A Distinct Reactive Oxygen Species Profile Confers Chemoresistance in Glioma-Propagating Cells and Associates with Patient Survival Outcome

Abstract

Aims: We explore the role of an elevated O₂ ⁻:H₂O₂ ratio as a prosurvival signal in glioma-propagating cells (GPCs). We hypothesize that depleting this ratio sensitizes GPCs to apoptotic triggers. Results: We observed that an elevated O₂ ⁻:H₂O₂ ratio conferred enhanced resistance in GPCs, and depletion of this ratio by pharmacological and genetic methods sensitized cells to apoptotic triggers. We established the reactive oxygen species (ROS) Index as a quantitative measure of a normalized O₂ ⁻:H₂O₂ ratio and determined its utility in predicting chemosensitivity. Importantly, mice implanted with GPCs of a reduced ROS Index demonstrated extended survival. Analysis of tumor sections revealed effective targeting of complementarity determinant 133 (CD133)- and nestin-expressing neural precursors. Further, we established the Connectivity Map to interrogate a gene signature derived from a varied ROS Index for the patterns of association with individual patient gene expression in four clinical databases. We showed that patients with a reduced ROS Index demonstrate better survival. These data provide clinical evidence for the viability of our O₂ ⁻:H₂O₂-mediated chemosensitivity profiles. Innovation and Conclusion: Gliomas are notoriously recurrent and highly infiltrative, and have been shown to arise from stem-like cells. We implicate an elevated O₂ ⁻:H₂O₂ ratio as a prosurvival signal in GPC self-renewal and proliferation. The ROS Index provides quantification of O₂ ⁻:H₂O₂-mediated chemosensitivity, an advancement in a previously qualitative field. Intriguingly, glioma patients with a reduced ROS Index correlate with longer survival and the Proneural molecular classification, a feature frequently associated with tumors of better prognosis. These data emphasize the feasibility of manipulating the O₂ ⁻:H₂O₂ ratio as a therapeutic strategy. Antioxid. Redox Signal. 19, 2261–2279.

Introduction

Glioblastoma multiforme (GBM) represents the most severe form of adult malignant brain tumor. Stem-like glioma-propagating cells (GPCs) can be isolated that initiate and sustain tumors in immune-compromised mice (20). These cells exhibit phenotypic, karyotypic, and transcriptomic hallmarks that mirror the primary tumor (33, 44). As GPCs possess extensive self-renewal capacity, they are the plausible culprits of tumor recurrence.

Innovation

We implicate the elevated O₂ ^-:H₂O₂ ratio as a tumor cell proliferation signal in glioma-propagating cell self-renewal and growth. Depletion of this ratio confers sensitivity to apoptotic triggers and extends survival in an orthotopic mouse model. The reactive oxygen species (ROS) Index, a measure of normalized O₂ ^-:H₂O₂ ratio, provides quantification of O₂ ^-:H₂O₂-mediated chemosensitivity, an advancement in a previously qualitative field. Importantly, glioma patients with a reduced ROS Index correlate with longer survival and the Proneural molecular classification, a feature frequently associated with tumors of better prognosis. These data emphasize the feasibility of manipulating the O₂ ^-:H₂O₂ ratio as a therapeutic strategy.

Reactive oxygen species (ROS) is implicated in signaling mechanisms of normal and cancer cells, and in oxidative damage-induced cell death. Although low levels of ROS regulate cellular signaling and normal cell proliferation, recent studies have indicated that tumor cells exhibit a persistent elevation of ROS (specifically the superoxide anion [O₂ ⁻]) and show prosurvival advantage via a redox-based mechanism (52). Such opposing roles of ROS seem paradoxical, yet detailed understanding of the roles of individual ROS molecules is lacking. Little is known about the redox status in cancer stem-like cells that initiate and sustain tumors in mouse models, and are purportedly enriched in human derived stem-like cultures (20). Recently, Naka et al. showed that the ability of leukemia-initiating cells (LICs) to cause disease was significantly decreased by FoxO3a deficiency, a critical mediator of hematopoietic stem cell resistance to oxidative stress (43, 65). Further, the TGFβ-PI3K/AKT signaling axis was implicated, illustrating for the first time how LICs utilize redox-based mechanisms to feed into well-established oncogenic pathways. In addition, work by Diehn and colleagues described ROS as a critical mediator of radioresistance in breast cancer stem cells (CSCs), defined by CD44+ CD24−/low Lin−(16). Specifically, the authors showed that reduced ROS, measured as hydrogen peroxide (H₂O₂), conferred breast CSCs their prosurvival advantage. Further elevation of ROS rendered them susceptible to death by oxidative damage.

The redox status of a cell can have dual effects—effete tumor cell removal by oxidative stress, or survival via a slight prooxidant microenvironment that does not trigger oxidative stress (50). Both seemingly diverse mechanisms have a central theme; redox levels control cell fate, and the avoidance of apoptosis can be harnessed by cancer cells to transform. Clement and Pervaiz previously implicated specifically O₂ ⁻ and H₂O₂, the main species in the cytosolic milieu, as central modulators of chemoresistance (11). These two species are interconvertible via the superoxide dismutase 2 (SOD2) enzyme, a potent source of ROS generation in the mitochondrial electron transport chain (26). An elevated O₂ ⁻:H₂O₂ ratio confers tumor cells their prosurvival advantage, while a reduced ratio triggers apoptosis (52). By inference, these findings present the exciting notion that cancer cells can be eradicated simply by depleting the O₂ ⁻:H₂O₂ ratio, a process termed as reductive stress (11). Intriguingly, a most recent study implicated an elevated O₂ ⁻:H₂O₂ ratio as the mechanism effecting angiopoietin-4-mediated tumor cell anoikis resistance in several cancer types (73). Collectively, these findings suggest that the O₂ ⁻:H₂O₂ ratio may play a more-significant central modulator role in mediating the tumor phenotype.

We ask if heterogeneity in GPC chemoresistance correlates with differences in the O₂ ⁻:H₂O₂ ratio. We hypothesize that an elevated O₂ ⁻:H₂O₂ ratio confers GPC prosurvival advantage against chemotherapeutic agents, while the inverse triggers sensitivity, thus suggesting a therapeutic strategy based on the redox status. We formulate a quantitative and reproducible ROS Index to reflect this ratio and predict the propensity of GPCs to undergo drug-induced apoptosis. Importantly, we ask if the ROS Index has a significant impact in patient disease progression and clinical outcome, and identify patient cohorts showing patterns of association with the ROS Index. This represents a novel effort to translate the ROS Index in a clinical setting; further, our data highlight the molecular heterogeneity of the disease and implicate the ROS resistance gene signature as a novel prognostic indicator.

Results

Chemoresistance correlates with increased O₂ ⁻:H₂O₂ ratio

GPCs demonstrate greater resistance to common chemotherapeutic agents than non-GPCs, likely due to the overexpression of genes such as the DNA repair enzyme (O(6)-methyl guanine–DNA methyltransferase [MGMT]) and antiapoptosis genes such as B-cell lymphoma 2 (BCL-2), B-cell lymphoma-extra large (BCL-XL), and survivin (18, 37). We chose complementarity determinant 133 (CD133) as the initial marker to focus on as it defines tumor-initiating capacity in a subset of GBM GPCs (3). Further, the molecular profile of our CD133(+) cell fractions correlated with stemness gene sets identified in GPCs (3) and hematopoietic stem cells (30) (Supplementary Fig. S1A; Supplementary Data are available online at www.liebertpub.com/ars). We first assessed the effects of 2 commonly used chemotherapeutic agents in GBM recurrence, cisplatin (CIS) and etoposide (VP-16), because of their effective in vitro concentrations and role as radiomimetics (9), on the sphere-forming ability of total sorted, CD133(+) and CD133(−) GPCs. The neurosphere assay reflects the GPC frequency, as opposed to transient-amplifying progenitors (55). We utilized 4 patient GPC lines characterized in our previous work: NNI-1, NNI-4, NNI-8, and NNI-11 (10, 19, 44). Except for NNI-1, CD133(+) cells displayed a greater sphere-forming capacity than CD133(−) cells in response to treatment with CIS or VP-16 (Fig. 1A) at their respective half-maximal inhibitory concentration (IC₅₀) values (Supplementary Fig. S1B). A similar trend was observed for sphere proliferation (Supplementary Fig. S1C). The trend was reversed in the NNI-1 line, where CD133(−) was the more-resistant population, supported by a similar trend of viability results of the individual cell fractions (Fig. 1B).

FIG. 1.

Glioma-propagating cells (GPCs) are heterogeneous in chemoresistance profiles (see also Supplementary Fig. S1). (A) Sphere-forming capacity was measured in both CD133(+) and CD133(−) cells in four GPC lines. CD133(+) cells were more resistant after cisplatin and VP-16 treatments except in NNI-1 where the reverse trend was observed. Data were normalized to the respective CD133-sorted populations treated with DMSO. *p<0.05; **p<0.01. (B) Cell viability was measured. CD133(+) cells were more proliferative after cisplatin and VP-16 treatments except in NNI-1 where the reverse trend was observed. Data were normalized to the respective CD133-sorted populations treated with DMSO. *p<0.05; **p<0.01; ***p<0.001; n.s., not significant. (C) Data summarized for both O₂ ⁻ (MitoSOX^™ Red; top panel) and H₂O₂ (Amplex^® Red; bottom panel) levels in four GPC lines. *p<0.05; **p<0.01. (D) O₂ ⁻:H₂O₂ ratios calculated for four GPC lines. An elevated O₂ ⁻:H₂O₂ ratio correlated with enhanced resistance to cisplatin and VP-16. *p<0.05; **p<0.01; ***p<0.001. CD133, complementarity determinant 133; DMSO, dimethyl sulfoxide; H₂O₂, hydrogen peroxide; O₂ ⁻, superoxide anion; VP-16, etoposide.

To determine the O₂ ⁻ and H₂O₂ levels, we utilized a well-established flow cytometry-based method with MitoSOX^™ Red or Amplex^® Red dyes, respectively (70, 74). This method is necessary to interrogate individual ROS levels in independent fractions of a heterogeneous cell mix. We determined that the flow-based method has a similar sensitivity to the conventional lucigenin-based method (Supplementary Fig. S2A–D). We observed that in NNI-4, 8, and 11, CD133(+) cells exhibited ∼10%–20% higher O₂ ⁻ levels than CD133(−) cells (Fig. 1C, top panel). Again, the trend was reversed in NNI-1 cells, where the CD133(−) O₂ ⁻ level was ∼10% higher than that of the CD133(+) fraction. In contrast, the H₂O₂ levels were ∼10%–20% lower in the CD133(+) fractions of NNI-4, 8, and 11 compared to their respective CD133(−) fractions; and CD133(−) of NNI-1 compared to its CD133(+) fraction (Fig. 1C, bottom panel). Figure 1D reflects an elevated O₂ ⁻:H₂O₂ ratio in the more-resistant CD133(+) cells of NNI-4, 8, and 11, and CD133(−) cells of NNI-1, while a reduced ratio correlated with the more-sensitive CD133(−) cells of NNI-4, 8, and 11, and CD133(+) cells of NNI-1, consistent with our hypothesis that an elevated O₂ ⁻:H₂O₂ ratio confers prosurvival advantage and resistance to chemotherapeutic agents.

Pharmacological manipulation of redox status results in either sensitivity to drug-induced apoptosis (reduced O₂ ⁻:H₂O₂) or resistance (elevated O₂ ⁻:H₂O₂)

To quantify the O₂ ⁻:H₂O₂ ratios of various drug-treated GPCs, we introduced the ROS Index, a log transformation of absolute values of O₂ ⁻:H₂O₂ as measured by the MitoSOX^™ Red:Amplex^® Red percentages (Fig. 2A, Equation 2.1). The normalized ROS Index is represented by Equation 2.2 (Fig. 2A). To specifically implicate O₂ ⁻:H₂O₂-mediated chemoresistance, we utilized two well-established pharmacological agents: diphenylene iodonium (DPI) to deplete intracellular O₂ ⁻, thereby decreasing the ROS Index to below zero; or diethyl dithiocarbamate (DDC) to elevate intracellular O₂ ⁻, thereby increasing the ROS Index to above zero. DPI has frequently been used to inhibit ROS production mediated by flavoenzymes, particularly reduced nicotinamide adenine dinucleotide phosphate (NADPH) oxidase. At concentrations that inhibit NADPH oxidase (this study), DPI diminishes the production of O₂ ⁻ by mitochondrial respiration (34). We confirmed this mechanism by measuring the O₂ ⁻ levels in the presence of mitochondrial respiratory chain inhibitors—antimycin, carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone (FCCP), pyruvate, or rotenone (Supplementary Fig. S2E). DDC is an efficient inhibitor of the O₂ ⁻ scavenger SOD (28). Consistent with the formula we proposed in Figure 2A, increasing DPI treatment dose-dependently decreased the ROS Index, while increasing DDC treatment dose-dependently increased the Index of both the CD133(+) and CD133(−) fractions (Fig. 2B). As cell death by reductive stress (reduction of ROS Index) occurs primarily through apoptosis (50), we also assessed the levels of cleaved poly(ADP-ribose) polymerase (PARP)-expressing cells after treatments with DPI or DDC, followed by CIS or VP-16. DPI or DDC treatments alone did not affect cell viability (<5%), thus ruling out nonspecific toxicity (Supplementary Fig. S2F). Depletion of the O₂ ⁻:H₂O₂ ratio sensitized both the CD133(+/−) fractions to drug-induced apoptosis, compared to the drug-alone treatment (Fig. 2C). Conversely, DDC treatment provided additional protection against drug-induced apoptosis, compared to drug-alone treatment, although reduction in apoptotic cell levels did not reach those in the dimethyl sulfoxide (DMSO) cells, likely because tumor cell resistance may involve other signaling pathways in addition to redox modification (Fig. 2D). Finally, the ROS Index correlated with the level of drug-induced apoptosis in a linear manner (Fig. 2E), suggesting that the extent of drug-induced apoptosis could possibly be predicted by the relative O₂ ⁻:H₂O₂ ratio within the tumor cells. Figure 2E also demonstrates that this redox mechanism affected both CD133(+) and CD133(−) cells equally, suggesting the targeting of both GPCs and bulk tumor cells.

FIG. 2.

Extent of apoptosis correlates with O₂ ⁻ :H₂O₂ ratio (see also Supplementary Fig. S2). (A) ROS Index is represented in Equation 2.1. A normalized ROS Index is represented in Equation 2.2. (B) ROS Indices were calculated after evaluating the O₂ ⁻:H₂O₂ ratios of DPI- and DDC-treated cells. *p<0.05; **p<0.01; ***p<0.001 (*, **, ***; vs. DMSO control). (C) DPI treatment sensitized GPCs to apoptotic triggers by cisplatin and VP-16. Both CD133(+) and CD133(−) fractions were sensitized to apoptosis after DPI + drug addition, displaying increased cleaved PARP(+) cells compared to the DMSO, DPI, or drug-alone groups. *p<0.05; **p<0.01 (*, **; vs. DMSO control); ⁺ p<0.05; ⁺⁺ p<0.01 (+, ++; vs. corresponding drug-alone group). (D) DDC treatment enhanced resistance of GPCs to chemotherapeutic agents, cisplatin, and VP-16, resulting in decreased cleaved PARP(+) cells. The extent of enhancement was not comparable to the initial DMSO control, most likely indicating that additional signaling mechanisms besides O₂ ⁻:H₂O₂ elevation are needed to confer the resistance phenotype. *p<0.05; **p<0.01 (*, **; vs. DMSO control); ⁺ p<0.05; ⁺⁺ p<0.01 (+, ++; vs. corresponding drug-alone group). (E) Linear relationship between normalized ROS Index and extent of apoptosis triggered by (i) VP-16 and (ii) cisplatin treatments illustrates utility of quantified O₂ ⁻:H₂O₂ ratios at predicting the susceptibility to apoptotic triggers. DDC, diethyl dithiocarbamate; DPI, diphenylene iodonium; PARP, poly(ADP-ribose) polymerase; ROS, reactive oxygen species.

Mitochondrial SOD2 lentiviral overexpression and knockdown correlate with DPI and DDC treatments, respectively

A potent source of ROS generation is the mitochondrial electron transport chain, where O₂ ⁻ and H₂O₂ are the major species, and interconvertible via the MnSOD (SOD2) enzyme (7). Coupled to SOD2 is the large body of work highlighting its tumor suppressor activity where lower expression and the inability to induce its expression upon oxidant challenge are invariably observed in tumor cells (14, 47, 69, 72). The obvious consequence of the deficiency/imbalance in the enzymatic activity of SOD2 is the build-up of the O₂ ⁻ anion, thereby altering the cellular redox status and favoring tumor cell survival. Further, re-expression of SOD2 reverses the invasive phenotype (15, 47, 69, 72). We thus focused on the genetic manipulation of SOD2 to specifically implicate its role in the interconversion between O₂ ⁻ and H₂O₂. We constructed a doxycycline (dox)-inducible SOD2 lentiviral vector for overexpression (Fig. 3A), or utilized pLKO.1-based lentiviral shSOD2 knockdown vectors (Fig. 3E). Parameters for SOD2 overexpression or knockdown were earlier optimized by verifying the effects on the normalized ROS Index (Supplementary Fig. S3A–C). Overexpression or knockdown of SOD2 in two GPC lines after stable clone selection was verified by immunoblot analyses and SOD2 activity assay (Figs 3B, F and Supplementary Fig. S3D). NNI-8 and NNI-11 were chosen to profile, as they formed orthotopic tumors at experimentally feasible timelines (∼3–6 months). As expected, SOD2 overexpression resulted in depletion of O₂ ⁻ in both the CD133(+) and CD133(−) fractions (Fig. 3C, top panel), while H₂O₂ levels were elevated, consistent with the role of SOD2 as the interconverting enzyme between the two species (Fig. 3C, bottom panel). Conversely, SOD2 knockdown led to elevation of O₂ ⁻ levels (Fig. 3G, top panel), while H₂O₂ levels were reduced (Fig. 3G, bottom panel). The resulting normalized ROS Indices were calculated in Figure 3D and H, compared to the values derived from DPI and DDC treatments, respectively. No significant variation was observed between pharmacologically generated O₂ ⁻:H₂O₂ ratios and the SOD2-mediated ratios. These data confirm that lentiviral-mediated dox-inducible SOD2 overexpression and pLKO.1-based knockdown occurred at levels representing earlier DPI and DDC treatments, respectively, for which variations in chemoresistance were observed.

FIG. 3.

Overexpression and knockdown of mitochondrial superoxide dismutase 2 ( SOD2 ) in GPCs (see also Supplementary Fig. S3). (A) SOD2 overexpression vector construct. SOD2 is driven by the dox-inducible promoter; eCFP enables visualization and quantification of transduction efficiency. (B) SOD2 overexpression. Overexpression efficiency was measured. Densitometric values of SOD2 protein normalized to β-actin are shown from the representative immunoblot. (C) O₂ ⁻ (top panel) and H₂O₂ (bottom panel) levels were measured after SOD2 overexpression at levels that mimicked pharmacological depletion of O₂ ⁻ by DPI (Supplementary Figure S3B). *p<0.05; **p<0.01. (D) O₂ ⁻ and H₂O₂ levels in (C) were quantified and represented as ROS Indices normalized to their respective DOX controls. An Index<0 indicates anticipated tumor cell sensitivity to chemotherapeutic agents. Dox-mediated SOD2 overexpression was comparable to DPI treatment (no significant p-values). The ratio of 1:3 (NNI-8) or 1:4 (NNI-11) represents the pLVX-Tet-On Advanced vector to pLVX–Tight-Puro-SOD2-IRES-eCFP plasmid ratio, previously optimized to yield similar ROS Index values to DPI treatment (Supplementary Fig. S3B). (E) SOD2 knockdown vector construct. A pLKO.1-based shSOD2 knockdown vector was utilized. (F) SOD2 knockdown. Knockdown efficiency was measured. Densitometric values of SOD2 protein normalized to β-actin are shown from the representative immunoblot. (G) O₂ ⁻ (top panel) and H₂O₂ (bottom panel) levels were measured after SOD2 knockdown at levels that mimicked pharmacological elevation by DDC (Supplementary Fig. S3C). *p<0.05; **p<0.01. (H) O₂ ⁻ and H₂O₂ levels in (G) were quantified and represented as ROS Indices normalized to their respective nontargeting (NT) controls. An Index >0 indicates anticipated tumor cell resistance to chemotherapeutic agents. pLKO.1-based shSOD2 knockdown was comparable to DDC treatment (no significant p-values). dox, doxycycline.

SOD2 overexpression sensitizes GPCs to drug-induced apoptosis, while SOD2 knockdown confers a prosurvival signal

To mimic the effects of DPI treatment, we lentivirally overexpressed SOD2, which would deplete O₂ ⁻. We then subjected the cells to CIS or VP-16 treatments. As expected, both NNI-8 and NNI-11 GPCs, regardless of the CD133 status, exhibited enhanced sensitivity to the chemotherapeutic agents (Fig. 4A). GPC sphere-forming capacity was decreased, with concomitant reduction in sphere size (Fig. 4B). Enhanced apoptotic levels were detected by measuring cleaved PARP-expressing cells (Fig. 4C). This suggests that GPCs can be sensitized to drug-induced apoptotic triggers by tilting the redox status balance toward H₂O₂ production (i.e., depleting the O₂ ⁻:H₂O₂ ratio).

FIG. 4.

SOD2 over expression and knockdown display phenotypic changes in GPCs. (A) SOD2 overexpression reduced GPC frequency. Dox-mediated SOD2 overexpression followed by drug treatment sensitized GPCs to cell death, resulting in reduction of spheres formed. *p<0.05; **p<0.01; ***p<0.001 (*, **, ***; vs. DMSO control); ⁺ p<0.05; ⁺⁺ p<0.01 (+, ++; vs. corresponding drug-alone group). (B) SOD2 overexpression reduced GPC proliferation, measured by the decreasing sphere size. (C) SOD2 overexpression resulted in increased apoptosis of GPCs to chemotherapeutic agents, cisplatin, and VP-16. *p<0.05; **p<0.01; ***p<0.001 (*, **, ***; vs. DMSO control); ⁺ p<0.05; ⁺⁺ p<0.01 (+, ++; vs. corresponding drug-alone group). (D) SOD2 knockdown increased the GPC frequency. SOD2 knockdown followed by drug treatment enhanced GPC resistance to cell death, resulting in increased sphere formation (see drug + NT compared to drug + shSOD2). *p<0.05; **p<0.01 (*, **; vs. NT control). (E) SOD2 knockdown increased GPC proliferation, measured by increasing sphere size. (F) SOD2 knockdown resulted in enhanced resistance and consequently reduced apoptosis of GPCs to chemotherapeutic agents, cisplatin, and VP-16. *p<0.05; **p<0.01; ***p<0.001 (*, **, ***; vs. NT control).

We next assessed the outcome of SOD2 knockdown that would shift the balance from H₂O₂ toward O₂ ⁻ production. The sphere-forming capacity was increased in both GPC lines regardless of the CD133 status (Fig. 4D). GPC proliferation was increased (Fig. 4E), and the level of apoptosis (Fig. 4F) was reduced. Our data indicate that SOD2 knockdown results in the production of O₂ ⁻, resulting in prosurvival/elevated O₂ ⁻:H₂O₂. We implicate mitochondrial SOD2 as a mediator of intracellular O₂ ⁻ and H₂O₂ production.

A reduced cellular redox status and intracellular pH implicate reductive stress and PI3K/AKT signaling

Earlier work elucidated the mechanism of H₂O₂-induced apoptosis (12). Cells exposed to apoptotic concentrations of H₂O₂ displayed a significant decrease in intracellular O₂ ⁻, which was associated with a reduction of the cytosolic milieu, as defined by an increase in the reduced glutathione (GSH)/oxidized glutathione disulfide (GSSG) ratio and a decrease in intracellular pH. In contrast, cells exposed to necrotic H₂O₂ concentrations exhibited a significant reduction of the GSH/GSSG ratio accompanied by cytosolic alkalinization. Accordingly, to distinguish the mechanism of reductive stress from oxidative damage in our GPCs, we assessed three parameters: intracellular pH, cellular redox potential, and the redox status [measured by GSH/GSSG, nicotinamide adenine dinucleotide phosphate (NADP)/NADPH ratios, and thioredoxin reductase activity]. We observed in all GPC lines that DPI treatment resulted in cytosolic acidification (pH∼4.5–6.0) with a concomitant increase in the GSH/GSSG ratio, an environment conducive for apoptosis, in contrast to necrotic concentrations of H₂O₂ (>10 mM H₂O₂) (Fig. 5A, Bi) (12). Upon DDC treatment, the intracellular pH rose to ∼7.0–7.5 with a reduced GSH/GSSG ratio. The enzyme glutathione disulfide reductase catalyzes an equilibrium that greatly favors formation of reduced GSH (40). Thus, the intracellular microenvironment is normally in a reducing state, with most GSH present in the cell in the thiol form. Any situation that leads to the depletion of intracellular O₂ ⁻:H₂O₂ may be expected to augment the baseline cellular reducing state and therefore the GSH/GSSG ratio. Conversely, metabolic changes that result in enhanced O₂ ⁻ production will lower the GSH/GSSG ratio (24). The reduced cellular redox environment upon DPI treatment was further corroborated by the determination of the half-cell reduction potential of the 2GSH/GSSG couple using the Nernst Equation (Fig. 5Bii) (56). To further affirm the intracellular redox system, two additional independent redox systems, namely, the NADP/NADPH ratio and the thioredoxin reductase 2 (TrxR2) activity, were ascertained. Our data showed that DPI treatment significantly reduced, while DDC treatment significantly elevated the NADP/NADPH ratio (Fig. 5Biii), consistent with cell death through reductive stress. TrxRs have been implicated in playing a role in protecting against oxidative injury, cell growth, and transformation (2, 61). We now observe that DDC treatment elevated TrxR2 activity, consistent with its tumor cell prosurvival role, while DPI diminished TrxR2 activity (Fig. 5Biv). Collectively, our data strongly suggest that depleting the O₂ ⁻:H₂O₂ ratio creates a reduced environment conducive for drug-induced apoptosis.

FIG. 5.

Intracellular pH changes and GSH/GSSG ratios implicate reductive stress mechanism. (A) DPI resulted in cytosolic acidification, a process that implicates cell death by reductive stress (as opposed to alkalinization in oxidative stress). In contrast, DDC elevated cytosolic pH. *p<0.05; **p<0.01 (*, **; vs. DMSO control). (B) (i) GSH/GSSG ratio implicates reductive stress. DPI treatment raised the GSH/GSSG ratio compared to the DMSO control, while DDC treatment reduced the ratio. This is consistent with DPI establishing a reduced microenvironment conducive for apoptotic triggers, hence reductive stress. In contrast, necrotic concentrations of H₂O₂ (>10 mM) led to a reduced GSH/GSSG ratio. (ii) A reduced cellular redox potential is implicated by a reduced O₂ ⁻:H₂O₂ ratio. The Nernst equation was used to calculate the cellular redox potential. In contrast, necrotic H₂O₂ concentrations resulted in a more oxidized state. (iii) A reduced O₂ ⁻:H₂O₂ ratio (DPI treatment) diminished the NADP/NADPH ratio. (iv) A reduced O₂ ⁻:H₂O₂ ratio diminished TrxR2 activity. *p<0.05; **p<0.01; ***p<0.001 (*, **, ***; vs. DMSO control). GSH, glutathione; GSSG, glutathione disulfide; NADP, nicotinamide adenine dinucleotide phosphate; NADPH, reduced nicotinamide adenine dinucleotide phosphate; TrxR2, thioredoxin reductase 2.

Elevated O₂ ⁻:H₂O₂ status has been associated with activation of phosphatidylinositol-3-kinase (PI3K) signaling in tumors (36, 38, 73), and may play a role here in GPC chemoresistance. Further, our subsequent ROS gene signature significantly enriched for the PI3K/AKT signaling module by GeneGo analysis (Supplementary Fig. S5). A constitutively active AKT2 lentiviral vector was constructed for overexpression, while pLKO.1-based lentiviral shAKT2 vectors were utilized for knockdown experiments. We observed that Akt2 overexpression partially restored viability, sphere-forming activity, and sphere size in DPI-treated cells compared to DMSO Akt2 (Fig. 6A and Supplementary Fig. S4A); conversely, AKT2 knockdown partially reduced viability, sphere formation, and sphere size in the presence of DDC (Fig. 6B and Supplementary Fig. S4B). To further strengthen our finding that an elevated O₂ ⁻:H₂O₂ ratio activates the PI3K/AKT signaling axis, we determined intracellular PI3K activity upon DPI or DDC treatments. We observed that cells treated with DDC displayed significantly elevated PI3K activity compared to cells treated with DPI (Fig. 6C). These data demonstrate a link between the redox status specified by the O₂ ⁻:H₂O₂ ratio and GPC chemoresistance, by implicating a key pathway in gliomagenesis and PI3K-AKT signaling axis (1).

FIG. 6.

Reductive stress mechanism implicates AKT signaling (see also Supplementary Fig. S4). (A) AKT2, an isoform implicated in gliomagenesis (41), was lentivirally overexpressed and transduced in NNI-8 and NNI-11 GPCs. Viability was determined after immunoblot confirmation of protein overexpression. ***p<0.001 versus DMSO mCherry-vector; ⁺⁺⁺ p<0.001 versus DPI mCherry-vector. (B) Lentiviral AKT2 knockdown was confirmed by immunoblot analysis in NNI-8 and NNI-11 GPCs, followed by viability determination. *p<0.05; ***p<0.001 (*, ***; vs. NT control); ⁺⁺⁺ p<0.001 versus corresponding shAKT2(C1) or (C2) clone. (C) DPI treatment resulted in reduced PI3K activity compared to DMSO control. In contrast, DDC treatment elevated PI3K activity. *p<0.05; **p<0.01; ***p<0.001 (*, **, ***; vs. DMSO control). AKT, protein kinase B; PI3K, phosphatidylinositol-3-kinase.

Depletion of intracellular O₂ ⁻:H₂O₂ sensitizes GPCs to drug-induced apoptosis, consequently extending survival

We utilized an orthotopic mouse model of glioma implanted with the following cell groups treated with various chemotherapeutic drugs at their respective IC₅₀ values: CIS, CIS+DPI, CIS+DDC, VP-16, VP-16+DPI, VP-16+DDC, and DMSO control. Significant extension of survival was observed in all mice implanted with NNI-8 treated with DPI+VP-16 (Fig. 7B, p=0.0347), and NNI-11 treated with either DPI + CIS (Fig. 7C, p=0.0173) or DPI+VP-16 (Fig. 7D, p=0.0011), compared to drug-alone (i.e., either CIS or VP-16). No survival advantage was observed in mice implanted with NNI-8 treated with CIS + DPI compared to CIS alone at the time of termination, 300 days (Fig. 7A). In contrast, no consistent trend was observed in mice implanted with CIS/VP-16 + DDC, compared to drug-alone mice, likely because tumor cell proliferation advantage and consequently enhanced in vivo growth involve additional signaling mechanisms other than O₂ ⁻:H₂O₂ elevation alone. We further observed that Nestin- and CD133-expressing cells were significantly diminished in DPI-treated tumors compared to drug-alone tumors (Fig 7E, F), supporting that GPC-like cells were effectively targeted via induction of apoptosis (Fig. 7G).

FIG. 7.

In vivo tumorigenicity after pharmacological manipulation of O₂ ⁻ and H₂O₂ levels. (A–D) DPI-sensitized GPCs to chemotherapeutic agents (except NNI-8 cisplatin): Cisplatin and VP-16 in NNI-11 and VP-16 in NNI-8, resulting in extended survival (p<0.05 vs. corresponding drug alone). In contrast, only moderate differences (p>0.05) were observed between the groups: DMSO control, drug alone, or DDC-drug (except for NNI-8 cisplatin), suggesting likely that other mechanisms beside the redox status are needed to confer the resistance and tumorigenic phenotype. (E, F) DPI-treated tumors were consistently depleted of (E) Nestin-expressing cells and (F) CD133-expressing cells compared to drug-alone tumors, while DDC treatment generated no consistent trend compared to the DMSO or drug-alone groups. These data suggest that Nestin- and CD133-expressing neural precursors can be targeted by reductive stress. No tumors could be obtained from NNI-8 cisplatin and DPI-cisplatin animals, hence precluding analysis. *p<0.05; **p<0.01; ***p<0.001 (*, **, ***; vs. DMSO control); ⁺ p<0.05; ⁺⁺ p<0.01; ⁺⁺⁺ p<0.001 (+, ++, +++; vs. drug-alone tumors). (G) DPI-treated tumors showed increased apoptosis compared to drug-alone tumors. In contrast, no consistent trend was observed between the DMSO, DDC, and drug-alone tumors. *p<0.05; **p<0.01 (*, **; vs. DMSO control); ⁺ p<0.05; ⁺⁺ p<0.01; ⁺⁺⁺ p<0.001 (+, ++, +++; vs. corresponding drug-alone tumors.

Elevated O₂ ⁻:H₂O₂ ratio identifies patients with poor prognosis, correlating with a higher tumor grade and the mesenchymal molecular subtype

We utilized the Connectivity Map (31) to interrogate the strength of association between an in vitro derived ROS gene signature, and individual patient's gene expression profile in two large, independent glioma datasets. We previously successfully used the Connectivity Map to correlate the tumor suppressor function of Parkin with patient survival outcome (71), as well as associated GPC subtypes with major glioma variants (44). Differentially regulated genes from the union of resistant CD133 fractions and DDC versus DPI fractions (ROS-resistance gene signature) were mapped onto REMBRANDT (39) and Gravendeel (25) (Supplementary Table S1 and Supplementary Fig. S6, additional clinical databases—Freije and Nutt). The ROS signature stratified patient survival independently of the current clinical indicators, age and histology (Fig. 8Ai, Bi and Supplementary Tables S2–S4), suggesting that ROS activation pathways contribute to glioma molecular heterogeneity. ROS(+) patients correlated with higher tumor grade (mainly GBM) and the Mesenchymal molecular classification with predominance of astrocytic cells, a molecular feature consistent with GBM recurrence (8, 53) (Fig. 8Aii, Bii; Supplementary Tables S5 and S6). In contrast, ROS(−) patients displayed lower tumor grades (mainly oligodendroglioma) and the proneural feature (oligodendroglial and neuronal cells). Gene Set Enrichment Analysis (GSEA) revealed that ROS(+) patients were enriched in Myc and embryonic stem cell (ESC) programs, while ROS(−) patients showed features of oligodendrocytes and neurons (Fig. 8Aiii, Biii; Supplementary Fig. S7 and Supplementary Table S7). Interestingly, oligodendroglial cells were implicated as the more sensitive cells to temozolomide (TMZ) compared to astrocytes or neural stem cells (NSCs) in a mouse model of oligodendroglioma (49). Our data provide clinical evidence for the ROS hypothesis.

FIG. 8.

ROS gene signature stratifies patient survival and is enriched in patients with poor prognosis (see also Supplementary Figs. S5–S7). (Ai, Bi) Kaplan–Meier survival plots are shown in two major, independent glioma databases, REMBRANDT and Gravendeel. Red, ROS(+) describes a positive association of the gene signature with patient gene expression data, while blue, ROS(−) represents an inverse correlation. The ROS signature represents enrichment of resistance genes over the sensitivity genes [i.e., DDC vs. DPI and CD133(+) in NNI-4, 8, 11; CD133(−) in NNI-1 vs. other fraction]. (Aii, Bii) Tumor grade and Phillips molecular classification of primary tumors (i.e., proneural, proliferative, and mesenchymal) are illustrated beneath the activation score graphs. ROS(+) patients correlated with higher tumor grades and the mesenchymal feature. ROS(−) patients correlated with lower tumor grades and the proneural feature. (Aiii, Biii) Heatmaps illustrating the association of the weighed score computed from Cahoy signature (8) and ROS classes (+and −). ROS(+) patients were enriched in the cultured astroglia cell type while ROS(−) patients were enriched in the neuronal and oligodendrocytic cell type.

Discussion

ROS molecules mediate different roles in cancer and stem cell biology. Diehn et al. demonstrated that breast CSCs possess reduced H₂O₂ levels, thereby mitigating their response to oxidative damage (16). Additionally, work in oligodendrocyte precursor cells showed that the more reduced cells exhibited enhanced self-renewal properties, whereas cells that were relatively oxidized had a higher probability of differentiating into nondividing oligodendrocytes (46). The common central theme of ROS till then is that at excessive levels, and it is harmful and is associated with oxidative damage, a rather nondiscriminatory event between different cell types. In contrast, recent work by Le Belle et al. has illuminated the seemingly reverse, prosurvival role of ROS in normal NSCs (32). Accordingly, exogenous agents that elevated ROS levels increased production of neurospheres, a key assay for NSC activity. Such findings emphasize that analysis of cell signaling function purely in terms of phosphorylation cascades, transcriptional regulation etc. provides only partial understanding of the means by which signaling regulates the cellular precursor function, and indicate that the roles of redox modulation are quite specific, lending support to the idea that ROS molecules may not just act as mere cofactors, but rather the redox state can act as a specific regulator of stem/progenitor cell function. We have specifically implicated an elevated O₂ ⁻:H₂O₂ ratio, both pharmacologically and by genetically overexpressing/knocking down SOD2, as a GPC tumor cell prosurvival signal, while depleting this ratio sensitizes cells to apoptotic triggers, thus lending support to its role in mediating the tumor phenotype (73). We recognize that there may be other intracellular sources such as NADPH oxidase contributing to varying the O₂ ⁻:H₂O₂ ratio that remains to be explored (50). In addition, the prosurvival O₂ ⁻:H₂O₂ ratio may trigger the proliferative pathways such as extracellular signal-regulated kinase (ERK) in addition to AKT, supported by our GeneGo analysis enriched for the V-Ki-ras2 Kirsten rat sarcoma viral oncogene homolog (KRAS) network. Our study does not contradict previous work, but instead extends on the roles of different ROS molecules—oncogenic O₂ ⁻ versus the apoptotic H₂O₂ at a slight prooxidant intracellular microenvironment where oxidative damage has not yet come into play. Importantly, the linear correlation between the normalized ROS Index and the extent of apoptosis we observed, regardless of patient cell line, marker expression, or drug [further, the NNI-1-resistant fraction was CD133(−)], strongly supports that heterogeneity in chemoresistance is a manifestation of differing O₂ ⁻:H₂O₂ ratios. This is a significant advance to a previously largely qualitative ROS field. The utility of this quantitative and reproducible ROS Index for drug screening thus holds promise for the evaluation of drug candidates targeting the O₂ ⁻:H₂O₂ ratio.

Although we did not detect tumor cell proliferation advantage due to an elevated O₂ ⁻:H₂O₂ ratio (DDC treatment) in vivo, and many reasons could have contributed to this, such as the inadequacy of xenograft models in immune-compromised mice to model the tumorigenic process, nevertheless, the significance of the ROS Index was further validated in a clinical setting when we mapped ROS pathway activation scores in two large, public glioma databases, REMBRANDT (N=298) (39) and Gravendeel (N=276) (25). We recognize that these glioma databases consist largely of patients treated with TMZ concomitant with radiotherapy and adjuvant TMZ as the standard care for initial treatment of GBM—the Stupp regimen (59). In some cases, the supplemental regimen included platinum/thalidomide/tamoxifen/CPT11/vincristine/cyclophosphamide. The high in vitro concentrations of TMZ (250–2000 μM) preclude their use in our earlier analyses (4). CIS and etoposide are clinically relevant investigational tools because of their expanded roles in the TMZ and bevacizumab era (9), and their mechanism as radiomimetics, that is, induce DNA double-strand breaks (57, 67). Our ROS gene signature enriched for the KRAS and PI3K/AKT signaling networks, both of which are implicated as central pathways mutated in ∼88% of GBM patients (1). This provides clinical evidence for our in vitro observations that demonstrated that the interconversion between the O₂ ⁻ and H₂O₂ species involved at least in part the PI3K/AKT signaling axis, a key pathway in gliomagenesis (1). Previous efforts by others have uncovered a role for NAPDH oxidase enzymes in the regulation of normal NSC proliferation and self-renewal in a PI3K-AKT-FoxO-dependent manner (32, 48). Our work extends by providing evidence that other major mitochondrial-based mechanisms can exert specific tumor-proliferative, and, in this case, GPC survival effects by acting on core signaling pathways such as PI3K/AKT. Other convincing evidence linking a slight prooxidant intracellular milieu with tumor promotion comes from studies linking cell survival pathways, such as the PI3K/AKT (54) and RAS (29, 51) circuits, with an increase in intracellular ROS generation.

Strikingly, patients who demonstrate strong concordance with an elevated O₂ ⁻:H₂O₂ ratio fared poorly, while patients with the inverse correlation tended to have a better prognosis. To shed more light into possible target pathways of the ROS Index-driven mechanism, we utilized genome-wide transcript profiles of ROS-resistant patient classes from REMBRANDT and Gravendeel to determine enrichment for the NFκB and HIF1α molecular gene sets identified from the Molecular Signatures DataBase (MSigDB), for the reason that these transcription factors are frequently regulated by the redox status (6), and whose expression of dependent genes promotes the glioma phenotype (35, 66). We identified that, from Supplementary Table S8, indeed, the NFκB- and HIF1α-related pathways are enriched in ROS(+) patients (poorer prognosis), implying that these master transcription factors could also possibly influence the generation of the malignant phenotype. Further, these regulatory modules are also enriched in the CD133(+) class of patients compared to the CD133(−) group (as defined by the CD133 signature), lending support to the importance of CD133 as a negative prognostic marker in glioma (Supplementary Table S9) (5).

The ROS(+) class of patients was enriched in the Myc and ESC programs. Clement et al. recently implicated an ESC signature in the progression and malignancy of glioblastoma tumors (13). Additionally, Shats et al. showed that the core ESC programs defined poor prognosis in patients with breast cancer, lung adenocarcinoma, and medulloblastoma (58). Further, c-Myc has been shown to promote the survival of CD133(+) GPCs and consequently tumorigenic growth (68). These findings highlight that Myc and ESC core modules play essential roles in mediating the O₂ ⁻:H₂O₂-driven chemoresistance of GPCs. Interestingly, our Cox regression analysis indicated that the gene signature prognosticated survival better than the current clinical indicators, age and grade, implying that a combination of stem-like traits and redox status is vital to disease progression and survival outcome, and importantly, contribute to the molecular heterogeneity of the disease. These findings highlight the ROS resistance gene signature as a novel molecular predictor of glioma disease outcome that cannot be accounted for by histology alone, and thus underscores the limitation of relying solely on morphology-based methods to diagnose and subsequently treat patients. Further, we are now able to identify patient cohorts (with their corresponding genetic, transcriptomic patterns, and clinical history) most likely to benefit from ROS Index-related small-molecule therapeutics. Although our ROS gene signature came from a limited number of GPC lines as with all studies till date, we leveraged the unavoidable small sample size with cross-validation in four large, independent patient databases. Thus, clinically amenable applications may be developed by profiling unsorted bulk tumor cells, because chemoresistance and disease progression are manifestations of the activation of redox–GPC pathways.

In summary, our work provides strong evidence for the clinical relevance of the ROS Index. We further suggest that redox-based therapies in brain tumors can be designed to favor a reduced O₂ ⁻:H₂O₂ ratio, thereby sensitizing tumor cells, and especially GPCs, to apoptotic triggers (proposed model, Supplementary Fig. S8).

Materials and Methods

Tissue collection and primary GPC culture

Graded brain tumor specimens were obtained with informed consent, as part of a study protocol approved by the SingHealth Centralised Institutional Review Board A. GPC lines were described in our previous work (10, 19, 44). All experiments were conducted with low-passage GPCs (within 20 passages) for which we previously demonstrated maintenance of the phenotypic, transcriptomic, and karyotypic features similar to the primary tumor.

Viability and sphere assays; antibodies used

Viability and sphere assays were carried out as previously described (10). Antibodies used for immunoblot analyses included anti-pan-Akt (11E7, 1:1000; Cell Signaling, #4685), anti-MnSOD (1:1000; Upstate, Millipore, 06-984), and anti-β-actin (AC-15, 1:10,000; Sigma-Aldrich, A5441). Antibodies for flow cytometric analyses included anti-CD133/2-allophycocyanin (Clone 293C, 1:10; Miltenyi MACS, 130-090-854); and anti-cleaved PARP conjugated to phycoerythrin (Asp214, 1:5; BD Pharmingen, 552933). Dead cells were distinguished by 7-aminoactinomycin D (7-AAD; BD Pharmingen, 559925) staining. Flow analyses were carried out on the FACSCalibur (BD Biosciences); flow sorting was carried out using the FACSAria (BD Biosciences). Data were plotted using FlowJo software (Treestar).

Detection of O₂ ⁻ and H₂O₂

Cells treated with DPI or DDC for 1 h (viability unaffected; Supplementary Fig. S2E) were dissociated with Accutase (eBioscience) and stained with the Amplex^® Red (Invitrogen) and MitoSOX^™ Red (Invitrogen) reagents, which measure H₂O₂ and O₂ ⁻, respectively. Amplex^® Red and MitoSOX^™ Red dyes were both detected using the FL-2 channel on FACSCalibur (BD Biosciences), with excitation of 488 nm and emission wavelengths of 585 nm for Amplex^® Red and 580 nm for MitoSOX^™ Red.

Measurement of intracellular pH with 2′,7′-bis(2-carboxyethyl)-5,6-carboxyfluorescein

Intracellular pH (pHi) was measured by loading cells with a membrane-impermeant dye 2′,7′-bis(2-carboxyethyl)-5,6-carboxyfluorescein (BCECF; Sigma-Aldrich) as described elsewhere (42). Cells were analyzed on FACSCalibur (BD Biosciences) with excitation of 488 nm. A minimum of 10,000 events was analyzed, and the ratio of BCECF fluorescence at 525 and 610 nm was used to obtain the pHi from a pH calibration curve. To generate a pH calibration curve, GPCs were loaded with BCECF as above, washed once with HBSS, and then resuspended in a high K⁺ buffer (135 mM KH₂PO₄, 20 mM NaCl, and 110 mM KH₂PO₄, and 20 mM NaCl with a range of pH between 6.0 and 8.0). Immediately before flow cytometry, cells were loaded with 20 mM nigericin (1 mM stock in absolute ethanol; Sigma-Aldrich), and the fluorescence ratio measurements (525 nm/610 nm) of cells in nigericin-containing buffers of a range of pH were then used to relate the histogram channel numbers to pHi.

Determination of the cellular redox state—GSH/GSSG measurement

Analysis of cellular GSH after manipulation of intracellular redox state was determined using the recycling assay of the Glutathione Assay kit (Cayman Chemical Co.). The reduced/oxidized glutathione ratio (GSH/GSSG) was calculated using the formula [(GSHt − GSSG)/GSSG].

Determination of the cellular redox state—NADP/NADPH measurement

Colorimetric NADP/NADPH assays were performed after manipulation of the intracellular redox state using a commercially available kit from Abcam (ab65349), according to the manufacturer's protocols. The NADP/NADPH ratio was quantified using the sample's OD_450nm reading after 4 h using the formula, [(NADP_total − NADPH)/NADPH].

Determination of the cellular redox state—TrxR2 measurement

TrxR2 activity in redox-manipulated GPCs was determined using the Thioredoxin Reductase Colorimetric Assay Kit (Cayman Chemical Co.), which is based on the reduction of 5,5′-dithiobis-2-nitrobenzoic acid (DTNB) by NADPH to yield thionitrobenzoate (TNB) that is detectable at 412 nm.

SOD2 overexpression and shSOD2 vector construction

The pLVX-Tight-puro-SOD2-IRES-eCFP overexpression vector was generated using the Lenti-X^™ Tet-On^® Advanced Inducible Expression System (Clontech). The full-length SOD2 cDNA was PCR-amplified, digested with Bglll and BamHI, and subsequently ligated into the pYIC vector (45) (Addgene plasmid 18673). Blunt ends of the resultant SOD2-IRES-eCFP fragment into the pLVX-Tight-puro vector (Clontech) were ligated, and resultant plasmids were fully sequenced. Lentiviral particles were generated as per the manufacturer's protocol. Inducibility and viral cotransduction ratio between the regulator pLVX-Tet-On Advanced vector and the response pLVX-Tight-Puro-Luc vector were optimized using the luciferase assay system (Promega) (Supplementary Fig. S3B, top panel). A range of ratios was selected, and GPCs were cotransduced with the pLVX-Tet-On Advanced vector and the pLVX-Tight-Puro-SOD2-IRES-eCFP vector (X:Y ratio, respectively). The ratio that best mimicked that of the DPI ROS Index was determined by flow cytometry (Supplementary Fig. S3B, middle and bottom panels). Cotransduced GPC lines were maintained under 2 μg/ml puromycin and 250 μg/ml geneticin (Gibco) selection. The overexpression efficiency of SOD2 was verified by immunoblot analysis (Fig. 3B) and the superoxide dismutase assay kit (Supplementary Fig. S3D, left panel) (Cayman Chemical).

Lentiviral-compatible shRNA clones were obtained from the Open Biosystems pLKO.1 shRNA library against SOD2 (TRCN0000005941 [shSOD2 (C1)] and TRCN0000005943 [shSOD2 (C2)]). Lentiviral particles were generated as described above. GPCs were transduced with the shRNA-bearing lentiviral particles and analyzed via flow cytometry to establish the ratio that best mimicked that of the DDC ROS Index (Supplementary Fig. S3C). The knockdown efficiency was determined by immunoblot analysis (Fig. 3F) and the superoxide dismutase assay (Supplementary Fig. S3D, right panel). Briefly, SOD2 activity was determined using the commercially available Superoxide Dismutase Assay kit (Cayman Chemical), according to the manufacturer's instructions. A tetrazolium salt was utilized for the detection of superoxide radicals generated by xanthine oxidase and hypoxanthine. As this assay measures all 3 forms of SOD (namely, Cu/Zn, Mn, and FeSOD), 3 mM potassium cyanide was incorporated into the assay mixture to inhibit both Cu/Zn-SOD and extracellular SOD, thereby resulting in the sole detection of MnSOD activity. After transduction, cells were harvested and sonicated in a cold buffer containing 20 mM HEPES, (pH 7.2), 1 mM EGTA, 210 mM mannitol, and 70 mM sucrose. The supernatant was then collected after centrifugation at 1500 g for 5 min at 4°C. Twenty microliters of xanthine oxidase was added to each of the wells to initiate the reactions and incubated for 20 min at room temperature. Absorbance of each standard and sample was analyzed spectrophotometrically at 450 nm using the Sunrise^™ microplate reader (Tecan). SOD activity of the samples was subsequently determined using the equation obtained from the linear regression of the standard curve. One unit is defined as the amount of enzyme needed to exhibit 50% dismutation of the superoxide radical.

AKT2 overexpression and shAKT2 vector construction

The pLenti6mCWmyrHAAkt2IRESmCherry overexpression vector was constructed from a modified pLenti6/V5-D-TOPO vector (ViraPower^™ Lentiviral Expression Systems; Invitrogen), named pLenti6mCW. Full-length AKT2 cDNA was PCR-amplified, digested with EcoRV and EcoRI, and subsequently ligated into the pLenti6mCW vector. After, the myrHA tag was PCR-generated, digested with EcoRV and SalI, and ligated into the resultant pLenti6mCWAKT2IRESmCherry plasmid. The consequent plasmid was sequenced. Lentiviral particles were generated as per above. The overexpression efficiency was determined by immunoblot analysis (Fig. 6A). GPCs were transduced with either the pLenti6mCWmCherry vector lentiviral particles or the pLenti6mCWmyrHAAkt2IRESmCherry overexpression vector lentiviral particles at the optimal MOI of 50 for 144 h before cell viability assays and sphere assays and maintained under 1 μg/ml blasticidin selection.

Lentiviral-compatible shRNA clones were obtained from the Open Biosystems pLKO.1 shRNA library against AKT2 (TRCN0000009820 [shAKT2 (C1)] and TRCN0000039971 [shAKT2(C2)]). Lentiviral particles were generated as per above. GPCs were transduced with the shRNA-bearing lentiviral particles, and the knockdown efficiency was determined by immunoblot analysis (Fig. 6B).

PI3-K p85-fast activated cell-based ELISA assay

PI3 Kinase activity of redox-manipulated GPCs was assessed using the fast activated cell-based ELISA (FACE^™) PI3 Kinase p85 ELISA Kit (Active Motif), according to the manufacturer's protocols. Total/phosphorylated PI3 kinase levels were determined via a colorimetric readout at OD₄₅₀ and normalized to the relative cell number in each well through the use of crystal violet absorbance at OD₅₉₅.

Stereotaxic intracranial implantations of NOD-SCID-gamma mice

Animal experimentation was performed according to the protocols approved by the Institutional Animal Care and Use Committee. Implantation was carried out as previously described (10). NOD.Cg-Prkdc^scid Il2rg^tm1Wjl /SzJ mice were used (The Jackson Laboratory). Coordinates were anteroposterior=+1.0 mm; mediolateral=+2 mm; dorsoventral=−2.5 mm. Mice were euthanized by means of transcardiac perfusion with 4% paraformaldehyde upon presentation of neurological deficits with ataxia, cachexia, lethargy, or seizure. Animal survival was assessed using the log-rank test in GraphPad Prism software (GraphPad Software, Inc.). Hematoxylin and eosin staining and immunohistochemistry were performed on 5-μm-thick paraffin sections. Tissue sections were stained with mouse monoclonal anti-Nestin antibody (1:500; Chemicon, MAB5326) or rabbit polyclonal anti-CD133 antibody (1:500; Abcam, ab19898) according to the manufacturer's protocol. The extent of apoptosis was determined via terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) assay using the ApopTag^® Peroxidase in situ apoptosis detection kit (Millipore) according to the manufacturer's protocols. A minimum of 200 cells were scored in each of 3 separate fields of view for each specimen.

Statistical analysis

Data are expressed as means±standard error of the mean of at least 3 independent experiments. Student's t- or Mann–Whitney U test was used where appropriate. The one way analysis of variance across multiple groups was evaluated using Kruskal–Wallis test. p≤0.05 was accepted as statistically significant.

Microarray data processing

Raw files from the Illumina BeadScanner were converted to text files using GenomeStudio software, and all downstream processing were conducted using R statistical software and the Bioconductor packages (23, 62). In this study, data from the Human Ref-8v2 beadchips were processed using the Lumi package (17). To process the raw data, a variance-stabilizing transformation was applied, and then probes with the following criteria were removed: (i) detection call ≥0.01 in all samples and (ii) nonexonic probes. After that, the data were quantile normalized, and an ExpressionSet was created using the Biobase package (23) for further analysis in R.

Signature generation and validation

Differentially expressed genes between positive ROS index samples (DDC-treated) and negative ROS index samples (DPI-treated) were generated using the Limma package (22). Probes with a p-value of 0.01 were considered significant, and an ROS index signature was obtained by further filtering of the fold change value >1.3. Subsequently, this set of genes was tested using a supervised classification method (Prediction Analysis of Microarrays) to predict the DDC-treated and DPI-treated samples. The pamr package (27) was utilized to cross-validate the ROS index signature using a leave-one-out method. On the other hand, microarray data of CD133+ and CD133- fractions were used to generate the second gene signature—the CD133 signature. Differentially expressed genes between resistant (NNI-1 CD133−, NNI-4 CD133+, and NNI-8 CD133+) and sensitive (NNI-1 CD133+, NNI-4 CD133−, and NNI-8 CD133−) samples were generated using the samr package (64). Probes with a fold change of 1.3 were considered significant (FDR 0.1). The same cross-validation method mentioned above was applied to the CD133+/− dataset using the CD133 signature. Finally, a combined ROS resistance gene signature was obtained by merging the above 2 optimized gene signatures. All array platform gene annotation was derived from Biomart (www.biomart.org). Raw and processed data are available on the NCBI Gene Expression Omnibus (GEO) Website: www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=dhcfjoqoismsuze&acc=GSE32335

Pathway analysis

Pathway and biological process enrichment analysis were carried out in MetaCore from GeneGo, Inc. The combined ROS signature was used as input for the analysis, and significant pathways (p-value≤0.05) were reported.

Connectivity map analysis

The Connectivity Map analysis (31) was adapted to score the patient samples in glioma gene expression databases according to the extent of pathway activation associated with the ROS signature. Public glioma datasets with clinical data, in terms of survival length, histology, age, and sex, were obtained from either GEO or the database Website. The following datasets were used in this analysis: REMBRANDT (http://caintegrator-info.nci.nih.gov/rembrandt) (39) and Gravendeel (GSE16011) (25). Reference profiles were generated from these two datasets as a means to interrogate patient samples with and without ROS pathway activation.

The ROS resistance gene signature (union of ROS index signature and CD133 signature) was compared to each sample in the reference profiles, and an activation score was calculated using the Connectivity Map approach. This method uses a nonparametric, rank-based pattern-matching procedure to score gene expression profiles based on the strength of association to the ROS signature. The activation scores are then sorted in descending order and two groups of patients identified: (+) and (−), where a positive activation score indicates that the sample is positively associated to the gene signature and vice versa. To assess the significance of the activation scores, two-tailed test p-values were estimated by permutation as described in Lamb et al. (31). p-values≤0.05 were considered significant.

Reference profile generation for Connectivity Map analysis

Public glioma datasets with clinical data, in terms of survival length, histology, grade, and age, were obtained from the REMBRANDT database (https://caintegrator.nci.nih.gov/rembrandt/) (39) and the GEO database in the case of the Gravendeel dataset (GSE16011) (25). To generate the reference profiles, all raw files were processed separately using the mas5 algorithm with the Bioconductor affy package (21). Expression values less than the threshold value of 50 were replaced with the threshold value. Next, the data were quantile normalized, and the gene expression values were row-wise median centered. Probe expression values were median centered, because not all cancer datasets contain matching normal or pooled normal samples that can be used as a reference for comparing the tumor samples. Hence, we have taken the approach of analyzing the range of probe values against the median of all patient samples. We believe that this approach is reasonable, given the large number of samples in the two datasets.

Survival analysis

Kaplan–Meier and Cox regression analysis of (+) and (−) groups were done in R using the survival package (63). For the REMBRANDT dataset, only survival ranges were available. Hence, the lower limit of the range was used in this analysis.

Subtype analysis

Differentially expressed genes in cell types relevant to brain development were obtained from Cahoy et al. (8). Mouse gene sets that are differentially expressed in cultured astroglia, neurons, oligodendrocytes, and astrocytes were converted to human genes using Ensembl. The extent of enrichment of these signatures in the ROS(+) and ROS(−) samples was scored using GSEA (60). In addition, we preselected 181 stem cell-related gene sets and performed GSEA to identify the significantly enriched modules.

Prediction of Phillips classification in REMBRANDT and Gravendeel datasets

To classify the REMBRANDT and Gravendeel samples according to the Phillips et al.'s classification (53), the Affymetrix U133A probes for the Phillips molecular subtypes were extracted from the publication. A shrunken centroid model was trained and tested on the Phillips dataset using the R package pamr (27). Next, the REMBRANDT and Gravendeel datasets were classified using the trained model.

Footnotes

Acknowledgments

The authors acknowledge financial support from competitive grants awarded by the Biomedical Research Council, A*STAR to C.S.L. Tang (07/1/33/19/530, 09/1/33/19/611), and institutional grant funding to B.T. Ang at the Singapore Institute for Clinical Sciences, A*STAR. We are grateful to A/Professor Kah Leong Lim and Dr. Chou Chai, National University of Singapore (Department of Physiology), and Duke-NUS Graduate Medical School, for the generous gift of pLenti6mCWmyrHAAkt2IRESmCherry and pLenti6mCWmCherry vectors. We thank the laboratory members, N.S. Tan of Nanyang Technological University, and Jonathon D. Sedgwick of Lilly Singapore Centre for Drug Discovery for critical review of the manuscript.

Author Disclosure Statement

No competing financial interests exist.

Abbreviations Used

References

Atlas

TCG

. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature, 455: 1061–1068, 2008.

Becker

, Gromer

, Schirmer

, and Muller

. Thioredoxin reductase as a pathophysiological factor and drug target. Eur J Biochem, 267: 6118–6125, 2000.

Beier

, Hau

, Proescholdt

, Lohmeier

, Wischhusen

, Oefner

, Aigner

, Brawanski

, Bogdahn

, and Beier

. CD133(+) and CD133(−) glioblastoma-derived cancer stem cells show differential growth characteristics and molecular profiles. Cancer Res, 67: 4010–4015, 2007.

Beier

, Schulz

, and Beier

. Chemoresistance of glioblastoma cancer stem cells—much more complex than expected. Mol Cancer, 10: 128, 2011.

Beier

, Wischhusen

, Dietmaier

, Hau

, Proescholdt

, Brawanski

, Bogdahn

, and Beier

. CD133 expression and cancer stem cells predict prognosis in high-grade oligodendroglial tumors. Brain Pathol, 18: 370–377, 2008.

Bonello

, Zahringer

, BelAiba

, Djordjevic

, Hess

, Michiels

, Kietzmann

, and Gorlach

. Reactive oxygen species activate the HIF-1alpha promoter via a functional NFkappaB site. Arterioscler Thromb Vasc Biol, 27: 755–761, 2007.

Cadenas

and Davies

. Mitochondrial free radical generation, oxidative stress, and aging. Free Radic Biol Med, 29: 222–230, 2000.

Cahoy

, Emery

, Kaushal

, Foo

, Zamanian

, Christopherson

, Xing

, Lubischer

, Krieg

, Krupenko

, Thompson

, and Barres

. A transcriptome database for astrocytes, neurons, and oligodendrocytes: a new resource for understanding brain development and function. J Neurosci, 28: 264–278, 2008.

Carrillo

and Munoz

. Alternative chemotherapeutic agents: nitrosoureas, cisplatin, irinotecan. Neurosurg Clin N Am, 23: 297–306, ix, 2012.

10.

Chong

, Toh

, Zaiden

, Poonepalli

, Leong

, Ong

, Yu

, Tan

, See

, Ng

, Hande

, Kon

, Ang

, and Tang

. Cryopreservation of neurospheres derived from human glioblastoma multiforme. Stem Cells, 27: 29–39, 2009.

11.

Clement

and Pervaiz

. Intracellular superoxide and hydrogen peroxide concentrations: a critical balance that determines survival or death. Redox Rep, 6: 211–214, 2001.

12.

Clement

, Ponton

, and Pervaiz

. Apoptosis induced by hydrogen peroxide is mediated by decreased superoxide anion concentration and reduction of intracellular milieu. FEBS Lett, 440: 13–18, 1998.

13.

Clement

, Sanchez

, de Tribolet

, Radovanovic

, and Ruiz i Altaba

. HEDGEHOG-GLI1 signaling regulates human glioma growth, cancer stem cell self-renewal, and tumorigenicity. Curr Biol, 17: 165–172, 2007.

14.

Cullen

, Weydert

, Hinkhouse

, Ritchie

, Domann

, Spitz

, and Oberley

. The role of manganese superoxide dismutase in the growth of pancreatic adenocarcinoma. Cancer Res, 63: 1297–1303, 2003.

15.

Darby Weydert

, Smith

, Xu

, Kregel

, Ritchie

, Davis

, and Oberley

. Inhibition of oral cancer cell growth by adenovirusMnSOD plus BCNU treatment. Free Radic Biol Med, 34: 316–329, 2003.

16.

Diehn

, Cho

, Lobo

, Kalisky

, Dorie

, Kulp

, Qian

, Lam

, Ailles

, Wong

, Joshua

, Kaplan

, Wapnir

, Dirbas

, Somlo

, Garberoglio

, Paz

, Shen

, Lau

, Quake

, Brown

, Weissman

, and Clarke

. Association of reactive oxygen species levels and radioresistance in cancer stem cells. Nature, 458: 780–783, 2009.

17.

, Kibbe

, and Lin

. lumi: a pipeline for processing Illumina microarray. Bioinformatics, 24: 1547–8, 2008.

18.

Eramo

, Ricci-Vitiani

, Zeuner

, Pallini

, Lotti

, Sette

, Pilozzi

, Larocca

, Peschle

, and De Maria

. Chemotherapy resistance of glioblastoma stem cells. Cell Death Differ, 13: 1238–1241, 2006.

19.

Foong

, Ng

, Phong

, Toh

, Chong

, Tucker-Kellogg

, Campbell

, Ang

, and Tang

. Cryopreservation of cancer-initiating cells derived from glioblastoma. Front Biosci (Schol Ed), 3: 698–708, 2011.

20.

Galli

, Binda

, Orfanelli

, Cipelletti

, Gritti

, De Vitis

, Fiocco

, Foroni

, Dimeco

, and Vescovi

. Isolation and characterization of tumorigenic, stem-like neural precursors from human glioblastoma. Cancer Res, 64: 7011–7021, 2004.

21.

Gautier

, Cope

, Bolstad

, and Irizarry

. affy—analysis of Affymetrix GeneChip data at the probe level. Bioinformatics, 20: 307–315, 2004.

22.

Gentleman

, Carey

, Dudoit

, Irizarry

, Huber

(eds), and Smyth

. Limma: linear models for microarray data. In: Bioinformatics and Computational Biology Solutions using R and Bioconductor. New York: Springer, 2005, pp. 397–420.

23.

Gentleman

, Carey

, Bates

, Bolstad

, Dettling

, Dudoit

, Ellis

, Gautier

, Ge

, Gentry

, Hornik

, Hothorn

, Huber

, Iacus

, Irizarry

, Leisch

, Li

, Maechler

, Rossini

, Sawitzki

, Smith

, Smyth

, Tierney

, Yang

, and Zhang

. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol, 5: R80, 2004.

24.

Gilbert

. Thiol/disulfide exchange equilibria and disulfide bond stability. Methods Enzymol, 251: 8–28, 1995.

25.

Gravendeel

, Kouwenhoven

, Gevaert

, de Rooi

, Stubbs

, Duijm

, Daemen

, Bleeker

, Bralten

, Kloosterhof

, De Moor

, Eilers

, van der Spek

, Kros

, Sillevis Smitt

, van den Bent

, and French

. Intrinsic gene expression profiles of gliomas are a better predictor of survival than histology. Cancer Res, 69: 9065–9072, 2009.

26.

Halliwell

and Gutteridge

. Role of free radicals and catalytic metal ions in human disease: an overview. Methods Enzymol, 186: 1–85, 1990.

27.

Hastie

, Tibshirani

, Narasimhan

, and Chu

. pamr: Pam: prediction analysis for microarrays. R package version 1.51. http://CRAN.R-project.org/package=pamr. 2011.

28.

Heikkila

, Cabbat

, and Cohen

. In vivo inhibition of superoxide dismutase in mice by diethyldithiocarbamate. J Biol Chem, 251: 2182–2185, 1976.

29.

Irani

, Xia

, Zweier

, Sollott

, Der

, Fearon

, Sundaresan

, Finkel

, and Goldschmidt-Clermont

. Mitogenic signaling mediated by oxidants in Ras-transformed fibroblasts. Science, 275: 1649–1652, 1997.

30.

Jaatinen

, Hemmoranta

, Hautaniemi

, Niemi

, Nicorici

, Laine

, Yli-Harja

, and Partanen

. Global gene expression profile of human cord blood-derived CD133+ cells. Stem Cells, 24: 631–641, 2006.

31.

Lamb

, Crawford

, Peck

, Modell

, Blat

, Wrobel

, Lerner

, Brunet

, Subramanian

, Ross

, Reich

, Hieronymus

, Wei

, Armstrong

, Haggarty

, Clemons

, Wei

, Carr

, Lander

, and Golub

. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science, 313: 1929–1935, 2006.

32.

Le Belle

, Orozco

, Paucar

, Saxe

, Mottahedeh

, Pyle

, Wu

, and Kornblum

. Proliferative neural stem cells have high endogenous ROS levels that regulate self-renewal and neurogenesis in a PI3K/Akt-dependant manner. Cell Stem Cell, 8: 59–71, 2011.

33.

Lee

, Kotliarova

, Kotliarov

, Li

, Su

, Donin

, Pastorino

, Purow

, Christopher

, Zhang

, Park

, and Fine

. Tumor stem cells derived from glioblastomas cultured in bFGF and EGF more closely mirror the phenotype and genotype of primary tumors than do serum-cultured cell lines. Cancer Cell, 9: 391–403, 2006.

34.

and Trush

. Diphenyleneiodonium, an NAD(P)H oxidase inhibitor, also potently inhibits mitochondrial reactive oxygen species production. Biochem Biophys Res Commun, 253: 295–299, 1998.

35.

, Bao

, Wu

, Wang

, Eyler

, Sathornsumetee

, Shi

, Cao

, Lathia

, McLendon

, Hjelmeland

, and Rich

. Hypoxia-inducible factors regulate tumorigenic capacity of glioma stem cells. Cancer Cell, 15: 501–513, 2009.

36.

Lim

and Clement

. Phosphorylation of the survival kinase Akt by superoxide is dependent on an ascorbate-reversible oxidation of PTEN. Free Radic Biol Med, 42: 1178–1192, 2007.

37.

Liu

, Yuan

, Zeng

, Tunici

, Ng

, Abdulkadir

, Lu

, Irvin

, Black

, and Yu

. Analysis of gene expression and chemoresistance of CD133+ cancer stem cells in glioblastoma. Mol Cancer, 5: 67, 2006.

38.

Luo

, Kaur Kumar

, and Clement

. Redox control of cytosolic Akt phosphorylation in PTEN null cells. Free Radic Biol Med, 53: 1697–1707, 2012.

39.

Madhavan

, Zenklusen

, Kotliarov

, Sahni

, Fine

, and Buetow

. Rembrandt: helping personalized medicine become a reality through integrative translational research. Mol Cancer Res, 7: 157–167, 2009.

40.

Meister

. Glutathione metabolism. Methods Enzymol, 251: 3–7, 1995.

41.

Mure

, Matsuzaki

, Kitazato

, Mizobuchi

, Kuwayama

, Kageji

, and Nagahiro

. Akt2 and Akt3 play a pivotal role in malignant gliomas. Neuro Oncol, 12: 221–232, 2010.

42.

Musgrove

and Hedley

. Measurement of intracellular pH. Methods Cell Biol, 33: 59–69, 1990.

43.

Naka

, Hoshii

, Muraguchi

, Tadokoro

, Ooshio

, Kondo

, Nakao

, Motoyama

, and Hirao

. TGF-beta-FOXO signalling maintains leukaemia-initiating cells in chronic myeloid leukaemia. Nature, 463: 676–680, 2010.

44.

, Toh

, Ting

, Koh

, Sandanaraj

, Phong

, Wong

, Leong

, Kon

, Tucker-Kellogg

, Ng

, Tang

, and Ang

. Progenitor-like traits contribute to patient survival and prognosis in oligodendroglial tumors. Clin Cancer Res, 18: 4122–4135, 2012.

45.

Nie

and Htun

. Different modes and potencies of translational repression by sequence-specific RNA-protein interaction at the 5′-UTR. Nucleic Acids Res, 34: 5528–5540, 2006.

46.

Noble

, Smith

, Power

, and Mayer-Proschel

. Redox state as a central modulator of precursor cell function. Ann N Y Acad Sci, 991: 251–271, 2003.

47.

Oberley

. Anticancer therapy by overexpression of superoxide dismutase. Antioxid Redox Signal, 3: 461–472, 2001.

48.

Paik

, Kollipara

, Chu

, Ji

, Xiao

, Ding

, Miao

, Tothova

, Horner

, Carrasco

, Jiang

, Gilliland

, Chin

, Wong

, Castrillon

, and DePinho

. FoxOs are lineage-restricted redundant tumor suppressors and regulate endothelial cell homeostasis. Cell, 128: 309–323, 2007.

49.

Persson

, Petritsch

, Swartling

, Itsara

, Sim

, Auvergne

, Goldenberg

, Vandenberg

, Nguyen

, Yakovenko

, Ayers-Ringler

, Nishiyama

, Stallcup

, Berger

, Bergers

, McKnight

, Goldman

, and Weiss

. Non-stem cell origin for oligodendroglioma. Cancer Cell, 18: 669–682, 2010.

50.

Pervaiz

. Pro-oxidant milieu blunts scissors: insight into tumor progression, drug resistance, and novel druggable targets. Curr Pharm Des, 12: 4469–4477, 2006.

51.

Pervaiz

, Cao

, Chao

, Chin

, and Clement

. Activation of the RacGTPase inhibits apoptosis in human tumor cells. Oncogene, 20: 6263–6268, 2001.

52.

Pervaiz

and Clement

. Superoxide anion: oncogenic reactive oxygen species?. Int J Biochem Cell Biol, 39: 1297–1304, 2007.

53.

Phillips

, Kharbanda

, Chen

, Forrest

, Soriano

, Wu

, Misra

, Nigro

, Colman

, Soroceanu

, Williams

, Modrusan

, Feuerstein

, and Aldape

. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell, 9: 157–173, 2006.

54.

Qin

and Chock

. Implication of phosphatidylinositol 3-kinase membrane recruitment in hydrogen peroxide-induced activation of PI3K and Akt. Biochemistry, 42: 2995–3003, 2003.

55.

Reynolds

and Weiss

. Generation of neurons and astrocytes from isolated cells of the adult mammalian central nervous system. Science, 255: 1707–1710, 1992.

56.

Schafer

and Buettner

. Redox environment of the cell as viewed through the redox state of the glutathione disulfide/glutathione couple. Free Radic Biol Med, 30: 1191–1212, 2001.

57.

Sevim

, Parkinson

, and McDonald

. Etoposide-mediated glioblastoma cell death: dependent or independent on the expression of its target, topoisomerase II alpha?. J Cancer Res Clin Oncol, 137: 1705–1712, 2011.

58.

Shats

, Gatza

, Chang

, Mori

, Wang

, Rich

, and Nevins

. Using a stem cell-based signature to guide therapeutic selection in cancer. Cancer Res, 71: 1772–1780, 2011.

59.

Stupp

, Mason

, van den Bent

, Weller

, Fisher

, Taphoorn

, Belanger

, Brandes

, Marosi

, Bogdahn

, Curschmann

, Janzer

, Ludwin

, Gorlia

, Allgeier

, Lacombe

, Cairncross

, Eisenhauer

, and Mirimanoff

. Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N Engl J Med, 352: 987–996, 2005.

60.

Subramanian

, Tamayo

, Mootha

, Mukherjee

, Ebert

, Gillette

, Paulovich

, Pomeroy

, Golub

, Lander

, and Mesirov

. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A, 102: 15545–15550, 2005.

61.

Sun

, Wu

, Zappacosta

, Jeang

, Lee

, Hatfield

, and Gladyshev

. Redox regulation of cell signaling by selenocysteine in mammalian thioredoxin reductases. J Biol Chem, 274: 24522–24530, 1999.

62.

Team RDC. R: A Language and Environment for Statistical Computing, version 2.10.1 Vienna, Austria: R Foundation for Statistical Computing, ISBN 3-900051-07-0, 2009.

63.

Therneau

. survival: Survival analysis, including penalised likelihood. R package version 2.36-9. http://CRAN.R-project.org/package=survival. Terry Therneau and original Splus->R port by Thomas Lumley, 2011.

64.

Tibshirani

, Chu

, Narasimhan

, and Li

. samr: SAM: Significance Analysis of Microarrays. R package version 2.0. http://CRAN.R-project.org/package=samr. 2011.

65.

Tothova

, Kollipara

, Huntly

, Lee

, Castrillon

, Cullen

, McDowell

, Lazo-Kallanian

, Williams

, Sears

, Armstrong

, Passegue

, DePinho

, and Gilliland

. FoxOs are critical mediators of hematopoietic stem cell resistance to physiologic oxidative stress. Cell, 128: 325–339, 2007.

66.

Verhaak

, Hoadley

, Purdom

, Wang

, Qi

, Wilkerson

, Miller

, Ding

, Golub

, Mesirov

, Alexe

, Lawrence

, O'Kelly

, Tamayo

, Weir

, Gabriel

, Winckler

, Gupta

, Jakkula

, Feiler

, Hodgson

, James

, Sarkaria

, Brennan

, Kahn

, Spellman

, Wilson

, Speed

, Gray

, Meyerson

, Getz

, Perou

, and Hayes

. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell, 17: 98–110, 2010.

67.

Wang

, Zhang

, Wang

, Lu

, Wu

, Weng

, Chen

, Zhang

, Lu

, Yang

, Chen

, Zhang

, Chen

, Xi

, Lu

, and Zhao

. REV3L confers chemoresistance to cisplatin in human gliomas: the potential of its RNAi for synergistic therapy. Neuro Oncol, 11: 790–802, 2009.

68.

Wang

, Wang

, Li

, Wu

, Lathia

, McLendon

, Hjelmeland

, and Rich

. c-Myc is required for maintenance of glioma cancer stem cells. PLoS One, 3: e3769, 2008.

69.

Weydert

, Roling

, Liu

, Hinkhouse

, Ritchie

, Oberley

, and Cullen

. Suppression of the malignant phenotype in human pancreatic cancer cells by the overexpression of manganese superoxide dismutase. Mol Cancer Ther, 2: 361–369, 2003.

70.

Whiteman

, Dogra

, Winyard

, and Armstrong

. Detection and measurement of reactive oxygen intermediates in mitochondria and cells. Methods Mol Biol, 476: 28–49, 2009.

71.

Yeo

, Ng

, Chai

, Tan

, Koh

, Chong

, Koh

, Foong

, Sandanaraj

, Holbrook

, Ang

, Takahashi

, Tang

, and Lim

. Parkin pathway activation mitigates glioma cell proliferation and predicts patient survival. Cancer Res, 72: 2543–2553, 2012.

72.

Zhao

, Xue

, Oberley

, Kiningham

, Lin

, Yen

, Majima

, Hines

, and St Clair

. Overexpression of manganese superoxide dismutase suppresses tumor formation by modulation of activator protein-1 signaling in a multistage skin carcinogenesis model. Cancer Res, 61: 6082–6088, 2001.

73.

Zhu

, Tan

, Huang

, Tan

, Chong

, Pal

, Lam

, Boukamp

, Pan

, Tan

, Kersten

, Li

, Ding

, and Tan

. Angiopoietin-like 4 protein elevates the prosurvival intracellular O(2)(−):H(2)O(2) ratio and confers anoikis resistance to tumors. Cancer Cell, 19: 401–415, 2011.

74.

Zielonka

, Vasquez-Vivar

, and Kalyanaraman

. Detection of 2-hydroxyethidium in cellular systems: a unique marker product of superoxide and hydroethidine. Nat Protoc, 3: 8–21, 2008.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

2.49 MB

A Distinct Reactive Oxygen Species Profile Confers Chemoresistance in Glioma-Propagating Cells and Associates with Patient Survival Outcome

Abstract

Introduction

Results

Chemoresistance correlates with increased O2 −:H2O2 ratio

Pharmacological manipulation of redox status results in either sensitivity to drug-induced apoptosis (reduced O2 −:H2O2) or resistance (elevated O2 −:H2O2)

Mitochondrial SOD2 lentiviral overexpression and knockdown correlate with DPI and DDC treatments, respectively

SOD2 overexpression sensitizes GPCs to drug-induced apoptosis, while SOD2 knockdown confers a prosurvival signal

A reduced cellular redox status and intracellular pH implicate reductive stress and PI3K/AKT signaling

Depletion of intracellular O2 −:H2O2 sensitizes GPCs to drug-induced apoptosis, consequently extending survival

Elevated O2 −:H2O2 ratio identifies patients with poor prognosis, correlating with a higher tumor grade and the mesenchymal molecular subtype

Discussion

Materials and Methods

Tissue collection and primary GPC culture

Viability and sphere assays; antibodies used

Detection of O2 − and H2O2

Measurement of intracellular pH with 2′,7′-bis(2-carboxyethyl)-5,6-carboxyfluorescein

Determination of the cellular redox state—GSH/GSSG measurement

Determination of the cellular redox state—NADP/NADPH measurement

Determination of the cellular redox state—TrxR2 measurement

SOD2 overexpression and shSOD2 vector construction

AKT2 overexpression and shAKT2 vector construction

PI3-K p85-fast activated cell-based ELISA assay

Stereotaxic intracranial implantations of NOD-SCID-gamma mice

Statistical analysis

Microarray data processing

Signature generation and validation

Pathway analysis

Connectivity map analysis

Reference profile generation for Connectivity Map analysis

Survival analysis

Subtype analysis

Prediction of Phillips classification in REMBRANDT and Gravendeel datasets

Footnotes

Acknowledgments

Author Disclosure Statement

Abbreviations Used

References

Supplementary Material

Chemoresistance correlates with increased O₂ ⁻:H₂O₂ ratio

Pharmacological manipulation of redox status results in either sensitivity to drug-induced apoptosis (reduced O₂ ⁻:H₂O₂) or resistance (elevated O₂ ⁻:H₂O₂)

Depletion of intracellular O₂ ⁻:H₂O₂ sensitizes GPCs to drug-induced apoptosis, consequently extending survival

Elevated O₂ ⁻:H₂O₂ ratio identifies patients with poor prognosis, correlating with a higher tumor grade and the mesenchymal molecular subtype

Detection of O₂ ⁻ and H₂O₂