PiggyBac mutagenesis and exome sequencing identify genetic driver landscapes and potential therapeutic targets of EGFR-mutant gliomas

Background Glioma is the most common intrinsic brain tumor and also occurs in the spinal cord. Activating EGFR mutations are common in IDH1 wild-type gliomas. However, the cooperative partners of EGFR driving gliomagenesis remain poorly understood. Results We explore EGFR-mutant glioma evolution in conditional mutant mice by whole-exome sequencing, transposon mutagenesis forward genetic screening, and transcriptomics. We show mutant EGFR is sufficient to initiate gliomagenesis in vivo, both in the brain and spinal cord. We identify significantly recurrent somatic alterations in these gliomas including mutant EGFR amplifications and Sub1, Trp53, and Tead2 loss-of-function mutations. Comprehensive functional characterization of 96 gliomas by genome-wide piggyBac insertional mutagenesis in vivo identifies 281 known and novel EGFR-cooperating driver genes, including Cdkn2a, Nf1, Spred1, and Nav3. Transcriptomics confirms transposon-mediated effects on expression of these genes. We validate the clinical relevance of new putative tumor suppressors by showing these are frequently altered in patients’ gliomas, with prognostic implications. We discover shared and distinct driver mutations in brain and spinal gliomas and confirm in vivo differential tumor suppressive effects of Pten between these tumors. Functional validation with CRISPR-Cas9-induced mutations in novel genes Tead2, Spred1, and Nav3 demonstrates heightened EGFRvIII-glioma cell proliferation. Chemogenomic analysis of mutated glioma genes reveals potential drug targets, with several investigational drugs showing efficacy in vitro. Conclusion Our work elucidates functional driver landscapes of EGFR-mutant gliomas, uncovering potential therapeutic strategies, and provides new tools for functional interrogation of gliomagenesis.


Figure S25
Cdkn2a   Examples of the formation of small tumors in the ventricular system and subarachnoid space. A, B, C, tumor growth in the lateral ventricle, the base of the frontal brain and the subventricular zone (SVZ) adjacent to the lateral ventricle. D, E, F, formation of the hypercellular myxoid intrinsic tumor in the third ventricle (D) the lateral ventricle (E) and the base of the pons (F, arrows). G, H, I, hypercellular cluster (dark nuclei of expanded SVZ stem/progenitor cells (green arrows) and adjacent a small glial neoplasm (blue arrows). H, small glioma protruding from the floor of the 3 rd ventricle and I, subarachnoid spread of a glial neoplasm on the base of the pons, in a "sugarcoat" fashion (arrows). Lettering on sides of panels reflect mouse IDs from which these tumor originated. Scale bar corresponds to 150m for A and B, 75m for C, 100m for D, E and F, 50m for G, 100m for H and I.   Figure S4. Expression of human EGFRvIII is limited to tumor cells. A, B, overview and detail images demonstrating EGFRvIII immunostaining is positive across glioma cells but not normal mouse brain in EGFRvIII; nes-cre mice (n=4). C, D, overview and detail imaged demonstrating EGFRvIII is expressed in smaller glioma nests (precursors to larger tumors) in these mice. Scale bar corresponds to 1mm for A and C, and 100m for B and D.

Figure S10. Comparative genomics of mouse EGFRvIII glioma mutations with human gliomas.
Using the TCGA dataset of 283 patients demonstrates that several of the most frequently mutated genes in EGFRvIII mouse gliomas are found methylated or genetically altered with high frequency in human low-grade gliomas (LGGs). A. Plot showing genes NLRP1, SUB1, CES1 and ITGA6 are commonly methylated in human LGGs (medians and interquartile ranges are displayed; the methylation levels for the genes are significantly higher than for EGFR, p < 0.0001, paired t-test). B. Oncoprint demonstrating the high frequency of genetic alterations across all tumors in this dataset for TP53, TEAD2, NT5C2, ADGRL2, and UIMC1.   Hoxc8 and Hoxa2 are amongst the top 5 most overexpressed genes in EGFRvIII-mouse brain tumors, and this analysis shows that overexpression of these genes correlates with significantly worse overall survival in GBM patients (p < 0.05, log-rank test). Figure S14. Gene ontology (DAVID) analysis of differentially expressed genes in EGFRvIII mouse brain and spinal gliomas. In brain tumors, there is enrichment for gene sets reflecting brain developmental processes, whereas in spinal tumors the gene sets reflect processes intrinsic to the spinal cord. These data reflect the different tissue origins of these tumors; note the absence of gene sets for oncogenic pathways here, as these are largely shared between the two types of tumor. FDR -false discovery rate.

Figure S15. Clinical Phenotype of EGFRvIII-PB mice.
A. Photograph of an EGFRvIII-PB mouse with enlarged head due to an underlying brain glioma. B, C. Macroscopic photographs of areas of the brain from the same mouse showing the presence of a tumor on the brain surface. Figure S16. Typical example of a grade III brain glioma from an EGFRvIII-PB mouse. Overview (A) and high power (B) H&E-stained histological sections of a brain tumor from the mouse previously shown in Supplementary Figure S13. C. This brain tumor has a higher proliferative index (Ki67 staining) than earlier microneoplasias, in this case estimated to be 10-20% across the tumor. This tumor contains many cells expressing neural stem and progenitor markers PDGFR (D), Olig2 (E) and nestin (F), with some cells also expressing Sox2 (G). Scale bar corresponds to 2.5mm for A, D, E, F, G; 200m for B; 150m for C.

Figure S17. Transposon mobilization in EGFRvIII-PB mice and survival times.
A. PCR assay to detect mobilization or lack of excision of the ATP1S2 transposon. B. Examples of PCR results showing mobilization of the transposon in GBMs from mice containing PB, and absence of transposition in brains from mice without PB; only genotyping from sites 2 and 3 (referenced in A) are shown here. Fragment sizes: 220bp for site 2; 182bp for site 3; 100bp ladder. C. Kaplan-Meier survival curves of EGFRvIII-only (n=48) and EGFR-PB (n=72) mice, with no significant difference between them (p = 0.95, log-rank test). No differences in survival or pathology were observed between EGFRvIII; nes-cre and EGFRvIII; nes-cre; ATP1S2 mice. Tumors were not observed in PB-only (TSPB; ATP1S2; nes-cre, n=20) or nes-cre (n=10) mice after 60 weeks. Figure S18. Plots showing focal copy number variations across EGFRvIII-only and EGFRvIII-PB mice. Significant focal deletions as determined by GISTIC2 are displayed in A, and significant focal amplications are displayed in B. Lower x-axis represents q-value (significance at < 0.05) and top x-axis represents the G-score.

Figure S19. Evidence of replication stress and activation of DNA damage response pathways.
A. Immunostaining for H2AX in EGFRvIII-mouse gliomas. High power view of a typical immunostain of a tumor; green arrow shows pan-nuclear staining of a tumor cell and red arrow indicates a typical pattern of focal nuclear staining. Scale bar 25m B. Quantification of H2AX in EGFRvIII-mouse gliomas, including focal and pan-nuclear staining. Bars represent mean values (n=2 tumors with three views per tumor) +/-SEM. C. RNA-seq analysis using gene set enrichment analysis (GSEA) of mouse EGFRvIII-gliomas shows significant enrichment for pathways involved in the DNA damage response, as indicated. Each enriched pathway has an FDR q-value < 0.001. Figure S20. Key genes from PiggyBac mutagenesis and whole-exome-sequencing cluster into oncogenic pathways. A. Oncogenic pathways driving EGFR-mutant glioma progression are displayed. Blue boxes depict the percentage of tumors containing a PiggyBac insertion for a particular CIS gene; red boxes show the percentage of tumors with a mutation or CNV detected by WES. All pathways have at least one gene targeted by transposons (p < 0.0001, Kernel convolution analysis). Analysis using Panther showed that key genes are grouped into oncogenic pathways. B. David gene ontology (GO) analysis of all 281 glioma CIS genes shows significant enrichment for pathways including neurogenesis and mesenchymal stem cell differentiation, suggesting these pathways are important in driving EGFR-mutant gliomagenesis (FDR = false discovery rate). Figure S21. Network analysis of all interacting CIS transposon genes. An analysis, performed using STRING, to determine the functional connectivity between CIS genes demonstrates there are 253 interactions between their proteins, showing PiggyBac mutagenesis has identified mutations in functionally interacting proteins (Benjamini-Hochberg adjusted p = 4.88 x 10 -13 , Hypergeometric test). Color coding: colored nodes are proteins from CIS genes; connecting lines are known or predicted interactions between proteins; see https://string-db.org for further details.

Figure S22. Insertional pattern consistent with Pdgfr gene activation in brain tumors.
PiggyBac transposons from all EGFRvIII-PB gliomas are largely at the start of the gene in the forward orientation, with only two at the last exons of the gene (likely to be of lesser functional significance), suggesting the transposons are driving transcriptional activation. Figure S23. EGFRvIII-PB gliomas display intratumor heterogeneity, and PB insertions identify their evolutionary routes. A. Overview of the experiment: two gliomas were sampled from three independent regions each, and their DNA was subjected to QI-seq to determine their insertions. Only insertions in CIS genes (determined to be significant across all 96 tumors) were included in this analysis. B. The insertional patterns from tumor A (a low grade glioma on histopathology) and tumor B (a glioblastoma) from all three regions are displayed on this oncoprint, with clonal PB insertions (found in all regions of the tumor) colored red and subclonal ones (found in some regions of the tumor but not all) colored blue. C. Tumor A shows branching evolution, with truncal clonal insertions in genes including Map7, Csmd3, Nav3 and Exosc9. *Subclones 1 and 3 have different Cdkn2a insertions, implying these arose later and independently in evolution. D. Tumor B similarly shows branching evolution, with distinct clonal and subclonal PB insertions. *Subclones 1 and 3 have the same Pdgfra insertion, but subclone 2 does not suggesting Pdgfra was likely a truncal insertion that subclone 2 later lost due to continued PB transposition. Figure S24. Comparative genomic analysis confirms the presence of genetic alterations in human GBMs for the top CIS genes. Data were analyzed from the TCGA high-grade glioma dataset of 273 patients, confirming that for most of the top CIS genes there is a high frequency of alterations (particularly copy-number changes) in these patients. LGGs and n=136 GBMs) in order to provide adequate sample sizes. Boxes span the third (Q3) quartile to the first (Q1) quartile (interquartile range, IQR), with the line at the median; whiskers extend to Q3 + 1.5 x IQR and Q1 -1.5 x IQR. Outliers are plotted as individual points. Spearman's rank correlation was used to calculate correlation coefficients () and P values. These data imply deletions of these genes result in loss of their expression, supporting their roles as tumor suppressors. Figure S26. PiggyBac mutagenesis identifies EGFRvIII cooperative genes in brain and spinal tumors. A. Oncoprint showing the top CIS genes for spinal tumors, ranked according to the total number of insertions. B. Oncoprint for the top CIS genes in brain tumors. Note that Pten ranks very highly in spinal tumors but ranks lower in brain tumors (not seen in this oncoprint), where in contrast there are some alternative drivers ranking highly such as Sox6 and Pik3r1. Figure S27. Treatment of gliomaspheres with afatinib in the presence of CRISPR-Cas9 induced mutations. A. EGFRvIII-mouse gliomaspheres were treated with afatinib at the concentrations shown. Wild-type cells (with a non-targeting sgRNA, NT sgRNA), as well as cells with Nav3 and Spred1 loss-of-function mutations were treated. No significant difference in cell viability was observed between these conditions. Significance testing was done with the two-sided t-test, and p < 0.05 was deemed significant. B. TIDE confirms on-target indels for Nav3 created by CRISPR-cas9, and also for Tead2 ( C ). The bar plots indicate the percentage of sequences with indels at the loci indicated around the sgRNA target site.  Table S4. RNA-seq in brain tumors. Results of analysis of RNA-seq data from brain tumors of EGFRvIII-only mice. Table S5. RNA-seq in spinal tumors. Results of analysis of RNA-seq data from spinal tumors of EGFRvIII-only mice. Table S6. CIS genes. All common integration sites from piggyBac in both brain and spinal tumors. Table S7. Fusion Transcripts. Results from RNA-seq analysis identifying fusion transcripts between piggyBac and endogenous genes. Table S8. CanSAR analysis for druggable genes. All druggable genes from the glioma network are presented. Dark green = targets with clinically approved drugs available; light green = targets of investigational drugs; yellow = targets of drugs under chemical investigation; light red = targets that are predicted to be druggable by chemistry-based assessment. Table S9. Glioma drug targets identified from CanSAR analysis of PB CIS driver genes and recurrently mutated genes. Proteins in bold typeface are those that have been targeted with drugs in human glioma cell lines with at least partial efficacy. Table S10. GDSC analysis for sensitivity of human glioma cell lines to drugs targeting the glioma network. Table S11. Sequences for CRISPR sgRNAs used in this study.