Preview

PDF, English Download (11MB) | Terms of use

Abstract

Project 1: Comparison of human & murine PA/PXA characteristics

Pediatric low-grade gliomas (pLGGs) are the most common brain tumors in this age group. Despite their favorable prognosis, the young patients often have to suffer from long-term sequalae of the (repeated) therapy or the tumor itself for the rest of their lives. Recent large-scale sequencing studies have advanced the molecular characterization of pLGGs to a new level and are driving the development of a tumor classification based on molecular features rather than primarily histology. However, this is hampered by the enormous heterogeneity of pLGGs, and even the separation of established tumor types is typically not acknowledged, for example in clinical trial stratification. Pilocytic astrocytoma (PA; WHO grade I) and pleomorphic xanthoastrocytoma (PXA; WHO grade II) are two of these pLGG types. They can be difficult to distinguish based on histology alone, but PXA tumors show a clearly worse clinical course than PA tumors and are in many cases lethal. Nevertheless, in clinical trials, both are often simply referred to as “pLGG” and thus treated in the same way. In the first part of this dissertation, I therefore investigated and compared the molecular characteristics of PA and PXA in more detail.

Using molecular data from 89 human pediatric tumor samples, I identified considerable differences between the methylome and transcriptome profiles of PA and PXA. The differentially expressed genes between both tumor types were highly enriched for cell cycle and developmental processes. My results confirm the distinctively more proliferative nature of PXA and suggest fundamental differences in the regulatory circuits that are implicated in tumor development and growth. While most PXA samples in the analysis cohort harbored the typical BRAF V600E mutation, I also identified NTRK fusions as a previously underappreciated genetic driver event for this tumor type and, in a separate case, discovered an EGFR:BRAF fusion for the first time. In addition, some tumors apparently retained expression of the CDKN2A/B locus, which is typically deleted in PXA. Given the important role of the tumor microenvironment, I extracted information on the immune cell content from the bulk sequencing data of PA and PXA. While both tumor types displayed similarly strong signs of overall immune cell infiltration (about 20%), I found a distinctive upregulation of a CD8 T cell gene signature in PXA tumors, which might represent a vulnerability for immunotherapeutic intervention.

In addition, I further developed a genetic mouse model for PXA tumors on the basis of previous work in the group. It relies on viral introduction of the human gene sequence for the BRAF V600E oncogene into neural progenitor cells of Cdkn2a knockout mice and nicely complements an existing PA mouse model. I demonstrate that tumors from both models faithfully resemble their human counterparts at the levels of growth behavior, histology, gene expression and immune cell infiltration. Strikingly, in contrast to the PA model, the murine PXA-like tumors were lethal for the mice, again indicating that the designation of this tumor type as “low-grade” might be inappropriate and should be reconsidered. In addition, these tumors showed the same characteristic CD8 T cell signature that I had observed in human PXA. My work thus paves the way for further molecular and preclinical research on these pLGG entities and suggests that the potential use of immunotherapy should not be rejected prematurely. (Sommerkamp et al., in preparation)

Project 2: KIAA1549:BRAF fusion detection from RNA-Seq data

The most common genetic alteration observed in human PA is the KIAA1549:BRAF fusion. It shows high specificity for this tumor type and is therefore of great diagnostic and prognostic value. In addition, it constitutes a vulnerability for targeted therapy. Thus, it is of major importance to reliably detect the KIAA1549:BRAF fusion in a clinical setting. RNA sequencing (RNA-Seq) is now widely used in research, and has also become increasingly popular for diagnostic purposes. In addition to providing gene- and isoform-specific expression data, RNA-Seq can be used to identify expressed fusion genes in an unbiased (i.e. not pre-selected) way. The KIAA1549:BRAF fusion, however, seems to be expressed at low levels and has previously proven difficult to detect by RNA-Seq. In the second part of this dissertation, I therefore examined the detection reliability of this fusion and developed an optimized workflow for detection from RNA-Seq data.

By analyzing the RNA-Seq data from known fusion-positive PA tumor samples in my analysis cohort, I found that deep sequencing alone is not sufficient to reliably detect the fusion. Instead, the detectability is influenced by different factors, including RNA integrity / library size, tumor cell content, KIAA1549 expression levels as well as the library preparation protocol. Strikingly, all samples in which the fusion was initially not identified did harbor supporting reads in their raw data that could in principle provide a basis for fusion detection. I identified bottlenecks in standard workflows of the alignment algorithm STAR and the fusion caller Arriba, and show that proper alignment of split reads is a major hurdle for reliable detection of the KIAA1549:BRAF fusion. An optimized workflow based on adjusted STAR parameters and a new version of Arriba significantly improved confidence of already detected fusions and allowed identification of fusions that had previously been missed. In an independent diagnostic cohort from Montreal, I demonstrate that this workflow considerably outperforms the previously used standard analysis tools. Moreover, I prove that the higher detection sensitivity is not accompanied by the identification of false positive fusions in an RNA-Seq dataset of > 1000 formalin-fixed, paraffin-embedded (FFPE) samples of diverse origin.

In summary, my work provides a novel workflow that substantially improves the detection of this important fusion from RNA-Seq data, and which will most likely also result in increased fusion detection performance in other tumor contexts. (Sommerkamp et al., 2020)