EGFR is frequently amplified, mutated, and overexpressed in malignant gliomas. Yet the EGFR-targeted therapies have thus far produced only marginal clinical responses, and the underlying mechanism remains poorly understood. Using an inducible oncogenic EGFR-driven glioma mouse model system, our current study reveals that a small population of glioma cells can evade therapy-initiated apoptosis and potentiate relapse development by adopting a mesenchymal-like phenotypic state that no longer depends on oncogenic EGFR signaling. Transcriptome analyses of proximal and distal treatment responses identified TGFβ/YAP/Slug signaling cascade activation as a major regulatory mechanism that promotes therapy-induced glioma mesenchymal lineage transdifferentiation. Following anti-EGFR treatment, TGFβ secreted from stressed glioma cells acted to promote YAP nuclear translocation that stimulated upregulation of the pro-mesenchymal transcriptional factor SLUG and subsequent glioma lineage transdifferentiation toward a stable therapy-refractory state. Blockade of this adaptive response through suppression of TGFβ-mediated YAP activation significantly delayed anti-EGFR relapse and prolonged animal survival. Together, our findings shed new insight into EGFR-targeted therapy resistance and suggest that combinatorial therapies of targeting both EGFR and mechanisms underlying glioma lineage transdifferentiation could ultimately lead to deeper and more durable responses.

Significance:

This study demonstrates that molecular reprogramming and lineage transdifferentiation underlie anti-EGFR therapy resistance and are clinically relevant to the development of new combinatorial targeting strategies against malignant gliomas with aberrant EGFR signaling.

Malignant glioma is the most common and lethal type of primary brain tumor (1). In its most aggressive form, glioblastoma (GBM) has a median survival of only 12–15 months even after maximum surgical resection and chemoradiotherapy, a statistic that has barely changed over the past 20 years (2). But contrary to the relative lag of clinical advancement, the world has witnessed over the same time period an explosion of knowledge in glioma biology and basic science discovery. Particularly, the breakthrough in sequencing technology has made a reality of unraveling the complete genomic landscape of GBMs (3, 4). Among the complex genetic and genomic events, EGFR has attracted arguably the most attention due to the fact that its amplification plus mutation and/or rearrangement were identified in approximately 60% of GBM patient samples (4). The success of EGFR tyrosine kinase inhibitors (TKI) in treatment of patients with non–small cell lung carcinoma (NSCLC) carrying active EGFR mutations further made it an appealing target for therapeutic interventions (5, 6).

Aberrant EGFR activation triggers prosurvival and proliferative signaling cascades in GBM. Despite its prevalence and the demonstrated role, EGFR-targeted interventions by strategies such as small-molecule TKIs, antibodies, and vaccines, have all failed to achieve tangible clinical benefit, even in patients with EGFR amplification/mutations (7). A variety of resistance mechanisms have been proposed, such as incomplete target suppression (8, 9), intratumoral heterogeneity (10), activation of downstream effectors in the same signaling pathway or engagement of alternative survival pathways (11, 12). As such, the vast majority of current efforts have been focused on developing better drugs or drug combinations to more vigorously suppress EGFR and its downstream surrogates, even though the clinical outcome of deep EGFR suppression and potential resistance development afterwards have been poorly understood.

A major hurdle against investigating effects of deep EGFR suppression is the dearth of relevant experimental model systems. In human GBMs, high copy EGFR amplification occurs within extra chromosomal DNA known as double minutes (13). As a consequence, the amplified EGFR copies are numerically unstable and often lost in cultured tumor cells (14, 15). This intrinsic instability also poses as an immense barrier for in vitro genetic manipulation, and thus necessitates alternative experimental systems to study EGFR functions and treatment responses.

In an attempt to address this challenge, we previously developed a glioma animal model driven by inducible expression of a truncated oncogenic EGFR mutant (EGFR*) deleted from exon 2–7 in EGFR extracellular domain (16). By using this model system, we revealed that development of resistance to EGFR-targeted therapy against malignant glioma occurred through the combination of EGFR-dependent and -independent mechanisms. In this study, we set out to further exploit this experimental platform and probe for the molecular mechanism(s) underlying EGFR-targeted therapeutic response and resistance development.

Mice

All mouse strains were housed in a barrier facility under protocols approved by the Institutional Animal Care and Use Committee of Weill Cornell Medicine. Rag1−/− mice were purchased from The Jackson Laboratory (JAX #002216). Both male and female mice were used for graft or xenograft between 8 and 12 weeks of age.

Cell culture

Sphere culture of iEIP cells was established and maintained in serum-free DMEM/F12 medium (Sigma), containing ITS (Invitrogen), EGF (20 ng/mL, Peprotech), and basic fibroblast growth factor (bFGF; 20 ng/mL; Peprotech) as previously described (17). The human glioma initiating cells (GS7–11, GSC280) were provided by Dr. Erik Sulman (NYU Langone Health, New York, NY) and maintained in serum-free DMEM/F12 medium (Sigma), containing B27 (Invitrogen), 20 ng/mL EGF, and 20 ng/mL bFGF. Low passage cultures were frozen down and used within 2–3 passages when thawed, for all experiments. All cultures were routinely subject to Mycoplasma testing and authentication.

Xenografting and drug treatment

Low passage iEIP or patient-derived glioma cells were retrovirally transduced with luciferase cassette. 8- to 12-week-old RAG1−/− mice were anesthetized and restrained using a stereotaxic instrument (Stoelting). 50,000 iEIP glioma cells in 2 μL PBS were injected into right caudate nucleus 2.2 mm below the surface of the brain at 1-mm anterior and 1-mm lateral from the bregma. For subcutaneous grafting, iEIP glioma cells were resuspended in 50% Matrigel (BD Biosciences; #356231) in PBS and approximately 10,000,000 cells were injected into each flank of Rag1−/− mice. Doxycycline (Research Product International Inc. #D43020) was administered as 2 g/L with 5% sucrose in drinking water. Erlotinib (LC Laboratories, #E4007) was administered as 150 mg/kg/d by gavage. SB525334 (medchemexpress, #HY-12043) was treated as 10 mg/kg/d by intraperitoneal injection. Tumor growth was monitored and measured every 7 days by caliper, and volume was calculated by the formula: V = {\frac{{4{\rm{\pi }}}}{3}}*{\frac{a}{2}}*{\frac{b}{2}}*{\frac{c}{2}}$(V, tumor volume; a, tumor length; b, tumor width; c, tumor height). Relative tumor volume changes were calculated by dividing by the tumor volume at the beginning of treatment.

Data analysis

For GBM molecular subtype analysis and primary-recurrent paired data analysis the datasets from ref. 18 were used. Analysis was performed as previously described as in ref. 19. Briefly, U133A array profiles for 533 primary GBMs were obtained from the Cancer Genome Atlas (TCGA) portal https://tcga-data.nci.nih.gov/tcga/. Mutation calls and DNA methylation profiles were obtained for all samples, where available. GBMs were identified as IDH wild-type whereas both mutation calls on IDH1 genes were wild-type and GCIMP status inferred using DNA methylation profile was negative. A set of 369 TCGA GBMs were identified as IDH wild-type according to this procedure. Processed primary/recurrence expression data could be analyzed through GlioVis portal http://recur.bioinfo.cnio.es/ (18), which includes 124 primary/recurrent pairs of gliomas and 91 pairs identified as IDH wild-type GBM pairs.

The YAP core signature genes were obtained from the molecular signature database (MSigDB; http://software.broadinstitute.org/gsea/index.jsp), filtered by in-house iEIP tumor expression profiles, and consequently contained: SERPINE1, TNNT2, SH2D4A, MDFIC, FGF2, TGFB2, NDRG1, FSTL1, DAB2, HEXB, AMOTL2, PDLIM2, ANKRD1, ITGB2, FLNA, LHFP, TNS1, EMP2, SLIT2, TGM2, ITGB5, and AXL. The signature score of YAP core genes was then calculated using ssGSEA (18).

Gene set enrichment analysis (GSEA) was used to identify altered cellular processes and oncogenic pathways of primary and recurrent tumors based on gene sets by Verhaak and colleagues (20). GESA was performed using c5 and c6 libraries from version 6.0 of MSigDB.

Statistical analysis

We determined experimental sample sizes on the basis of preliminary data. All results are expressed as mean ± SEM. GraphPad Prism software (version 7) was used for all statistical analysis. Normal distribution of the sample sets was determined before applying the unpaired Student two-tailed t test for two group comparisons. One-Way ANOVA was used to assess the differences between multiple groups. The mean values of each group were compared by the Bonferroni's post-hoc procedure. Differences were considered significant when P < 0.05.

Data and materials availability

RNA sequencing (RNA-seq) and whole-exome sequencing (WES) data can be found on Gene Expression Omnibus (GEO) database with accession number GSE108658 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE108658). Antibodies and oligonucleotides used are listed under Supplementary Tables S1 and S2I. For specific material request, please contact the corresponding author.

No overt genetic changes in iEIP-relapsed tumors following EGFR* suppression

By using animals engineered with doxycycline (dox)-off oncogenic EGFR* transgene and conditional Ink4a/Arf and Pten alleles (Nestin-CreERT2; Ink4aL/L; PtenL/L; hGFAP_tTA; tetO-EGFR*, designated iEIP), we previously demonstrated that sustained oncogenic EGFR signaling was required for maintenance of EGFR-driven glioma progression and suppression of oncogenic EGFR* transgene expression induces tumor regression (16). But despite a robust initial response, the regressed tumors inevitably relapsed that led to mortality eventually. To investigate underlying mechanisms of the tumor recurrence, we orthotopically grafted a group of immunocompromised nu/nu mice with luciferase-expressing iEIP glioma cells. Bioluminescence imaging (BLI) revealed formation of significant tumor burdens by 5 to 7 weeks. Consistent with our previous study (16), extinction of oncogenic EGFR* transgene expression by doxycycline treatment elicited a deep tumor regression, with virtually no BLI-detectable tumor mass by 2 weeks following doxycycline switch (Fig. 1A). After a period of relative indolence that lasted approximately 6 to 10 weeks, however, the animals started to show signs of relapse even under continued doxycycline treatment. Eventually, all of them succumbed to tumor progression ranging from 12 to 20 weeks (Fig. 1B). As compared with the treatment-naïve primary tumors, immunoblot analysis of the relapsed tumors found significantly diminished EGFR and its downstream ERK signaling (Fig. 1C), suggesting that the relapse was unlikely driven by compensatory EGFR signaling. Remarkably, despite their silenced EGFR* expression, the relapsed tumors retained comparable levels of phosphor-AKT and phosphor-4EBP1 compared with their treatment-naïve controls (Fig. 1C), indicating that PTEN deletion by itself can sustain high PI3K–AKT–mTOR signaling.

Figure 1.

The relapsed tumor upon EGFR* inhibition is driven by oncogenic EGFR-independent pathway(s). A, Representative BLI on indicated days after doxycycline (Dox) treatment (N = 5). B, Survival plot for mice treated with control (N = 4) or doxycycline water (N = 6). C, Representative images of Western blot analysis for indicated proteins in primary and relapsed tumors. Samples were run on two gels to detect multiple proteins, and the most representative loading control blot is shown. The experiment was independently repeated three times to ensure the reproducibility of the findings. D, Relative tumor growth of regrafted relapsed tumors after control or doxycycline water treatment for the indicated time. Relapsed tumors from doxycycline treatment (first relapsed) were regrafted subcutaneously into Nu/Nu mice and maintained off-Dox. After the tumors (second transplant) reached analyzable size (marked, ∼200 mm3), the animals were separated into control (N = 3) or doxycycline (N = 4) groups. E, WES copy-number results for three relapse and two matched parental lines. Segmented copy-number data are shown for each chromosome by genomic position in columns and by cell lines in rows.

Figure 1.

The relapsed tumor upon EGFR* inhibition is driven by oncogenic EGFR-independent pathway(s). A, Representative BLI on indicated days after doxycycline (Dox) treatment (N = 5). B, Survival plot for mice treated with control (N = 4) or doxycycline water (N = 6). C, Representative images of Western blot analysis for indicated proteins in primary and relapsed tumors. Samples were run on two gels to detect multiple proteins, and the most representative loading control blot is shown. The experiment was independently repeated three times to ensure the reproducibility of the findings. D, Relative tumor growth of regrafted relapsed tumors after control or doxycycline water treatment for the indicated time. Relapsed tumors from doxycycline treatment (first relapsed) were regrafted subcutaneously into Nu/Nu mice and maintained off-Dox. After the tumors (second transplant) reached analyzable size (marked, ∼200 mm3), the animals were separated into control (N = 3) or doxycycline (N = 4) groups. E, WES copy-number results for three relapse and two matched parental lines. Segmented copy-number data are shown for each chromosome by genomic position in columns and by cell lines in rows.

Close modal

In targeted therapies, a subpopulation of drug-tolerant cells is known to survive through potentially lethal exposures and drive relapse development. But these therapy-tolerant states are often transient and can be reversed back to drug-sensitive state upon removal of the treatment. To assess whether the observed anti-EGFR resistance was a stable feature, freshly isolated iEIP-relapsed tumor cells were re-transplanted subcutaneously into recipient nude mice. The animals were kept off-dox until tumors reached analyzable sizes. Remarkably, IHC analyses of the treatment-naïve controls revealed that only a small fraction (<5%) of tumor cells in the re-grafted gliomas reactivated their EGFR* transgene expression (Supplementary Fig. S1A). Re-initiation of doxycycline treatment exerted no visible effect on tumor growth (Fig. 1D), indicating that progression of the relapsed tumors no longer relies on oncogenic EGFR* signaling.

The finding that relapsed tumors had escaped oncogenic EGFR signaling addiction promoted us to search for potential genetic events that might fuel the resistance development. Three relapsed and two matched treatment-naïve tumors derived from the same parental line were analyzed by WES (Supplementary Fig. S1B). First analysis of copy-number variations recovered no recurrent regions of copy-number alterations (Fig. 1E). To identify the potential relapse-specific mutations, we first excluded the SNPs that appeared in both parental and relapsed tumor cells. The remaining sequence data were then filtered to subtract the variants present in less than 25% of the reads. The rare variants that eluded the threshold filter and appeared only in relapsed tumors affect 15 genes. However, further verification analysis of the raw data using integrated genomic viewer revealed that all of the 15 SNPs preexisted in the parental lines, even though in less abundant levels (between 5% and 25%). These findings together suggest that genetic alterations might not be the major driving force behind EGFR-independent relapse development.

The relapse is accompanied by glioma subtype reprogramming

The lack of evident genetic alterations in the iEIP-relapsed tumors raised a possibility that epigenetic factors might fuel the oncogenic EGFR-independent tumor growth. To explore this prospect, we next characterized molecular and phenotypic changes along relapse development. As compared with matched treatment-naïve control tumors, immunoblot analysis of relapsed iEIP tumors revealed a stark loss of the proneural/classical subtype-enriched OLIG2 and ASCL1 expression and a concurrent upregulation of mesenchymal subtype (MES) signature genes, VIM and TGFBI (Fig. 2A; ref. 19). IHC further confirmed a ubiquitous silencing of OLIG2 with a concerted VIM upregulation in all relapsed tumors analyzed (Fig. 2B), suggesting a tumor subtype transdifferentiation during the relapse.

Figure 2.

Relapsed tumors following EGFR* inhibition show subtype transition. A, Representative images of Western blot analysis for indicated proteins in primary and recurrent tumors. Samples were run on two gels to detect multiple proteins, and the most representative loading control blot is shown. B, Representative images of IHC staining performed on sections of primary and relapsed tumors. Scale bars, 50 μm. C, One-sided GSEA plots showing positive enrichment of the mesenchymal signature (Mes) gene set and negative enrichment of the proneural (PN) and classical signature (CL) gene sets in relapsed tumors when compared witth primary tumors. D, Gene expression comparison between mesenchymal (MES) and nonmesenchymal (non-MES) GBMs. The expression profiles of 369 IDH wild-type GBMs were compiled from TCGA GBM dataset. Cases were classified into mesenchymal, proneural, and classical GBMs according to the latest glioma-intrinsic subtype signatures. Green and pink left bar indicate fold change values (MES vs. non-MES) less and greater than 1, respectively. Heatmap represents the normalized gene expression of 13 genes, and red and green color indicate high and low expression, respectively. E and F, qPCR analysis of primary (P) and relapsed (R) tumors for PN or CL (E) or MES (F) signature genes in tumor tissues. Mean ± SEM of 4 to 10 tumors. Statistical significance was determined by the unpaired t test for E and F. *, P < 0.05; **, P < 0.01; ****, P < 0.0001. Experiments for A and B were independently repeated three times to ensure the reproducibility of the findings.

Figure 2.

Relapsed tumors following EGFR* inhibition show subtype transition. A, Representative images of Western blot analysis for indicated proteins in primary and recurrent tumors. Samples were run on two gels to detect multiple proteins, and the most representative loading control blot is shown. B, Representative images of IHC staining performed on sections of primary and relapsed tumors. Scale bars, 50 μm. C, One-sided GSEA plots showing positive enrichment of the mesenchymal signature (Mes) gene set and negative enrichment of the proneural (PN) and classical signature (CL) gene sets in relapsed tumors when compared witth primary tumors. D, Gene expression comparison between mesenchymal (MES) and nonmesenchymal (non-MES) GBMs. The expression profiles of 369 IDH wild-type GBMs were compiled from TCGA GBM dataset. Cases were classified into mesenchymal, proneural, and classical GBMs according to the latest glioma-intrinsic subtype signatures. Green and pink left bar indicate fold change values (MES vs. non-MES) less and greater than 1, respectively. Heatmap represents the normalized gene expression of 13 genes, and red and green color indicate high and low expression, respectively. E and F, qPCR analysis of primary (P) and relapsed (R) tumors for PN or CL (E) or MES (F) signature genes in tumor tissues. Mean ± SEM of 4 to 10 tumors. Statistical significance was determined by the unpaired t test for E and F. *, P < 0.05; **, P < 0.01; ****, P < 0.0001. Experiments for A and B were independently repeated three times to ensure the reproducibility of the findings.

Close modal

To search for the molecular components underlying the relapse, we re-analyzed our previously published expression profiles of three relapsed tumors versus their paired treatment-naïve controls (GSE64751; ref. 16). Using the classifying gene list developed by the TCGA group (20), GSEA revealed that relapsed tumors were strongly enriched for MES signatures, whereas the treatment-naïve tumors exhibited features of the classical and proneural subtypes (Fig. 2B and C). Gene expression heatmap and quantitative real-time PCR (qPCR) further identified a set of significantly upregulated mesenchymal-subtype–associated genes (Vim, Tgfbi, Ykl40, and Serpine1) as well as a panel of silenced proneural- or classical-subtype–enriched genes (Olig2, Dll3, Bcan, Ncam1, and Ascl1; Fig. 2D–F).

Enforced mesenchymal transcription factor expression relieves EGFR dependency

In agreement with their mesenchymal transdifferentiation, qPCR and immunoblot analyses of the relapsed tumors revealed a significant upregulation of core mesenchymal transcriptional factors SLUG and TWIST, as compared with their treatment-naïve controls (Fig. 3A–C). IHC analysis for SLUG or TWIST indicated that the enhanced expression in the relapsed tumors was pervasive and not confined to a particular pocket area (Fig. 3D). Given that mesenchymal transdifferentiation has been a touted resistance mechanism against anti-EGFR therapies in both clinic and model systems of lung cancer treatment (21, 22), our observation raised a possibility that dysregulated mesenchymal transcriptional factor expression might underlie the relapse development upon oncogenic EGFR* suppression.

Figure 3.

Mesenchymal transcription factor expression-reprogrammed tumors progress independently of EGFR signaling. A and B, qPCR analysis of primary (P) and relapsed (R) tumors for Slug (A) or Twist1 (B). Mean ± SEM of 8 tumors. C, Representative images of Western blot analysis for indicated proteins in primary and relapsed tumors. D, Representative images of IHC staining performed on sections of primary and relapsed tumors. Scale bars, 50 μm. E, Representative images of Western blot analysis for SLUG in iEIP cells expressing empty vector (EV) or Slug. F, Schema for orthotopic transplantation of EV and Slug-expressing iEIP cells, doxycycline treatment, and IVIS imaging. G, Representative BLI on indicated days after doxycycline (Dox) treatment. H, Tumor growth was measured at indicated times and calculated relative to initial tumor volume. Mean ± SEM of 3 to 5 biological replicates. Day 0 represents the day when treatment was initiated. I, Heat map of C, CD, S, and SD showing PN, CL, and MS enrichment. The signature score was calculated using ssGSEA. J, BLI of representative tumor-bearing animals is shown. Bottom right, Western blot analysis for SLUG in GSC280 cells expressing empty vector (Ctrl) or Slug before injection. K, Survival plot for mice bearing GSC280 control (Ctrl) or SLUG-expressing tumors treated with vehicle (N = 3) or erlotinib (Ctrl, N = 3; SLUG, N = 4). Statistical significance was determined by the unpaired t test for A. **, P < 0.01; ****, P < 0.0001. Experiments for C, D, E, and J were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Figure 3.

Mesenchymal transcription factor expression-reprogrammed tumors progress independently of EGFR signaling. A and B, qPCR analysis of primary (P) and relapsed (R) tumors for Slug (A) or Twist1 (B). Mean ± SEM of 8 tumors. C, Representative images of Western blot analysis for indicated proteins in primary and relapsed tumors. D, Representative images of IHC staining performed on sections of primary and relapsed tumors. Scale bars, 50 μm. E, Representative images of Western blot analysis for SLUG in iEIP cells expressing empty vector (EV) or Slug. F, Schema for orthotopic transplantation of EV and Slug-expressing iEIP cells, doxycycline treatment, and IVIS imaging. G, Representative BLI on indicated days after doxycycline (Dox) treatment. H, Tumor growth was measured at indicated times and calculated relative to initial tumor volume. Mean ± SEM of 3 to 5 biological replicates. Day 0 represents the day when treatment was initiated. I, Heat map of C, CD, S, and SD showing PN, CL, and MS enrichment. The signature score was calculated using ssGSEA. J, BLI of representative tumor-bearing animals is shown. Bottom right, Western blot analysis for SLUG in GSC280 cells expressing empty vector (Ctrl) or Slug before injection. K, Survival plot for mice bearing GSC280 control (Ctrl) or SLUG-expressing tumors treated with vehicle (N = 3) or erlotinib (Ctrl, N = 3; SLUG, N = 4). Statistical significance was determined by the unpaired t test for A. **, P < 0.01; ****, P < 0.0001. Experiments for C, D, E, and J were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Close modal

To test this hypothesis, a cohort of immunocompromised Rag1−/− recipient mice were orthotopically grafted with luciferase- expressing iEIP glioma cells that were additionally transduced with either a lentiviral vector control (EV) or Slug expressing construct (Fig. 3E). The animals were kept off-dox and tumor growth was monitored weekly by BLI. After the BLI signals reached approximately 5 × 105, the animals were switched to doxycycline-containing drinking water to turn off the EGFR* transgene (Fig. 3F). As expected, the EV-transduced control tumors exhibited a robust initial response upon doxycycline treatment, followed by an extended period of dormancy before the eventual recurrence (Fig. 3G and H). By contrast, the growth of Slug-transduced tumors showed little response to doxycycline-induced EGFR* suppression, indicating that upregulation of mesenchymal transcription factors is able to drive oncogenic EGFR-independent iEIP glioma progression. Further RNA-seq analysis of the four groups of tumor samples (C, control EV-transduced doxycycline treatment-naïve; S, Slug-transduced doxycycline treatment-naïve; CD, control EV-transduced relapse under doxycycline treatment; and SD, Slug-transduced under doxycycline treatment) confirmed that enforced Slug expression was sufficient to promote mesenchymal transdifferentiation of iEIP tumor cells (Fig. 3I). But notably, although enforced Slug expression in iEIP glioma cells was able to overcome their EGFR* addiction, knockdown of Slug in relapsed iEIP tumor cells was not sufficient to restore their EGFR* dependency (Supplementary Fig. S2A). Instead, we found increased TWIST1 expression in these Slug-depleted tumors (Supplementary Fig. S2B and S2C), suggesting a functional redundancy among mesenchymal transcriptional factors.

To determine whether mesenchymal transdifferentiation also promoted oncogenic EGFR independency in human gliomas, patient-derived EGFRhighGSC280 glioma cells were transduced with lentivirus encoding either vector control or SLUG before orthotopically grafted into immunocompromised recipient animals. In accordance with the findings from iEIP mouse glioma cells, the tumors derived from the SLUG-transduced cells failed to respond to EGFR kinase inhibitor erlotinib treatment (Fig. 3J and K). To further demonstrate that mesenchymal transdifferentiation is capable of promoting oncogenic EGFR-independent human glioma growth, we transduced another patient-derived GS7–11 glioma cells with lentiviral control or EGFR-targeting sgRNA (EGFRc). These cells were next infected with lentiviral vector control or SLUG and subcutaneously transplanted into immunocompromised recipient animals (Supplementary Fig. S3A). As expected, the EGFR-depleted glioma cells exhibited a significantly diminished tumorigenicity relative to their EV-transduced controls (Supplementary Fig. S3B and S3C). By contrast, transduction of a SLUG construct completely nullified the tumor-suppressive effect of EGFR depletion on tumor progression, although SLUG expression by itself did not have a significant growth effect on control tumors. These findings support the role of mesenchymal core transcriptional factors in promoting glioma fate transdifferentiation and oncogenic EGFR signaling-independent relapse.

Relapse-inducing cells emerge during the proximal EGFR inhibition response phase

Tumor relapse from targeted therapies often results from a rare subpopulation of treatment-resistant cells that persist through therapy. Indeed, immunofluorescence analysis of doxycycline-treated tumors from orthotopically transplanted iEIP cells identified residues of GFP-positive tumor cells that survived through a 10-day treatment (Fig. 4A). Bromodeoxyuridine incorporation analysis indicated that these EGFR-depleted residual tumor cells remained a relatively non-proliferative or dormant state after surviving through the initial apoptosis (Fig. 4B). Their non-clustering distribution pattern also suggested that these relapse-inducing cells were unlikely originated from a subclone of tumor cells with pre-existing resistance-conferring mutations.

Figure 4.

Relapse-inducing cells emerge during the proximal EGFR inhibition response phase. A and B, Representative immunofluorescence analysis for BrdU (bromodeoxyuridine) incorporation (A) and EGFR expression (B) in GFP-labeled iEIP tumors treated with doxycycline for indicated time. Scale bars, 100 μm. C, Representative Western blot analysis for indicated proteins in iEIP tumors; R, relapsed tumor. Samples were run on two gels to detect multiple proteins, and the most representative loading control blot is shown. D and H, Representative IHC images for indicated proteins in iEIP tumors following doxycycline treatment for indicated time. Scale bars, 50 μm (D), 10 μm (H). E–G, qPCR analysis. Data are presented as mean ± SEM of 8 biological replicates. Statistical significance was determined by one-way ANOVA for E–G. ***, P < 0.002; ****, P < 0.0001. Experiments for A, B, C, D, and H were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Figure 4.

Relapse-inducing cells emerge during the proximal EGFR inhibition response phase. A and B, Representative immunofluorescence analysis for BrdU (bromodeoxyuridine) incorporation (A) and EGFR expression (B) in GFP-labeled iEIP tumors treated with doxycycline for indicated time. Scale bars, 100 μm. C, Representative Western blot analysis for indicated proteins in iEIP tumors; R, relapsed tumor. Samples were run on two gels to detect multiple proteins, and the most representative loading control blot is shown. D and H, Representative IHC images for indicated proteins in iEIP tumors following doxycycline treatment for indicated time. Scale bars, 50 μm (D), 10 μm (H). E–G, qPCR analysis. Data are presented as mean ± SEM of 8 biological replicates. Statistical significance was determined by one-way ANOVA for E–G. ***, P < 0.002; ****, P < 0.0001. Experiments for A, B, C, D, and H were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Close modal

To address whether mesenchymal transdifferentiation occurred at the early phase of treatment, we next examined acute response of the orthotopically transplanted iEIP tumors upon doxycycline-induced EGFR* suppression. Immunoblot analysis of tumor samples revealed a rapid reduction of transgene-encoded EGFR* protein expression by day 3 of doxycycline treatment (Fig. 4C). This prompt reduction of EGFR* expression correlated with a rapid increase of cleaved caspase-3 levels, an indicative of apoptosis (Fig. 4C and D). Along this line, qPCR and IHC analyses revealed a rapid diminution of proneural marker Olig2 expression upon doxycycline administration (Fig. 4E). Importantly, this process was accompanied by a sporadic upsurge of surviving tumor cells that expressed mesenchymal transcription factor Slug or Twist1 (Fig. 4F–H), suggesting that the relapse is likely driven by a stochastic process of mesenchymal transdifferentiation.

YAP1 activation drives mesenchymal transdifferentiation in response to acute EGFR* suppression

To characterize the transition state following EGFR* deprivation, we performed RNA-seq of orthotopically transplanted iEIP tumors following control or doxycycline treatment. GSEA revealed YAP (Yes‐associated protein) signaling (NES = −2, FWER P = 0) as the top increased gene expression signature of doxycycline-treated tumors (Fig. 5A and B). Interestingly, although qPCR assay revealed no consistent Yap1 gene upregulation upon acute EGFR* suppression (Fig. 5C), IHC analysis of tumor samples following 1- or 3-day doxycycline treatment revealed a progressively increased nuclear YAP1 localization (Fig. 5D and E). This is consistent with the notion that intracellular localization is a key determinant of YAP activity (23). Notably, nuclear-localized YAP1 was mostly limited to tumor cells labeled by their luciferase expression (Supplementary Fig. S4A).

Figure 5.

YAP1 activation drives mesenchymal transdifferentiation in response to acute EGFR* suppression. A and B, GSEA of C versus CD, S, SD tumors. Enriched oncogenic pathways (A). One-sided GSEA for negative enrichment of YAP-related gene sets in primary tumors (B). C, qPCR analysis in iEIP tumors. Mean ± SEM of 4 to 12 biological replicates. D, Representative images of YAP1 IHC. Bottom panels are of higher magnification than upper ones; scale bars, 50 μm. E, Summary of subcellular localization of YAP1 on control (N = 263), doxycycline 1 day (N = 212), 3 day (N = 210), and relapse tumor (N = 168) cells. Results are from four biological replicates. F, Representative Western blot analysis of iEIP cells infected with lentivirus expressing respective YAP1 constructs. G and H, qPCR analysis in iEIP cells expressing YAP1 or YAP1 mutant (DA: YAP1S127A, DN: YAP1S94A). Mean ± SEM of four independent experiments. I, Relative tumor volume of iEIP subcutaneous tumor. Mice bearing subcutaneously transplanted iEIP cells expressing YAP1S127A (N = 6), YAP1S94A (N = 6), or control (N = 8) were treated with either vehicle or doxycycline containing water. Percentage of survival of mice sacrificed by the tumor size reached to 200∼400 mm3. Mean ± SEM of 6 to 8 tumors. Statistical significance was determined by one-way ANOVA for C, G, H, and I. **, P < 0.01; ****, P < 0.0001. Experiments for D and F were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Figure 5.

YAP1 activation drives mesenchymal transdifferentiation in response to acute EGFR* suppression. A and B, GSEA of C versus CD, S, SD tumors. Enriched oncogenic pathways (A). One-sided GSEA for negative enrichment of YAP-related gene sets in primary tumors (B). C, qPCR analysis in iEIP tumors. Mean ± SEM of 4 to 12 biological replicates. D, Representative images of YAP1 IHC. Bottom panels are of higher magnification than upper ones; scale bars, 50 μm. E, Summary of subcellular localization of YAP1 on control (N = 263), doxycycline 1 day (N = 212), 3 day (N = 210), and relapse tumor (N = 168) cells. Results are from four biological replicates. F, Representative Western blot analysis of iEIP cells infected with lentivirus expressing respective YAP1 constructs. G and H, qPCR analysis in iEIP cells expressing YAP1 or YAP1 mutant (DA: YAP1S127A, DN: YAP1S94A). Mean ± SEM of four independent experiments. I, Relative tumor volume of iEIP subcutaneous tumor. Mice bearing subcutaneously transplanted iEIP cells expressing YAP1S127A (N = 6), YAP1S94A (N = 6), or control (N = 8) were treated with either vehicle or doxycycline containing water. Percentage of survival of mice sacrificed by the tumor size reached to 200∼400 mm3. Mean ± SEM of 6 to 8 tumors. Statistical significance was determined by one-way ANOVA for C, G, H, and I. **, P < 0.01; ****, P < 0.0001. Experiments for D and F were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Close modal

YAP is the main effector of Hippo signal transduction pathway. Its activation has been implicated in resistance to various targeted therapies, including EGFR TKIs in NSCLC (22, 24). To determine whether YAP1 activation contributes to treatment-induced mesenchymal transdifferentiation and relapse development, we transduced primary iEIP glioma cells with retrovirus encoding YAP wild-type (YAPWT), nuclear-localized active mutant (YAPS127A; ref. 25), or mutant (YAPS94A) defective of TEAD binding and nuclear translocation (Fig. 5F; ref. 26). qPCR analysis of glioma cells transduced with either YAPWT or active YAPS127A, but not YAPS94A mutant, significantly upregulated the expression of Slug and other MES markers, including Tgfbi, Serpine1 and Vim (Fig. 5G and H), indicating that YAP activation is capable of driving mesenchymal transdifferentiation.

To determine whether activated YAP could promote oncogenic EGFR* independent glioma growth in vivo, primary iEIP cells transduced with vector control or construct encoding YAPS127A or YAPS94A were subcutaneously transplanted into immunocompromised recipient animals. The animals were kept off doxycycline until the tumor reached a palpable size (∼200 mm3). Notably, expression of YAPS127A or YAPS94A did not affect primary tumor growth (Fig. 5I). However, the proliferation of active YAPS127A-expressing tumors showed little response to doxycycline-induced EGFR* suppression, similar to the Slug-expressing iEIP tumors. By contrast, dominant negative YAPS94A-expressing tumors experienced a much deeper initial regression upon doxycycline treatment and also took significantly longer to recur, indicating that YAP activity is required for EGFR*-independent relapse.

Treatment-induced TGFβ secretion activates YAP

Targeted therapy has been shown to induce a network of secreted signals that can instigate resistance development by supporting relapse-inducing cells (27, 28). To determine whether the anti-EGFR treatment-induced secretion of signaling factors could also promote YAP activation, we generated conditioned media (CM) from control and doxycycline-treated iEIP cultures. Immunoblot and immunofluorescence analyses revealed that iEIP cells treated with dox-CM had a significantly increased nuclear YAP fraction as compared with the control CM-treated samples (Fig. 6A and B). These results suggest that EGFR* suppression-stressed iEIP cells might secret YAP-activating factor(s).

Figure 6.

Treatment-induced TGFβ secretion activates YAP. A and B, Representative images of Western blot analysis for nuclear/cytosolic fraction (A) and immunofluorescence analysis (B) in iEIP cells treated with conditioned medium (CM) of iEIP cells cultured with or without doxycycline (Dox). Scale bars, 10 μm. C, qPCR for Tgfb1 in iEIP tumors. Mean ± SEM of 4 to 6 tumors. Statistical significance was determined by one-way ANOVA. ***, P < 0.002; ****, P < 0.0001. D and E, Representative images of immunofluorescence analysis (D) and Western blot analysis for nuclear/cytosolic fraction (E) in iEIP cells treated with 10 ng/mL of TGFβ1. Scale bars, 10 μm. F, Representative images of Western blot analysis for nuclear/cytosolic fraction in iEIP cells treated with conditioned medium of iEIP cells cultured with or without 10 μmol/L of SB525334. The bar graphs under Western blot images of A, E, and F are quantification for each blot. G, Tumor growth was measured at indicated time points and calculated relative to the initial tumor volume. NT, N = 4; SB, N = 4; DOX, N = 3; DOX+SB, N = 5. Data are expressed as mean ± SEM. Statistical significance was determined by two-tailed t test for week 7 tumors comparing DOX versus DOX+SB groups (**, P = 0.001). Experiments for A, B, D, E, and F were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Figure 6.

Treatment-induced TGFβ secretion activates YAP. A and B, Representative images of Western blot analysis for nuclear/cytosolic fraction (A) and immunofluorescence analysis (B) in iEIP cells treated with conditioned medium (CM) of iEIP cells cultured with or without doxycycline (Dox). Scale bars, 10 μm. C, qPCR for Tgfb1 in iEIP tumors. Mean ± SEM of 4 to 6 tumors. Statistical significance was determined by one-way ANOVA. ***, P < 0.002; ****, P < 0.0001. D and E, Representative images of immunofluorescence analysis (D) and Western blot analysis for nuclear/cytosolic fraction (E) in iEIP cells treated with 10 ng/mL of TGFβ1. Scale bars, 10 μm. F, Representative images of Western blot analysis for nuclear/cytosolic fraction in iEIP cells treated with conditioned medium of iEIP cells cultured with or without 10 μmol/L of SB525334. The bar graphs under Western blot images of A, E, and F are quantification for each blot. G, Tumor growth was measured at indicated time points and calculated relative to the initial tumor volume. NT, N = 4; SB, N = 4; DOX, N = 3; DOX+SB, N = 5. Data are expressed as mean ± SEM. Statistical significance was determined by two-tailed t test for week 7 tumors comparing DOX versus DOX+SB groups (**, P = 0.001). Experiments for A, B, D, E, and F were independently repeated three times with three different sets to ensure the reproducibility of the findings.

Close modal

To identify the relevant YAP-activating factor(s), we analyzed gene expression changes and found a panel of cytokines (i.e., GRN, FIGF, TNFA, SLIT2, and TGFB1) whose expression was significantly augmented in relapsed iEIP tumors compared with their treatment-naïve controls (Supplementary Fig. S5A and S5B). To determine their role in YAP1 activation and mesenchymal transdifferentiation upon EGFR* suppression, we treated the cultured primary iEIP cells with individual factors and assessed their mesenchymal transdifferentiation potential. qPCR analysis revealed that treatment of TNFα, GRN, FIGF, or SLIT2 did not significantly affect MES gene expression (Supplementary Fig. S5C–S5E). By contrast, TGFβ stimulation markedly augmented Serpine1, Tgfbi and Vim mRNA transcription. Remarkably, Tgfb1 expression in iEIP tumors was greatly elevated in the acute phase of EGFR* inhibition, tracking with the timeline of YAP1 activation (Fig. 6C). Immunofluorescence and immunoblot analyses of subcellular fractionates of cultured iEIP cells confirmed a strong increase of YAP1 nuclear translocation upon TGFβ treatment (Fig. 6D and E). Conversely, inhibition of TGFβ signaling by TGFβR1 inhibitor SB525334 compromised dox-CM–induced YAP nuclear accumulation (Fig. 6F). Moreover, treatment of the patient-derived GSC280 glioma cells with EGFR kinase inhibitor erlotinib also induced TGFB1 mRNA expression upregulation and subsequent YAP activation (Supplementary Fig. S6A–S6D). Together, these findings support the TGFβ/YAP/SLUG signaling axis as a crucial mediator of treatment-induced mesenchymal reprogramming and anti-EGFR resistance development. Consistently, combined TGFβ receptor inhibitor (SB525334) with doxycycline-mediated EGFR* inhibition in treatment of subcutaneously grafted iEIP tumors markedly delayed the relapse development and prolonged animal survival, as compared with the doxycycline treatment alone (Fig. 6G).

Activation of YAP in recurrent gliomas predicts poor patient survival

Having established that YAP activation stimulates mesenchymal transdifferentiation and relapse from treatment, we next asked whether YAP activation also correlated with mesenchymal transition in recurrent clinical samples. We first established a panel of total 22 YAP core signature genes from MSigDB that were also significantly upregulated in EGFR*-independent iEIP tumors (CD, S, and SD) as compared with their treatment-naïve controls (C) (Fig. 7A). We next re-analyzed one of our previously published cohorts (19). 39 of total the 91 recurrent GBM patient samples were classified into MES, whereas the rest 52 as non-MES. The Wilcoxon rank-sum test of the MES subtype versus to the non-MES group indicated a significantly elevated YAP1 expression and also increased signature scores for YAP core target genes (ssGSEA; Fig. 7B). IHC analysis of 22 recurrent GBM samples further revealed a strong correlation between nuclear-localized YAP1 levels and core mesenchymal transcriptional factor SLUG protein expression (P < 0.0001, R2 = 0.78, Fig. 7C and D).

Figure 7.

Activation of YAP in recurrent gliomas predicts poor patient survival. A, Heat map of 22 YAP core signature genes significantly upregulated in EGFR*-independent tumors. B, The comparison of the average score for the YAP pathway or expression of YAP1 between MES and non-MES in recurrent tumors. *, P < 0.05; **, P < 0.01. C and D, Recurrent GBM patient samples (N = 22) were scored for YAP1 and SLUG/SNAIL nuclear expression. C and D, Representative images of YAP1 IHC (C) and H-score plot (D). E and F, Survival analysis of paired IDH1 WT patients with GBM (N = 91). Overall survival (E) and PFS (F) analyses of recurrent patients with either high or low YAP core signature expressions. G, Schema for mechanism of TGFβ-YAP1 activation SLUG-induced mesenchymal transition under EGFR-targeted therapy.

Figure 7.

Activation of YAP in recurrent gliomas predicts poor patient survival. A, Heat map of 22 YAP core signature genes significantly upregulated in EGFR*-independent tumors. B, The comparison of the average score for the YAP pathway or expression of YAP1 between MES and non-MES in recurrent tumors. *, P < 0.05; **, P < 0.01. C and D, Recurrent GBM patient samples (N = 22) were scored for YAP1 and SLUG/SNAIL nuclear expression. C and D, Representative images of YAP1 IHC (C) and H-score plot (D). E and F, Survival analysis of paired IDH1 WT patients with GBM (N = 91). Overall survival (E) and PFS (F) analyses of recurrent patients with either high or low YAP core signature expressions. G, Schema for mechanism of TGFβ-YAP1 activation SLUG-induced mesenchymal transition under EGFR-targeted therapy.

Close modal

Finally, we evaluated the effect of YAP activation on patient survival in 54 cases for whom annotation on overall survival time and time to disease progression (progression-free survival, PFS) were available. Patients whose recurrent tumors were classified as high YAP-ssGSEA (top 40%) trended toward adverse overall survival (log rank test P = 0.0192 with HR = 0.499; Fig. 7E) and PFS (log rank test P = 0.0112 with HR = 0.526; Fig. 7F). Collectively, these results suggest that YAP-dependent mesenchymal program represents a highly relevant molecular pathway that determines disease recurrence, progression, and prognosis.

Targeted-therapy is the standard care for many cancer types that harbor an activated oncogene. But even with effective therapies, rare tumor cells always survive and eventually re-initiate the malignant disease (29). In this study, we applied an inducible oncogenic EGFR-driven glioma mouse model to investigate anti-EGFR therapeutic resistance development. By analyzing the proximal and distal response upon oncogenic EGFR* deprivation, our study identified treatment-induced cell plasticity as an underlying mechanism that drives EGFR-targeted therapy evasion and relapse development. Our findings revealed that anti-EGFR therapy activated a TGFβ/YAP/SLUG signaling cascade that subsequently instigated mesenchymal lineage transdifferentiation in a rare population of glioma relapse–initiating cells. We further demonstrated that this process enabled reprogramming of the oncogenic EGFR-addicted glioma cells toward a phenotypic state no longer relying on EGFR signaling (Fig. 7G). Inhibition of this adaptive response significantly delayed relapse from anti-EGFR treatment and prolonged animal survival.

EGFR is amplified and mutated in a majority of human malignant gliomas; yet various treatment strategies targeting EGFR have thus far failed in clinical trials (30). One major reason might be insufficient target efficacy due to either suboptimal mutant EGFR inhibition (8, 9), or inefficient drug penetration and distribution in the CNS (31, 32). In addition, a body of reports also pointed that intratumoral heterogeneity or compensatory upregulation of other RTKs, including PDGFRA and MET, could contribute to anti-EGFR therapy resistance (10, 11, 33, 34). Similarly, activation of downstream effectors or engagement of alternative survival pathways could also mediate EGFR-targeted therapy evasion (12, 35, 36). But aside from its classical function as receptor tyrosine kinase (RTK), it is worthy to note that EGFR can also proceed pro-growth and -survival functions independent of its kinase activities (37, 38). Given the fact that EGFR is highly overexpressed in a majority of malignant glioma patient samples, it is tempting to speculate that kinase-independent activities of EGFR may also contribute to the therapeutic resistance, particularly against RTK inhibitors.

In addition to the aforementioned anti-EGFR resistance mechanisms that rely on activation of compensatory signaling effectors of the same or alternative survival pathways, our current study points to the importance of glioma cell plasticity as another avenue of therapy evasion. Indeed, even with near complete oncogenic EGFR deprivation that leads to deep response, we found that a small population of iEIP glioma cells would survive and undergo treatment-induced transdifferentiation toward a mesenchymal-like state that no longer depends on oncogenic EGFR*. These findings indicate that anti-EGFR therapeutic resistance in malignant glioma might result, at least in part, from treatment-induced cell plasticity and enhancement of mesenchymal transdifferentiation. Mesenchymal switching, an important biological process that enables cells to reprogram toward distinct phenotypes in response to their environmental changes, is identified not only during development such as mesoderm and neural tube formation, but also upon injury and disease (39, 40). In cancers, epithelial-to-mesenchymal transition (EMT) has been widely associated with tumor stemness, metastasis, and drug resistance (29, 41). The development of EMT as a drug-resistant mechanism has been observed in various in vitro and in vivo model systems of epithelia cancers, including EGFR mutant lung cancer (22, 42), KRAS mutant colon and pancreatic cancers (43, 44), and PyMT-driven breast cancer (45). In malignant gliomas, transcriptome comparison of paired primary and recurrent GBM samples following standard therapy also revealed a trend of phenotypic and molecular shift toward a mesenchymal state during recurrence (46), suggesting that mesenchymal tumor cells may intrinsically lack sensitivity to many conventional and targeted therapies. Besides mesenchymal transformation, other forms of lineage plasticity such as drug-induced neuroendocrine lineage transdifferentiation, has also been implicated in therapeutic resistance development in NSCL and prostate cancer (21, 47, 48), further underscoring cell plasticity as a general resistance mechanism against conventional and targeted therapies.

Our study reveals that anti-EGFR treatment induces glioma mesenchymal lineage transdifferentiation and oncogenic EGFR-independent relapse through activation of the TGFβ/YAP/SLUG signaling axis. TGFβ, a well-known EMT inducer, has been associated with reduced treatment effectiveness in many types of cancers (49, 50). Our findings indicate that therapy-triggered TGFβ secretion promotes YAP nuclear shuffling and subsequent upregulation of mesenchymal transcriptional factor SLUG. This is consistent with recent reports that causally link the mesenchymal transformation with YAP-mediated bypass of EGFR or KRAS inhibition (22, 43). Similarly, we observed that although inhibition of YAP signaling by itself had little effect on primary iEIP glioma progression, combinatorial suppression of oncogenic EGFR* with TGFβ or YAP signaling dampened the anti-EGFR treatment–induced mesenchymal transdifferentiation and delayed relapse development. Together, these indicate that activation of YAP signaling can drive mesenchymal transdifferentiation to promote therapy evasion in cancer treatment.

Relapse under targeted therapy often results from rare treatment-refractory tumor cells that persist through treatment (29). In this study, we identified a small portion of Slug and/or Twist-positive iEIP glioma cells that survived initial EGFR* deprivation-induced apoptosis and potentiate later tumor regrowth. Interestingly, those mesenchymal-like relapse-initiating cells do not seem to pre-exist in treatment-naïve tumors but rather arise stochastically following the treatment. These findings support the facultative transient resistance model in which a small portion of tumor cells transiently acquire drug-refractory state by epigenetic modification (29). One possible scenario is that primary iEIP glioma cells dynamically express fluctuating levels of mesenchymal transcription factors such as SLUG and/or TWIST1. Only cells that display high levels of expression at the time of treatment can survive and assume a transient drug-refractory state. And the final relapse would rely on establishment of a stable anti–EGFR-resistant state, a process that is presumably enabled by TGFβ/YAP/SLUG signaling-mediated mesenchymal transdifferentiation. Thus, mechanism-based blocking of lineage transdifferentiation may aid the development of future therapeutic approaches against malignant gliomas.

No disclosures were reported.

H. Oh: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft. I. Hwang: Data curation, formal analysis, validation, investigation, writing–original draft, writing–review and editing. J.-Y. Jang: Data curation. L. Wu: Formal analysis. D. Cao: Data curation. J. Yao: Formal analysis, investigation, visualization, writing–review and editing. H. Ying: Conceptualization, resources, writing–review and editing. J.Y. Li: Resources. Y. Yao: Resources, data curation, supervision, writing–review and editing. B. Hu: Conceptualization, supervision, methodology. Q. Wang: Resources, formal analysis, supervision. H. Zheng: Conceptualization, resources, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J. Paik: Conceptualization, resources, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

We thank Dr. Jan Koster for support with data analysis. This work was supported by the Irma T. Hirschl Award and the Feldstein Medical Foundation (to J. Paik).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Dunn
GP
,
Rinne
ML
,
Wykosky
J
,
Genovese
G
,
Quayle
SN
,
Dunn
IF
, et al
Emerging insights into the molecular and cellular basis of glioblastoma
.
Genes Dev
2012
;
26
:
756
84
.
2.
Wen
PY
,
Weller
M
,
Lee
EQ
,
Alexander
BA
,
Barnholtz-Sloan
JS
,
Barthel
FP
, et al
Glioblastoma in adults: a Society for Neuro-Oncology (SNO) and European Society of Neuro-Oncology (EANO) consensus review on current management and future directions
.
Neuro Oncol
2020
;
22
:
1073
113
.
3.
Parsons
DW
,
Jones
S
,
Zhang
X
,
Lin
JC
,
Leary
RJ
,
Angenendt
P
, et al
An integrated genomic analysis of human glioblastoma multiforme
.
Science
2008
;
321
:
1807
12
.
4.
Brennan
CW
,
Verhaak
RG
,
McKenna
A
,
Campos
B
,
Noushmehr
H
,
Salama
SR
, et al
The somatic genomic landscape of glioblastoma
.
Cell
2013
;
155
:
462
77
.
5.
Lynch
TJ
,
Bell
DW
,
Sordella
R
,
Gurubhagavatula
S
,
Okimoto
RA
,
Brannigan
BW
, et al
Activating mutations in the epidermal growth factor receptor underlying responsiveness of non–small cell lung cancer to gefitinib
.
N Engl J Med
2004
;
350
:
2129
39
.
6.
Paez
JG
,
Jänne
PA
,
Lee
JC
,
Tracy
S
,
Greulich
H
,
Gabriel
S
, et al
EGFR mutations in lung cancer: correlation with clinical response to gefitinib therapy
.
Science
2004
;
304
:
1497
500
.
7.
Thorne
AH
,
Zanca
C
,
Furnari
F
. 
Epidermal growth factor receptor targeting and challenges in glioblastoma
.
Neuro Oncol
2016
;
18
:
914
8
.
8.
Vivanco
I
,
Robins
HI
,
Rohle
D
,
Campos
C
,
Grommes
C
,
Nghiemphu
PL
, et al
Differential sensitivity of glioma- versus lung cancer-specific EGFR mutations to EGFR kinase inhibitors
.
Cancer Discov
2012
;
2
:
458
71
.
9.
Barkovich
KJ
,
Hariono
S
,
Garske
AL
,
Zhang
J
,
Blair
JA
,
Fan
QW
, et al
Kinetics of inhibitor cycling underlie therapeutic disparities between EGFR-driven lung and brain cancers
.
Cancer Discov
2012
;
2
:
450
7
.
10.
Nathanson
DA
,
Gini
B
,
Mottahedeh
J
,
Visnyei
K
,
Koga
T
,
Gomez
G
, et al
Targeted therapy resistance mediated by dynamic regulation of extrachromosomal mutant EGFR DNA
.
Science
2014
;
343
:
72
6
.
11.
Stommel
JM
,
Kimmelman
AC
,
Ying
H
,
Nabioullin
R
,
Ponugoti
AH
,
Wiedemeyer
R
, et al
Coactivation of receptor tyrosine kinases affects the response of tumor cells to targeted therapies
.
Science
2007
;
318
:
287
90
.
12.
Zanca
C
,
Villa
GR
,
Benitez
JA
,
Thorne
AH
,
Koga
T
,
D'Antonio
M
, et al
Glioblastoma cellular cross-talk converges on NF-κB to attenuate EGFR inhibitor sensitivity
.
Genes Dev
2017
;
31
:
1212
27
.
13.
Humphrey
PA
,
Wong
AJ
,
Vogelstein
B
,
Friedman
HS
,
Werner
MH
,
Bigner
DD
, et al
Amplification and expression of the epidermal growth factor receptor gene in human glioma xenografts
.
Cancer Res
1988
;
48
:
2231
8
.
14.
Bigner
SH
,
Humphrey
PA
,
Wong
AJ
,
Vogelstein
B
,
Mark
J
,
Friedman
HS
, et al
Characterization of the epidermal growth factor receptor in human glioma cell lines and xenografts
.
Cancer Res
1990
;
50
:
8017
22
.
15.
Pandita
A
,
Aldape
KD
,
Zadeh
G
,
Guha
A
,
James
CD
. 
Contrasting in vivo and in vitro fates of glioblastoma cell subpopulations with amplified EGFR
.
Genes Chromosomes Cancer
2004
;
39
:
29
36
.
16.
Klingler
S
,
Guo
B
,
Yao
J
,
Yan
H
,
Zhang
L
,
Vaseva
AV
, et al
Development of resistance to EGFR-targeted therapy in malignant glioma can occur through EGFR-dependent and -independent mechanisms
.
Cancer Res
2015
;
75
:
2109
19
.
17.
Kim
DY
,
Hwang
I
,
Muller
FL
,
Paik
JH
. 
Functional regulation of FoxO1 in neural stem cell differentiation
.
Cell Death Differ
2015
;
22
:
2034
45
.
18.
Bowman
RL
,
Wang
Q
,
Carro
A
,
Verhaak
RG
,
Squatrito
M
. 
GlioVis data portal for visualization and analysis of brain tumor expression datasets
.
Neuro Oncol
2017
;
19
:
139
41
.
19.
Wang
Q
,
Hu
B
,
Hu
X
,
Kim
H
,
Squatrito
M
,
Scarpace
L
, et al
Tumor evolution of glioma-intrinsic gene expression subtypes associates with immunological changes in the microenvironment
.
Cancer Cell
2017
;
32
:
42
56
.
20.
Verhaak
RG
,
Hoadley
KA
,
Purdom
E
,
Wang
V
,
Qi
Y
,
Wilkerson
MD
, et al
Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1
.
Cancer Cell
2010
;
17
:
98
110
.
21.
Sequist
LV
,
Waltman
BA
,
Dias-Santagata
D
,
Digumarthy
S
,
Turke
AB
,
Fidias
P
, et al
Genotypic and histological evolution of lung cancers acquiring resistance to EGFR inhibitors
.
Sci Transl Med
2011
;
3
:
75ra26
.
22.
Kurppa
KJ
,
Liu
Y
,
To
C
,
Zhang
T
,
Fan
M
,
Vajdi
A
, et al
Treatment-induced tumor dormancy through YAP-mediated transcriptional reprogramming of the apoptotic pathway
.
Cancer Cell
2020
;
37
:
104
22
.
23.
Shreberk-Shaked
M
,
Oren
M
. 
New insights into YAP/TAZ nucleo-cytoplasmic shuttling: new cancer therapeutic opportunities?
Mol Oncol
2019
;
13
:
1335
41
.
24.
Chaib
I
,
Karachaliou
N
,
Pilotto
S
,
Codony Servat
J
,
Cai
X
,
Li
X
, et al
Co-activation of STAT3 and YES-associated protein 1 (YAP1) pathway in EGFR-mutant NSCLC
.
J Natl Cancer Inst
2017
;
109
:
djx014
.
25.
Varelas
X
. 
The Hippo pathway effectors TAZ and YAP in development, homeostasis and disease
.
Development
2014
;
141
:
1614
26
.
26.
Zhao
B
,
Ye
X
,
Yu
J
,
Li
L
,
Li
W
,
Li
S
, et al
TEAD mediates YAP-dependent gene induction and growth control
.
Genes Dev
2008
;
22
:
1962
71
.
27.
Sun
Y
,
Campisi
J
,
Higano
C
,
Beer
TM
,
Porter
P
,
Coleman
I
, et al
Treatment-induced damage to the tumor microenvironment promotes prostate cancer therapy resistance through WNT16B
.
Nat Med
2012
;
18
:
1359
68
.
28.
Obenauf
AC
,
Zou
Y
,
Ji
AL
,
Vanharanta
S
,
Shu
W
,
Shi
H
, et al
Therapy-induced tumour secretomes promote resistance and tumour progression
.
Nature
2015
;
520
:
368
72
.
29.
Boumahdi
S
,
de Sauvage
FJ
. 
The great escape: tumour cell plasticity in resistance to targeted therapy
.
Nat Rev Drug Discov
2020
;
19
:
39
56
.
30.
Eskilsson
E
,
Røsland
GV
,
Solecki
G
,
Wang
Q
,
Harter
PN
,
Graziani
G
, et al
EGFR heterogeneity and implications for therapeutic intervention in glioblastoma
.
Neuro Oncol
2018
;
20
:
743
52
.
31.
Heffron
TP
. 
Challenges of developing small-molecule kinase inhibitors for brain tumors and the need for emphasis on free drug levels
.
Neuro Oncol
2018
;
20
:
307
12
.
32.
Elmeliegy
MA
,
Carcaboso
AM
,
Tagen
M
,
Bai
F
,
Stewart
CF
. 
Role of ATP-binding cassette and solute carrier transporters in erlotinib CNS penetration and intracellular accumulation
.
Clin Cancer Res
2011
;
17
:
89
99
.
33.
Snuderl
M
,
Fazlollahi
L
,
Le
LP
,
Nitta
M
,
Zhelyazkova
BH
,
Davidson
CJ
, et al
Mosaic amplification of multiple receptor tyrosine kinase genes in glioblastoma
.
Cancer Cell
2011
;
20
:
810
7
.
34.
Szerlip
NJ
,
Pedraza
A
,
Chakravarty
D
,
Azim
M
,
McGuire
J
,
Fang
Y
, et al
Intratumoral heterogeneity of receptor tyrosine kinases EGFR and PDGFRA amplification in glioblastoma defines subpopulations with distinct growth factor response
.
Proc Natl Acad Sci U S A
2012
;
109
:
3041
6
.
35.
Wykosky
J
,
Hu
J
,
Gomez
GG
,
Taylor
T
,
Villa
GR
,
Pizzo
D
, et al
A urokinase receptor-Bim signaling axis emerges during EGFR inhibitor resistance in mutant EGFR glioblastoma
.
Cancer Res
2015
;
75
:
394
404
.
36.
Guo
G
,
Gong
K
,
Ali
S
,
Ali
N
,
Shallwani
S
,
Hatanpaa
KJ
, et al
A TNF–JNK–Axl–ERK signaling axis mediates primary resistance to EGFR inhibition in glioblastoma
.
Nat Neurosci
2017
;
20
:
1074
84
.
37.
Weihua
Z
,
Tsan
R
,
Huang
WC
,
Wu
Q
,
Chiu
CH
,
Fidler
IJ
, et al
Survival of cancer cells is maintained by EGFR independent of its kinase activity
.
Cancer Cell
2008
;
13
:
385
93
.
38.
Tan
X
,
Thapa
N
,
Sun
Y
,
Anderson
RA
. 
A kinase-independent role for EGF receptor in autophagy initiation
.
Cell
2015
;
160
:
145
60
.
39.
Thiery
JP
,
Acloque
H
,
Huang
RY
,
Nieto
MA
. 
Epithelial–mesenchymal transitions in development and disease
.
Cell
2009
;
139
:
871
90
.
40.
Tata
PR
,
Rajagopal
J
. 
Cellular plasticity: 1712 to the present day
.
Curr Opin Cell Biol
2016
;
43
:
46
54
.
41.
Shibue
T
,
Weinberg
RA
. 
EMT, CSCs, and drug resistance: the mechanistic link and clinical implications
.
Nat Rev Clin Oncol
2017
;
14
:
611
29
.
42.
Zhang
Z
,
Lee
JC
,
Lin
L
,
Olivas
V
,
Au
V
,
LaFramboise
T
, et al
Activation of the AXL kinase causes resistance to EGFR-targeted therapy in lung cancer
.
Nat Genet
2012
;
44
:
852
60
.
43.
Shao
DD
,
Xue
W
,
Krall
EB
,
Bhutkar
A
,
Piccioni
F
,
Wang
X
, et al
KRAS and YAP1 converge to regulate EMT and tumor survival
.
Cell
2014
;
158
:
171
84
.
44.
Zheng
X
,
Carstens
JL
,
Kim
J
,
Scheible
M
,
Kaye
J
,
Sugimoto
H
, et al
Epithelial-to-mesenchymal transition is dispensable for metastasis but induces chemoresistance in pancreatic cancer
.
Nature
2015
;
527
:
525
30
.
45.
Fischer
KR
,
Durrans
A
,
Lee
S
,
Sheng
J
,
Li
F
,
Wong
ST
, et al
Epithelial-to-mesenchymal transition is not required for lung metastasis but contributes to chemoresistance
.
Nature
2015
;
527
:
472
6
.
46.
Phillips
HS
,
Kharbanda
S
,
Chen
R
,
Forrest
WF
,
Soriano
RH
,
Wu
TD
, et al
Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis
.
Cancer Cell
2006
;
9
:
157
73
.
47.
Zou
M
,
Toivanen
R
,
Mitrofanova
A
,
Floch
N
,
Hayati
S
,
Sun
Y
, et al
Transdifferentiation as a mechanism of treatment resistance in a mouse model of castration-resistant prostate cancer
.
Cancer Discov
2017
;
7
:
736
49
.
48.
Mu
P
,
Zhang
Z
,
Benelli
M
,
Karthaus
WR
,
Hoover
E
,
Chen
CC
, et al
SOX2 promotes lineage plasticity and antiandrogen resistance in TP53- and RB1-deficient prostate cancer
.
Science
2017
;
355
:
84
8
.
49.
Huang
S
,
Hölzel
M
,
Knijnenburg
T
,
Schlicker
A
,
Roepman
P
,
McDermott
U
, et al
MED12 controls the response to multiple cancer drugs through regulation of TGF-β receptor signaling
.
Cell
2012
;
151
:
937
50
.
50.
Oshimori
N
,
Oristian
D
,
Fuchs
E
. 
TGF-β promotes heterogeneity and drug resistance in squamous cell carcinoma
.
Cell
2015
;
160
:
963
76
.