Imp2 controls oxidative phosphorylation and is crucial for preserving glioblastoma cancer stem cells

Abstract

Growth of numerous cancer types is believed to be driven by a subpopulation of poorly differentiated cells, often referred to as cancer stem cells (CSCs), that have the capacity for self-renewal, tumor initiation, and generation of nontumorigenic progeny. Despite their potentially key role in tumor establishment and maintenance, the energy requirements of these cells and the mechanisms that regulate their energy production are unknown. Here, we show that the oncofetal insulin-like growth factor 2 mRNA-binding protein 2 (IMP2, IGF2BP2) regulates oxidative phosphorylation (OXPHOS) in primary glioblastoma (GBM) sphere cultures (gliomaspheres), an established in vitro model for CSC expansion. We demonstrate that IMP2 binds several mRNAs that encode mitochondrial respiratory chain complex subunits and that it interacts with complex I (NADH:ubiquinone oxidoreductase) proteins. Depletion of IMP2 in gliomaspheres decreases their oxygen consumption rate and both complex I and complex IV activity that results in impaired clonogenicity in vitro and tumorigenicity in vivo. Importantly, inhibition of OXPHOS but not of glycolysis abolishes GBM cell clonogenicity. Our observations suggest that gliomaspheres depend on OXPHOS for their energy production and survival and that IMP2 expression provides a key mechanism to ensure OXPHOS maintenance by delivering respiratory chain subunit-encoding mRNAs to mitochondria and contributing to complex I and complex IV assembly.

Figure 1.

Imp2 expression in GBM. (A, top panel) Imp2-positive cells in GBM localize predominantly to perinecrotic areas (N) and in the vicinity of blood vessels (denoted by asterisks). (Bottom panel) Immunohistochemical staining of paraffin GBM sections shows the absence of Imp2 expression in normal brains and grade II and III astrocytomas. (B) Imp2 expression in astrocytomas of different grades in an independent data set (Sun et al. 2006); t-test P-values: GBM versus normal brain: P = 9.678 × 10⁻¹¹; GBM versus grade II/III astrocytoma: P = 4.310 × 10⁻⁵. Whiskers represent the 95th percentile. (C) IMP2 expression differs significantly among TCGA molecular subtypes (Kruskal-Wallis test P-value < 10⁻⁷) (Verhaak et al. 2010). (D) Kaplan-Meier survival curves for proneural TCGA samples with “high” or “low” IMP2 expression (see the Materials and Methods). Median survival of the “Imp2 high” group is 13.3 mo compared with 24.9 for patients in the “Imp2 low” group; log-rank test P-value <0.06. (E) Imp2 expression in GBM CD133⁺ cells. (Left panel) Cells from two surgical biopsies of GBM patients were sorted for CD133 expression. Imp2 transcripts as measured by qRT–PCR were significantly higher in CD133⁺ populations. (****) P < 0.0001 (unpaired two-tailed t-test). (Right panel) Higher level of Imp2 protein expression in freshly isolated CD133⁺ cells was confirmed by Western blot analysis. (F) Imp2 expression in GBM CSCs in situ. Double immunofluorescence staining shows that GBM cells expressing nestin are also positive for Imp2. (G) Costaining of Imp2 and stem/progenitor cell marker nestin in the SVZ of fetal brains.

Figure 2.

Gliomaspheres as a model of GBM CSC-enriched populations. (A) Morphology of sphere-forming and adherent primary GBM cultures. (B) FACS analysis of CSC-associated marker expression in the two types of primary GBM cultures. Only GBM cultured as spheres express CD133 (top panel) and SSEA1 (bottom panel). Representative results for one out of three GBM samples are shown. (C) Survival of mice after intracranial injection of GBM cells cultured as spheres and adherent cells. Five-hundred spherogenic cells (three independent cell batches, BT1–3) were sufficient to initiate tumor growth, whereas 100,000 adherent cells derived from spheres exposed to serum (BT-1adh) failed to form tumors (data for BT2–3adh are not shown); injection of 50 or 100 spherogenic cells was insufficient for tumor growth initiation (data for BT2–3 are not shown); six animals per condition were injected. (D) Immunofluorescence staining of a GBM spheroid (nonreconstituted image [top panel] and three-dimensional [3D] reconstruction [middle panel]) compared with adherent GBM cells (bottom panel) shows high Imp2 expression levels in a CSC-enriched sphere. (Green immunofluorescence) Imp2 staining; (white) Imp2 channel 3D reconstruction (signal above threshold); (blue) DAPI staining. (E) Imp2 expression in gliomasphere cells sorted for CD133 expression. Imp2 transcripts as measured by qRT–PCR were significantly higher in CD133⁺ populations. (***) P = 0.0003 and 0.001 (unpaired two-tailed t-test).

Figure 3.

Effect of Imp2 depletion on gliomasphere cells. (A) Imp2 silencing in gliomasphere cells stably expressing two different shRNAs. (B) Morphology of spheres formed by control shRNA-expressing and Imp2-depleted GBM cells in culture. (C) Sphere morphology change upon Imp2 depletion shown as sphere diameter. P < 0.0001 (n = 50), unpaired two-tailed t-test. Whiskers represent the 95th percentile. (D) Cell viability as measured by FACS analysis of Calcein AM staining in three different cell batches. Bars represent the percentage of positively stained cells. Low viability of ctrl cells is due to puromycin selection. (**) P = 0.0094 and 0.0068, respectively; paired two-tailed t-test. Error bars represent the standard deviation (SD). (E) Cell proliferation in spherogenic culture conditions upon Imp2 knockdown as assessed by BrdU incorporation. (***) P = 0.0008 and 0.0007; (**) P = 0.0024, 0.0057, 0.002, and 0.0027, respectively; unpaired two-tailed t-test. Error bars represent the SD. (F) Single-cell clonogenic assay performed on three independent cell batches; single cells were plated in 96-well plates, and three plates per condition were used; statistical analysis for each cell batch was performed using unpaired two-tailed t-test comparing control cells with Imp2-depleted cells. (*) P = 0.0227; (**) P = 0.0015; (***) P = 0.0008; error bars represent the SD. (G) Imp2 depletion decreases CD133, SOX2, OCT4, and NANOG transcripts as measured by qRT–PCR. (*) P = 0.0443; (**) P = 0.0098; (*) P = 0.013; (**) P = 0.0034; (***) P = 0.0002. Representative values from two independent experiments are shown. (H) Imp2 shRNA expression decreases GBM CSC tumor-forming capacity. The experiment was repeated using three batches of cells. Six animals were used per condition. The differences between Imp2- and ctrl shRNA-expressing cells for each cell batch were significant (log-rank [Mantel-Cox] test). P-values for BT-1, BT-2, and BT-3 were 0.0088, 0.0036, and 0.0056, respectively. The median survival of the animals is presented in the right panel.

Figure 4.

mRNA and proteins bound by Imp2 suggest a functional role in mitochondrial processes. (A) Specificity of the RIP assay is illustrated in a heat map that compares transcripts obtained by immunoprecipitation using anti-Imp2 and isotype-matched control antibody; results for three batches of primary GBM cultures in spherogenic conditions are shown. (B) Gene ontology (GO) analysis of Imp2-bound mRNA shows enrichment in transcripts associated with mitochondrial function. A GO term was considered overrepresented if the nominal P of the exact one-tailed Fisher test was <10⁻⁵. The length of the bar for each term is proportional to the number of bound mRNAs annotated to the term. The shaded part of the bar represents the number of annotated mRNAs expected by chance. (C) Imp2 knockdown decreases expression of Imp2-bound transcripts (transcripts associated with mitochondrial functions are shown). (D) Identification of several mitochondrial respiratory CI proteins by mass spectrometry analysis of anti-Imp2 antibody pull-down material. (E) Immunoprecipitation of Imp2 followed by Western blotting using an anti-NDUFS3 antibody (top panel) and immunoprecipitation of NDUFS3 followed by Western blotting using an anti-Imp2 antibody (bottom panel) confirm the interaction between the two proteins. (Ctrl) Immunoprecipitation with isotype-matched control antibody; (PB) post-beads unbound fraction. (F) Imp2 depletion decreases the level of Imp2-bound proteins. (G) Densitometric quantification of the Imp2-bound proteins observed in F. Data were normalized to actin.

Figure 5.

Imp2 depletion affects mitochondrial respiration in gliomasphere cells. (A) MitoTracker Red staining of mitochondria decreases in cells depleted of Imp2, as analyzed by FACS. DAPI-stained dead cells were excluded from the analysis. Bars represent mean values for three independent cell batches. (**) P = 0.0071 and 0.0081 (paired two-tailed t-test), respectively. Error bars represent the SD. (B) OCR is decreased by Imp2 depletion. OCR values were normalized to viable cell counts as assessed by post-measurement FACS analysis of Calcein AM staining. Results for three cell batches are shown. Imp2 shRNA1: (**) P = 0.005, (*) P = 0.005, (**) P = 0.0038; Imp2 shRNA2: (*) P = 0.0114, (**) P = 0.0021, (***) P = 0.0004 (unpaired two-tailed t-test). (C) The OCR is higher in freshly isolated CD133⁺ than in CD133⁻ cells. Results for two freshly obtained surgical biopsies are shown. (*) P = 0.0355; (****) P < 0.0001 (unpaired two-tailed t-test). (D) Citrate synthase activity upon Imp2 knockdown. Representative values of two experiments with three different cell batches are shown. P > 0.05, not significant. (E) ATP content upon Imp2 knockdown. Representative values of two experiments with two different cell batches are shown. (**) P = 0.0024 (two-tailed t-test). (F) Respiratory CI and CIV activity is lowered by Imp2 knockdown, as shown by BN-PAGE and in-gel activity assay. Forty micrograms of the mitochondrial extracts was loaded onto the gel. Representative results of three independent experiments with three different cell batches are shown. Coomassie staining shows band intensity before the activity assay. Error bars represent the SD. (G) Quantification of CI and CIV Coomassie-stained bands. No significant difference was observed. (H) Quantification of CI and CIV in-gel activity. (***) P = 0.0008; (*) P = 0.0321 (two-tailed t-test).

Figure 6.

Inhibition of OXPHOS impairs clonogenicity and survival of gliomasphere cells. (A) Inhibition of OXPHOS with 1 μM rotenone affects gliomasphere morphology in the same way as Imp2 silencing. (B) Morphological changes in response to rotenone (ROT), oxamic acid (OXA), and Imp2 knockdown shown as sphere diameter measurements. (****) P < 0.0001 (n = 50), unpaired two-tailed t-test. (C) Clonogenic potential of gliomasphere cells is decreased by rotenone treatment. (**) P = 0.0024, unpaired two-tailed t-test. In contrast, anaerobic glycolysis inhibition by 25 mM oxamic acid does not significantly affect the clonogenic potential of GBM CSCs. Representative results for one out of three cell batches are shown. (D) Addition of 1 μM rotenone (indicated by the vertical line) decreases the OCR of control cells to a level comparable with that measured in Imp2-depleted cells. (E) Rotenone decreases ATP content in spheres but not in adherent cells, whereas oxamic acid affects adherent cells but not spheres. (***) P = 0.0001; (**) P = 0.0074, unpaired two-tailed t-test. Representative results for one out of two cell batches are shown. (F) Oxamic acid affects proliferation of only adherent cells as measured in a BrdU incorporation assay; rotenone inhibits proliferation of both gliomasphere and adherent cells. (***) P = 0.0006 (DMSO vs. rotenone in spheres), 0.0008 (DMSO vs. rotenone in adherent cells), and 0.0001 (DMSO vs. oxamic acid in adherent cells), unpaired two-tailed t-test. Representative results for one out of two cell batches are shown. Error bars represent the SD.

Figure 7.

Imp2 may provide mRNA delivery to mitochondria and regulate respiratory complex assembly. (A) Imp2 does not play a role in the stabilization of mRNA that it binds; degradation of Imp2-bound transcripts was measured by qRT–PCR 0, 4, 8, and 12 h after actinomycin D addition. Lines represent linear regression; the slopes of the linear regression are not significantly different between Imp2-expressing and Imp2-depleted cells. Representative data from two independent overexpression/knockdown experiments and for two representative transcripts are shown. (B) Imp2 localizes to the surface of mitochondria as illustrated by subcellular fractionation. (CYT) Cytosolic fraction; (MT) isolated mitochondria. Western blot analysis of Imp2 and NDUFS3 shows the disappearance of the Imp2 band following proteinase K treatment of intact mitochondria. (C) Assembly of respiratory CI and CIV is impaired by Imp2 knockdown; BN-PAGE followed by Western blot (WB) using anti-NDUFS3, anti-ND2, and anti-COX7b antibodies. Imp2 knockdown depleted the high-molecular-weight CI (NDUFS3 WB) and increased accumulation of its intermediate species (ND2 WB). CIV assembly (COX7b WB) was also altered. (>) CI holocomplex; (*) CI subcomplexes containing NDUFS3 (pattern corresponding to that published by Vogel 2007); (o) CI subcomplexes containing ND2; (m) monomeric ND2; (>>) CIV holocomplex; (#) CIV subcomplexes. Holocomplex band localization was verified with Native Mark marker (Invitrogen) and the in-gel activity assays. (D) Imp2 delivers mRNA to mitochondria-bound polysomes; qRT–PCR was performed on a mitochondria-bound polysomal fraction and cytosolic polysomes isolated from SVGp12 astrocytes overexpressing Imp2 or infected with empty vector. Ratios of transcript content in cytosolic to that in mitochondria-bound polysomes are presented. The ratio value for UCP2, a mitochondria-bound polysome-associated transcript, was set as the 0 value. Values below 0 indicate higher mitochondrial than free polysome association of a given transcript. Controls were 12S (mitochondrial rRNA for mitochondria purification efficiency), UCP2 (a mitochondrial protein that is not a direct target of Imp2), and nestin (a cytosolic protein that is not a direct target of Imp2). Representative values of two independent Imp2 overexpression experiments are shown.