ERK-Mediated Loss of miR-199a-3p and Induction of EGR1 Act as a “Toggle Switch” of GBM Cell Dedifferentiation into NANOG- and OCT4-Positive Cells

The cancer stem cell concept sustains the idea of a hierarchical organization of tumor cells where only a subpopulation of stem-like cells, the tumor propagating cells (TPC), shares with normal adult stem cells the property to self-renew and to differentiate (1–3). Due to these stem-like properties, the TPCs are the type of tumor cell responsible for tumor initiation, maintenance, resistance, and recurrences (1–3). Glioblastoma (GBM), the most common form of primitive brain tumors in adults (4), exemplifies this concept and harbors such tumor cells referred as glioma stem–like cells (GSC; refs. 5–8). These cells play important contributions in intratumor heterogeneity, architecture, and homeostasis, by producing various tumor differentiated cells including endothelial- and pericyte-like cells (9, 10). As a result, GSC and tumor differentiated cells coexist in the tumor microenvironment, either intermingled or segregated as functionally divergent tumor territories (11), and contribute to tumor maintenance, growth, and propagation in function of their differentiation status and their ability to resist to conventional treatments. The GBM ability to resist to conventional treatments and to recur in nodular pattern most often localized at the primary site of the tumor strongly suggests the presence of persistent tumor cells endowed with clonal amplification properties. Therefore, the question of the maintenance and the enrichment of the glioma stem–like/progenitor cell compartment becomes crucial for prospective therapeutic applications against TPCs.

GSCs have long been thought to follow a unidirectional path of differentiation, leading to proliferating committed progenitors and at the end, fully differentiated cells (3). However, results accumulated during the last decades challenge this view. GBM cells have the potential to dedifferentiate into stem-like cells when cultured in EGF/bFGF-supplemented defined medium or exposed to various stresses such as hypoxia, acidification, or genotoxic stress (12–14). These data indicate that the GSC compartment may be enriched within tumors, by the dedifferentiation of GBM cells, likely, in response to clues from the microenvironment. Therefore, in terms of therapeutic targeting, it is of importance to confirm that GSC could also correspond to a functional state that may be acquired by many tumor cells when placed in the adequate environment.

Importantly, most of the studies showing GBM cell reprogramming into stem-like cells were performed using either tumorspheres differentiated in serum medium (13, 14) or a CD133⁻ cell population maintained in serum culture prior to be exposed to reprogramming conditions (15). It remains to be determined whether or not differentiated GBM cells freshly dissociate from tumors and which cells that have never been cultured in serum media have the intrinsic capacity to reprogram into GSC.

In the present study, we demonstrate that GBM cells originally negative for pluripotency marker expression, freshly dissociated from resected human GBM positive for EGFR expression, have an intrinsic property to spontaneously acquire OCT4 and NANOG expression as well as long-term clonal expansion capability and aggressiveness, when exposed to tumor environment enriched in EGF and favoring GSC maintenance and amplification. We confirm that GSC obtained from in vitro dedifferentiation maintain this reprogramming capability in culture as previously described (14). This mechanism is dependent on ERK-mediated repression of miR-199a-3p and induction of EGR1, constituting a “toggle switch” that orientates GBM cells toward a stem-like state. Our findings constitute one more milestone leading to a better understanding of the control of tumor cell fates and behaviors that may influence tumor functional heterogeneity and aggressiveness.

We have obtained written-informed consent from all the patients to the use of their tumor for research purposes. The studies were conducted in accordance with recognized ethical guidelines (e.g., Declaration of Helsinki, CIOMS, Belmont Report, U.S. Common Rule) and standard of our institution (CHU de Nice and INSERM) and were approved by our Institutional Review Board (INSERM scientific commission).

Patient-derived cells were isolated from surgical resection of human primary GBM provided by the department of neurosurgery of the University Hospital of Nice (GB1, GB2, GB3, GB5, GB9, GB15, and GB16). TG1 cells were provided by Hervé Chneiweiss, University of Pierre and Marie Curie, Paris. The dissociation of tumor cells was performed as described elsewhere (16). The resulting primary cultures were used immediately for experiments. Neurospheres were grown in NS34+ medium containing EGF and bFGF (DMEM-F12 1/1 ratio, 10 mmol/L glutamine, 10 mmol/L Hepes, 0.025% Sodium bicarbonate, N2, G5, and B27). Defined medium for dedifferentiation experiments was only supplemented by EGF or when indicated FGF or PDGFA. Differentiation experiments were performed in serum medium (DMEM-F12, glutamine 10 mmol/L, Hepes 10 mmol/L, sodium bicarbonate 0.025%, and FCS 0.5%). Mycoplasma was tested by PCR using primers that detect cell-culture–relevant Mycoplasma strains (forward primer: cttcatcgactttcaga; reverse primer: acaccatgggagctggtaat). When indicated, the cells were treated with 5 μg/mL cetuximab (EGFR blocking antibody, kindly provided by Centre Antoine Lacassagne) or gefitinib (2 μmol/L), lapatinib (20 μmol/L; inhibitor of EGFR tyrosine kinase), AG1296 (5 μmol/L, PDGFR inhibitor), BGJ398 (FGFR inhibitor), or U0126 (15 μmol/L) and ulixertinib (2 μmol/L), two specific inhibitors of ERKs. When indicated, cells were transfected using the Xfect Protein Transfection Reagent (ref 631323 Takara) with Recombinant Human Active ERK1 Protein (CF, ref 1879-KS R&D) and Recombinant Human Active ERK2 Protein (CF, ref 1230-KS R&D). The dedifferentiation was triggered with defined medium supplemented with EGF (10–20 ng/mL). When indicated PDGFA and bFGF were tested in place of EGF.

Following tumor dissociation, the cell suspensions were treated with a red cell lysis buffer kit and with Dynabeads CD31 Endothelial Cell, Dynabeads CD45, and Dynabeads CD15 in order to remove red cells, endothelial, leucocytes, and myeloid cells.

Patient-derived cell authentication is conducted by observing cellular morphology at low and high densities, analysis of their growth curve, detection of Mycoplasma, and comparative genomic hybridization analysis.

Cell death was assessed by trypan blue staining. At least 500 cells were counted in three independent experiments.

Proteins were extracted (buffer: 50 mmol/L Tris-Hcl, ph7.6, 150 mmol/L NaCl, 5 mmol/L EDTA, and 1% NP40). After migration, the proteins were transferred to Immobilon P membranes (Millipore Corporation) and probed with antibodies to pERK1/2 (sc16982; Santa Cruz Biotechnology) or mouse antibody to ERK1 (sc94; Santa Cruz Biotechnology).

GB1 and GB8 cells (2 × 10⁴) were resuspended in 5 μL of Hanks balanced salt solution (Invitrogen) and stereotaxically implanted into the right striatum of male NOD.CB17-Prkdcscid/NCrHsd mice (Harlan, [email protected]). Mice were sacrificed upon suffering sign appearance. The brains were dissected, sliced, and analyzed by IHC with a human vimentin antibody. TG1 and GB1 were engineered in order to stably express a luciferase reporter gene (TG1-luc and GB1-luc) to allow tumor detection in living animals. Three groups [undifferentiated (ND), differentiated (Diff), or EGF-dedifferentiated (De-diff)] of 5 mice [NOD.CB17-Prkdcscid/NCrHsd mice (Harlan, [email protected])] were orthotopically injected with TG1-luc (2 × 10⁵). Cell survival and tumor growth were monitored and quantified in the living animals up to 90 days after grafting using the IVIS Lumina system (Caliper Life Sciences). GB1-luc cells were engineered to stably express miR-199a-3p (miR-199, n = 4) or sh-EGR1 (sh-EGR1, n = 4) or a nonrelevant small noncoding RNA (CTL, n = 4). Note that 2 × 10⁴ cells were injected in the brain of nude mice (Charles River). Three groups of 5 nude mice (Charles River) were orthotopically injected with 2 × 10⁴ GB1-luc cells. When indicated the mice were treated or not (i.p. injection) with cetuximab (50 mg/kg) or U0126 25 μg/kg every other day. For each experiment, tumor growth was followed after grafting using the IVIS Lumina system (Caliper Life Sciences). All animal procedures were approved by the local ethics committee (CIEPAL; PEA 14, PEA 510).

Living CD133⁺ and CD133⁻ populations were separated from fresh tumor, using an anti–CD133-PE (1/100°; Miltenyi Biotec) and a cell sorter (BDFACS ARIA). Alternatively, cell populations were separated using CD133 microbeads (Miltenyi Biotec). Cells were then fixed with 4% paraformaldehyde for 15 minutes at room temperature. For the FACS analysis, the cells were permeabilized, blocked, and hybridized in DMEM containing 10% FCS and 0.5% TritonX100 with the following antibodies: anti-OCT4 (1/100°; Ab19857, Abcam), anti-NANOG (1/250°; Ab21624, Abcam), or anti–CD133-PE (1/100°; Miltenyi Biotec). Cell staining was analyzed on FACS Calibur (Becton Dickinson). NANOG- and OCT4-positive cells were also counted and quantified using fluorescent microscopy (TIE NIKON).

Cells were seeded on polylysine-coated glass slides in EGF medium or serum medium. At indicated times, cells were fixed with methanol for 5 minutes at −20°C. Blocking and hybridization were realized in PBS containing 10% FCS and 0.1% Triton X100 with the following primary antibodies: anti-SOX2 at 1/200 (AB15766, Upstate), anti-Oct4 at 1/50° (H-134; sc9081, Santa Cruz Biotechnology), anti-Nanog at 5 μg/mL (AF1997; R&D Systems), anti-GFAP (1/250°; 2203PGF, R&D Systems), anti-phosphoactive-ERK (V803A, Promega) at 1/100°, anti-CD133 at 1 μg/mL (Ab19898, Abcam), EGR1 (sc-110, Santa Cruz Biotechnology; dilution 1:50), and phosphorylated histone 3 serine 10 (antibody 5176–100, Abcam; dilution 1:100). All patients were informed and signed their consent for the use of all human samples. Brain tumor samples were sliced and subjected to immunofluorescence analysis with anti-NANOG antibody (AF1997; R&D Systems).

Cells were seeded at 1 cell/well in 96-well plates. The number of wells containing tumorspheres was counted 3 weeks after seeding. Experiments were repeated 3 independent times.

All patients were informed and signed their consent for the use of all human samples. Immunolabeling was performed using a DAKO automat with the following primary antibodies: anti-EGR1 (sc-110, Santa Cruz Biotechnology) diluted 1:100, anti–MIB-1 (IR62661, DAKO, catalog no. IR62661) diluted 1:1,000. Deparaffinization, rehydration, and antigen retrieval were performed using the pretreatment module PTlink (Dako). Five μm sections of formalin-fixed paraffin-embedded tissue in tumor microarrays were tested for the presence of EGR1, using a Benchmark Ventana autostainer (Ventana Medical Systems SA). Immunostaining was independently scored by a pathologist (F. Burel-Vandenbos) and a researcher (T. Virolle).

Cells were seeded in 6-well plates at a density of 0.5 × 10⁶ cells/well and transiently transfected by using Lipofectamine 2000 reagent (Life Technologies) according to the manufacturer's instructions. Briefly, a 10 nmol/L concentration of Silencer RNAi (Life Technologies, si-EGR1-1, si-EGR1-2, si-EGFR) was diluted in 50 μL of Opti-MEM medium, and 1 μL of Lipofectamine was diluted in 50 μL of Opti-MEM medium. After 5 minutes of incubation, the diluted Silencer RNAi and the diluted Lipofectamine were combined, mixed gently, and incubated for 20 minutes at room temperature prior to adding the complexes to cells. Cells were lysed, and RNA or proteins were extracted for experiments. Similar protocol was used for the transfection of synthetic miR-199a, anti–miR-199a-3p, and control miRNA. For stable transfection in GSC, miR-199a-3p sequences or sh-EGR1 were cloned downstream H1 polIII promoter in 2K7 lentiviral vector, as previously described (14, 16). GSC were transduced and then selected in blasticidin. For orthotopic xenograft experiments, the cells were transduced with a luciferase vector. Sequences of miRNA, and shRNA are available upon request.

EGF-dedifferentiated GB1 cells were treated or not with 15 μmol/L of U0126. Forty-eight hours following EGF stimulation, cells were lysed and total RNA was extracted. Samples were then used for transcriptomic studies using pan genomic human Agylent chips (Agilent-014850:Whole Human Genome Microarray 4 × 44K). The miRnome profiling was performed by QRT-PCR using TLDA arrays (Applied Biosystems).

To monitor the behavior of GBM cells when placed for the first time in EGF-enriched medium, we dissociated freshly resected EGFR-amplified and positive (Supplementary Fig. S1) GBM samples (GB1, GB2, GB3, GB5, and GB9) and plated the cell suspension in serum-free medium supplemented with EGF only (10 ng/mL). The cells exhibited various shapes immediately after seeding, going from simple round small soma to large soma with multiple extensions (Fig. 1A). Progressive formation of cell aggregates was observed over the following 48 hours, culminating in the formation of neurosphere-like structures (Fig. 1A; Supplementary Fig. S2). Cell proliferation rate was evaluated over 15 days of culture in EGF-supplemented defined medium using newly formed floating spheroids, which were harvested, dissociated, and seeded at a density of 10,000 cells/well in 24-well plates. The cells exhibited a slow proliferation rate with a doubling time of about 10 days (Fig. 1A). This result indicates that cell aggregation prevails over cell proliferation in cell spheroids formation. Remarkably, GBM cells freshly dissociated from the tumor were mostly negative for stemness markers and strongly positive for glial fibrillary acidic protein (GFAP) considered as a marker of glial differentiation (Fig. 1B, time-point 4 hours). On the opposite, the spheroids formed after 48 hours of culture in presence of EGF and contained numerous cells strongly immunoreactive for the pluripotency markers SOX2, OCT4, NANOG, and the adult neural stem cell marker CD133 (Fig. 1B; Supplementary Fig. S3 for magnified pictures) and only rare GFAP-immunoreactive cells. Conversely, the adherent cells surrounding the spheres did not express adult stem cell and pluripotency markers but exhibited strong GFAP expression (Fig. 1B, bottom plots). Interestingly, the cells that compose the resulting spheroids displayed long-term and stable clonal expansion ability when diluted at a single-cell level (Fig. 1C; Supplementary Fig. S4A–S4C).

We determined also the respective tumorigenic potential of spheroid-derived cells (48-hour EGF treatment) and of the original GBM cell suspension (4-hour EGF treatment). Orthotopic xenografts of 25,000 of each cell types were performed in NOD/SCID mice. Mice grafted with the OCT4/NANOG-positive cell population died within 2 months, whereas the mice implanted with freshly dissociated GBM cells survived beyond 6 months (Fig. 1D). Similar results were obtained with GB9 tumor cells (Supplementary Fig. S4D).

We then sought to determine whether the conversion into tumor cells positive for OCT4 and NANOG requires cell division. Using GB1 spheroids, we observed that loss of stemness properties due to serum-mediated differentiation (i.e., sphere formation, pluripotency marker expression, resistance to temozolomide, and clonal and tumorigenic properties; refs. 13, 14) was fully reverted after switching cells from differentiation medium to defined medium supplemented with as low as 10 ng/mL EGF (Fig. 2A and B; Supplementary Fig. S5A–S5D and Supplementary Fig. S6A), confirming that cell dedifferentiation ability still occurs following in vitro differentiation (14). Upon differentiation, GB1 cells underwent a drop in mitosis (H3Pser10), CYCLIN A production, and pRB phosphorylation, exemplifying a cell-cycle arrest (Fig. 2C and D). Importantly, the cells expressed pluripotency markers at 2 days following the start of dedifferentiation (Fig. 2A–C), 1 day prior resuming cell cycle, which was indicated by pRB hyperphosphorylation and a strong CYCLIN A production (Fig. 2C and D). These data indicate that reacquisition of stemness markers in these cells under these experimental conditions is not a feature of the progeny of the cell submitted to the dedifferentiation process, but directly affects the original differentiated cell.

We then determined whether mobilization of the EGFR pathway upon exposure to EGF-containing medium was instrumental in the dedifferentiation process using the monoclonal antibody cetuximab, which targets EGFR-binding domain thus inhibiting its activation. Immunoblotting assays showed a drastic increase in the activated, phosphorylated form of EGFR 15 minutes after transfer of the differentiated GSC in serum-free medium containing EGF (Fig. 2E). As expected, cetuximab prevented EGFR phosphorylation (Fig. 2E). Cetuximab blockade of EGFR activation was also accompanied with inhibition of ERK phosphorylation, one of the protein kinases transducing the EGFR signal (Fig. 2E). Cetuximab blockade of EGFR pathway suppressed differentiated cell ability to form spheroids (Fig. 2F) and to re-express pluripotency markers (Fig. 2G and H). The use of other EGFR inhibitors, lapatinib and gefitinib, or specific siRNA that efficiently knocked down EGFR has provided identical results (Supplementary Fig. S6B–S6F). According to EGFR-mediated activation of ERK (Fig. 2E), the use of U0126 or ulixertinib, which efficiently blocked ERK activation, exerted the same inhibitory effects on EGFR-driven dedifferentiation (repression of OCT4 and NANOG) as cetuximab did (Supplementary Table S1; Supplementary Fig. S7A–S7C). On the other hand, the rescue of ERK activation by transfecting recombinant p-ERK proteins rescued NANOG expression (Supplementary Fig. S7D). These results showed that ERK was likely one of the main downstream kinase of EGFR pathway for acquisition and maintenance of GSC properties. Accordingly, i.p. injection of Cetuximab or U0126 in nude mice, following GSC orthotopic xenografts, prevented tumor formation and growth (Supplementary Fig. S7E).

In order to further strengthen the molecular relevance of EGFR/ERK axis in GBM, we have identified through a transcriptomic analysis genes that are upregulated under the control of p-ERK during EGF-mediated dedifferentiation (Supplementary Fig. S8A and S8B). Among the 189 genes identified, we found three clusters of genes functionally linked (string analysis: http://string-db.org/cgi/input.pl; Supplementary Fig. S8C; Supplementary Tables S2 and S3): One major cluster and two smaller ones, which contained respectively 62.4%, 10.6%, and 10% of activated genes (Fig. S8C; Supplementary Tables S2 and S3). Interestingly, according to the “Glioblastoma Bio Discovery Portal” (https://gbm-biodp.nci.nih.gov), patients having higher expression level of each of these clusters of genes displayed significant lower survival (Supplementary Fig. S9).

EGFR dependency for GBM cell reprogramming was further verified using cells freshly isolated from additional GBM surgical resections (GB15 and GB16, Fig. 3A–E; Supplementary Fig. S10A–S10E). We used CD133 expression to isolate a tumor cell population impoverished in NANOG- and OCT4-positive cells (Fig. 3A). As shown by FACS analysis, the GB15 and GB16 CD133⁻ population contained 1.9% to 7.3% NANOG- and 3.2% to 4.1% OCT4-positive cells, respectively (Fig. 3B; Supplementary Fig. S10A–S10C). For comparison, FACS analysis of the CD133⁺ population in GB15 revealed 85% NANOG and 86% OCT4-positive cells (Fig. 3B). In order to separate the OCT4- and NANOG-positive cells from the negative one, the CD133⁻ cells were then seeded at 1 cell/well in defined medium containing EGF supplemented or not with cetuximab (Fig. 3C; Supplementary Fig. S10D). An average of 64% and 80% of the wells, respectively for GB15 and GB16, displayed spheroid formation, positive for OCT4 and NANOG (Fig. 3D and E; Supplementary Fig. S10D). Although only a minority of the CD133⁻ cells was NANOG and OCT4 positive before incubation in EGF medium, FACS analysis performed on the cell population generated by pooling the newly formed spheroids showed that the vast majority of cells became positive for these pluripotency markers (Fig. 3D and E; Supplementary Fig. S10E). Interestingly, none spheroids were formed in the presence of cetuximab (Fig. 3C; Supplementary Fig. S10D) confirming the importance of the EGFR pathway.

In addition to EGF, and given that multiple receptor tyrosine kinases activate ERKs and can contribute to stemness, we have tested bFGF and PDGFA ability to trigger cell dedifferentiation. Although bFGF failed to induce dedifferentiation, PDGFA, albeit less efficient than EGF because it was used 5 times more concentrated, triggered the formation of NANOG-positive spheroids (Supplementary Fig. S11A–S11C). Accordingly, although no effects were observed when the cells were treated with FGFR inhibitor (BGJ398) in the presence of PDGFA, the use of PDGFR inhibitor (AG1296) or ERK inhibition abrogated PDGFA-dependent dedifferentiation process (Supplementary Fig. S11A–S11C). These results showed that EGFR and PDGFR are both capable to trigger cell dedifferentiation through ERK activation. Interestingly, we observed in self-renewing GSC that constitutive EGFR/ERK activation also contributed in maintaining their clonal proliferation (Supplementary Fig. S11D).

To further dissect out the molecular pathways controlling the switch between stemness loss and acquisition by GBM cells, we focused on the transcription factor EGR1 that we previously found to promote GSC proliferation and self-renewal (17). EGR1 expression was monitored over the differentiation and dedifferentiation processes, using two patient-derived GSC cultures (TG1 and GB1). The cells were serum-differentiated for 4 days, prior to be further cultured in serum-free medium containing EGF. As expected, EGR1 expression was strongly reduced in GSC after 4 days of differentiation and upregulated upon further dedifferentiation, as shown by immunoblotting and immunofluorescence (Fig. 4A and B; Supplementary Fig. S12A). Blockade of EGFR or inhibition of ERK phosphorylation by U0126 or ulixertinib prevented EGR1 re-expression upon GBM cell dedifferentiation (Fig. 4A and B; Supplementary Fig. S12A and S12B). Interestingly, EGR1 protein expression occurred before the formation of NANOG-positive spheroids, strongly suggesting EGR1 as one of the triggers of dedifferentiation process (Fig. 4C and D). Remarkably, the drop in EGR1 protein occurring upon serum differentiation (Fig. 4B) was accompanied with a strong increase of EGR1 mRNA level (Fig. 4E). Conversely, the increase in EGR1 protein levels observed upon dedifferentiation was accompanied with a decrease in EGR1 mRNA levels, thus becoming comparable with the level observed in self-renewing GSC (Fig. 4E), which was strongly decreased upon ERK inhibition and restored with the rescue of ERK activation (Fig. 4E). This opposition between protein and mRNA levels suggests a possible miRNA-mediated posttranscriptional regulation of EGR1 protein production. To identify miRNA that may be involved in the regulation of EGR1 expression, we used qPCR and TAQMAN probes in a microfluidic device (TLDA). We thus measured the expression of about 600 human miRNA in differentiated and freshly dedifferentiated GBM cells, in the presence and absence of the ERK inhibitor U0126. Around 10% of the 600 human miRNAs measured were either down- or upregulated during dedifferentiation (Fig. 5A). Among the 10% that were downregulated during dedifferentiation, we focused on the 42% of the miRNA, the downregulation of which was prevented by ERK inhibition (Fig. 5A and B). One of them, miR-199a-3p, recognized a binding site very well conserved in the 3′untranslated region (UTR) of EGR1 mRNA (Supplementary Fig. S13), suggesting its possible involvement in the blockade of EGR1 production in differentiated GBM cells. Although EGFR and ERK inhibition prevented miR-199a-3p repression during dedifferentiation, p-ERK rescue restored its repression (Fig. 5C).

To test a possible regulation of EGR1 production at protein level by the miR-199a-3p, we took advantage of the inverse correlation between expression of EGR1 protein and miR-199a-3p in two different serum-differentiated patient-derived GSC (TG1 and GB1). Comparison of EGR1 mRNA and miR-199a-3p expression patterns in self-renewing GB1 or TG1, upon differentiation and at 24 hours and 48 hours of dedifferentiation, confirmed both their coexpression and the repression of miR-199a-3p upon dedifferentiation (Fig. 5D and E). To further reveal the importance of miR-199a-3p repression for EGR1 protein induction upon dedifferentiation, we transfected, in GB1 differentiated cells, either a scramble synthetic miRNA (GB1-miR-CTL) or miR-199a-3p synthetic sequences (GB1-miR-199a-3p). As expected, GB1-miR-CTL cells regulated miR-199a-3p expression like naïve GB1 cells upon differentiation and dedifferentiation treatment (Supplementary Fig. S14A). Conversely, GB1-miR-199a-3p cells, which harbored miR-199a-3p levels above to that determined in naïve and control differentiated GB1 cells (Supplementary Fig. S14A), did not decrease miR-199a-3p expression upon dedifferentiation. Although GB1-miR-CTL cells stimulated EGR1 expression at 24 and 48 hours of dedifferentiation, the GB1-miR199a-3p, unable to decrease miR-199a-3p levels, did not display any stimulation of EGR1 protein in similar conditions (Fig. 5F). In addition, transfection of naïve GB1 cells with miR-199a-3p inhibitors (antagomir, am-199a3p) prevented miR-199a-3p upregulation, induced by cell differentiation (Supplementary Fig. S14B). In a coherent manner, we observed maintenance of EGR1 protein expression in GB1 cells expressing miR-199a-3p antagomir upon the differentiation process (Fig. 5G). Taken together, these data demonstrate that miR-199a-3p efficiently contributes to EGR1 repression in differentiated GSC and that the decrease of miR-199a-3p level is required for EGR1 re-expression upon dedifferentiation. On the other hand, siRNA-mediated invalidation of EGR1 during dedifferentiation (Supplementary Fig. S14C) did not display any effects at 15 hours following EGF addition neither on ERK activation nor on miR-199a-3p level, which were comparable with the control cells. However, at 48 and 72 hours following EGF treatment, EGR1 invalidation repressed ERK activation, which was accompanied by an increase of miR-199a-3p expression (Fig. 5H and I). These results show that EGR1 can influence miR-199a-3p level in GSC through the regulation of ERK activity.

To further characterize EGR1 contribution to GBM cell dedifferentiation, we triggered the dedifferentiation of serum-differentiated TG1 and GB1 transfected with either siRNA control or siRNA targeting EGR1 (siEGR1-1 or siEGR1-2; ref. 17). EGR1 knockdown was assessed by immunoblotting (Supplementary Fig. S14C). Spheroid formation and enrichment in OCT4- and NANOG-positive cells following dedifferentiation of TG1 and GB1 cells expressing siRNA control was verified by immunocytochemistry and cell counting, as compared with their differentiated counterparts (Fig. 6A, right and left). On the opposite, EGR1 knockdown prevented both the formation of spheroid and OCT4 and NANOG expression, thereby strongly altering cell dedifferentiation (Fig. 6A). In addition, EGR1-deficient cells did not resume their cell-cycle progression contrary to control cells (Fig. 6B). Furthermore, ectopic expression of miR-199a-3p in GB1 (GB1-miR-199a-3p) and TG1 (TG1-miR-199a-3p), which maintained constant artificial miR-199a-3p expression and prevented EGR1 expression (Fig. 5F; Supplementary Fig. S14A) upon dedifferentiation, behaved similarly than EGR1-deficient cells and lost the ability to form spheroids and to re-express OCT4 and NANOG proteins (Fig. 6A). Accordingly, GB1 cells that were engineered to promote stable and constitutive miR-199a expression (GB1-miR-199a) did not form regular spheroids and failed to express NANOG expression (Supplementary Fig. S14D and S14E). These results demonstrate that miR-199a-3p repression, which is concomitant with EGR1 re-expression, at the onset of dedifferentiation, is mandatory for induction of pluripotency marker expression (Fig. 6C) and subsequent cell-cycle progression. Accordingly, GB1-miR-199a cells as well as GSC stably deficient for EGR1 expression (GB1-sh-EGR1; ref. 17; Supplementary Fig. S14F) failed to develop tumors following orthotopic xenograft in nude mice, whereas control GSC, stably expressing a scramble small noncoding RNA, developed brain tumors (Fig. 6D).

We previously reported the coexistence of functionally divergent territories within the GBM-dense tumor (11). These tumor territories were exemplified by either an enrichment in mitotic, OLIG2- and NANOG-positive cells, or conversely by an enrichment in nonmitotic cells negative for OLIG2 and NANOG expression (Fig. 7A–C). We previously reported a preferential nuclear detection of EGR1-immunoreactive cells in mitotic OLIG2-positive cells (17). Here, we quantified EGR1 immunoreactive cells in human GBM tumors using a collection of 20 samples representative of mitotic and nonmitotic territories. Nuclear expression of EGR1 was detected only in mitotic territories and was strongly correlated to OLIG2 and NANOG expression (Fig. 7A–E; Supplementary Fig. S15). These results confirm our previous results obtained from GBM samples representative of the whole tumors (17). Expression of miR-199a-3p was assessed by qPCR in the corresponding block mirrors that were snap frozen and used for miRNA extraction. Conversely to EGR1, the miR-199a-3p strongest level was detected in nonmitotic territories (Fig. 7A and C). Indeed, correlation analysis from the 20 GBM samples clearly showed a significant inverse correlation between miR-199a-3p level and EGR1, OLIG2, NANOG, and Ki67 protein expression (Fig. 7F–I). These results demonstrate that miR-199a-3p expression is expressed within human GBM tumors with a pattern coherent with its repressive action on EGR1 expression.

Our study demonstrates that GBM cells, negative for stemness and pluripotency markers and freshly isolated from surgical resections, have the potential to undergo an efficient spontaneous conversion into aggressive tumor cells expressing pluripotency markers such as OCT4 and NANOG when placed in a defined environment enriched in EGF or PDGFA. Such plastic properties of glial tumor cells may account for the maintenance of the GSC population and thus may likely contribute to GBM homeostasis as well as to the propagation and initiation of tumor foci at a distance. Our results further support the idea that stem-like properties in GBM are not restricted to original GSCs but may be a functional state that can be acquired by any tumor cell in response to changes in their environment. This paradigm contrasts with the unidirectional nature originally described in the cancer stem cell model (3, 18) and rather defines a bidirectional commitment, which allows tumor cells to undergo transition from a differentiated state to stemness without cell division. Conversion of differentiated cells into stem-like cells was first described in normal astrocytes and reported as sensitizer to cell transformation (19, 20). In human mammary epithelia, a subpopulation of basal-like cells is capable of spontaneously converting into stem-like cells (21). In that model, oncogenic transformation enhances cell dedifferentiation so that nonstem cancer cells give rise to cancer stem–like cells (21). This process was further associated with epithelial-to-mesenchymal transition (EMT), where a direct link between gain of epithelial stem-cell properties and EMT transcription factors such as SNAIL, TWIST, or activation of the Ras–MAPK pathway was clearly established (21–25). Genotoxic stress provided by anticancer treatments may promote dedifferentiation. Indeed, in the context of breast, head, and neck cancer and GBMs, compelling results show that radiation may trigger the reprogramming of differentiated cancer cells into cancer stem cells with re-expression of stemness-related genes such as Oct4, Nanog, and Klf4 (13, 26, 27), providing one explanation on the increased number of cancer stem cells after anticancer therapies. The promotion of a stem-like state may also be done in vitro by hypoxia (15, 28) in GBM cells cultured in differentiating conditions.

Upon the switch to the stem-like phenotype, we observed that acquisition of pluripotency markers such as OCT4 and NANOG occurs before cell-cycle resumption so that dedifferentiation process does not necessarily require any cell division. This important finding reveals that the dedifferentiation of GBM cells into stem-like cells relies on epigenetic changes and not necessarily on cell division leading to production of particular committed progenies. These results obtained experimentally in GSC, contrast with the mathematical model of the replicator dynamics of cancer stem cells previously described in breast cancer cells (29). This model based on the concept of symmetric or asymmetric division to produce committed progeny or to sustained CSC population hypothesized that dedifferentiation of partially differentiated tumor cells may occur through asymmetric division and, therefore, the production of a progeny committed to dedifferentiate.

From a mechanistic point of view, direct chromatin remodeling using HDAC inhibitors in breast cancer can lead to the dedifferentiation of ALDH-negative into ALDH-positive cells, mainly through induction of the WNT pathway (21, 30). Similarly, in GBM, expression of the chromatin regulator PRC2 may influence GSC plasticity by inducing their reprogramming after having been differentiated in serum culture (31). Gene regulators such as transcription factors and miRNA play also crucial role in cancer cell dedifferentiation and aggressiveness. Indeed, OCT4 overexpression in melanoma cells is sufficient to promote tumor cell dedifferentiation, increasing malignancy as well as NANOG and KLF4 expression (32). Similarly, forced expression of NANOG in HeLa cells led to increased invasion, resistance to chemotherapeutic agents, and tumorigenicity (33), which are features of CSCs. This NANOG-dependent process was also associated with increased expression of OCT4, SOX2, and KLF4, suggesting that expression of at least one of these pluripotency markers, originally described as core transcriptional circuitry in embryonic stem cells involved in pluripotency and self-renewal regulation (34), is a prerequisite for the conversion of tumor cells into aggressive stem-like cells. Supporting this information, engraftment of embryonic stem cells, which express high levels of NANOG and OCT4 in immunocompromised mice, leads to the formation of teratocarcinoma, whereas using normal adult stem cells, which do not express these markers, does not induce any tumor formation (35, 36).

The activation of EGFR, frequently overexpressed in a variety of tumors including glioma (37, 38), and more precisely the axis EGFR/ERK, is sufficient to spontaneously promote the conversion of GBM cells originally negative for OCT4 and NANOG into GSC positive for pluripotency markers. The dedifferentiation process is strongly altered when EGFR^WT activation is prevented by the inhibition of EGF binding. In this context, the mutant EGFRVIII, which is frequently expressed when the gene is amplified, do not compensate for EGFR^WT inactivation. This result supports reports describing that EGFRVIII activity is highly dependent on EGF-activated EGFR^WT, which heterodimerizes with the mutant form EGFRVIII (39, 40). This finding extends EGFR physiologic role beyond its mitogenic activity (41) originally described for the maintain of GSC self-renewal and proliferation (42) or for transit amplifying C cells (TAC) of adult brain subventricular zone, which, when exposed to EGF, stops the production of neurons and becomes highly proliferating and infiltrating (43). Therefore, according to their highly proliferating nature, which may be favorable for mutation accumulations, TAC dedifferentiation may account for CSC formation and clonal expansion (44). In a mechanistic point of view, EGFR activation obviously leads to ERK phosphorylation, which plays a crucial role in allowing proper GBM cell reprogramming (14, 38). Interestingly, the constitutive EGFR/ERK axis activation in self-renewing GSC, although much weaker as compared with the level observed at the onset of the reprogramming process, is also involved in maintaining GSC clonal amplification, showing the contribution of this signaling axis, both in stemness initiation and maintenance. However, EGFR/ERK axis likely cooperates, in this context, with other previously reported signaling pathways (7, 14), such as Notch, BMP, NF-κB, SHH, and Wnt. In this study, EGFR-mediated ERK activation is the trigger of a complex dedifferentiation mechanism inducing in the same time the upregulation of the transcription factor EGR1 and the downregulation of miR-199a-3p. Indeed, we show that the increase of EGR1 protein level in GBM cells is necessary to chronologically induce the expression of OCT4 and NANOG expression, the re-enter in cell-cycle progression, as well as the formation of spheroids when the cells are placed in EGF-enriched medium. However, the stimulation of EGR1 protein expression cannot occur in presence of miR-199a-3p, which efficiently blocks EGR1 protein production. Accordingly maintaining artificially miR-199a-3p in cells during dedifferentiation prevents EGR1 induction and consequently prevents also the expression of pluripotency markers and the capacity to form spheroids. This interplay between miR-199a-3p and EGR1 mRNA constitutes a “toggle switch” of GBM cell plasticity, which prevents any cell dedifferentiation as long as the miR-199a-3p remains expressed. The functional duality between EGR1 and miR-199a-3p is further confirmed by their inversely correlated expression pattern in mitotic and nonmitotic GBM samples. As already mentioned, both EGR1 stimulation and miR-199a-3p repression are under the control of EGFR/ERK activation, which gives to this axis a central role for triggering the dedifferentiation mechanism. In a therapeutic point of view, whether or not this dedifferentiation mechanism contributes along with previously described mechanisms (13), to ionizing radiation-induced GBM cell plasticity remains an important point to determine in order to contribute to the development of novel strategies targeting therapeutic resistance.

In conclusion, our results sustain the idea of a strong GBM cell plasticity under the control of EGR1/miR-199a-3p interplay. This confers to GBM cell a bidirectional commitment that may likely have an important contribution to the maintenance of GSC subpopulation and functional intratumor heterogeneity.

No potential conflicts of interest were disclosed.

Conception and design: L. Turchi, F. Burel-Vandenbos, T. Virolle

Development of methodology: L. Turchi, D.N. Debruyne, F. Burel-Vandenbos, T. Virolle

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Almairac, L. Turchi, N. Sakakini, S. Elkeurti, E. Gjernes, L. Bianchini, D. Fontaine, F. Burel-Vandenbos, T. Virolle

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Almairac, L. Turchi, N. Sakakini, D.N. Debruyne, E. Gjernes, H. Chneiweiss, P. Verrando, F. Burel-Vandenbos, T. Virolle

Writing, review, and/or revision of the manuscript: F. Almairac, H. Chneiweiss, F. Burel-Vandenbos, T. Virolle

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Turchi, E. Gjernes, B. Polo, P. Paquis, M.-P. Junier, T. Virolle

Study supervision: F. Burel-Vandenbos, T. Virolle

Others (scientific discussions): P. Verrando

This work was supported by grants from the Association Sauvons Laura, Association Dimitri Bessières, Agence Nationale pour la Recherche (ANR Jeunes Chercheurs, Jeunes Chercheuses, « GLIOMIRSTEM project »), Fondation de France, Association pour la Recherche sur le Cancer, INCA PLBIO, ITMO CANCER plan cancer, Association pour le développement de la Recherche sur les tumeurs urologiques, cérébrales et pulmonaires (ADeRTU), INSERM, CNRS, UNSA, UCA, Conseil départemental 06, Cancéropole PACA.

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.