Abstract 8- 0900-0915
Category: Clinical

At the end of the session,
participants will be able to:

  1. Review the impact of methylation profiling on CNS tumor classifications.
  2. Understand some clinical issues and pitfalls with implementing a methylation assay in the clinical neuropathology workflow.

COI Disclosure:

None to disclose


Adrian Levine completed his medical school at the University of Western Ontario (2016) and neuropathology residency at the University of British Columbia (2021). He is currently in the Clinician Investigator Program, doing a research fellowship at the Hospital for Sick Children in Toronto, under the supervision of Dr Cynthia Hawkins. His clinical interests include tumor neuropathology and molecular pathology, and research interests include cancer genetics and bioinformatics/machine learning.


Adrian Levine,1,2 Michelle Ku,1 Scott Ryall,1 and Cynthia Hawkins1

  1. Department of Pediatric Laboratory Medicine, Hospital for Sick Children, Toronto, ON
  2. Clinician Investigator Program, University of British Columbia, BC

Target Audience:
Pathologists, Residents

Medical Expert (the integrating role), Communicator, Scholar, Professional

Clinical implementation of methylation profiling: the SickKids experience


Over the past decade, methylation profiling and machine learning based classification has had a major impact on the diagnostic framework for CNS tumors, however there remains inconsistent and limited access to clinical testing for this analysis. At SickKids, we have implemented a clinically validated methylation assay using the Illumina EPIC platform for the Heidelberg CNS tumor classifier, copy number profiling, and MGMT promoter methylation status.

 We have now run 136 samples for clinical use (plus additional samples for validation), which have been classified through both v11 and v12.5 of the Heidelberg classifier. Many external samples were not received with a histologic diagnosis, those that do have a diagnosis include 34 low grade gliomas (LGG), 19 high grade gliomas (HGG), 10 ependymomas, 4 pineal tumors, and 4 medulloblastoma/embryonal tumor.

 Overall, only 34 samples (25%) classified with confidence score >0.85 on the v11 classifier and 77 samples (57%) on v12.5. The medulloblastomas all classified with high confidence scores, as did 7/10 ependymomas, and 2/3 papillary tumors of the pineal region. 9/19 HGGs classified with high confidence – most of which were methylation-defined subtypes of IDH-mutant and IDH-wildtype astrocytomas or glioblastomas, with unclear clinical significance. Notably, though 2 of the cases that were referred as HGG were in fact other tumor types, one being an ATRT, and the other an ependymoma with ZFTA-fusion. The classifier had particular challenge with LGGs, of which only 17 (50%) classified with high confidence, including one that was erroneously called “Control tissue, reactive tumor microenvironment”.