br Methods br Results br Discussion Differential methylation
Discussion Differential methylation of specific CpG loci within the promoter region of ANRIL at the CDKN2A gene locus was predictive of measures of adiposity in four independent cohorts. In the UK SWS cohort (n=247), umbilical cord CDKN2A methylation was negatively associated with %fat mass at ages 4 and 6-years, in the Singapore GUSTO cohort (n=305), it was negatively associated with skinfold thickness or ponderal index at ages 7days and 18months, in the Australian RAINE cohort (n=814), whole blood CDKN2A methylation was negatively associated with concurrent BMI in 17year olds and in the UK BIOCLAIMS cohort (n=81) obese individuals had lower adipose tissue CDKN2A methylation compared to lean individuals. Although the negative association between CDKN2A methylation and adiposity measures was observed across the 4 cohorts, in 3 tissues and 3 age groups, there were differences in the nature of the relations. In the SWS cohort, lower umbilical cord CDKN2A DMR methylation predicted increased adiposity at ages 1, 4 and 6years but not at birth, suggesting that CDKN2A methylation may be linked to the gain of fat mass postnatally. Two developmental pathways to obesity have been suggested, one associated with fetal undernutrition, followed by rapid postnatal weight gain, while the second pathway is linked to fetal overnutrition and greater fat mass at birth (Crozier et al., 2012, 2010). Interestingly, fetal growth faltering in late NSC 74859 was associated with lower CDKN2A methylation at birth, suggesting that CDKN2A methylation may be a marker for the former pathway. In the GUSTO cohort, there were associations between the methylation of CDKN2A with measures of adiposity from 7days after birth. The timing of the associations between the cohorts may reflect differences in fat deposition or rate of body fat accumulation between birth and early infancy, potentially due to genetic and epigenetic ancestral differences between the different cohorts that can affect timing of associations. The associations between the methylation of the individual CpGs within the CDKN2A DMR and adiposity were also not identical across the cohorts. The methylation status of CpG sites is often closely aligned to that of their neighbors, however developmentally induced changes are often CpG site specific (Lillycrop et al., 2008). In the DMR examined, the CpGs were divided into distinct groupings based upon strong inter-correlation, but these groupings had distinct differences between them, suggestive of tissue or population differences. This is re-iterated by the EMSAs, which showed that different protein complexes bind across the CpG sites within the DMR, suggesting they are likely to be differentially regulated. This is also supported by the in vitro studies, which showed that mutagenesis of the individual CpG sites had differential quantitative effects on expression. An inverse association between CDKN2A methylation in umbilical cord and measures of adiposity was observed in the SWS and GUSTO cohorts. Surrogate variable analysis to control for cellular heterogeneity in cord tissue had little effect on our findings suggesting that, in cord, differences in CDKN2A methylation are not dependent upon variation in cellular heterogeneity. An inverse association between CDKN2A methylation and obesity was found in adipose tissue from adults, suggesting that differential CDKN2A methylation may be a robust marker of adiposity across cord and adipose tissue. The negative association between CDKN2A methylation and BMI was again observed in peripheral blood samples from the RAINE cohort. Blood is a heterogeneous material with dynamic cellular proportions and DNA methylation is differential across cell types, making it challenging to resolve the causal relationships between methylation of whole tissue, cellular mix and phenotype. Cellular proportions were measured in the RAINE cohort, however in this (Table S11A) and other studies, cellular proportions strongly associate with BMI. As would be expected methylation levels also differ across cell types (Table S11B). Multicollinearity was seen between all cell types and log BMI for all CpG clusters within the CDKN2A DMR. Therefore adjustment of the association between methylation and BMI for cellular proportions would violate the assumption of non-collinearity implicit in regression analysis. Consequently, we were unable to assess the dependence on cellular heterogeneity of the association of CDKN2A methylation in blood and BMI, and therefore cannot exclude differences in cellular composition as the explanation for the association between CDKN2A methylation in peripheral blood and adiposity in late adolescence. However; the existence of this association in very different tissues, which are (in the case of adipose) more homogenous in their cellular composition, and the fact that in umbilical cord it survived adjustment for cellular heterogeneity, suggests that the relationship is not completely explained by the confounding influence of cell type.