Summary Accurate pathological diagnosis is crucial for optimal management of cancer patients. For the ~100 known central nervous system (CNS) tumour entities, standardization of the diagnostic process has been shown to be particularly challenging - with substantial inter-observer variability in the histopathological diagnosis of many tumour types. We herein present the development of a comprehensive approach for DNA methylation-based CNS tumour classification across all entities and age groups, and demonstrate its application in a routine diagnostic setting. We show that availability of this method may have substantial impact on diagnostic precision compared with standard methods, resulting in a change of diagnosis in up to 12% of prospective cases. For broader accessibility we have designed a free online classifier tool (www.molecularneuropathology.org) requiring no additional onsite data processing. Our results provide a blueprint for the generation of machine learning-based tumour classifiers across other cancer entities, with the potential to fundamentally transform tumour pathology.

Ependymal tumors across age groups are currently classified and graded solely by histopathology. It is, however, commonly accepted that this classification scheme has limited clinical utility based on its lack of reproducibility in predicting patients’ outcome. We aimed at establishing a uniform molecular classification using DNA methylation profiling. Nine molecular subgroups were identified in a large cohort of 500 tumors, 3 in each anatomical compartment of the CNS, spine, posterior fossa, supratentorial. Two supratentorial subgroups are characterized by prototypic fusion genes involving RELA and YAP1, respectively. Regarding clinical associations, the molecular classification proposed herein outperforms the current histopathological classification and thus might serve as a basis for the next World Health Organization classification of CNS tumors.

SUMMARY Despite extensive study, few therapeutic targets have been identified for glioblastoma (GBM). Here we show that patient derived glioma sphere cultures (GSCs) that resemble either the proneural (PN) or mesenchymal (MES) transcriptomal subtypes differ significantly in their biological characteristics. Moreover, we found that a subset of the PN GSCs undergo differentiation to a MES state in a TNFα/NF-κB dependent manner with an associated enrichment of CD44 subpopulations and radio-resistant phenotypes. We present data to suggest that the tumor microenvironment cell types such as macrophages/microglia may play an integral role in this process. We further show that the MES signature, CD44 expression, and NF-κB activation correlate with poor radiation response and shorter survival in patients with GBM.

Immune checkpoint blockade represents a major breakthrough in cancer therapy, however responses are not universal. Genomic and immune features in pre-treatment tumor biopsies have been reported to correlate with response in patients with melanoma and other cancers, but robust biomarkers have not been identified. We studied a cohort of metastatic melanoma patients initially treated with cytotoxic T-lymphocyte-associated antigen-4 (CTLA-4) blockade (n=53) followed by programmed death-1 (PD-1) blockade at progression (n=46), and analyzed immune signatures in longitudinal tissue samples collected at multiple time points during therapy. In these studies, we demonstrate that adaptive immune signatures in tumor biopsy samples obtained early during the course of treatment are highly predictive of response to immune checkpoint blockade, and also demonstrate differential effects on the tumor microenvironment induced by CTLA-4 and PD-1 blockade. Importantly, potential mechanisms of therapeutic resistance to immune checkpoint blockade were also identified. Significance These studies demonstrate that adaptive immune signatures in early on-treatment tumor biopsies are predictive of response to checkpoint blockade, and yield insight into mechanisms of therapeutic resistance. These concepts have far-reaching implications in this age of precision medicine, and should be explored in immune checkpoint blockade treatment across cancer types.

Immune checkpoint blockade produces clinical benefit in many patients. However better biomarkers of response are still needed, and mechanisms of resistance remain incompletely understood. To address this, we recently studied a cohort of melanoma patients treated with sequential checkpoint blockade against cytotoxic T lymphocyte antigen-4 (CTLA-4) followed by programmed death receptor-1 (PD-1), and identified immune markers of response and resistance. Building on these studies, we performed deep molecular profiling including T-cell receptor sequencing (TCR-seq) and whole exome sequencing (WES) within the same cohort, and demonstrated that a more clonal T cell repertoire was predictive of response to PD-1 but not CTLA-4 blockade. Analysis of copy number alterations identified a higher burden of copy number loss in non-responders to CTLA-4 and PD-1 blockade and found that it was associated with decreased expression of genes in immune-related pathways. The effect of mutational load and burden of copy number loss on response was non-redundant, suggesting the potential utility of a combinatorial biomarker to optimize patient care with checkpoint blockade therapy.