Glioblastoma Subtypes Research Articles

EPCO-15. LINKING HISTOLOGICAL GLIOBLASTOMA PHENOTYPES TO TRANSCRIPTIONAL SUBTYPES AND PROGNOSIS USING DEEP LEARNING

Abstract Deep learning is increasingly utilized in medical image analysis and shows promise to substantially inform treatment decisions in clinical practice. While it finds broad applications in radiology or genomics, there are still relatively few applications of deep learning to the analysis of histological images. A disease in which a better understanding of underlying phenotype-genotype correlations seems crucial is glioblastoma, the most common malignant brain tumor in adults that features a dismal prognosis and limited therapeutic options. Using a large cohort of 276 glioblastoma patients with available histological, immunohistochemical, clinical and molecular data, we trained an Xception convolutional neural network to predict patient survival and transcriptional subtypes on histological images alone. We used 5-fold cross validation to evaluate the performance of the trained network. We achieve a high accuracy for predicting overall survival on univariable (p < 0.001) and multivariable (p = 0.01) analysis. Moreover, we observe a high accuracy for predicting the predominant transcriptional subtype with an AUC of 0.84, 0.81, and 0.71 for classical, mesenchymal and proneural subtypes, respectively (all p < 0.001). Finally, we plot heatmaps of the predicted transcriptional subtype and risk scores to directly link them to the histological phenotypes and the tumor microenvironment. Reassuringly, we find an association of the mesenchymal subtype with the perinecrotic niche and of the proneural subtype with the infiltrating edge. In conclusion, we introduce an easily adaptable convolutional neural network model for working with histological data, which can accurately predict outcomes and transcriptional subtypes in glioblastoma.

MODL-16. SOMATICALLY ENGINEERED MOUSE MODELS RECAPITULATING CELLULAR AND MOLECULAR FEATURES OF HUMAN GBM

Abstract INTRODUCTION Mouse models are instrumental in advancing our understanding of glioblastoma (GBM), however many commonly used models have features that limit their practicality or applicability. These limitations can include an incomplete immune context, a requirement for complex breeding strategies, high cost of specialized lines, and unpredictability in where tumors arise. We sought to develop and validate a mouse modeling system for some of the major human GBM subtypes that overcame these practical challenges. METHODS In vivo electroporation was used to introduce genetic alterations into periventricular cells of early postnatal C57Bl6/j mice. PiggyBac transposon/transposase and CRISPR-Cas9 systems were used to overexpress (OE) or knock out (KO) genes associated with mesenchymal (Nf1-KO/Pten-KO/p53-KO), classical (EGFRvIII-OE/Cdkn2a-KO/Pten-KO), and proneural (Pdgfra-OE/Cdk4-OE/p53-KO) IDH-WT GBM subtypes. KO or OE was confirmed by Indel Detection by Amplicon Analysis and immunofluorescent staining (IF). Tumours were analyzed for histologic, immunohistochemical, and transcriptional (RNAseq) features. RESULTS Tumors arose with near complete penetrance and median survivals of 36 to 116 days amongst models. All tumors were GFAP-expressing high-grade gliomas, with distinct phenotypes associated with different genetic combinations. Nf1-KO/Pten-KO/p53-KO tumors were composed of spindle cells. EGFRvIII-OE/Cdkn2a-KO/Pten-KO and Pdgfra-OE/Cdk4-OE/p53-KO tumors showed prevalence of small, round cells. Molecularly, Nf1-KO/Pten-KO/p53-KO tumors were enriched for human mesenchymal signature and displayed more differentiated phenotype. The Nf1-KO/Pten-KO/p53-KO model also showed increased stromal cells and macrophages. The other models had mixed proneural and classical profiles and were enriched for OPC- and NPC-like gene signatures found in human GBM. Once tumors formed, cells from each model were transplantable into C57Bl6/j mice to generate subsequent tumors. CONCLUSION We developed and validated a rapid, versatile, and reproducible system to model GBMs. These models allow for controlled study of GBM pathogenesis, progression, and treatment response, and allow for robust generation of syngeneic implantable mouse models that can serve as valuable tools for preclinical testing.

NIMG-37. JOINT LEARNING OF IMAGING AND GENOMIC DATA REVEALS DISTINCT GLIOBLASTOMA SUBTYPES

Abstract PURPOSE The significant heterogeneity of glioblastoma is typically displayed on both phenotypical and molecular levels. Non-invasive in vivo approaches to characterize this heterogeneity would potentially facilitate personalized therapies. Here we leverage advanced unsupervised machine learning to integrate radiomic imaging features and genomics to identify distinct subtypes of glioblastoma. METHODS A retrospective cohort of 571 IDH-wildtype glioblastoma patients were collected with pre-operative multi-parametric MRI (T1, T1CE, T2, T2-FLAIR, DSC, DTI) scans (available in 462/571 patients) and targeted next-generation sequencing (NGS) data (available in 355/571 patients). Radiomic features (n= 971) were extracted from these MRI scans and a subset of 12 features were selected by L21-norm minimization. A total of 14 key driver genes in the 5 main pathways that are most frequently altered in glioblastoma were chosen. Subtypes were identified by a joint learning approach called Anchor-based Partial Multi-modal Clustering (APMC) on both radiomic and genomic modalities. RESULTS Three distinct glioblastoma subtypes were discovered by APMC based on 14-dimension NGS data together with 12 selected radiomic features representing characteristics from histograms, shape, and volumetric measures for different tumor sub-regions. The identified subtypes were 1) high-risk; 2) medium-risk; and 3) low-risk, in terms of their overall survival outcome in Kaplan-Meier analysis (p= 5.52e-6, log-rank test; HR= 1.51, 95%CI:1.20-1.74, Cox proportional hazard model). The three subtypes also displayed different molecular characteristics: subtype 1 exhibited increased frequency of mutation in [EGFR, PIK3CA, PTEN, NF1], subtype 3 showed frequently mutated [PDGFRA, ATRX], while subtype 2 did not show significant differences for mutations in any of these genes. CONCLUSION Our results revealed the synergistic value of integrated radiomic signatures and molecular characteristics for glioblastoma subtyping. Joint learning on both modalities could help better understand the molecular basis of phenotypical signatures of glioblastoma and further provide insights into the biologic underpinnings of tumor formation and progression.

TMIC-24. SINGLE-CELL ANALYSIS OF TUMOR-ASSOCIATED MICROGLIA AND MACROPHAGES FROM HUMAN GLIOBLASTOMA

Abstract INTRODUCTION Patients with glioblastoma have a very poor prognosis. Tumor-associated microglia and macrophages (TAMs) constitute up to 30 % of the cells in glioblastoma, and they secrete cytokines, chemokines and growth factors that influence the microenvironment. The existence of different TAM subtypes appears to be more complex than the established M1 and M2 phenotypes, but their role in glioblastoma is not fully understood and rarely considered therapeutically. This could explain why many clinical trials fail despite of promising preclinical results. This project aims to interrogate the existence and characteristics of different TAM subtypes in human glioblastoma in order to identify novel subpopulations and therapeutic targets. MATERIALS AND METHODS Live CD11b+ TAMs were isolated from patient glioblastoma tissue, and single-cell RNA sequencing (scRNA-seq) was performed using the 10X Genomics Chromium platform. The data was processed and analyzed with the R-package Seurat. RESULTS We have sequenced 74,000 TAMs/microglia from three glioblastomas and two control brain biopsies. In the controls, we detected mostly microglia, while the primary glioblastomas showed a predominance of monocyte-derived TAMs. We identified 11 TAM subtypes, such as hypoxic, proliferating, interferon-induced, chemokine-producing and TNF-producing TAMs, as well as a novel subtype potentially involved in tumor progression. CONCLUSION AND PERSPECTIVES We have detected a spectrum of TAM subtypes, which is more complex than the established M1 and M2 phenotypes. Our findings confirm a recent TAM scRNA-seq study, and in addition, we identify a novel subpopulation, which express known tumor-promoting genes, normally expressed by cancer cells. We are currently validating our findings. This study contributes to the elucidation of the glioblastoma microenvironment. It will potentially lead to identification of novel clinically relevant targets affecting the TAM-glioblastoma interactions.

EPCO-27. REVEALING TUMOR HETEROGENEITY AND IMMUNE MICROENVIRONMENT IN NF-1 GLIOMA BY SINGLE-CELL GENE EXPRESSION PROFILING

Abstract Germline mutations in the NF1 tumor suppressor gene cause neurofibromatosis type 1 (NF-1), the most common human genetic cancer predisposition syndrome. NF1 patients are prone to develop peripheral and central nervous system tumors, with glioma occurring in 15-20% of patients. The molecular profiles of glioma from patients with NF-1 have been previously explored by large scale DNA and RNA bulk analyses, providing the first comprehensive genetic landscape of NF-1 glioma. However, the intra-tumoral architecture and the microenvironment organization remained largely concealed within the global bulk profile. Here, to characterize the complexity of molecular states of both malignant and non-malignant cells, we profiled NF-1 glioma gene expression at individual cell resolution. We performed single-nuclei RNA sequencing (Nucseq) on frozen NF-1 glioma specimens, including high-grade and low-grade tumors. The functional classification of the transcriptional programs activated in the individual tumor cells identified three main molecular states that partially overlapped with the cellular subtypes of sporadic glioblastoma. Tumor cells of low-grade NF-1 glioma exhibited marked enrichment of neuronal functions, suggesting that the gliomagenesis process in the context of NF1 inactivation might be supported by specialized neuronal activities. Conversely, the proliferative state characterized high-grade NF-1 glioma, indicating that cell cycle-associated pathways are drivers of tumor progression. The single-cell deconvolution of immune microenvironment in NF-1 low grade glioma revealed infiltration of CD8+ T lymphocytes and the active crosstalk of these cells with pro-inflammatory macrophages and mature dendritic cells, supporting a possible anti-tumor cell-mediated response. The identification of divergent tumor cell states as well as the characterization of the immune infiltrates provide new insights for accurate stratification of NF-1 glioma patients and to inform personalized therapeutics.

CNSC-12. NEURONAL DIFFERENTIATION DRIVES THE ANTI-TUMOR EFFICACY OF MEK INHIBITION IN GLIOBLASTOMA

Abstract Multiple genomic aberrations and distinct molecular subtypes of glioblastoma (GBM) have been identified through multi-omics analyses including single-cell genomic and epigenomic analysis. Epidermal Growth Factor Receptor (EGFR) amplification is found in nearly 50% of GBM tumors. Several EGFR inhibitors have been tested in GBM but failed to demonstrate long-term therapeutic benefit, presumably due to target heterogeneity, escape signal pathways, and acquired resistance. Therefore, we investigated downstream signaling, mitogen-activated protein kinase (MEK1/2) as a potential therapeutic target for GBM. We utilized an unbiased high-throughput screening with a panel of glioma stem-like cells (GSCs) and identified that GBM cells harboring EGFR amplification showed anti-proliferative and apoptotic cell death (sensitive) to multiple MEK inhibitors. RNA-seq analysis of cells pre- and post-treatment with trametinib, a potent MEK inhibitor, revealed an upregulation of neurogenesis and neuronal differentiation genes such as β-III tubulin, ASCL1, DLL1, DLL3, NeuroD4, PAX6 and NCAM1. In addition, downregulation of the several MEK targets including erythroblast transformation specific (ETS) family genes, particularly ETV5 was observed. We demonstrated that MEK inhibition increased neuronal differentiation demonstrated by immunofluorescence, western blot, and RT-qPCR experiments using selected differentiation markers (β-III tubulin, ASCL1, DLL3, and NeuroD4). In addition, ETV-5 knockout in GSCs by CRISPR/Cas-9 promoted the expression of neuronal differentiation genes. Oral treatment of trametinib in an orthotopic GSC xenograft model significantly improved animal survival, in which 25-30% of mice are long-term survivors. This was accompanied by decreased MEK/ERK signal and cell proliferation in tumor tissues. Thus, we demonstrated here that MEK1/2 inhibition by trametinib induces neuronal differentiation mediated through downregulation of ETV-5 in GBM, a potential novel mechanism of action of MEK1/2 signal inhibition.

TMIC-17. DISCOVERING DOMINANT TUMOR CHARACTERISTICS IN CYSTIC GLIOBLASTOMA

Abstract BACKGROUND As a rare subtype of glioblastoma (GBM), the genetic features and outcomes in cystic GBM (cGBM) are largely unknown. This study aimed to evaluate the survival of cystic features, investigate genomic patterns and the immune microenvironment in patients with cGBM. METHODS Gd T1-weighted, T2-weighted, T2-Flair, DWI or ADC images were used to classify cGBM or noncystic GBM (non-cGBM). Clinical information from IvyGAP, EGA databases and Chinese cohort were used for survival analysis. Fresh cystic fluid and residual tumor tissue were collected to evaluate immunological components, pathological and genomic differences between cGBM and non-cGBM. Data from IvyGAP were included to determine difference of molecular feature among substructure. RESULTS 143 cases with cGBM and non-cGBM were screened to compare survival outcomes. Cystic features were to confer a survival benefit. It was an independent prognostic factor irrespective of IDH, MGMT and therapeutic regimen. cGBM patients have a hypomethylation state compared to non-cGBM. Active immune cells such as CD4+, CD8+ T cells were present in the cystic fluid and infiltrated into tumor. 10 cytokines/chemokines were only detected in cystic fluid, 8 cytokines/chemokines increased two-fold in cystic fluid than in peripheral blood serum (PBPs). Much differential methylated position (DMPs), differentially expressed genes (DEGs), Gene Set Enrichment Analysis (GSAE), Functional Epigenetic Modules (FEM) networks and molecular feature of substructural regions were different between cGBM and non-cGBM. CONCLUSION These results imply that cGBM could be a new subtype and has distinctive immune microenvironment which may have potential value correlating with prognosis.

TMIC-84. MULTIREGIONAL ANALYSIS OF GLIOBLASTOMA REVEALS TOPOLOGICAL DIVERGENCE OF CANCER CELLS AND THEIR SURROUNDING MICROENVIRONMENT

Abstract Tissue organization plays an important role in tumor development and evolution. Systematic investigation of the differences in the local composition of the tumor microenvironment (TME) in glioblastoma (GBM) in distinct regions of the tumor is needed for identification of traits driving tumor recurrence. Thus, we performed an analysis of three tumors, six spatially distant biopsies each, representing tumor surface, core, and deep margin. GBM cell fluorescence generated by the 5-aminolevulinic acid (5-ALA) metabolic labeling prior to surgery allowed us to enrich for the tumor cell fractions and the cells of the microenvironment, respectively, from each biopsy. Using RNA-seq and scRNA-seq we compared the transcriptomic profiles of GBM and tumor microenvironment (TME) at distinct locations within the tumor. We observed a gradient of GBM subtypes across the tumor regions, with a molecular identity progressively less defined from the tumor core towards the margins. Interestingly, the single-cell based copy number analysis revealed a subset of GBM cells that appeared to evade the 5-ALA labelling. The high prevalence of these 5-ALA-negative GBM cells within the deep tumor margins points to a metabolically distinct subpopulation that could be responsible for local tumor recurrence. The phenotypic gradient of cancer cells from tumor core towards the margins was also mirrored by tumor microenvironment-enriched fractions. Deep margin samples from the profiled tumors were transcriptionally more divergent than biopsies from other regions. Moreover, they were depleted from major populations of TME cells. In situ analysis revealed that TERT promoter mutant cells are not equally distributed across the tumor indicating the coexistence of GBM sub-clones with distinct evolutionary potential. Altogether, our results provide a novel insight into co-evolution of tumor microenvironment and GBM clonal heterogeneity within the context of the tumor tissue.

BIOM-26. GENOME-WIDE SOMATIC COPY NUMBER ALTERATION SIGNATURES HAVE PROGNOSTIC IMPLICATIONS FOR ADULT ASTROCYTIC GLIOMA

Abstract BACKGROUND The molecular landscape of adult diffuse glioma has been extensively characterized by gene expression and DNA methylation profiling, but less attention has been paid to somatic copy number alteration (SCNA) data. This study aimed to give a rigorous, survival-focused analysis of glioma genome-wide SCNA data that builds on our previous work. METHODS Detailed survival analyses were conducted on the substructure of UMAP projections of all TCGA glioma (Nf1092), exclusively astrocytic glioma (Nf914), and exclusively IDH-wildtype glioma (Nf528). Results were validated with data from the Glioma Longitudinal Analysis Consortium (GLASS) (Nf224). Clinical factors such as age and MGMT methylation were tested in multivariate survival analyses. RESULTS A UMAP projection of TCGA glioma SCNA data generated three distinct clusters composed of entirely oligodendroglioma, predominantly IDH-mutant astrocytoma (C-IDHmut-astro), and predominantly IDH-wildtype glioma (C-IDHwt), respectively. For astrocytic tumors, cluster assignment was independently prognostic of IDH status (p< 0.001): TCGA IDH-mutant astrocytomas that clustered in C-IDHwt had poorer outcomes than their counterparts (p< 0.04) and IDH-wildtype tumors that clustered in C-IDHmut-astro fared better than those that did not (p< 0.01). The distribution of GLASS astrocytic tumors, which is skewed for better survival, supported our results in IDH-wildtype glioblastoma (p< 0.001, Fisher’s). Among four distinct subclusters of TCGA IDH-wildtype glioblastomas, the largest was significantly or marginally significantly negatively prognostic compared to each other cluster (p=0.048, p=0.059, p=0.027) and their combination (p=0.002). In the GLASS dataset, inclusion in the largest subcluster was also prognostic (p=0.013) and similar trends were observed between individual clusters (p=0.21, p=0.036, p=0.083). Furthermore, membership to the largest cluster was independently prognostic of MGMT methylation status and several published IDH-wildtype glioblastoma subtypes in the TCGA. CONCLUSIONS Unsupervised learning of genome-wide SCNA has prognostic implications for astrocytic glioma. SCNA cluster membership is independently prognostic of MGMT methylation status.

DNA5mC Regulator-Mediated Molecular Clusters and Tumor Microenvironment Signatures in Glioblastoma.

Growing evidence links DNA methylation to tumor immunity. The impact of DNA methylation (5mC) on the microenvironment surrounding tumors and immunotherapy remains to be clarified. Through clustering gene expression of 20 DNA methylation regulators, this study aimed at systematically analyzing DNA methylation regulator patterns and tumor microenvironment characteristics of TCGA-GBM patients. Various subtypes of glioblastoma exhibit different tumor microenvironments and DNA methylation patterns. Each DNA methylation modification was then assigned a DNA methylation score (DMS). High DMS was associated with a good prognosis. In contrast, the low DMS group had a relatively low survival rate. A correlation was also found between high DMS and enhanced immunotherapy efficacy in two immune checkpoint blocking treatment cohorts. To conclude, identifying DNA methylation regulation patterns may prove critical to understanding glioblastoma progression and differentiation, as well as future therapeutic targets.

Identification of tumor-associated antigens and immune subtypes of lower-grade glioma and glioblastoma for mRNA vaccine development

BackgroundmRNA became a promising therapeutic approach in many diseases. This study aimed to identify the tumor antigens specifically expressed in tumor cells for lower-grade glioma (LGG) and glioblastoma (GBM) patients.MethodsIn this work, the mRNA microarray expression profile and clinical data were obtained from 301 samples in the Chinese Glioma Genome Atlas (CGGA) database, the mRNA sequencing data and clinical data of 701 samples were downloaded from The Cancer Genome Atlas (TCGA) database. Genetic alterations profiles were extracted from CGGA and cBioPortal datasets. R language and GraphPad Prism software were applied for the statistical analysis and graph work.ResultsPTBP1 and SLC39A1, which were overexpressed and indicated poor prognosis in LGG patients, were selected as tumor-specific antigens for LGG patients. Meanwhile, MMP9 and SLC16A3, the negative prognostic factors overexpressed in GBM, were identified as tumor-specific antigens for GBM patients. Besides, three immune subtypes (LGG1-LGG3) and eight WGCNA modules were identified in LGG patients. Meanwhile, two immune subtypes (GBM1–GBM2) and 10 WGCNA modules were selected in GBM. The immune characteristics and potential functions between different subtypes were diversity. LGG2 and GBM1 immune subtype were associated with longer overall survival than other subtypes.ConclusionIn this study, PTBP1 and SLC39A1 are promising antigens for mRNA vaccines development in LGG, and MMP9 and SLC16A3 were potential antigens in GBM. Our analyses indicated that mRNA vaccine immunotherapy was more suitable for LGG2 and GBM1 subtypes. This study was helpful for the development of glioma immunotherapies.

The Impact of Chemoradiotherapy on Biophysical Properties of Glioblastoma Utilizing a Novel Murine Model

<h3>Purpose/Objective(s)</h3> Glioblastoma, historically viewed as a single disease entity, is now known to be composed of various subtypes. Here, we present our work evaluating the effects of chemoradiotherapy on cellular biophysical processes using a novel murine model of glioblastoma developed by our group. <h3>Materials/Methods</h3> Primary tumor cell lines were established from a murine glioma model generated from a synthetic DNA transposon system delivered oncogenic drivers SV40-LgTA and either NRASV12 or PDGFB. We have previously shown these cell lines, termed NRAS and PDGF, respectively, recapitulate the mesenchymal and proneural tumor subtypes. Treatment was administered with temozolomide (TMZ) to a maximum of 100µM, single fraction radiation therapy (RT) to a maximum of 10 Gy, or both. Cells were stained with Hoechst dye, imaged with an epifluorescence microscope with 10X objective lens, and counted with an image segmentation algorithm daily for 4 days to estimate cell survival and proliferation. On post-treatment day 4, surviving cells were plated on Matrigel coated 9.3kPa polyacrylamide gel and imaged with an inverted, transmitted light microscope with 10X objective lens every 5 minutes for 4 hours. Single cell mean squared displacement (MSD) was measured from time lapse images via an image segmentation algorithm. The random motility coefficient (µ) was calculated by the relationship MSD(t) = 4µt, where t is time. Data was compared via the Kruskal-Wallis test with Bonferroni correction for multiple comparisons when appropriate. <h3>Results</h3> There was no significant difference in cellular death or proliferation between treatment and control arms for the PDGF cell lines. In the NRAS cell lines, TMZ alone did not induce cell death while RT (P < 0.01) and combined modality therapy (P < 0.01) yielded cellular killing in a dose dependent manner. Furthermore, RT-treated NRAS cells proliferated more slowly than non-treated cells (0.05 vs. 0.20 hr-1, P=0.04). Control NRAS cell lines migrated faster than their PDGF counterparts at baseline (3.82 vs. 0.73 µm2/hr, P<1E-5). Within each cell line, the use of combined modality therapy significantly reduced cellular motility (PDGF: 0.73 vs. 0.21 µm2/hr, P=0.008; NRAS: 0.65 vs. 3.82 µm2/hr, P<1E-5). RT monotherapy additionally reduced motility in NRAS cell lines (1.07 vs. 3.82 µm2/hr, P<1E-5) while no difference was observed in PDGF cells. TMZ monotherapy did not affect motility within either cell line. <h3>Conclusion</h3> We evaluated biophysical properties of a novel murine glioma model utilizing oncogenic drivers overexpressing NRASV12 or PDGFB which we have previously shown recapitulates the mesenchymal and proneural subtypes of human glioblastoma, respectively. Here, we observed phenotypic differences in cellular migration following RT and TMZ regimens which could have subtype-specific ramifications for radiotherapy dose and target volume delineation.

Pre-operative MRI radiomics model non-invasively predicts key genomic markers and survival in glioblastoma patients.

Although glioblastoma (GBM) is the most common primary brain malignancy, few tools exist to pre-operatively risk-stratify patients by overall survival (OS) or common genetic alterations. We developed an MRI-based radiomics model to identify patients with EGFR amplification, MGMT methylation, GBM subtype, and OS greater than 12months. We retrospectively identified 235 patients with pathologically confirmed GBMs from the Cancer Genome Atlas (88; TCGA) and MD Anderson Cancer Center (147; MDACC). After two neuroradiologists segmented MRI tumor volumes, we extracted first-order and second-order radiomic features (gray-level co-occurrence matrices). We used the Maximum Relevance Minimum Redundancy technique to identify the 100 most relevant features and validated models using leave-one-out-cross-validation and validation on external datasets (i.e., TCGA). Our results were reported as the area under the curve (AUC). The MDACC patient cohort had significantly higher OS (22months) than the TCGA dataset (14months). On both LOOCV and external validation, our radiomics models were able to identify EGFR amplification (all AUCs > 0.83), MGMT methylation (all AUCs > 0.85), GBM subtype (all AUCs > 0.92), and OS (AUC > 0.91 on LOOCV and 0.71 for TCGA validation). Our robust radiomics pipeline has the potential to pre-operatively discriminate common genetic alterations and identify patients with favorable survival.

Current Insights into Mesenchymal Signatures in Glioblastoma.

Glioblastoma (GBM) is a fatal primary malignant brain tumor in adults. Despite decades of research, the prognosis for GBM patients is still disappointing. One major reason for the intense therapeutic resistance of GBM is inter- and intra-tumor heterogeneity. GBM-intrinsic transcriptional profiling has suggested the presence of at least three subtypes of GBM: the proneural, classic, and mesenchymal subtypes. The mesenchymal subtype is the most aggressive, and patients with the mesenchymal subtype of primary and recurrent tumors tend to have a worse prognosis compared with patients with the other subtypes. Furthermore, GBM can shift from other subtypes to the mesenchymal subtype over the course of disease progression or recurrence. This phenotypic transition is driven by diverse tumor-intrinsic molecular mechanisms or microenvironmental factors. Thus, better understanding of the plastic nature of mesenchymal transition in GBM is pivotal to developing new therapeutic strategies. In this review, we provide a comprehensive overview of the current understanding of the elements involved in the mesenchymal transition of GBM and discuss future perspectives.

The proteomic landscape of glioblastoma recurrence reveals novel and targetable immunoregulatory drivers.

Glioblastoma (GBM) is characterized by extensive cellular and genetic heterogeneity. Its initial presentation as primary disease (pGBM) has been subject to exhaustive molecular and cellular profiling. By contrast, our understanding of how GBM evolves to evade the selective pressure of therapy is starkly limited. The proteomic landscape of recurrent GBM (rGBM), which is refractory to most treatments used for pGBM, are poorly known. We, therefore, quantified the transcriptome and proteome of 134 patient-derived pGBM and rGBM samples, including 40 matched pGBM-rGBM pairs. GBM subtypes transition from pGBM to rGBM towards a preferentially mesenchymal state at recurrence, consistent with the increasingly invasive nature of rGBM. We identified immune regulatory/suppressive genes as important drivers of rGBM and in particular 2-5-oligoadenylate synthase 2 (OAS2) as an essential gene in recurrent disease. Our data identify a new class of therapeutic targets that emerge from the adaptive response of pGBM to therapy, emerging specifically in recurrent disease and may provide new therapeutic opportunities absent at pGBM diagnosis.

Identification and validation of SNHG gene signature to predict malignant behaviors and therapeutic responses in glioblastoma.

Glioblastoma (GBM) patients exhibit high mortality and recurrence rates despite multimodal therapy. Small nucleolar RNA host genes (SNHGs) are a group of long noncoding RNAs that perform a wide range of biological functions. We aimed to reveal the role of SNHGs in GBM subtypes, cell infiltration into the tumor microenvironment (TME), and stemness characteristics. SNHG interaction patterns were determined based on 25 SNHGs and systematically correlated with GBM subtypes, TME and stemness characteristics. The SNHG interaction score (SNHGscore) model was generated to quantify SNHG interaction patterns. The high SNHGscore group was characterized by a poor prognosis, the mesenchymal (MES) subtype, the infiltration of suppressive immune cells and a differentiated phenotype. Further analysis indicated that high SNHGscore was associated with a weaker response to anti-PD-1/L1 immunotherapy. Tumor cells with high SNHG scores were more sensitive to drugs targeting the EGFR and ERK-MAPK signaling pathways. Finally, we assessed SNHG interaction patterns in multiple cancers to verify their universality. This is a novel and comprehensive study that provides targeted therapeutic strategies based on SNHG interactions. Our work highlights the crosstalk and potential clinical utility of SNHG interactions in cancer therapy.

P02.07.B Patient-derived glioblastoma organoids: Elucidating the mechanisms of glioblastoma therapeutic resistance in the context of tumor microenvironment

Abstract Background Intratumoral heterogeneity plays an important role in glioblastoma (GB) resistance to standard therapy consisting of irradiation and chemotherapy with temozolomide (TMZ). However, classical in vitro GB models fail to represent the complex cellular composition of tumors in vivo, which hinders relevant examination of GB therapeutic response. To overcome these limitations, we studied the effects of irradiation and TMZ in a novel patient-derived organoid model. Material and Methods We established a patient-derived GB organoid model by a protocol recently published by Jacob et al. Original tumor tissue and tissue-derived organoids were compared by immunofluorescence staining of selected cell type markers and qPCR analysis of expression levels of a panel of selected target genes, including 15 genes defining GB subtypes. To analyze GB therapeutic response, organoids from 11 patients were exposed to a single dose of irradiation (10 Gy), one-week treatment with TMZ (50 µM) or their combination. The effects of therapy were assessed by viability and invasion assays. Expression levels of a number of genes related to GB subtypes, epithelial-mesenchymal transition, stemness, DNA damage responses, cell cycle, cytokines, and cell markers of the tumor microenvironment (TME) were compared between treated organoids and untreated controls. In addition, the heterogeneity of the TME and its responses to treatment were investigated by spatially resolved transcriptomics with in situ sequencing (ISS) methodology. Results Organoids recapitulate inter-patient variability and reflect the cellular composition and gene expression levels of the tumor tissue from which they were derived. GB stem cells and differentiated cancer cells are present in organoids along with various cells of the TME, e.g., macrophages and microglia, lymphocytes, and endothelial cells. Irradiation and TMZ showed no significant effects on organoid viability and invasion. However, some target genes were differentially expressed in the treated organoids, such as E3 ubiquitin-protein ligase MDM2 and cyclin-dependent kinase inhibitor 1A (CDKN1A). To our knowledge, we are the first to apply spatially resolved transcriptomics (ISS) to formalin-fixed, paraffin-embedded sections of (un)treated GB organoids. Our results elucidate the role of the TME in GB therapeutic response and shed light on potential mechanism underlying GB therapy resistance. Conclusion Patient-derived GB organoids recapitulate the key characteristics and complex composition of patient’s tumor tissue, providing a valuable platform for studies of GB therapeutic response and resistance.

P11.69.B IDH1/2wildtype gliomas grade 2 and 3 with molecular glioblastoma-like profile have a distinct course of epilepsy compared toIDH1/2wildtype glioblastomas

Abstract Background IDH1/2 wildtype (IDHwt) glioma WHO grade 2 and 3 patients with pTERT mutation and/or EGFR amplification and/or +7/−10 chromosome gain/loss have a similar overall survival time as IDHwt glioblastoma patients, and are both considered glioblastoma IDHwt according to the WHO 2021 classification. However, differences in seizure onset have been observed. This study aimed to compare the course of epilepsy in the two glioblastoma subtypes. Material and Methods We analyzed epilepsy data of an existing cohort including IDHwt histologically lower-grade glioma WHO grade 2 and 3 with molecular glioblastoma-like profile (IDHwt hLGG) and IDHwt glioblastoma patients. Primary outcome was the incidence proportion of epilepsy during the disease course. Major secondary outcomes included onset of epilepsy, number of seizure days and antiepileptic drug (AED) polytherapy. Results Out of 254 patients, 78% (50/64) IDHwt hLGG and 68% (129/190) IDHwt glioblastoma patients developed epilepsy during the disease course (p=0.121). Epilepsy onset before histopathological diagnosis occurred more frequently in IDHwt hLGG compared to IDHwt glioblastoma patients (90% versus 60%, p&amp;lt;0.001), with a significantly longer median time to diagnosis (3.5 versus 1.3 months, p&amp;lt;0.001). Median total seizure days was also longer for IDHwt hLGG patients (7.0 versus 3.0, p=0.005), and they received more often AED polytherapy (32% versus 17%, p=0.028). Conclusion Although the incidence proportion of epilepsy during the entire disease course is similar, IDHwt hLGG patients show a significantly higher incidence of epilepsy before diagnosis and a significantly longer median time between first seizure and diagnosis compared to IDHwt glioblastoma patients, indicating a distinct clinical course.

P16.03.A Epithelioid gliobastoma requires rapid treatment and BRAF inhibitors should be made readily available for their treatment

Abstract Background Epithelioid glioblastoma is a rare subtype of Glioblastoma. We examine two cases who presented acutely with symptoms of headache, neck stiffness and an eye squint. The purpose of this case review is to look at their management, the spread of the disease and propose the availability of BRAF inhibiting agents be made readily available for this subtype. Material and Methods The clinical records including pathology and surgical reports, multi-disciplinary team meeting documents, oncology plans and inpatient notes have all been reviewed alongside the literature on epithelioid GBM and BRAF V600e mutations and inhibitors. Results Patients were females aged 25 and 32 presenting with acute onset headache and neck stiffness to emergency Department. The 25 year old had developed a right eye squint within seven days of the other symptoms, her tumour was right frontal with midline shift. The 32 year old had symptoms for 10 days prior to presentation, the tumour was right temporal. They both underwent craniotomies. The frontal tumour was totally resected, while the temporal lobe tumour was haemorrhagic in nature and minimally de-bulked. Pre-operative stealth imaging showed that there had been an increase in the size of the temporal lesion in the fourteen days since presentation. Histology proved these to be epithelioid GBM’s with BRAF V600e mutations, IDH wild-type and TERT promoter mutant. Full pathology reports with molecular markers were available within ten days. The frontal lobe patient began chemo-radiation sixteen days from her surgical date. On day two she was admitted with severe headache and nausea. She became agitated, confused, and transferred back to the neurosurgery unit for management of hydrocephalus. She was diagnosed with PRES and simultaneously treated for this and hydrocephalus. Clinically she suffered storming, passing away exactly eight weeks from presentation. Seven days after surgery the temporal lobe tumour patient suffered a seizure and admitted for symptom management and expedite oncology treatment. She passed away six days later suffering continual neurological deterioration and the tumour progression with leptomeningeal spread. This was exactly four weeks from initial presentation. Conclusion The prognosis for epithelioid Glioblastoma is limited to weeks to short months. Extent of resection in these case studies demonstrates benefit in delaying progression though it is clear that time is of the essence from presentation, to surgery, to adjuvant treatment. Neither of these tumours were methylated meaning the standard treatment for glioblastoma is likely to lack efficacy. BRAF inhibitors should be made readily available for this rare sub-type to commence treatment expediently. Both patients suffered distressing neurological symptoms in their final days which require expert management and are best managed in a neurosurgical centre.

P06.01.A Deconvolution of immunotherapy-treated glioblastoma identifies cellular heterogeneity and plasticity at the single-cell level

Abstract Background Glioblastoma is the most aggressive cancer originating in the brain with an average survival of 15 months. One of the characteristics of glioblastoma is the high level of intra-tumour heterogeneity (ITH), but the composition and complexity at the single-cell level is poorly understood. Here, we aimed to assess the effects and consequences of immune checkpoint inhibitor (ICI) on the cellular and molecular heterogeneity of glioblastoma tumours using at the single cell level. Material and Methods In collaboration with the phase I trials unit at Rigshospitalet, we performed paired molecular analysis of glioma cells from primary and relapse surgery after ICI treatment. Samples were analysed using single-cell RNA sequencing (scRNA-seq) as well as bulk RNA sequencing and whole exome DNA sequencing. Results In an effort to trace cellular lineages we developed and refined methods to a identify copy number changes using scRNA-seq. To this end, we identified clonal and subclonal tumour cell populations in each sample. We found high levels of ITH prior to treatment, both with respect to the glioblastoma subtype enrichment and the cell type-specific gene expression. Using expression-based cell-type classification, we found defined recurrent cell-type populations present at both surgery time points. The immune checkpoint treatment had consequences on the cellular phenotypes and proportions of tumour cells, suggesting a level of plasticity in the neoplastic cells. Moreover, we identified examples of clonal dynamics and sweeps following ICI treatment, pointing to potential treatment response and resistance in these population. Conclusion In summary, we pursued single cell-focused analysis of ICI treated glioblastoma patients to study the cellular and molecular heterogeneity within and between glioblastoma patients, which pointed to recurrent patterns of cellular responses following ICI treatment.