In vivo functional characterization of EGFR variants identifies novel drivers of glioblastoma

Abstract

Background

Glioblastoma is the most common and aggressive primary brain tumor. Large-scale sequencing initiatives have cataloged its mutational landscape in hopes of elucidating mechanisms driving this deadly disease. However, a major bottleneck in harnessing this data for new therapies is deciphering “driver” and “passenger” events amongst the vast volume of information.

Methods

We utilized an autochthonous, in vivo screening approach to identify driver, EGFR variants. RNA-Seq identified unique molecular signatures of mouse gliomas across these variants, which only differ by a single amino acid change. In particular, we identified alterations to lipid metabolism, which we further validated through an unbiased lipidomics screen.

Results

Our screen identified A289I as the most potent EGFR variant, which has previously not been characterized. One of the mechanisms through which A289I promotes gliomagenesis is to alter cellular triacylglycerides through MTTP. Knockout of Mttp in mouse gliomas, reduces gliomagenesis in multiple models.

Conclusions

EGFR variants that differ by a single amino acid residue differentially promote gliomagenesis. Among the identified mechanism that drives glioma growth include lipid metabolism through MTTP. Understanding triacylglyceride accumulation may present a prospective therapeutic pathway for this deadly disease.

Key Points.

• Identification of novel EGFR driver variants in glioblastoma.

• Tumors driven by similar variants have distinct gene expression profiles.

• Altered lipid metabolism is a defining feature of key EGFR driver variants.

Importance of the Study.

Our functional genomics study is the first to test an allelic series of EGFR variants in an autochthonous, immunocompetent, mouse model. We screened through an array of 36 EGFR variants, and validated 6 of these through survival, histological, and molecular profiling studies. While the hotspot theory would suggest the most over-represented mutations are the strongest drivers, our study finds this may not be the case, as rarer variants significantly drive gliomagenesis. Among the mechanisms that promote glioma growth, we identified altered lipid metabolism to be of key importance for our most potent driver variants. Specifically, these tumors altered the activity of Mttp, to reduce triacylglyceride levels. Elevated triacylglycerides are a key factor for nonalcoholic fatty liver disease, and our studies could suggest that tumors increase MTTP activity to avoid a toxic, hepatic steatosis-like states. This, in turn, provides a survival advantage for rapidly dividing cells.

Glioblastoma (GBM) is one of the most lethal forms of cancer and is characterized by highly infiltrative and aggressive cellular phenotypes. Despite advances in our understanding of basic, disease-driving mechanisms, survival rates for GBM have remained stagnant for the past 50 years, highlighting the urgent need to find new targetable pathways. The emergence of genomic technologies led to large-scale sequencing initiatives for a host of cancers, including GBM, with the hopes of identifying novel targetable pathways.¹ These efforts identified a vast spectrum of mutated genes and associated variants; however, the functional significance of these individual genetic mutations remains poorly defined. Critically, while genetic mutations are a principle driver of malignancy, they are also a consequence, as genomic instability is a hallmark of cancer.² Therefore, in end-stage tumors it is difficult to discern whether the accumulated genetic aberrations contribute to malignancy (i.e. drivers), or are a by-product of genomic instability (i.e. passengers). Statistical and bioinformatic approaches have taken an evolutionary approach by correlating high-frequency “hotspot” mutations as drivers,³ with the overarching idea being the more frequent a mutation is present across cancers, the more likely it is to be functionally significant. However, the vast majority of mutations are not hotspots, and these statistical studies demonstrate that certain variants are only present in specific lineages suggesting a dependence on tissue or co-mutation context.³In vitro screens are capable of functionalizing a vast cohort of mutations but do not address appropriate, in vivo lineage or microenvironmental context. Together, these observations illustrate the dire need for context-specific, in vivo screening systems to functionalize nonhotspot variants to decipher their prospective driver or passenger activities.

The RTK/PI3K/MAPK pathway is one of the most frequently altered pathways in GBM with EGFR mutations and/or amplifications present in 57% of all samples.⁴ It remains the most frequently mutated oncogene in GBM. In an effort to identify and validate bona fide, driver mutations in GBM, we employed our recently developed in vivo competition screening system that utilizes an autochthonous mouse model of glioma.^5,6 Through this approach, we demonstrate that the tumorgenicity of rare EGFR variants can exceed that of more frequently occurring mutations and identify several new EGFR driver variants. Further molecular studies of EGFR-A298I identifies MTTP as a downstream effector of EGFR-signaling and novel regulator of glioma tumorigenesis. Together, our studies highlight the importance of functional genomics, whereby in silico findings need to be rigorously tested in biologically relevant settings to discern their contributions to disease pathogenesis.

Methods

Mouse IUE and Constructs

All mouse gliomas were generated in the ICR CD-1 mouse background as previously described.⁵ Briefly, IUEs were performed on embryonic day E14.5. The injection cocktail included CRISPR constructs (1.5 µg/µl), the piggyBac (PB) helper plasmid pGlast-PBase⁷ (2.0 µg/µl), and PB donor plasmids (1.0 µg/µl). PB donor constructs include EGFR expression cassette, a GFP expression cassette (for visualization), or a mCherry expression cassette as a control. CRISPR constructs targeting Tp53, Pten, and Nf1 as well as the PB helper are those previously described.⁶ All procedures were approved by the IACUC at Baylor College of Medicine and conform to the US Public Health Service Policy on Humane Care and Use of

Laboratory Animals

piggyBac (PB) donor vectors for mCherry and EGFR variants were generated through the previously described HiTMMoB approach^8,9 with the PB transposable PBCAG-EGFP-T2A-GWR1R4 destination vector.⁵ The MTTP (NM_000253.2) expression vector was similarly cloned in through Gateway cloning.

A CRISPR construct targeting the mouse Mttp gene was cloned with the pX330 vector. The Mttp guideRNA target sequence (GGAAAACCGCAAGACAGCGT) was generated using the CRISPOR online tool¹⁰ on the 6th exon (NM_001355052.1). Guide cloning was confirmed through Sanger sequencing. On-target (ON) specificity and absence of off-target (OT) effect was confirmed with the mismatch-cleavage SURVEYOR assay on the top 5 OT candidates. Primer pair sequences are as follows: CCCAGGACACTTTGAATGGATTCC/ GCAGGTCACACAACTGGCC (ON), GCTAGCACTGTCAGGCCTATGGG/ AACCTACATTGTCTGGTGTGGTC (OT1), TACTTGCCTGGCATGTGAGAGGC/ GAGATGAAAATTGAGCTCCCTGC (OT2), GGGGAAGTAGGTGAAAGGGCACT/ TCTGATGTCTGTCAGGGTGATGA (OT3), GGCCTGGCTCTTTGGGATGC/ GGGGAAGTTGTGGCA AGCTCC (OT4), and GCTCTCGTTCCAAGGGCATGTTG/ TGTGTGCACGTGTTCACATTTAT (OT5).

In vivo Barcode Competition Assay

Candidate driver EGFR variants were screened as previously described.⁵ Briefly, an EGFR variant library was filtered based on presence in GBM patients (as indexed in TCGA or COSMIC) which resulted in 36 variants (Supplementary Table 1). Controls included the nonmutated wild-type copy, 3 silent mutants, and 10 passenger mCherry constructs to achieve the 1:50 dilution. Each ORF was tagged with a genetic barcode (BC) sequence. This pooled library (at a final concentration of 1.0 µg per µl) was electroporated along with CRISPRs targeting Tp53 and Pten, pGlast-PBase, and PBCAG-EGFP. After mice demonstrated glioma symptomology (see below), mice were euthanized, tumors dissected, and genomic DNA prepped. Tumor samples were prepared with 5 biological replicated and technical triplicates. The IUE injection mixture served as input.

BC amplicons were synthesized as previously described¹ using the Ion Plus Library Kit (ThermoFisher, 4471252), Ion Xpress Barcode Adaptors (ThermoFisher, 4474517), and PGM sequencing (318 V2 Chip). Raw data were concatenated into one “reference” file and indexed using Burrows–Wheeler alignment tool for alignment of BC sequences (with parameters “-l7 -t12 -N -n3”) for counting each BC’s occurrence. BC enrichment was calculated by the ratio of each BC occurrence to the total BC reads.

Survival Study and Brain Collection for Histology

Mice were observed for symptoms suggestive of tumors including—but not limited to—lethargy, hunched posture, decreased appetite, decreased grooming, trembling/shaking, squinting eyes, partial limb paralysis, and abnormal gait, denoting the IACUC permitted endpoint. After symptom demonstration, mice were humanely euthanized, and brains were fixed through intracardial perfusion of 4% paraformaldehyde (PFA) in PBS (phosphate buffered saline), dissected, further fixed overnight in PFA, and then preserved in 70% ethanol for eventual paraffin embedding. Comparison of survival trends was assessed with survival (v3.2-10, https://cran.r-project.org/package=survival).

Histological Analysis

Paraffin embedded tumor brains were sectioned at 10 µm thickness. Hematoxylin (Poly Scientific R&D, S212A) and eosin (Poly Scientific R&D, S176) staining was performed as previously described.⁵ After staining, section were dehydrated and staining was preserved with Permount Mounting Media (Electron Microscope Sciences, 17986-01) under a coverslip. Histological diagnoses of mouse-IUE-generated tumors were validated across n ≥ 6 tumors per variant.

For immunohistolgoical analysis for MTTP, after deparaffinizing and rehydrating tissue sections, sections were processed with heat induced antigen retrieval (1mM sodium citrate, 10 minutes, 95°C). Endogenous peroxidase activity was quenched (3% H₂O₂, 1 hour). Slides were washed with PBS, blocked in NHS (Vector Laboratories, #S-2012-50), and stained with rabbit anti-MTTP antibody (ABCAM, ab75316, 1:50, Lot #GR3314313-5) overnight at 4°C. Following primary antibody incubation, HRP-conjugated anti-rabbit antibodies (Vector Laboratories, #MP-7401) were applied (1 hour, room temperature). Peroxidase activity was detected with DAB (Vector Laboratories, #SK-4105). Sections were then dehydrated and coverslipped as above.

RNA-Seq

RNA was extracted from bulk tumors using the RNeasy Plus Mini Kit (Qiagen, 74134) following the manufacturer’s instructions. ~700ng of total RNA with RIN ≥ 8.0 was used for mRNA library preparation with the TruSeq Stranded mRNA LT kit (Illumina, RS-122- 2101). Library size and concentration were assessed with the Standard Sensitivity NGS Fragment Analysis Kit (AATI, DNF-473-0500) on a 12-Capillary Fragment Analyzer and Quant-iT dsDNA Assay Kit (Q33120, Thermo Fisher) on a Perkin Elmer VICTOR X3 2030 Multilabel Plate Reader. Biological replicates, n ≥ 3 per variant genotype, were pooled at an equimolar concentration (2nM), diluted, denatured, and sequenced with approximately 20 million paired-end (2x75) reads per sample using the Mid Output v2 kit (20024904, Illumina) on a NextSeq500.

Bioinformatics Analysis

FASTQ files were quality controlled using fastQC (v0.10.1) and MultiQC (v0.9),¹¹ and aligned to the mm10 reference genome via STAR (v2.5.0a).¹² Count matrices and gene models were built from aligned files using Rsamtools (v2.0.0) and GenomicFeatures (v1.32.2) in R (v3.5.2). Using DESeq2 (v1.20.0) samples were normalized and analyzed for differential gene expression where each variant was compared to WT EGFR tumor individually. Genes were considered significantly differentially expressed with a fold change > ±1.5 at P < .01 compared to WT tumors. Variant-specific genes were determined by comparing DEGs of A289I to each other variant DEGs individually to identify the unique expression patterns (fold change > ±1.5 at P < .01). Data were visualized using ComplexHeatmap (v2.0.0).

Human expression and survival correlation analysis was performed utilizing the GEPIA online tool.¹³ Low and high expression cohorts were set at the bottom and top 50%, respectively. Log2 fold change cutoff was set at 1, P-value cutoff was set at.01, and match normal data were set to TCGA normal and GTEx data.

Synaptic analysis

Peritumoral synaptic analysis was performed as previously described.⁵ Tumor-bearing mice were humanely euthanized at postnatal day 30, and brains were fixed through intracardial perfusion of 4% PFA, dissected, cryopreserved in 20% sucrose in PBS overnight, and then embedded and frozen in OCT Compound (Optimal Cutting Temperature, 4583, Sakura Finetek). Frozen brains were sectioned to 40 µm thickness. After performing antigen retrieval (1mM sodium citrate, 15 minutes, 80°C), sections were blocked and stained with mouse anti-gephyrin (3B11, 1:500; Synaptic Systems, 147011; 1-64), mouse anti-PSD95 (7E3-1B8, 1:500; ThermoFisher, MA1-046; LJ147875), guinea-pig anti-VGAT (1:500; Synaptic Systems, 131004; 2-41), and guinea-pig anti-VGLUT1 (1:2000; Millipore, AB5905; 3193844), on a plate rocker overnight at 4°C. We used species-specific secondary antibodies tagged with Alexa Fluor 568 or 647 (1:1,000, ThermoFisher), along with Hoechst nuclear counterstaining (ThermoFisher, H3570, 1:50,000), and mounted slides with VECTASHIELD anti-fade mounting medium (Vector Laboratories, H-1000). Sections were imaged and synapses were quantified with the Synapse Counter plugin for ImageJ.

Unbiased Lipidomics Screen and Analysis

The unbiased lipidomics screens were performed through the Metabolomics Core at Baylor College of Medicine. Tumors were generated through IUE and mice were allowed to grow until postnatal day 30 when tumor tissue was collected. Tumor tissue was dissected under a fluorescence dissecting microscope, carefully collecting GFP-positive tissue. Samples were collected and submitted for screening as previously described.^14–16 3 biological replicates were submitted for each condition. 5 liver samples were used for control.

Following data acquisition, negative or positive normalization were performed to Internal Standard 18:1 (d7) Lyso PE (ISTD) or Internal Standard 19:1 (d7) Lyso PC (ISTD), respectively. Normalized values were compared for each lipid species across biological conditions using the one-way ANOVA test, where significant differences were set at P < .05. Lipid class comparisons were also performed with one-way ANOVA tests, comparing the averages of each lipid species in a given lipid class, against those of different experimental conditions.

Results

Identification of EGFR Variants That Drive GBM

In order to identify EGFR variants that function as drivers of GBM, we first tested the efficacy of our in utero electroporation (IUE) approach to functionalize EGFR variants found in GBM (Figure 1A). Mimicking previously established mouse model¹⁷ of malignant glioma we electroporated CRISPR constructs targeting Tp53 and Pten (henceforth denoted 2xCR, double CRISPR) in conjunction with a piggyBac (PB)-transposable cassette which constitutively expressed wild-type EGFR (biological control), mCherry (technical control), or a bona fide EGFR driver variant (i.e. A289V) (2xCR; EGFR, 2xCR; mCherry, and 2xCR; A289V, respectively) (Figure 1D–E). We observed that the survival of the 2 controls was not significantly different (P-value = .70) (Supplementary Table 2), suggesting that in this approach, copy number amplification of wild-type EGFR is not sufficient to drive gliomagenesis. As expected, the 2xCr; A289V tumor-bearing mice exhibited significantly increased mortality (median survival 118 days) compared to the technical and biological controls (median survival 322 and 287 days, respectively) (Figure 1D). Histological analyses of these brains confirmed the presence of a high-grade glioma (HGG) (Supplementary Figure 1).

Fig. 1.

Identification of EGFR Driver Variants in GBM. (A) Lollipop plot represents the distribution of mutated residues in GBM, index in TCGA Firehose Legacy cohort. Particular hotspot residues are emphasized and labeled. (B) Schematic of BC competition assay. ORFs are genetically tagged with unique BC sequences. With cancerous transformation and proliferation, stronger drivers outcompete for weaker drivers, which are detected through BC sequencing. (C) BC competition assay results show the relative abundance of each variant by percentage (y-axis). Tumor samples in red, input in black. N = 5 tumors. (D) Kaplan-Meier survival curve of mice overexpressing different EGFR variants. (E) Color-coded key and survival statistics of each tested variant.

Having validated that our approach can generate HGG, we compiled a barcoded (BC) pool of PB-transposable constructs, constitutively expressing EGFR variants where each variant is genetically linked with a unique BC sequence (as previously described⁵). This pool included a curated list of 36 EGFR variants and control constructs (see methods) (Supplementary Table 1), and we coelectroporated this pool in our 2xCR model (Figure 1B). We excluded the vIII variant (focusing on missense mutations) along with previously tested hotspots A289V and G598V^18,19 because these strong drivers could mask prospective signals from other variants. After tumors developed, we used targeted sequencing to quantify the abundance of each BC’s occurrence from bulk tumor DNA. We utilized BC abundance as a proxy for the associated variant’s relative driver strength in this in vivo competition assay. Through this approach, we identified A298I, T263P, and R108G as the top candidate drivers for further testing (Figure 1C).

Having screened and identified candidate missense, driver variants, we next individually validated each of these variants in the context of the 2xCR system (Figure 1D–E). We also included an analysis of 2 other variants which were not identified on our screen, R108K and R222C. In concordance with our IUE BC screen, the 3 candidate variants from our competition assay accelerated tumor-associated death in our 2xCR model, as did R108K and R222C (Figure 1D–E). To ensure that tumor burden was the cause of death, we performed histopathological analysis on the resulting brains and confirmed that all mouse brains harbored neoplasms histologically classifiable as HGG (Supplementary Figure 1). Collectively, these results confirm the driver potential of the tested EGFR variants towards gliomagenesis, many of which have not been previously characterized in vivo. They also reveal that rare mutations can contribute potently to tumor progression.

EGFR-A289I Driven Tumors Upregulate MTTP

In order to identify the differential molecular programs that are imparting lethality across the tested EGFR variants, we perform RNA sequencing (RNA-Seq) analysis on variant-driven tumors that were dissected from end-stage samples. Normalizing to nontumor, adult mouse brain samples, our analysis identified 2365 differentially expressed genes (DEG) across all variant tumors (Figure 2A andSupplementary Table 3). Given that our manipulations were likely to impact the EGFR-signaling pathway, we first analyzed the expression of EGFR pathway target genes^20–22 (Supplementary Figure 2andSupplementary Table 4). When compared to 4 different gene sets which identify targets of EGFR pathway activation, we observe altered expression patterns in target genes across these variants when compared to the 2xCR; mCherry profile (Supplementary Figure 2). When combined with the survival study of the various EGFR variant tumors, these analyses not only support that our manipulation affects the EGFR pathway, but each variant does so differentially.

Fig. 2.

Transcriptional Profiling Reveals Altered Lipid Metabolism in A289I Gliomas. ( A) Heatmap showing Log₂ normalized expression of each variant’s differentially expressed genes as compared to mCherry control. (B) KEGG Pathway analysis for R108K, A289V, and A289I variants demonstrated unique signatures in signal transduction, synaptic activity, and lipid metabolism, respectively, for these variants. (C) Gene Set Enrichment Analysis of 2 most significant variants, A289I and A289V. P-values for both plots: P-value < .00001. (D) Normalize Enrichment Score (NES) heatmap of selected ontologies categories from human phenotype (HP), KEGG, and GO Biological Processes.

Our survival studies revealed that tumors driven by R108K, A289I, and A289V were significantly different than our controls (Figure 1E andSupplementary Table 2), suggesting that these tumors may have unique molecular profiles that engender tumorigenesis. Further bioinformatics analysis of RNA-Seq data from these variant-driven tumors identified overrepresented KEGG pathways in the DEGs of each variant expression profile (Figure 2B andSupplementary Figure 3). While not mutually exclusive, we generally observed that the DEGs in R108K tumors are more associated with the regulation of core signaling pathways, A289I signatures corresponded to lipid metabolism pathways, and A289V is highly aligned with synaptic activity. We further validated these signatures through GSEA comparisons with associated gene ontologies (Figure 2C–D). While signaling pathway dysregulation is common in all tumors,^23,24 it has been recently appreciated that increased synaptic activity can potential tumor growth and may reflect a brain-specific mechanism utilized by A289V.²⁵ To test for alterations in the synaptic constituency in the brains of these tumor-bearing mice, we stained for excitatory and inhibitory synapses at the peritumoral margins. Comparing the A289V and A289I tumors, we observed a significant increase in excitatory synapses in A289V tumor margins, supporting this differential signature (Supplementary Figure 4). Collectively, these data suggest that each driver variant uses unique molecular mechanisms to drive GBM tumorigenesis.

We focused our analysis on the hyper-potent A289I variant which was the most aggressive driver, significantly accelerating tumor mortality compared to all other variants we tested, including the bona fide hotspot driver A289V (Figure 1E andSupplementary Table 1). We first sought to validate the altered lipid biosynthesis signatures through an unbiased lipidomics screen on bulk tumor tissue (Figure 3A–B andSupplementary Table 5), comparing A289I to A289V driven tumors. We analyzed the changes of each lipid class between the 2 variants, comparing all the species within each lipid type. Strikingly, only triacylglycerides were significantly different (Figure 3C-blue bar), with triacylglycerides demonstrating significant decreases in A289I tumors compared to A289V (Figure 3D andSupplementary Figure 5).

Fig. 3.

Triacylglycerides are Altered in A289I Gliomas. ( A) Table describing the unbiased lipidomics screen. Details listed include 17 different lipid classes, abbreviations used in C, the total number of different species per lipid class, and the color map for each lipid class in Figure 3B. (B) Heatmap of lipids assayed in lipidomics screen. Each row corresponds to a single lipid species. Lipids are grouped according to each lipid class, designated by the color code on left. The order is the same as that listed in Figure 3A. The heatmap shows the average across biological duplicates (N = 3) for each experimental condition, mapping the Z-score within each row. (C) Significance analysis of each lipid class, comparing the differences in lipids between A289I to A289V gliomas (blue) and A289I; Mttp^CR to A289I gliomas (orange). Graph charts the –log₁₀ of the p-values (y-axis) across all the different lipid classes (x-axis). (D) Volcano plot showing log₂ fold change of triacylglycerides (TG) when comparing A289I to A289V tumors. Dark green dots are those with P-values < .05 (horizontal line). (E) Volcano plot showing log₂ fold change of triacylglycerides (TG) when comparing A289I; Mttp^CR to A289I tumors. Dark green dots are those with P-values < .05.

In parallel, we further analyzed our RNA-Seq data, to identify unique molecular targets that could explain the survival phenotypes observed in A289I gliomas. To this end, we first identified the DEGs that were unique to A289I as compared to all other variant tumor signatures (Figure 4A). This yielded a list of 45 significantly differentially expressed candidate genes (Supplementary Table 6). We then analyzed the human orthologues of the candidates (for which 39 existed), screening for those that demonstrated a significant difference in survival between high and low expression (P < .05, log-rank test), which narrowed our list to 4 candidates (annotated in Figure 4A andSupplementary Figure 6). We then compared the expression-survival correlation of these 4 between A289I mouse tumors and human GBM. Specifically, we isolated candidates where high or low expression in both A289I mouse tumors and human tumors resulted in poorer survival in human patients. This resulted in 2 candidates, CD24 and MTTP (Microsomal Triglyceride Transfer Protein). MTTP transcripts are high in both A289I mouse tumors and human GBM samples, and higher expression of MTTP in human GBM samples significantly corresponded to poorer survival (P-value = .032) (Figure 4B). MTTP complexes with protein disulfide isomerase and is critical for lipoprotein assembly,²⁶ specifically the export of cholesterol esters and triacylglycerides from the liver through VLDL.^27–29 While MTTP has not been investigated in gliomas, it has been implicated as a downstream target of insulin signaling through PI3K and MAPK signaling.³⁰ Given this association between MTTP and RTK signaling, the function of MTTP in triacylglyceride processing, along with the aforementioned dysregulation in lipid metabolism pathways in A289I tumors, we surmised that MTTP may serve as a potential target in attenuating glioma growth through altering the cellular lipid content.

Fig. 4.

MTTP is Expressed in Aggressive Gliomas. ( A) Heatmap showing Log₂ normalized expression of DEGs unique to A289I. Labeled genes are those whose human orthologues demonstrated a significant difference in survival when comparing high and low-expression patients. (B) Kaplan-Meier survival curve of human glioblastoma (GBM) patients with relative low (blue) and high (red) MTTP expression. TPM – transcript per million. (C) Mttp IHC in EGFR variant tumor brains. N = 6 tumors per variant. Black scale bar = 25µm. (D) Comparison of MTTP expression in human tumor samples (red) to normal brain samples (gray) for GBM. * P-value < .01. E) Representative images of MTTP antibody staining from the human normal brain and GBM.

MTTP Loss Slows Gliomagenesis and Increases Tumor Triacylglycerides

Directing our efforts towards MTTP, we first assess whether Mttp was expressed in A289I tumors. Immunohistochemical analysis for Mttp demonstrated that it is selectively expressed in A289I mouse gliomas compared to controls (Figure 4C). As controls, we compared A289I sections to mCherry and A289V tumors, as the A289I and A289V variants represented mutations at the same residue. We then assessed MTTP expression in human glioma samples from TCGA, which revealed that MTTP transcripts are significantly higher (P-value < .01) in GBM samples compared to nontumor brains in (Figure 4D). Immunohistochemical analysis of human tissue microarray of GBM validated the presence of elevated MTTP protein in GBM compared to nontumor brain (Figure 4E). Together, these experiments reveal that MTTP expression and protein levels are correlated with highly aggressive mouse and human gliomas. This association with poorer survival and elevated presence supported the supposition that MTTP may have a critical role in gliomagenesis, warranting further testing.

Having identified and validated the presence of Mttp as a candidate gene, we next sought to assess its functional contributions to gliomagenesis. We tested the necessity of Mttp towards gliomagenesis. To this end, we first generated a CRISPR construct targeting the Mttp coding sequence in mice and coelectroporated it in our 2xCR; A289I model (2xCR; A289I; Mttp^CR). We chose this model, as we had already demonstrated MTTP is already present in these tumors. We first validated the targeted knockout of MTTP through the mismatch-cleavage assay analyses (Supplementary Figure. 7A). After extracting genomic DNA from these tumors, we observed that our guide sequence specifically induced indel mutations at our intended target site but not on the top 5 off-target sites. With this construct, we then assessed the functional significance of MTTP loss on gliomagenesis. Utilizing survival as a surrogate for tumorigenic potential, we observed that mice bearing 2xCR; A289I; Mttp^CR tumors exhibited significantly enhanced survival compared to 2xCR; A289I (Figure 5A,D), suggesting that MTTP loss suppresses glioma tumor progression. Histopathological analysis of these brains revealed that 2xCR; A289I; Mttp^CR tumors are HGG (Figure 5C). We next examined how the loss of MTTP would affect gliomagenesis in an aggressive mouse model of GBM that uses CRISPR/Cas9 technology to delete NF1, PTEN, and p53, herein termed 3xCR,⁵ observing abundant Mttp protein in these tumors (Supplementary Figure 7B). Similar to our observations in the 2xCR; A289I mode, elimination of MTTP in the 3xCR model (3xCR; Mttp^CR) also prolonged survival (Figure 5B,D), suggesting that its functional contributions are not limited to the A289I variant.

Fig. 5.

Mttp Loss Delays Glioma Associated Death. (A) Kaplan-Meier survival curve of mice with Mttp knockout (green) in the 3xCR model (control, black). (B) Kaplan-Meier survival curve of Mttp loss in A289I tumors (2xCR; A289I; Mttp^CR, magenta) compared to its control (blue). (C) Histological analysis of A289I tumors with Mttp loss. Yellow scale bar = 1mm; grey scale bar is 200µm; blue scale bar = 50µm. (D) Survival statistics for curves are present in this figure. Death and survival values constitute days. p-values were calculated by log-rank test against each group’s control (CTRL).

Lastly, to assess how the loss of Mttp impacts tumor lipids, we again performed a lipidomics screen on our 2xCR; A289I; Mttp^CR tumor, comparing these to our 2xCR; A289I tumors. Analysis of all lipid classes revealed that triacylglycerides were again the only lipid class that demonstrated a significant alteration (P-value = 9.9E-22) (Figure 3C-orange bar), where the loss of Mttp resulted in an increase in triacylglycerides (Figure 3E), indicating that Mttp is regulating triacylglyceride levels in A289I-driven tumors. Collectively, these studies suggest that MTTP can make oncogenic contributions to gliomagenesis, by altering the lipid constituency in gliomas. Genetic targeting MTTP reduces tumorigenesis.

Discussion

One hope for the large genomic profiling initiatives was to bridge the knowledge gap between disease biology and effective therapeutics. While this has translated well in certain populations of pediatric medulloblastoma,^31,32 the same cannot be said for GBM. Despite being among the first solid tumors to be molecularly characterized en masse,¹ survival statistics remain largely unchanged for over 50 years. Even as the volume of genetic and molecular information from patient tumors has exponentially grown, there remains a dearth in understanding the functional relevance of this critical information. In our functional genomics study, we sought to address this critical gap. In this second of our “allelic series” studies,⁵ we screened a library of EGFR variants that only differ by a single amino acid residue, and similar to our studies with PIK3CA variants, we found drastic differences in survival across EGFR variants which was not correlated to variant abundance. Moreover, our current study also reveals that differences at the same amino acid residue can have a profound effect on disease progression, as A289I is a significantly stronger driver than A289V. Despite its more potent, glioma transformation effect, the rarity of the A289I (compared to A289V) is likely in that 2 bases of the codon need to be mutated rather than just 1 (Supplementary Table 1). Interestingly, there have been 5 variant mutations identified at A289 (V/T/D/I/N) with a range of biochemical properties (hydrophobic vs hydrophilic vs charged). This could suggest that the steric influences of a larger side chain substitution (compared to a single methyl group in alanine) at A289 imparts a constitutively active form on the receptor. This study, along with others continues to highlight the need for functional genomics at both the network and individual variant levels.

To determine how A289I imparts its robust effect, we utilized RNA-Seq, identifying alterations to lipid metabolism pathways along with a candidate gene Mttp. We also considered sex as a factor (Supplementary Figure 8). Interestingly, while 2 variants significantly demonstrate sexually dimorphic trends in our mouse models, but they did not include A289I, thus we did not pursue this for the current study. Mttp is predominantly expressed in the liver and gastrointestinal tract,³⁰ is vital for lipoprotein assembly,^29,33,34 and is essential for viability as knockout mice die embryonically.³⁵ Mice with Mttp loss in the liver demonstrate reduced serum cholesterol and triacylglycerides and a buildup of lipids in the liver, which leads to hepatic steatosis.^28,36,37 Taken together with our study, this could suggest that in A289I gliomas, MTTP supports the general viability of the cell by reducing the toxic buildup of free lipids or safely sequestering them. The accumulations of lipids in the form of lipid droplets have been observed across various models of glioma, both in vitro and in vivo.^38,39 These lipid droplets likely serve as reserves for glioma cells as they expend an extraordinary amount of energy in cellular division.⁴⁰ Thus the elevated levels of MTTP in fast-growing glioma cells could serve to properly and safely store lipids, specifically triacylglycerides, in the cell until the high metabolic demands are present, thereby preventing steatosis-like conditions in tumor cells. Given that MTTP is upregulated in GBM and lower expression significantly trends with better survival (Figure 4), the importance of MTTP may not be restricted to just the rare A289I variant.

In summary, we have functionally screened and validated rare EGFR variants found in patient glioma samples. Our study reveals that these variants behaved differently in driving gliomagenesis, and that a rare A289I variant was the strongest driver in our cohort. Among the mechanism A289I employs to promote tumor growth, we identified alterations to lipid metabolism through a molecular target Mttp. While elevated MTTP transcripts are present in GBM, ours is the first study to implicate a functional contribution toward gliomagenesis. As we hope to move forward towards the ideal of patient-specific treatments, it becomes more critical to understand the functional relevance of patient—omics data along with any other contributing factors. For instance, our approach targets subventricular zone cells for tumor transformation, but it remains plausible that lineage and cell-of-origin may impact specific variants, as EGFR mutations in the extracellular domain are more represented in glioma compared to kinase domain mutations in lung cancer.⁴¹ There may also exist other genetic co-dependencies; however, these remain difficult to assess, as currently there are so few samples of some variants, underpowering these comparative analyses (Supplementary Table 1). In addressing these and other concerns, we look forward to bringing renewed hope to patients after over half a century of research.

Funding

This work was supported by grants from the Cancer Prevention Research Institute of Texas (RP150334 and RP160192 to BD, KLS, and CAM), National Cancer Institute-Cancer Therapeutic Discovery and Development (U01-CA217842 to BD, GBM, and KLS), National Institutes of Health (R01-CA223388 to BD; R01- NS124093 to BD and GR; R50-CA252125 to KY; T32-HL902332 to KY), and the American Cancer Society-Rob Rutherford Glioblastoma Research Postdoctoral Fellowship (PF-15-220-01-TBG to KY).

Conflict of Interest

The authors have no conflicts to declare.

Authorship Statement

KY, KK, KLS, and BD designed the experiments and interpreted results; KY performed the IUEs; KK cloned all EGFR clones; KK, KY, and AC completed the survival studies; KK and BL performed the bar coded competition assay; BL performed the RNA-Seq and bioinformatics analysis; ELF preformed synaptic staining and analysis; KY, BL, and ELF analyzed lipidomics data; RNC perform the human correlation study; KY and ELF sectioned tissue and performed H/E staining; RNC and KY performed the IHC for MTTP; KY and AC generated reagents for MTTP functionalization; CAM and GR assessed tumor neuropathology; AJ assisted in statistical analysis; KY performed sexual dimorphism survival analysis; KLS and GBM provided EGFR variant constructs; the unbiased lipidomics screen was performed by the Metabolimics Core at Baylor College of Medicine; the Pathology and Histology Core (HTAP) at Baylor College of Medicine provide tumor microarray slides; KY and BD wrote the manuscript; BD and KLS conceived the projected; BD and KY supervised all aspects of this work.