DNA methylation-based classification of hematolymphoid neoplasms. DNA methylation-based classification of hematolymphoid neoplasms

Accurate pathological diagnosis is crucial for optimal management of cancer patients. For a number of hematolymphoid tumor entities, standardization of the diagnostic process has been shown to be particularly challenging - with substantial inter-observer variability in the histopathological diagnosis of many tumor types. Genome-wide DNA methylation profiling has been shown to contribute to accurate and precise tumor classification and diagnosis in several tumor types, including central nervous system neoplasms. We herein present the development of a comprehensive approach for DNA methylation-based hematolymphoid tumor classification across most entities and age groups and demonstrate its application in a clinical setting, defining 44 methylation classes comprising a good proportion of recognized hematolymphoid tumor types. In several cases, methylation signatures separated recognized tumor types into subclasses that showed statistically significant and distinct patient outcomes. We validate the classifier in independent samples that were not utilized in classifier development and show that the reliability of this method is highly accurate (concordance 97%) with a substantial impact on diagnostic precision in addition to standard methods. We show a relationship of methylation class match scores with tumor purity, finding that high-tumor-purity samples are more likely to receive a high-confidence match with our classifier. Finally, we show the impact of the classifier in clinical practice, where classifier results resulted in a change in diagnosis in specific hematolymphoid neoplasms. For broader accessibility, a free and user-friendly online portal for access to the hematolymphoid tumor methylation classifier will be available for external users. Overall design: The reference set from previously published and inhouse data was used to train classifier. The validation dataset asembled from previously published data and new samples processed inhouse and acquired from collaborators. Classification perfomace estimated on validation dataset.