Here we have performed the first large scale (over 300 samples) genome-wide study of DNAm profiles using a state-of-the art platform that measures the methylation state of over 27,000 CpGs. We have shown that ovarian cancer has a significant impact on the DNAm pattern of peripheral blood cells. While the epigenetic signature we have presented still lacks the high specificity necessary for an immediate diagnostic application, the fact that active ovarian cancer could be predicted with a relative high accuracy (AUC
0.8) from a DNAm profile in blood clearly demonstrates the future potential of epigenetic profiling as a diagnostic tool.
In addition, we provided evidence for the existence of DNAm markers that may serve as early detection or predisposition markers for ovarian cancer. Of the 18 candidate risk markers, 11 and 7 showed hyper and hypomethylation in cancer, respectively, with TSG101
and the pre-mRNA splicing factor SFRS6
both undergoing hypermethylation in cancer. Further confirmation that the markers identified here may serve as early detection or predisposition markers for ovarian cancer will require a large prospective study, which is currently ongoing 
The observed DNAm patterns can be summarised in terms of two biologically distinct signatures. First, the observation that a DNAm signature for aging, characterised by differential methylation of genes with hematopoietic cell lineage and immune system functions is aggravated in the presence of ovarian cancer, suggests that an epithelial tumour and aging elicit common changes in the cellular composition of peripheral blood. This interpretation is further supported by the fact that the same biological terms were strongly enriched among genes differentially expressed between blood cell types 
), and that genes commonly upregulated in granulocytes and lymphocytes showed a differential methylation pattern (Table 1, Figure S4
) consistent with an increased granulocyte to lymphocyte ratio in response to aging or cancer. It is significant that there is independent evidence that both aging and cancer presence lead to a myeloid-skewing in the myeloid/lymphoid differentiation program, with a corresponding higher granulocyte/lymphocyte ratio 
, a pattern consistent with our observation that granulocyte and T-lymphocyte specific genes were enriched among CpGs hypo-and hypermethylated in cancer, respectively. Since this effect was inferred from DNAm profiles, it would appear that the expression of a substantial number of genes characterising bood cell types is under direct epigenetic regulation.
Another related question is whether the diagnostic epigenetic signature is specific to ovarian cancer. If different cancers elicit similar immune responses and thus similar changes in blood cell type composition, we could expect a proportion of the identified diagnostic signature to be non-cancer specific. However, to conclusively determine that this is the case and that age and cancer are not mediating immuno-compromising effects via epigenetic modification of particular cell-types, requires additional data not provided by our study. Similarly, whether parts of the observed cancer signatures may be causally involved in the disease must await further investigations.
A second DNAm signature was characterised by CpGs undergoing hypermethylation with age and was highly enriched for developmental genes and REST
-targets. Given that REST
) is involved in suppression of genes that are required for differentiation of embryonic and adult stem cells 
, the age-induced hypermethylation of REST
-targets, if confirmed in a stem-cell population, may represent a generic mechanism for age-associated loss of stem-cell function and increased predisposition to cancer 
Finally, our finding that most of the methylation variability is associated with niCpGs further supports the view that most of the phenotypically relevant DNAm variation is to be found in regions other than CpG islands 
. Confirming this further, the observed association between gene expression of different blood cell-types and the pattern of DNA hypo and hypermethylation was much stronger for niCpGs than iCpGs (data not shown).
In summary, this work demonstrates that DNAm profiling in blood holds significant promise as a future diagnostic or risk-prediction tool and warrants further higher CpG-coverage studies to fully elucidate this role.