized in the field[119].