Beschreibung:
<jats:title>Abstract</jats:title>
<jats:p>DNA methylation profiling led to the definition of ten molecular types of ependymomas with distinct biological and clinical features. The largest published reference cohort consists of 500 ependymomas encompassing nine molecular types that were known at this point. Our study aimed to confirm and refine molecular and clinical characteristics of ependymoma types and subtypes based on a considerably larger, well-annotated cohort of ependymal tumors. Here, we analyzed previously published and newly generated DNA methylation profiles, generated on the Illumina 450k and EPIC arrays, from a total of n=1,801 ependymomas. We looked at both global DNA methylation as well as inferred copy number profiles and correlated these data with clinical parameters. Patients with EPN-PFA, EPN-RELA, and EPN-MYCN tumors showed the worst outcome with 10 year-overall survival rates of 56%, 64%, and a 5 year-overall survival rate of 73%, respectively. EPN-PFA, the most frequent tumor type in our series, occurred in children in &gt;98% of cases, epigenetically split into two major subtypes, and harbored chromosome 1q gains and 6q losses as markers for worse survival (p&lt;0.0001 and p=0.042, respectively). In supratentorial EPN-RELA, a combined loss of CDKN2A/B indicated a worse survival compared to the wildtype (p=0.0004), but the loss of only CDKN2A or CDKN2B did not. Ten out of 169 EPN-RELA were located in the posterior fossa, and Kaplan Meier estimators showed that these tumors relapsed or progressed even earlier than supratentorial cases with a combined loss of CDKN2A/B (p=0.0062). SP-MPE and PF-SE, which are generally regarded as non-aggressive tumors, only had a 10-year progression-free survival of 60% and 63%, respectively, in our series. In ongoing analyses, we are training machine learning algorithms to more accurately predict survival and response to treatments. These methods shall help clinicians make informed decisions regarding treatment options on new patients.</jats:p>