A novel hypothesis-generating approach for detecting phenotypic associations using epigenetic data

<b>Aim:</b> Hypotheses about what phenotypes to include in causal analyses, that in turn can have clinical and policy implications, can be guided by hypothesis-free approaches leveraging the epigenome, for example. <b>Materials &amp; methods:</b> Minimally adjusted epigenome-wide association studies (EWAS) using ALSPAC data were performed for example conditions, dysmenorrhea and heavy menstrual bleeding (HMB). Differentially methylated CpGs were searched in the EWAS Catalog and associated traits identified. Traits were compared between those with and without the example conditions in ALSPAC. <b>Results:</b> Seven CpG sites were associated with dysmenorrhea and two with HMB. Smoking and adverse childhood experience score were associated with both conditions in the hypothesis-testing phase. <b>Conclusion:</b> Hypothesis-generating EWAS can help identify associations for future analyses. To inform policy and improve clinical practice, it is important that researchers who study people's health find out which traits might increase the risk of illness. However, it can be difficult to know which traits should be looked at. In this study, we wanted to look for traits that might increase the risk of painful and heavy periods, using data about the switches that turn our genes on and off. There are some people in the Children of the 90s study that have data on gene switches. We compared all the switches between those with and without painful or heavy periods. For painful periods, we found links with seven switches and for heavy periods, we found two. We then used another data source, called the EWAS Catalog, to see which traits were associated with these switches. The traits we found included body size, smoking and child abuse. Finally, when using data on traits from the wider Children of the 90s group, we found that smoking and more difficult childhoods were some of the traits related to painful and heavy periods. A good thing about this approach is that we could find new traits that might increase the risk of painful or heavy periods; these should be looked at in future studies. Leveraging EWAS data can identify novel risk factors for conditions such as menstrual symptoms in future, causally motivated analyses: a proof-of-concept study in the Children of the 90's cohort (ALSPAC) <b>Background</b>DNA methylation can be explored in an epigenome-wide association study (EWAS) context to identify causal mechanisms or confounding relationships between exposures that can alter the epigenome and phenotypes of interest.In a minimally adjusted EWAS acting as hypothesis-generating, associations can be identified that represent either one of these relationships to be further explored in causally motivated analyses for conditions that are understudied.In the present study, we demonstrated the utility of a hypothesis-generating EWAS approach followed by a hypothesis-testing phase using logistic regression, investigating two understudied conditions as examples: dysmenorrhea (painful periods) and heavy menstrual bleeding (HMB). DNA methylation can be explored in an epigenome-wide association study (EWAS) context to identify causal mechanisms or confounding relationships between exposures that can alter the epigenome and phenotypes of interest. In a minimally adjusted EWAS acting as hypothesis-generating, associations can be identified that represent either one of these relationships to be further explored in causally motivated analyses for conditions that are understudied. In the present study, we demonstrated the utility of a hypothesis-generating EWAS approach followed by a hypothesis-testing phase using logistic regression, investigating two understudied conditions as examples: dysmenorrhea (painful periods) and heavy menstrual bleeding (HMB). We used the Avon Longitudinal Study of Parents And Children (ALSPAC) to identify cases of adolescent dysmenorrhea and HMB; those with epigenetic data from ARIES were included in two hypothesis-generating EWAS where each condition served as the exposure. The hypothesis-generating phase consisted of two minimally adjusted EWASs of dysmenorrhea and HMB to identify differentially methylated CpGs. CpGs identified in the hypothesis-generating phase were then searched in the EWAS Catalog to find associated traits with the CpG and its resident gene. The hypothesis-testing phase involved taking identified traits and investigating their association with dysmenorrhea and HMB in the wider ALSPAC cohort. Having found seven differentially methylated CpGs for dysmenorrhea and two for HMB, we searched them in the EWAS Catalog and identified phenotypes associated with each of them. In the hypothesis-testing phase, we proxied the phenotypes found in the EWAS Catalog using variables from ALSPAC and included them in minimally adjusted logistic regression models where each condition served as the outcome. Using this approach, we found that smoking and alcohol use at age 13 was associated with dysmenorrhea and HMB; higher cotinine levels at age 7 was associated with HMB. Higher adverse childhood experience (ACE) score was associated with both conditions. We identified several potential targets of investigation for future research into risk factors for dysmenorrhea and HMB. Although temporality was not easily established in the present study and causality indeterminable, we leverage confounding to guide future causally motivated analyses in other cohorts with menstruation data.