Dedifferentiation of Glioma Cells to Glioma Stem-like Cells By Therapeutic Stress-induced HIF Signaling in the Recurrent GBM Model

Glioblastoma multiforme is the most common adult primary brain tumor and is universally lethal due to its high rate of recurrence (1). Despite aggressive therapeutic intervention, which consists of surgical resection followed by radio and chemotherapy, glioblastoma prognosis remains dismal with less than 10% of patients surviving longer than 5 years after treatment (2, 3). The poor prognosis of glioblastoma results from a high rate of disease recurrence as well as recurrent tumors, which are almost always more aggressive, infiltrative, and therapy-resistant than the original malignancy (4–7). To develop more effective treatments for glioblastoma, it is crucial to understand disease recurrence at the molecular level to develop an effective therapeutic strategy to prevent recurrence.

Recent models of tumorigenesis postulate that human malignancies arise from a rare subset of the cancer cells known as cancer stem cells (CSC), which possess enhanced abilities to self-renew, differentiate, and induce the formation of new tumors upon orthotopic implantation in mice (8, 9). It is believed that CSCs possess the inherent capacity to resist conventional therapy and as a result, they play important roles in driving disease recurrence (5, 10). In contrast to traditional models of hierarchical differentiation from the CSC to differentiated tumor cell populations, recent studies have shown that there exists a dynamic equilibrium between CSC populations and their lineage-committed counterparts (2, 9, 11–13). This equilibrium is regulated by the microenvironmental factors such as intratumoral hypoxia and pH that can influence the rate of tumor differentiation and the balance between asymmetric and symmetric cell division in the CSC compartment and is considered to be critical for disease progression as heterogeneous glioblastoma contain a small number of glioma stem cells (GSC) within a larger population of less tumorigenic differentiated tumor cells (14). Any shift in this equilibrium has the potential to influence clinical outcomes of specific tumors as such shifts may result in a larger number of therapy-resistant CSCs within the tumor that allow them to acquire more aggressive characteristics and to produce poorer prognoses in patients (13, 15). Our laboratory, along with others, has shown that therapeutic stress promotes cellular plasticity, enhancing the conversion of non-stem glioblastoma multiforme cells to highly infiltrative, tumor-initiating stem-like cells (16–18). These data argue against the unidirectional flow of cellular hierarchy, increasing the possibility that the fate of these cancer cells is rather a bidirectional, dynamic process (19, 20).

To understand how the bidirectional flow of cancer cells influences the stemness equilibrium in glioblastoma during anti-glioma chemotherapy and to elucidate the molecular mechanisms governing such equilibrium, we developed a chemo-induced glioblastoma recurrence model. A shift in the equilibrium towards a more stem-like state was observed in patient-derived glioblastoma multiforme tumors (PDX) post-therapy. To examine such conversion dynamics at the single-cell level GSC-specific reporter systems using promoter region of multiple GSC-associated genes have been developed, and the conversion was monitored in real time. To further investigate the molecular mechanisms governing such conversion, the HIF signaling axis has been identified as a key mediator in stimulating the bidirectional conversion of glioma cells, promoting the progression of the recurrent and refractory disease. Unveiling the relationship between therapy-induced HIFs and GSCs allow us to develop therapeutic strategies that will enhance current standards of care and eliminate the regeneration of recurrent glioblastoma post-therapy.

Patient-derived xenograft (PDX) glioma specimens GBM43 and GBM6 were provided by Dr. David James from Northwestern University (Chicago, IL) and maintained according to the published protocol with some modifications (21). For in vitro examination of therapy-induced reprogramming of non-GSC to GSC, the PDX glioblastoma multiforme cells were forced into differentiation using 10% FBS-containing media. They were modified to constitutively express a blue fluorescent protein (BFP) using lentivirus-mediated infection in culture and then propagated in vivo by serial passaging in flanks of athymic nude mice.

The human glioma cell lines U87 were purchased from the ATCC and U251 from Sigma in 2013. Cells were maintained in culture in DMEM with 2 mmol/L l-glutamine (Mediatech), 10% FBS (Atlanta Biologicals), and 2% penicillin–streptomycin antibiotic (CellGro). Cells were grown in 37°C, and 5% CO₂ humidified atmospheric conditions and were enzymatically dissociated using 0.25% trypsin/2.21 mmol/L EDTA solution (Mediatech).

Authentication of PDX and glioblastoma line performed bi-yearly by using Promega GenePrint 10 system, which allowed us to coamplify ten human loci (TH01, TPOX, vWA, Amelogenin, CSF1PO, D16S539, D7S829, D13S317, D5S818, and D21S11). The amplified DNA from each cell line was then sent to Northwestern University Sequencing Core Facility. The GeneMapper 5 software was used to analyze raw data that provide us series of DAN sequence that can be compared against a database in Northwestern Nervous System Tissue Bank, ATCC, and Sigma. This test collectively provided a genetic signature with a random match probability of 1 in 2.92 × 109, which exceeds the standards ASN-0002 issued by the ATCC Standards Development Organization Workgroup for cell line authentication.

GSC-specific promoters, SOX2, Oct4, Nanog, and CD133, in the Light Switch Promoter Reporter system were obtained from Switchgear Genomics (cat. nos. S706722, S722579, S722680). The promoters were subcloned into the pDC516 vector (cat. no. PD-01-36, Microbix Biosystems Inc.) and then inserted into the lentiviral shuttle vector pCMV-RTK upstream of the red fluorescent protein (RFP) gene (Systems Biosciences). Lentiviruses were produced using X293T human embryonic kidney cells for infection according to the published protocol (22). GBM43, U251, and U87 cell lines were simultaneously induced to express the BFP using lentivirus-mediated infection in culture and then the populations expressing high-intensity BFP were sorted by FACS.

The recurrent tumor model was developed using PDX models GBM6-BFP and GBM43-BFP. Mice were briefly anesthetized with ketamine (115 mg/kg), and the stereotactic injection was carried out using a 10 μL Hamilton syringe (Reno) with a 30-gauge needle, which was inserted through a burr hole to a depth of 2.5 mm (1,000 cells). Five days after implantation, mice were treated without and with temozolomide (2.5 mg/kg/day for 5 days; i.p. injection, n = 10); 3 days post-therapy, 5 mice without and 5 mice treated with temozolomide were sacrificed, leaving 5 temozolomide-treated mice to be observed and sacrificed upon sickness.

Analysis of GSC-specific promoter activity in vivo pre- and post-chemotherapy was performed by implanting U87-BFP cells engineered to express a RFP driven by Oct4, SOX2, CD133, and Nanog promoter into flanks of nude mice. Once the tumors were established mice were treated with temozolomide (2.5 mg/kg/day) for 5 days and imaged the same tumor using the two-photon intravital microscopy (IVM; post-therapy). To investigate the intratumoral localization of the RFP-tagged CSCs, the glioblastoma cells with stably expressing Oct4 reporter system were implanted orthotopically, and a cranial window was placed on the animal brain according to published protocol (23). After two weeks post-tumor implantation, two-photon IVM was performed to capture RFP fluorescence within the tumor bulk (pre-therapy).

GBM43 cells were seeded onto an 8-well glass slide chamber (cat. no. 1256518, Fisher Scientific). Cells were treated with or without temozolomide 50 μmol/L for 8 days, fixed with 4% paraformaldehyde, and then stained with primary anti-HIF1α antibody. Animal brains with implanted PDX glioma were removed, frozen in Tissue-Tek O.C.T. Compound (cat. no. 4583), and sectioned for analysis. Nuclei were stained with ProLong Gold Antifade Mountant with DAPI (cat. no. P-36931, Life Technologies). Images were captured using Zeiss LSM 510 Confocal Microscope at a magnification of 20× (Carl Zeiss Microscopy).

HIF1α and HIF2α lentiviral shRNA constructs were purchased from Genecopoeia (cat. no. HSH008832-LvU6, HSH005050-LVRU6GP) and used to infect U251-blue cells following the lentiviral infection protocol mentioned above. Knockdown efficiency was confirmed by Western blot analysis and flow cytometry.

GBM43 and U251-BFP cells with Sox2, Oct4, or Nanog promoters driving RFP expression were plated in 35-mm round live-cell dishes on 10-mm diameter glass-bottom insets (MatTek Corporation). Cells were treated with temozolomide. Seventy-two hours after treatment, dishes were placed inside the Olympus LCV110U VivaView incubator, which was kept at 37°C/5% CO₂ and housed an inverted fluorescence microscope (Olympus). Images were taken every hour for 24 hours at 20× magnification using MetaMorph image capture software (Olympus) and were analyzed using Fiji image processing program (NIH, Bethesda, MD).

Experimental data were analyzed using GraphPad Prism Software v4.0 (GraphPad Software). Data represent results performed in triplicate. Where applicable, unpaired t tests and survival analysis were applied. Statistical significance was defined as *, P < 0.05; **, P < 0.01; and ***, P < 0.001.

To investigate the behavior of the GSCs in recurrent diseases, two recurrent glioblastoma models were developed using the PDX lines GBM6 (classical, MGMT hypermethylated) and GBM43 (proneural, MGMT unmethylated) tagged with BFP (21, 24, 25). After intracranial implantation of the PDX lines into mice, we mimicked clinical recurrence by treating implanted tumors with alkylating agent temozolomide, the current standard chemotherapeutic care for glioblastoma. Temozolomide (2.5 mg/kg) was administered intraperitoneally every day for 5 days (Fig. 1A). The treatment decreased the tumor burden by approximately 95%, resulting in an improvement of 13 to 27 days in median survival (Supplementary Fig. S1) as compared with vehicle-treated animals (see Materials and Methods; ref. 26). However, within three to six weeks after treatment, tumors recurred in 100% of temozolomide-treated mice.

We could subsequently evaluate the status of GSC populations in tumors of three different groups: vehicle-control untreated animals, animals with primary tumors sacrificed 3 days post-treatment and animals with recurrent tumors as determined by symptoms of disease (Fig. 1A). The GSC populations were evaluated by flow cytometry analysis of two GSC-specific phenotypic markers, CD133, CD15, and one transcription factor SOX2. In GBM6 tumor model, there was a significant increase in CD133⁺, CD15⁺, and SOX2⁺ populations in the recurrent tumors as compared to both the primary and the control tumors (Fig. 1B). The greatest increase was seen in the CD15⁺ compartment with the CD15⁺ GSC population increasing from 11 ± 2.1% in the control group to 85% ± 2.4% for recurrent tumors (P < 0.0001, Fig. 1B). As there is considerable controversy over the use of single markers to accurately identify CSC populations, we verified our results by analyzing GSC populations expressing multiple markers. For GBM6 model, a significant increase was observed in all three double-positive GSC populations, CD133⁺CD15⁺ as well as CD133⁺, and CD15⁺ cells expressing SOX2 transcription factor, as comparing recurrent tumors to the control and non-recurrent groups (Fig. 1B). The largest increase was observed in the CD133⁺SOX2⁺ GSC compartment, with an increase from 28% ± 3.0% in the control group to 83% ± 2.3% for recurrent tumors (P < 0.0001, Fig. 1B). On the basis of the expression of stemness markers, some GSC populations decreased slightly, immediately after temozolomide-treated primary tumors (3-day post-therapy) suggesting that temozolomide is perhaps partially capable of depleting the GSC population during initial therapy (Fig. 1B and C). However, for GBM6 model, all recurrent tumors had significantly higher GSC populations than their non-recurrent and nontreated counterparts (Fig. 1B).

The results for the GBM43 model were similar. However, single-marker analysis of GBM43 tumors revealed no significant increases in the CD15⁺ or SOX2⁺ GSC populations in recurrent tumors. As single marker is not the ideal indicator of GSC populations, the dual-marker analysis was performed, and it was seen that all three double-positive populations, CD133⁺ and CD15⁺ expressing GSC-specific transcription factor SOX2 as well as the CD133⁺CD15⁺, were significantly enriched in recurrent tumors when compared with the control ones (Fig. 1C).

In particular, differences between the two models can be affected when the expansion of the GSC pool occurs, either during or immediately after therapy. Our observation revealed that GBM6 tumors exhibit a significant increase in GSC populations following disease recurrence while GBM43 tumors are more enriched in specific GSCs immediately after primary therapy (Fig. 1B and C). However, the observed increase in GSC populations in both PDX models indicates that tumor recurrence following chemotherapy shifts the equilibrium of glioma cells to a more stem-like state.

To investigate the biological consequence of such equilibrium shift we next created a single-cell suspension from two freshly isolated glioblastoma multiforme patients samples and exposed to 50 μmol/L temozolomide or vehicle control for 8 days. After day 8, cells were rechallenged with escalating doses of temozolomide to evaluate their ability to resist therapy as well as used in the tumorsphere assay to measure the CSC frequency. As demonstrated in Supplementary Fig. S2, exposure to temozolomide made both of the tested glioblastoma multiforme cells significantly more resistance to temozolomide (Supplementary Fig. S2A and S2B). Moreover, the stemness of both of the cells was significantly enhanced post-temozolomide exposure as the stem cell frequency was elevated at least 3-fold (Supplementary Fig. S2C and S2D). Finally, to investigate the tumor engraftment efficiency in vivo, we have treated the U251 cells with 50 μmol/L temozolomide for 8 days. Next, the CD133⁺ GSCs and CD133⁺CD15⁻ non-GSC glioma cells were sorted out by FACS, intracranially implanted (100 and 1,000 cells) in the nude mice and animals were monitored for the endpoint survival. As shown in the table in the Supplementary Fig. S2E, the tumor engraftment capacity of the non-GSC population was significantly enhanced in the temozolomide-treated groups in both 100 and 1,000 cells implanted groups as compared with DMSO-treated control. On the basis of these data along with our previous published data, we conclude that post-therapy glioblastoma multiforme cells are more resistance to rechallenge and enhanced self-renewing capacity (26).

Next, we set to investigate the role of cellular plasticity in promoting post-therapy glioma stem cell niche. Cellular plasticity is a dynamic process and requires monitoring of the changes in the GSC population over time. It is critical to investigate this process in real time to elucidate the conditions under and mechanisms through which this plasticity occurs. Therefore, we set to develop a GSC-specific reporter system to study glioblastoma cells plasticity in real time. One of the greatest challenges in studying GSCs is selecting the appropriate marker to identify the CSC population. To overcome this problem, we have designed multiple reporter systems by using the promoter element of four different CSC-specific genes: CD133, Sox2, Nanog, and Oct4, to ensure that reporter expression is selectively elevated in the GSC compartment (Fig. 2A). These promoters were cloned upstream from the RFP. The CD133 is one of the most commonly used markers for GSCs, while the other three genes are known as pluripotency factors, those reported to be also associated with CSC population (20, 27). These reporter systems were then cloned into a lentivirus system, and viral particles were generated to stably modify the glioblastoma multiforme cell lines as well as the PDX lines. To investigate whether these reporter systems can faithfully identify the GSCs population, total mRNA from each samples were compared with respect to the expression levels of several stemness promoting genes such as Sox2, Oct4, CD133, Noggin, Nestin, Musashi, BIII Tubulin, as well as GFAP mRNA transcripts in the RFP-positive versus RFP-negative populations. For CD133-p-RFP reporter cell line, we observed statistically significant upregulation of CD133, SOX2, Oct4, and Noggin mRNA expression in RFP-positive versus RFP-negative populations (Fig. 2B1). The RFP-positive population of the Sox2-p-RFP reporter cell line demonstrated statistically significant upregulation of Sox2, Oct4, and Noggin mRNA expression. However, as expected, GFAP transcripts were downregulated. In addition, no statistically significant difference in CD133, Nestin, Musashi, and B III Tubulin mRNA levels was observed when compared with the RFP-negative population (Fig. 2C1). Finally, the Nanog-p-RFP reporter cell line showed statistically significant upregulation of Sox2, Noggin, Nestin, and B III Tubulin mRNA expression in the RFP-positive versus RFP-negative population. However, there was no statistically significant difference in CD133, Oct4, Musashi, and GFAP mRNA levels between the two populations (Fig. 2D1). Such discrepancy of gene expression pattern between the different reporter systems is probably due to the reported heterogeneity within the GSC population (28, 29). In most cases, it is unclear whether phenotypically diverse CSC populations are also functionally distinct. To address this relationship, one needs to use the same tumor (cell) system to simultaneously compare the tumorigenic potential of different subsets. In theory, various CSC populations may or may not be hierarchically organized, analogous to how cancer cells, in general, may be organized. Taken together, we conclude that the RFP-positive cells, in general, were expressing an elevated level of CSC-associated genes.

Next, FACS analysis was performed on the four different cell lines: GBM43 SOX2-p-RFP, U251 SOX2-p-RFP, U251 CD133-p-RFP, and U251 Nanog-p-RFP. The highest RFP-positive population at 18.2% was observed for U251 CD133-p-RFP cell line (Fig. 2C). The GBM43 SOX2-p-RFP, U251 SOX2-p-RFP, and the U251 Nanog-p-RFP express RFP for 10.5%, 8.3% (Fig. 2B, data not shown), and 1.01% (Fig. 2D), respectively. Next, to test whether the reporter activity correlates with CD133⁺ GSC population, a FACS analysis was conducted to study colocalization of the RFP expression with the CD133 expression with each of the reporter cell lines. The RFP-positive population of all the reporter cell lines expressed high levels of the CD133 surface protein while the opposite was true for all the RFP-negative populations (Supplementary Fig. S3). In all four-reporter systems, the mean fluorescence intensity (MFI) for RFP signal ranged from 150 to 250 in CD133⁻ populations, while the RFP signal increased to 700–1,000 MFI on CD133⁺ populations (P < 0.0001, Supplementary Fig. S3).

After examining that the reports systems were capable of identifying GSC subpopulation phenotypically, the functional characteristics of the RFP⁺ population of cells in vitro were verified. The RFP-positive and RFP-negative populations from the reporter cell lines were sorted out on the BD FACSAria and subsequently used in a tumorsphere assay to evaluate their stem-like phenotype (Fig. 2B1 and C1). Tumorsphere assay is normally set up in non-adhere 96-well plates to enhance tumor-sphere formation (30). However, the tumorsphere assay was firstly conducted in a coated plate to confirm comparison of RFP expression in the neurosphere-forming cells and the adherent cells (with lower stemness) in one assay. As shown in Fig. 2B2 and C2, glioma cells that were capable of forming neurosphere also express high levels of RFP while the attached cells entirely lost RFP expression indicating that the reporter system did get activated only in the cells that were part of tumorspheres. Next, to investigate in vivo distribution and localization of CSC-specific reporter labeled GSC, the BFP-labeled GBM43-Sox2-p-RFP cells were implanted intracranially for eventual imaging by the IVM. Two-photon IVM (refer to Materials and Methods section) enable us to image the reporter-induced RFP expression in the BFP-positive tumor's anatomy in live animals (Fig. 2E1). Our observation revealed that in the Sox2 reporter tumor, the majority of the RFP-positive cells were localized near the vasculature labeled with the green dextran (Fig. 2E). Thus, our preliminary study suggests that RFP-labeled GSCs demonstrated perivascular niche as described previously in the literature (31). Finally, standard tumorsphere assays were performed in non-coated 96-well plate by plating a different number of sorted RFP-positive and negative cells in the neurosphere media. As demonstrated in Fig. 2G and H, the RFP-positive cells exhibited a significantly elevated level of self-renewing ability with CSC frequency 1 in 214 glioblastoma multiforme cells in the RFP-positive population as compared with 1 in 619 population in the RFP-negative population for SOX2 reporter cells (Fig. 2G). In the Oct4 reporter cells, the stem cell frequency was 1/189 and 1/554 for RFP-positive and RFP-negative cells, respectively (Fig. 2H). Collectively, these data indicated that these reporter systems could faithfully detect the GSCs in real time.

In our recurrent glioblastoma model, we observed a population-wide shift in fate cell equilibrium. To investigate whether similar dynamics occur in cell fate equilibrium as of our recurrent glioblastoma multiforme model, post-therapy stemness was monitored at the single-cell level. First, to confirm our in vivo findings of glioma cell “de-differentiation” in vitro, the GBM43 PDX line was treated with either 50 μmol/L temozolomide, the maximum concentration found in human blood plasma (32, 33), or vehicle-treated control (DMSO). The cells were sorted by flow cytometry for stem-like CD133⁺ and non-stem CD133⁻ populations of each group (14). The sorted populations were cultured in FBS, which forces differentiation to minimize contamination with the CSC populations during sorting (Supplementary Fig. S4; ref. 18). In this condition, the GSC CD133-expressing as well as CD133⁻ non-GSC populations were sorted every 48 hours for 20 days in the presence or absence of temozolomide (Fig. 3A). For a baseline control, the untreated CD133⁻ and CD133⁺ populations were sorted. Both populations together with an unsorted (“DMSO”) population reached an equilibrium percentage of 1%–2% CD133⁺ cells within 48–72 hours (Fig. 3A). When cells were pretreated with DMSO or temozolomide before sorting, all DMSO-treated cell populations returned to baseline levels within 72 hours regardless of their initial percentage after sorting (Fig. 3A). Approximately 1%–2% of sorted CD133⁻ populations started to expressed CD133 within 48 hours, while the sorted CD133⁺ population lost CD133 expression to reach an equilibrium of 1%–2% 72 hours post-sorting (Fig. 3A).

In contrast, temozolomide-treated sorted cell populations showed a significant change from baseline levels (Fig. 3A). The sorted temozolomide-treated CD133⁺ population took approximately 10 days to reach the equilibrium, and the percent of CD133⁺ GSCs was about 16%, which is about 9-fold higher than the control (Fig. 3A). In the sorted CD133⁻ population treated with temozolomide, cells immediately increased in CD133 protein expression, reached a peak at day 6 (14% ± 1), and then decreased at days 8–10, maintaining their downhill trend until 20 days post-sorting. This in vitro observation, therefore, confirmed results of in vivo observations, which helped us conclude that temozolomide therapy shifts the equilibrium proportion of GSCs in tumor cell populations both in vivo and in vitro.

To further investigate this phenomenon of cellular plasticity at the single-cell level and its contribution to cell fate equilibrium shift post-therapy, GSC-specific reporter systems were employed to track interconversion between glioma stem and non-stem cells. GBM43 and U251 glioblastoma multiforme cells stably expressing GSC-specific reporters were subsequently cultured in the presence or absence of temozolomide for 8 days and analyzed by flow cytometry to determine whether the cells with active promoters also expressed GSC-specific phenotypic markers as well as reprogramming transcription factors. Single-cell analysis of engineered cells using Olympus VivaView real-time sequential imaging showed active conversions of previously RFP-negative cells (non-GSCs) into RFP-positive cells (GSCs) in chemotherapy-treated samples 4–6 days post-therapy (Fig. 3B and C). The percentage of RFP-positive GBM43 cells expressing SOX2-p-RFP was consistently and significantly lower compared with temozolomide-treated cells (9.34% ± 0.81% vs. 17.91% ± 1.29, P < 0.001). For U251 cells expressing SOX2-p-RFP, the percentage of RFP-positive was consistently and significantly lower compared with temozolomide-treated cells (14% ± 2.2% vs. 35% ± 5.3%, P < 0.001). A very similar result was also observed by using Oct4 and Nanog reporter systems (Supplementary Fig. S5A–S5C). Baseline spontaneous conversion rates of 14% for RFP+Sox2-tagged, 11% for RFP+Oct4-tagged, and 22% for RFP+Nanog-tagged glioblastoma multiforme fell within the range of Sox2⁺, Oct4⁺, and Nanog⁺ glioma cells reported by other groups (26, 34, 35). After chemotherapy treatment, the frequency of RFP-negative cells that became RFP-positive increased to approximately 35%, 36%, and 43% for Sox2, Oct4, and Nanog promoters, respectively (P < 0.01). Next, to examine whether these post-therapy RFP-positive cells also express other GSC-specific markers, the temozolomide-treated U251 CD133-p-RFP cells were FACS analyzed (Fig. 3D). In this reporter system, a majority of the post-therapy RFP-positive cells were significantly colocalized with the CD133, CD15, and SOX2-expressing glioblastoma multiforme cells. Similar experiments were performed with the other three reporter systems, which demonstrated that post-therapy RFP⁺ cells were also expressing GSC-specific markers such as CD133 and SOX2 (Fig. 3E and Supplementary Fig. S5D).

Next, to validate these observations in vivo, the glioblastoma multiforme cells expressing RFP driven by Oct4 promoter were implanted intracranially, and once the tumor was engrafted, the RFP-positive tumor cells within the tumor bulk were detected by using cranial window and IVM. The tumor-bearing mice were then treated with chemotherapy for 5 days, and a visually significant increase in RFP expression scattered throughout the tumor bulk within the same animal was observed (Fig. 3F).

Finally, to investigate whether the observed expansion of CD133 population post-therapy is not due to proliferation of this specific subpopulation, the CD133⁻ and CD133⁺ populations were examined for their proliferative capacity post-therapy by Ki67 staining (Supplementary Fig. S5E and S5F). As demonstrated, the temozolomide treatment significantly decreased the Ki67 total protein expression as well as the percent of Ki67-positive cells, only in the CD133⁺ GSC compartment. This data indicated that the observed post-therapy induced expansion of the GSC compartment is not due to the proliferation of the CD133⁺ GSCs. Next, to investigate whether the post-therapy induced GSCs can be differentiated into the different, previously reported differentiation-inducing agent bone morphogenic protein 2 (BMP2) was added at day 6 after temozolomide treatment for 2 days (36). These cells were subsequently analyzed by FACS and observed that in the presence of BMP2, the temozolomide induced GSCs in the GBM43 PDX line cultured in serum-free neurobasal media (Supplementary Fig. S6A and S6B) or in the FBS-containing media (Supplementary Fig. S6C and S6D) significantly decreased the GSC-specific markers such as CD133 and CD15 as well as upregulation of astrocytic lineage marker GFAP. These data are another validation of the CSC authenticity of the therapy-induced GSCs. Taken together, these data showed that GSC-specific promoter activities, as well as GSC-specific phenotypic markers, were significantly enhanced during primary chemotherapy. This real-time observation of cellular plasticity at the population level combining with our functional data positively indicated the ability of non-GSCs to acquire GSC-like states in the presence of therapeutic stress.

Cellular responses to hypoxia are commonly regulated by hypoxia-inducible factors (HIF), which are key transcriptional factors active in the cell during oxygen tension. Our previous report indicated that anti-glioma chemotherapy induces the expression of HIFs and promotes hypoxia-like responses (26) and the intratumoral hypoxia niche was previously described as an important factor for the maintenance of the GSC subpopulation found within glioma (27, 37, 38). On the basis of this reasoning, the role of therapy-induced HIFs in regulating the equilibrium of cellular stemness during anti-glioma therapy was investigated.

Examining the levels of both HIF1α and HIF2α in a glioblastoma multiforme model after primary chemotherapy, a significant increase in levels of HIF1α and HIF2α expressions was observed in cells treated with temozolomide (50 μmol/L) as compared with the control (Fig. 4A). HIF1α expression was consistently elevated in chemotherapy-treated samples from day two through eight. On day two, HIF1α MFI was 845 ± 1.5 for DMSO-treated and 1190 ± 2.2 for temozolomide-treated (P < 0.001), while on day eight MFI levels were almost the same at 820 ± 1.2 for DMSO-treated and 1100 ± 2.7 for temozolomide-treated (P < 0.001; Fig. 4A). While HIF2α expression was similarly elevated in temozolomide-treated conditions as compared with control, the HIF2α increased over time and reached its peak 6–8 days after primary therapy. On day 2, HIF2α MFI for DMSO-treated was 248 ± 0.4 and 375 ± 0.6 for temozolomide-treated, and on day 8, MFI for temozolomide-treated samples increased to 730 ± 4.7 with DMSO at 252 ± 0.2 (P < 0.001; Fig. 4A).

This result opens a new insight into the dynamics of HIF transcriptional factors in the hypoxic microenvironment. The HIF1α transcriptional factor is expressed immediately after the oxygen tension is initiated, and HIF2α is expressed at the chronic phase of hypoxia (39). On the basis of our data, we propose that the immediate peak expression of HIF1α in PDX glioblastoma multiforme cells after exposure to chemotherapy may represent one of the necessary steps for the conversion of non-GSCs into GSCs. Meanwhile, the increase in HIF2α expression over an extended period may be required for the maintenance of the newly converted GSCs post-therapy.

To further evaluate the dynamics of chemotherapy-induced HIF activation, an HIF1α-responsive reporter assay was utilized (Fig. 4B). To stabilize itself, HIFα dimerizes with HIFβ and then binds to hypoxia-responsive elements (HRE) located in the promoters of hypoxia-regulated genes. The subsequent binding activates the transcription of in numerous genes responsible for modulating cell survival and metabolism (40).

The functionality and physiological relevance of the increase in HIF activity after chemotherapy were investigated by analyzing in vivo expressions of HIF1α and HIF2α in the PDX glioblastoma at 3 days post-primary chemotherapy as well as upon tumor recurrence (Fig. 5A and B). Experiments were performed in the recurrent model as described in Fig. 1 in two different PDX lines (GBM43 proneural and GBM6 classical). The GBM43 model presented a significantly higher expression of HIF1α at 3 days post-primary chemotherapy as compared with the control (Fig. 5A). There was a significant increase in the expressions of HIF1α and HIF2α in the CD133⁺ GSC compartment both post-therapy and at disease recurrence (P < 0.001). In the GBM6 PDX model, a higher expression of HIF1α was seen 3 days post-therapy, while HIF2α expression present in the GSC compartment was only observed when the disease recurred.

To further assess the roles of HIF1α and HIF2α in the conversion of non-GSCs into GSCs, shRNA-mediated knockdown of HIF1α or HIF2α in the cancer cells was used (Supplementary Fig. S7). Knockdown of both HIF1α and HIF2α resulted in a 3-fold decrease in the CD133⁺ GSC population as compared with control cells (P < 0.001, Fig. 6A). These findings were confirmed using a chemical inhibitor of HIF1α translation (Calbiochem, 400088; Fig. 6B). Using a concentration of inhibitor that was not toxic to cells (IC₅₀ of 20 μmol/L or lower as determined by 7-AAD staining), the PDX lines were concurrently treated with temozolomide and HIF1α inhibitor. As a result, HIF1α inhibition significantly blocked the amplification of CD133⁺, CD15⁺, Sox2+, CD133⁺Sox2⁺, and CD133⁺CD15⁺ GSCs after primary therapy (P < 0.001). Collectively both HIF1α and HIF2α are critical for conversion of non-GSCs to a GSC-like state after anti-glioma chemotherapy.

Despite the standard use of a combination of aggressive therapies (surgical resection followed by chemotherapy and radiation), glioblastoma tumors invariably recur (41, 42). The implementation of temozolomide has only improved the overall survival from 12 to 14.6 months in the clinical setting (2, 42). It is believed that a small population of cells within glioblastoma multiforme, known as glioma stem cells, is responsible for such a dismal disease prognosis (43, 44). The GSCs have the capacity to self-renew, differentiate, and give rise to a whole new heterogeneous tumor, and recent reports have demonstrated the existence of both spontaneous and therapy-induced GSC plasticity (15, 17, 26). This opens up the possibility of such bidirectional conversion that may preclude optimal efficacies of standard therapies for two main reasons (17, 20, 26, 45). First, as standard therapies mostly target rapidly dividing glioblastoma multiforme cells, the subpopulation of quiescent GSCs may largely spared, which leads to a more aggressive recurrent tumor driven by chemotherapy-resistant GSCs (46). Second, due to the invariable interconversion of non-GSGs to GSCs following therapeutic stress, aggressive therapeutic strategies that remove the majority of the tumor bulk and then target the GSC subpopulation are still doomed to fail. The increased occurrence of plasticity between non-GSCs and GSCs during and post-therapy will result in the replenishment of all glioblastoma subpopulations and consequent tumor regrowth (17, 20, 26, 47). In this article, our main goal is to investigate the dynamics of interconversion during therapy and understand the possible mechanisms behind this conversion process in the recurrent PDX glioma model. To avoid such risks of any contamination of small fraction of CD133⁺ cells during FACS sorting or quiescent CSCs, which may yet to express phenotypic markers for CSC, we have utilized culture condition media with 10% FBS as well as in the presence of differentiation promoting factor BMP2 that reported promoting differentiation in the CSCs population. In this manuscript, we demonstrate an increased interconversion of non-GSCs to GSCs even in the presence of differentiation condition in vitro as well as in the recurrent model in vivo. The understudied nature of cell fate equilibrium in recurrent glioblastoma multiforme allowed us to investigate the dynamics of inherent cellular behavior in response to primary chemotherapy (48), and we found that although cells acquire an equilibrium state over time both in the absence and presence of chemotherapy, there is a distinct shift in the equilibrium of therapy-treated cells to become more stem-like in behavior. Acquiring stem-like phenotype could be partially due to induction of stem phenotype by chemotherapy or by enhancing the transition of cell status from non-GSC to GSC phenotype. These observations were confirmed by analysis of the whole glioblastoma multiforme population as well as on the single-cell level by the development of multiple specific GSC-specific promoter-based reporter systems. Taken together, our results strongly support the concept that the plastic nature of glioblastoma cells promotes their acquisition of stem-like phenotypes post-therapy in addition to highly aggressive malignant characteristics. Further analysis have to be done in the future to determine if any GSC markers or molecular characteristics could be utilize to discriminate tumor forming versus non-tumor–forming cells. Also in accordance with previous publications (14, 25, 49, 50), we found that different subtypes of glioblastoma multiforme present variable responses to initial therapeutic insult. This is mostly seen as varied influences between self-renewal of GSCs and conversion of non-GSCs into GSCs in the recurrent model. The variability between the two models suggests that the unique genetic signature and origin of each PDX model, GBM6 is of classical origin, while GBM43 is proneural, affects their response to therapy (24, 25). For example, the classical subtype of glioblastoma multiforme contains an EGFR alteration as compared with a PDEGR alteration predominately observed in the proneural glioblastoma multiforme (25). Therefore, each subtype may respond differently to treatment, such as in the timing of GSC expansion, which can occur during the course of therapy, post-therapy, or as the disease burden progresses and promotes recurrence.

The importance of environmental cues as triggers for the activation of self-renewal in GSCs is undeniable. Data presented in this manuscript indicated that therapy-induced plasticity in the GSC population was at least in part HIF-mediated. Direct measurement of the percentages of HIF1α- and HIF2α-expressing cells in the whole recurrent tumor and assessment of the activities of both HIFs in CD133⁺ GSCs revealed that HIF2α had a direct effect on the maintenance of the GSC phenotype while HIF1α was mostly related to the initial activation of this plastic phenomenon. Blocking HIF1α activity using shRNA and a translation inhibitor led to significant decreases in the GSC population after primary chemotherapy. On the basis of these results, we identified HIF1α as the critical regulator of glioblastoma multiforme cell plasticity in our recurrent model. We were able to demonstrate that dedifferentiation of tumor cells by temozolomide-induced therapeutic stress is partly mediated by HIFs. The role of HIFs promoting glioblastoma cells stemness has been extensively examined (37, 51). However, some conflicting results have been reported regarding the role of intratumoral hypoxia-mediated CSC niche regulating the chemoresistance properties of glioblastoma multiforme. Pistollato and colleagues investigated the effects of the intratumoral hypoxia gradient on the TMZ resistance capacity of GSC and reported that GSCs located within the core of the tumor were highly resistance to temozolomide therapy as compared with GSCs from the tumor periphery. The CD133-expressing cancer cells mostly enriched in the tumor core also overexpress MGMT, a gene that is known be critical for temozolomide resistance (52). In a contesting study, Glas and colleagues demonstrated that even though a fraction of tested CSC lines derived from the core and the periphery of the same patient's glioblastoma multiforme tumors responded differentially to radiation, temozolomide, and CCNU, the difference was not consistent in between patients (53). The intratumoral hypoxia has been already linked to chemo- and radioresistant properties of tumor cells as hypoxic niche may reduce the oxidative DNA damage and allow the tumor cells to overcome the conventional radio- and chemotherapy (54). However, the link between hypoxia regulate CSC niche and post-therapy glioblastoma multiforme recurrence is much more complex than expected as the majority of the glioblastoma patient relapse even after the resection of the entire primary tumor mass including the core of the tumor burden that contains the majority of the hypoxic region (51). Data presented in this manuscript indicated that therapeutic stress could promote hypoxia-like response, which could be independent of intratumoral hypoxia (19, 26). Additional investigation will require elucidating the role of intratumoral hypoxia in post-therapy HIFs accumulation. Collectively, our data suggest that blocking glioblastoma cell plasticity that allows the glioblastoma cells to acquire CSC-like state during chemotherapy can be a potential therapeutic strategy to be used in combination with the standard of care to decrease malignant relapse.

There may be some limitations to our study. First, we have presented limited data regarding the functional aspect of the therapy induced GSCs. Even though we have demonstrated that our GSC-specific reporter systems could faithfully identify the GSC population in real time as the RFP-positive population showed significantly elevated level of self-renewing capacity as compared with RTK-negative cells, in the future, it would be prudent to validate this observation in vivo by using a limited dilution tumor formation assay. Secondly, we observed phenotypic heterogeneity within the induced GSC subpopulations. However, the functional characterization of the phenotypic heterogeneity or their contribution to disease recurrence is remaining largely unknown. A negative selection system regulated by our GSC-specific reporter system could be employed to address this critical question. Finally, the contribution of intratumoral hypoxia in post-therapy conversion of glioblastoma multiforme cells to GSCs is yet to be investigated. Although our results demonstrated that therapy-induced GSCs niche is independent of the intratumoral hypoxia as therapeutic stress induced HIF expression in the normoxic condition in vivo, further studies are required to elucidate the precious molecular mechanisms of therapy induced cellular plasticity and its role in disease recurrence.

Cancer cells within individual tumors exist in many distinct phenotypic states that vary significantly with respect to functional characteristics. The diverse cell populations within different tumors usually display a unique equilibrium of cells in various fate states. Such dynamic equilibrium is achieved by the cell fate decision orchestrated by interactions between an individual cell and its microenvironment. This is significant in disease progression because microenvironmental changes, such as extracellular stress, can directly influence survival fitness. As a summary, our results demonstrated that therapeutic stress can shift the intratumoral balance towards a more stem-like state by promoting bidirectional interconvertibility between non-CSC and CSC compartments, possibly promoting an aggressive and therapy-resistant phenotype in recurrent glioblastoma multiforme. Elucidating the molecular mechanisms that maintain the intratumoral cell fate equilibrium state and the effects of anti-glioma therapies on the CSC niche gives a new insight to understand how CSCs contribute to disease progression, identify novel therapeutic targets to prevent glioblastoma multiforme recurrence, and promote a better prognosis for patients with glioblastoma in the future.

No potential conflicts of interest were disclosed.

Conception and design: G. Lee, D. Guo, T. Hasan, M.S. Lesniak, A.U. Ahmed

Development of methodology: G. Lee, B. Auffinger, D. Guo, T. Hasan, M.S. Lesniak, A.U. Ahmed

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Lee, D. Guo, T. Hasan, M. Deheeger, A.L. Tobias

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Lee, B. Auffinger, D. Guo, T. Hasan, M. Deheeger, A.L. Tobias, L. Zhang, M.S. Lesniak

Writing, review, and/or revision of the manuscript: G. Lee, B. Auffinger, D. Guo, T. Hasan, M. Deheeger, A.L. Tobias, F. Atashi, C.D. James, A.U. Ahmed

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Auffinger, D. Guo, T. Hasan, M. Deheeger, J.Y. Kim, C.D. James, A.U. Ahmed

Study supervision: A.U. Ahmed

We thank Dr. Mejing Wu for her statistical assistance. The work of this manuscript is dedicated to the memory of Donna Guo.

This work was supported by the National Cancer Institute grant R00 CA160775 and the American Cancer Society (299503) grant (to A.U. Ahmed).

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.