Acidosis Acts through HSP90 in a PHD/VHL-Independent Manner to Promote HIF Function and Stem Cell Maintenance in Glioma

Rapidly growing tumors often outpace their blood supply generating a hypoxic microenvironment. Tumor hypoxia triggers a set of adaptive responses that ultimately promote a more aggressive tumor phenotype and are primarily controlled by the transcription factor system of the hypoxia-inducible factors (HIF; ref. 1). HIF abundance is tightly regulated by the prolyl hydroxylase domain (PHD) proteins. PHDs use O₂ to hydroxylate HIFs (2, 3), which enables binding of the VHL protein, a component of an E3 ubiquitin ligase complex that catalyzes HIF1/2α ubiquitination and degradation. One of the primary effects of hypoxia is the induction of a metabolic shift from oxidative phosphorylation to glycolysis, with lactic acid as the end product (4). This is accompanied by the upregulation of carbonic anhydrases, transporters for lactate, H⁺, HCO₃⁻, and other ions, leading to a net acidification of the extracellular tumor environment (5, 6). A decreased pH is a characteristic feature of many tumor types and a number of studies have started to delineate critical functions of the acidic extracellular tumor environment (7). For example, blockade of carbonic anhydrase IX (CA IX), a hypoxia upregulated transmembrane enzyme that converts H₂O and CO₂ into carbonic acid (H⁺ and HCO₃⁻), increased extracellular pH, decreased proliferation, enhanced apoptosis, and markedly suppressed tumor growth in vivo (5, 8). Acidic extracellular pH has also been shown to promote tumor cell migration, invasiveness, and metastasis (8–10). Although tumor hypoxia and acidosis are linked and both represent crucial aspects of the tumor microenvironment, their interplay in tumor progression remains poorly understood.

It has become clear that tumor growth and progression is driven by a subpopulation of cells with stem cell–like properties, termed cancer stem cells (CSC) or tumor-initiating cells (11). We and others have previously established that hypoxia plays a key role in promoting the CSC phenotype through HIFs (12, 13). Acidic stress has also been shown to promote a cancer stem cell phenotype (14); however, very little is known about the mechanisms through which microenvironmental pH may impact on hypoxic signaling and control CSC maintenance within the hypoxic niche. Interestingly, glioblastoma cells exposed to more acidic conditions show an increased resistance to chemotherapeutics (15), and elevated expression of VEGF (16), properties characteristic of CSCs. Thus, the contribution of acidosis to the regulation of CSCs, particularly within the hypoxic CSC niche, is of high interest and potential therapeutic relevance.

Here, we show that hypoxia and acidosis synergize to increase HIF levels and function in glioblastoma, resulting in a marked potentiation of CSC self-renewal, maintenance, and tumorigenicity. The pH-mediated regulation of HIF, CSC function, and tumorigenicity is not dependent on PHD/VHL activity, but is instead controlled by HSP90. Furthermore, increased HSP90 levels in human glioblastomas are associated with a higher expression of HIF target genes and CSC markers, highlighting the potential importance of HSP90 as a druggable pathway for the targeting of glioma CSCs.

The glioblastoma cell lines G55TL, G121, G141, and G142 were kindly provided by K. Lamszus and M. Westphal [Department of Neurosurgery, University Medical Center Hamburg-Eppendorf (UKE), Hamburg, Germany; ref. 17], the renal carcinoma line 786-O was obtained from ATCC and RCC11 (18) was obtained from Dr C. Bauer (Institute of Physiology, University of Zurich, Zurich, Switzerland) between 1998 and 2002. The primary glioblastoma line NCH644 was kindly provided by C. Herold-Mende (Heidelberg, Germany; ref. 19) in 2010. The primary glioblastoma lines GBM015 and GBM031 were obtained in 2007–2008 from patients undergoing surgery in accordance with a protocol approved by the institutional review board and cultured in neurosphere medium (13). G55TL, G121, G141, and G142 cells were cultured in DMEM (Invitrogen) supplemented with 10% FBS (PAN Systems). NCH644, GBM015, and GBM031 were propagated under neurosphere conditions in neurosphere medium (DMEM-F12; Invitrogen), 5 mmol/L HEPES, 2% B-27 serum-free supplement without vitamin A (Invitrogen) supplemented with 20 ng/mL bFGF and 20 ng/mL EGF (PeproTech)]. All cells were maintained at 37 °C in 5% CO₂. For experiments with different pH, G55TL, 786-O, and RCC11 cells were incubated in CO₂-independent medium (#18045-054, Invitrogen) supplemented with 2 mmol/L l-glutamine and 10% FBS; the primary glioma cells lines were incubated in CO₂-independent medium supplemented with 2% B-27 serum-free supplement without vitamin A, 2 mmol/L l-glutamine, 20 ng/mL bFGF, and 20 ng/mL EGF. The pH of the medium was adjusted to the values indicated in the figures with 1 mol/L HCl or 1 mol/L NaOH. For hypoxic treatment cells were grown at 1% O₂ for the indicated periods of time in a Hypoxic Workstation (Ruskinn Technology). NCH644, GBM031, and GBM015 were preincubated in medium with pH7.4 at normoxia and incubated in media with pH7.4 or 6.7 at hypoxia. For geldanamycin treatment, cells were preincubated with pH 7.4 at 21% O₂ for 4 days; in the last two hours of the preincubation 1 μmol/L geldanamycin (Enzo Life Sciences) or control DMSO was added, followed by incubation at 1% O₂ in medium with pH 7.4 or 6.7 supplemented with DMSO control or 1 μmol/L geldanamycin for 6 or 18 hours (for G55 and NCH644 cells, respectively). Further details on the experimental procedures are included in the in corresponding figure legends. For stem cell condition, the cells were incubated in neurosphere medium for the same time. Cells that showed unusual growth or morphology were tested for mycoplasma contamination and only noncontaminated cells were used for experiments. Cell lines obtained from ATCC were authenticated by the manufacturer. No additional authentication was performed by the authors for any of the cell lines.

Animal experiments were approved by the veterinary department of the regional council in Darmstadt, Germany. Xenograft transplantations were performed in athymic 6–8 week old female NMRI nu/nu mice (Janvier Labs) that were kept in a specific pathogen-free animal facility according to the institutional guidelines.

For intracranial tumor transplantation, mice were placed into a stereotactic apparatus, and 1 × 10⁵ cells in a volume of 2 μL (NCH644), or 7.5 × 10³ cells in a volume of 1 μL (G55) were resuspended in cold, CO₂-independent medium and slowly injected into the left striatum. At the onset of neurologic symptoms, mice were sacrificed at the same time point (at day 19 after transplantation for the experiments with G55 cells, and at day 22 posttransplantation for the experiments with NCH644 cells) through deep anesthesia with ketamine and xylazine. The chest was opened and the vasculature was perfused with a 0.9% NaCl solution for 2 minutes and fixative (4% PFA) for 4 minutes. Brains were removed and additionally fixed overnight in 4% PFA, dehydrated in 30% sucrose for 4 days, and rapidly frozen on dry ice for sectioning with a cryotome. The sections were stained with hematoxylin and eosin. Tumor volume was determined using stereological quantification of series of every twelfth 40-μm section (480 μm intervals) throughout the brain using ImageJ, as previously described (NCH644; ref. 20). G55 tumor–bearing mice were injected with 60 μg/g Hypoxyprobe intraperitoneally 90 minutes prior to cardiac perfusion with 0.9% NaCl solution. After removal of the brain and dissection of the tumors, one part of each tumor was snap-frozen in liquid nitrogen for subsequent protein isolation. Another part of the tumor was incubated in fixative overnight and embedded in paraffin for subsequent histologic and immunohistochemical analysis. Tumor volume was analyzed using the section with the biggest tumor area by measuring the largest diameter (L) and largest perpendicular diameter (W), using the formula L × W²/2 (G55).

For subcutaneous tumor injections, 7.5 × 10⁴ G55 cells suspended in 0.1 mL PBS/Matrigel were injected subcutaneously into the flanks of 6–8-week-old female nude (NMRI nu/nu) mice. The tumor size was measured at regular intervals using a caliper according to the formula V = L × W²/2. Mice were maintained until tumors exceeded a volume of 2,000 mm³ or upon the onset of morbidity symptoms (>20% weight loss, tumor ulceration). Tumors were excised and snap-frozen in liquid nitrogen for subsequent protein isolation.

G55TL cells were transiently transfected with plasmid DNA using Lipofectamine 2000 according to the manufacturer's instructions. siRNAs against HIF1α, HIF2α, or nontargeting control siRNA were obtained as a pool of 4 siRNA oligos (ON-TARGETplus SMART pool, Dharmacon). Reverse siRNA transfection of GBM015 cells was performed with 10 pmol siHIF1α and 50 pmol siHIF2α, or 60 pmol nontargeting siRNA, respectively, with Lipofectamine RNAiMAX (Thermo Scientific) in antibiotic-free neurosphere medium according to the manufacturer's instructions. Twenty-four hours after transfection, the cells were seeded in CO₂-independent neurosphere medium at pH 7.4 and preincubated under normoxic conditions for 48 hours. Subsequently, the cells were cultured under 1% O₂ neurosphere medium at pH 7.4 or 6.7 and harvested after 18 hours (protein and RNA) or 96 hours (FACS).

For immunoblotting, cells were harvested in PBS (4°C), cell pellets were lysed in 10 mmol/L Tris.HCl (pH 7.5), 2% SDS, 2 mmol/L EGTA, 20 mmol/L NaF, and 15–50 μg of protein lysates were subjected to SDS-PAGE and Western blot analysis using antibodies specific for CD133 (Abcam ab19898 and Miltenyi Biotec 130-092-395), GFAP (Dako Z 0334), HIF1α (Cayman Chemical 10006421 or BD Transduction Laboratories 610958), HIF2α (Novus Biologicals NB 100-122), HSP90 (Enzo/Stressgen ADI-SPA-830), V5 (Invitrogen R960-25), VHL (Abcam, ab11189), and Tubulin (Dianova DLN09992) as a loading control. Immunoreactive bands were visualized with the ECL system (Perkin Elmer or Pierce).

Spheres were dissociated into single-cell suspension by Accutase (PAA) treatment for 15 minutes at 37°C. After two washing steps with FACS Staining buffer (PBS, pH 7.2, 0.5% BSA, 2 mmol/L EDTA), single cells were blocked with 20 μL of normal mouse IgG (Invitrogen) for 20 minutes at 4°C and stained with 10 μL of CD133/2 (293C3)-PE–conjugated antibody (Miltenyi Biotec 130-090-853) or 5 μL of CD15-V450–conjugated antibody (BD Biosciences 642917) for 30 minutes at 4°C. The background staining was determined using matching isotype control antibodies from the same manufacturers at the same concentration as the specific antibodies. Five minutes before analysis, 1 μmol/L SYTOX Blue (Invitrogen; CD133/2-PE) or SYTOX Red (CD15) nucleic acid stain was added to exclude dead cells. Flow cytometry was performed using BD FACSCanto II (BD Biosciences). Data were analyzed using FlowJo v7/9 (Tree Star) by gating for live cells based on SYTOX staining, then for singlets using forward scatter area versus width and side scatter area versus width, followed by gating for the CD133 or CD15-positive populations, respectively.

A total of 1 × 10⁶ cells were seeded per 10-cm dish and cultured at pH 7.4 or 6.7 at 1% O₂ for 96 hours. For quantification of sphere forming units (SFU), the cells were then split in 12-well suspension culture plates (n = 12) at a density of 300 cells/well in 1 mL or 6-well suspension culture plates (n = 6) at a density of 500 cells/well in 2-mL neurosphere medium. After 7 days, the spheres were counted and the percentage of sphere-forming cells was calculated. For sphere formation experiments applying siRNA-mediated knockdowns, cells were transfected as described above. Twenty-four hours after transfection, medium was changed and the cells cultured at pH 7.4 or 6.7, 1% O₂ for 72 hours. For quantification of SFUs, the cells were then split in 6-well suspension culture plates (n = 6) at a density of 500 cells/well in 2-mL neurosphere medium. After 7 days, the spheres were counted and the percentage of sphere-forming cells was calculated.

RNA was extracted with the RNeasy Mini Kit (Qiagen), and reverse transcribed using standard protocols (RevertAid H Minus M-MuLV Reverse Transcriptase, Fermentas). cDNA was amplified using the ABsolute QPCR SYBR Green Mix. Gene-specific PCR products were measured continuously in a StepOnePlus real-time PCR system (Applied Biosystems) for up to 45 cycles. The difference in the threshold number of cycles between the gene of interest and HPRT (hypoxanthine phosphoribosyltransferase 1) was then normalized relative to the standard chosen for each experiment and converted into fold difference.

Immunohistochemical stainings of human biopsies were approved by the institutional review board and were carried out using antibodies against HIF1α (Cayman Chemical 10006421), HSP90 (Enzo/Stressgen ADI-SPA-830), and CD133 (Miltenyi Biotec 130-092-395) with an Autostainer (Ventana/Roche) according to the manufacturer's instructions. For Hypoxyprobe detection (Hypoxyprobe-1, NPI Inc.), paraffin sections were dewaxed in xylene (2 × 10 minutes) followed by a series of descending alcohol concentrations (2 × 5 minutes 100% EtOH, 2 × 5 minutes 96% EtOH, 2 × 5 minutes 70% EtOH) and rehydrated for 5 minutes in H₂O and TBS buffer each. For antigen retrieval, the sections were boiled in preheated citrate buffer (pH 6) for 10 minutes and subsequently cooled down at room temperature for 20 minutes, followed by washing steps in TBS (2 × 5 minutes). Endogenous peroxidase was blocked in 0.6% H₂O₂ for 30 minutes followed by washing steps with TBS (3 × 5 minutes). Unspecific antibody binding was blocked with 10% normal goat serum (NGS) in TBS/0.1% Tween (TBST) for 1 hour at room temperature, followed by incubation with FITC-Mab1 antibody (1:100 in 10% NGS/TBST, NPI Inc.) for 2 hours at room temperature. After three washing steps in TBST (5-minute each), sections were incubated with anti-FITC HRP secondary antibody (1:200 in 10% NGS/TBST, NPI Inc.) for 2 hours at room temperature followed by three washing steps in TBST. Chromogen reaction was performed using diaminobenzidine (DAB+, Dako) for 3 minutes. Counterstaining was performed in hematoxylin for 2 minutes. To achieve color development of the stained nuclei, sections were rinsed in tab water for 2 minutes followed by a rinse in H₂O. Sections were dehydrated in a series of ascending alcohol concentrations (2 × 5 minutes 70% EtOH, 2 × 5 minutes 96% EtOH, 2 × 5 minutes 100% EtOH) with a final incubation in xylene (2 × 5 minutes). Sections were permanently mounted using Cytoseal XYL (Thermo Scientific).

A total of 3 × 10⁴ G55 cells were plated in 24-well plates. Twenty-four later the cells were transfected using SuperFect (Qiagen) with VEGF promoter-firefly luciferase (21) and a SV40 Renilla luciferase constructs for normalization of transfection efficiency. To assess the effect of HIFs on VEGF promoter transcription, the cells were cotransfected with 300 ng pcDNA3-HIF1α or HIF2α overexpression constructs, or pcDNA3 as a control. The next day, the cells were placed at the indicated concentrations of O₂ and CO₂ and 24 hours later were lysed for analysis of luciferase expression downstream of the VEGF promoter, using a dual luciferase reporter system (Promega), according to the manufacturer's instructions.

pGIPZ lentiviral constructs against HSP90α, HSP90β, and pGIPZ nonsilencing control vectors were purchased from Open Biosystems (Thermo Scientific). Lentiviral HA-tagged dominant-negative mutant of HSP90β (HA-HSP90 DN) or a GFP-expressing control plasmids were constructed as described in the Supplementary Materials and Methods. Lentivirus was packaged by cotransfection of lentiviral vectors with the packaging plasmids pCI-VSVG and psPAX into 14-cm plates with HEK293T cells using Fugene HD (Roche). Medium was changed every 24 hours, the 48- and 72-hour supernatants were pooled, filtered through a 0.45-μm filter, and ultracentrifuged at 20,000 × g, 4°C for 4 hours. Titers were determined by counting the number of GFP-positive colonies or crystal violet–stained colonies. The shHSP90 and control lines were generated by lentiviral transduction and selection with 2 μg/mL puromycin to create the stable shRNA pools. The HA-HSP90 DN and GFP lines were generated by lentiviral transduction and selection with 6 μg/mL blasticidin to create the stable HA-HSP90 DN or GFP-overexpressing cells.

Gene expression data [z-scores (all genes)] for the glioblastoma cohort of The Cancer Genome Atlas (TCGA) Research Network were downloaded from the cBio portal (http://www.cbioportal.org/public-portal/index.do). Tumors with z-scores for HSP90α < −1 (n = 90) were considered as low HSP90-expressing, tumors with z-scores of > +1 (n = 64) were considered as high HSP90 expressing.

Results are presented as mean ± SEM. Sample size was chosen based on previous empirical experience with the respective types of assays or animal tumor models. Animals that unexpectedly died before tumors were collected from the remaining animals were excluded from the analysis. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment. For pairwise comparisons, statistical analysis was performed using the Student t test except for the expression analysis in the TCGA glioblastoma cohort, where the Mann–Whitney test was used due to non-normal distribution. For comparisons between multiple groups, ANOVA with a Bonferroni post test was performed using GraphPad Prism. Statistical significance was defined as *, P < 0.05; **, P < 0.01; ***, P < 0.001.

We were interested in analyzing to what extent metabolic parameters of the hypoxic tumor microenvironment other than pO₂ may regulate HIF levels and function in glioblastoma. As tumor hypoxia is frequently accompanied by acidosis and hypoglycemia, we cosubjected a panel of glioblastoma cell lines to different hypercapnic conditions (10% and 20% CO₂), as a physiologic means of lowering the pH, and different glucose levels (high glucose: 4.5 g/L; low glucose: 1.0 and 0 g/L glucose; Fig. 1A and B). Under normoxia, only very low-level of HIF1α and moderate levels of HIF2α were detectable, which were highly upregulated under hypoxia (Fig. 1A and B) with a concomitant increase in mRNA expression of the HIF-target genes VEGF, LDHa, Glut-1, and CAIX (Supplementary Fig. S1). Glucose deprivation mostly did not alter or even decreased HIF1α and HIF2α abundance under normoxia and hypoxia. In contrast, decreasing the pH through hypercapnia moderately increased HIF1α and HIF2α levels already under normoxia (Fig. 1A and B) and dramatically elevated the hypoxic upregulation of HIF1α and HIF2α, demonstrating that hypoxia and acidosis synergistically increase the HIF response (Fig. 1B). In line with the pH-dependent control of the HIF response, hypercapnia efficiently potentiated the hypoxia- and HIF1/2α–driven transcription downstream of the VEGF promoter (Fig. 1C), suggesting a posttranscriptional HIFα regulation by pH. As CO₂ can exert effects on cell physiology that are independent of acidification of the cellular environment (22), we next modulated the pH by addition of NaOH and HCl in CO₂-independent medium (Fig. 1D and E). Acidic pH was sufficient to upregulate HIF1α and HIF2α levels already under normoxia and even further potentiated the hypoxia-dependent increase in HIF1/2α expression (Fig. 1D). Decreasing the pH down to 6.6–6.8, conditions that are observed in tumors (23), efficiently induced HIF1α and HIF2α expression under normoxia and hypoxia, whereas further pH reduction decreased HIF1/2α abundance (Fig. 1E).

Previous work by us and others has demonstrated that hypoxia promotes the CSC phenotype through the induction of HIFs (12, 13). As expression and activity of HIFs were induced by low pH, we assessed whether acidosis would also promote the stem cell properties of glioblastoma cells. We first confirmed the pH-dependent regulation of HIF and HIF target genes in a panel of primary glioblastoma lines derived from patient biopsies and continuously propagated in serum-free sphere (stem cell) conditions (Fig. 2A and B). Importantly, acidosis concomitantly increased the expression of a panel of CSC-associated genes and the fraction of cells positive for the glioma CSC markers CD133 and CD15 (Fig. 2C and Supplementary Fig. S2A and S2B), as well as the capacity to form spheres as a measure of self-renewal (Fig. 2D). Notably, the stimulation of CSC maintenance by low pH was dependent on HIFs, as HIF1/2α silencing (Supplementary Fig. S3A and S3B) abrogated not only the acidosis-induced upregulation of HIF target gene expression (Fig. 2E), but also the increase in the fraction of cells positive for the CSC marker CD133 and in self-renewal capacity (Fig. 2F and G). Collectively, our results identify acidosis as a potent metabolic factor that regulates HIF function and the CSC phenotype synergistically with hypoxia within the pathophysiologic pH range observed in the tumor microenvironment.

We next aimed at identifying the molecular mechanism that allows acidosis to regulate HIF. It has been previously shown, that an acidic pH of 5.8–6.2 induces HIF1α by nuclear sequestration of VHL, resulting in a relative cytoplasmic lack of VHL and inhibiting VHL-dependent proteasomal degradation (24). To examine whether an acidic pH in the range of 6.7–6.9, values typically measured in tumors (14, 23), regulates HIF through VHL we first analyzed the influence of acidification on HIF levels in the VHL-deficient renal cell carcinoma lines 786-O and RCC11 (which express high HIF2α, but not HIF1α; refs. 18, 25). Importantly, lowered pH robustly increased HIF levels in these cells (Fig. 3A), demonstrating that the effect of acidosis is not VHL dependent. Consistently, increasing the intracellular levels of VHL by overexpression in glioblastoma cells did not affect the acidosis-induced hypoxic upregulation of HIFs (Fig. 3B). Furthermore, blockade of PHD function using the iron chelator dipyridyl (DP) and the PHD family inhibitor dimethyloxalylglycine (DMOG) did not affect the induction of HIF levels by acidosis (Fig. 3C). Further in line with a PHD/VHL-independent regulation of HIF proteins by acidosis, nonhydroxylatable mutant proteins of HIF1α and HIF2α with substitution of both prolyl residues that are hydroxylated by PHDs (mPPN mutants, ref. 13), were strongly increased by acidic pH (6.7; Fig. 3D). Taken together, these data provide compelling evidence that acidosis regulates HIF levels independently of the PHD/VHL-mediated degradation pathway in glioma.

Apart from the oxygen-dependent PHD/VHL pathway, one of the best characterized mechanisms controlling HIF stability is linked to HSP90. HSP90 interacts with HIF (26, 27) and competes with the protein RACK1 for association with HIF, thus protecting HIF from proteasomal degradation (28). Therefore, we next examined whether the acidosis-dependent upregulation of HIF levels and the CSC phenotype are coupled to HSP90. We found that lowered pH induced a marked upregulation of HSP90 in established and primary glioblastoma cell lines (Fig. 4A and B). Importantly, inhibition of HSP90 function with geldanamycin abolished the acidosis-induced increase in HIF1/2α and CD133 (Fig. 4C and D). Similarly, a dominant-negative HSP90 construct (29) blocked the increase of HIF levels at acidic pH (Fig. 4E), demonstrating that HSP90 activity is required to mediate HIF upregulation under conditions of acidic stress. To corroborate the HSP90 dependent control of HIF we tested whether increasing HSP90 levels would replicate the effects of acidosis. Indeed, overexpression of wild-type HSP90 already at physiologic pH led to an elevation of HIF1α and HIF2α levels similar to the one caused by low pH (Fig. 4F).

We next assessed the function of HSP90 in HIF regulation in a microenvironmental setting of hypoxia and acidosis in vivo. Importantly, HSP90 disruption by shRNA-mediated knockdown in glioblastoma cells significantly reduced intracranial tumor growth (Fig. 5A and B). While both control and shHSP90 glioblastomas displayed typical perinecrotic areas with pronounced hypoxia, as revealed by Hypoxyprobe staining (Fig. 5C), HIF1/2α levels and the expression of the HIF target gene VEGF were significantly reduced in shHSP90 tumors (Fig. 5D–F). A similar decrease of HIF levels and orthotopic tumor growth was elicited HPS90 inhibition through a dominant-negative HSP90 (Supplementary Fig.S4A–S4D). Moreover, HSP90 silencing increased survival in a subcutaneous transplantation model, again concomitant with a reduction in HIF1/2α levels (Supplementary Fig. S5A–S5D). Importantly, HPS90 disruption reduced the tumor-initiating capacity of the cells, indicating an involvement of the HSP90-mediated increase in HIF levels under acidosis in tumor initiation and CSC maintenance. Indeed, HSP90 inhibition with geldanamycin efficiently blocked the acidosis induced increase in the CD133-postive tumor cell fraction as well as in self-renewal (Fig. 6A and B). To further corroborate the role of acidosis and HSP90 in the promotion of CSC function, we examined the tumorigenic capacity of primary glioblastoma cells, a defining property of CSCs. Glioblastoma cells grown under pH 6.7 generated significantly larger intracranial tumors than cells cultured at physiologic pH (Fig. 6A, C, and D). Crucially, concomitant treatment of the cells with geldanamycin completely abolished the increased tumorigenicity under acidosis, but had no effect under physiologic pH (Fig. 6C and D). These findings demonstrate that acidosis increases HIF levels and CSC function through HSP90 and that HSP90 inhibition can efficiently suppress HIF induction, CSC maintenance, and tumorigenicity.

We next wanted to assess the relevance of our findings to human glioblastomas. Analysis of gene expression in the glioblastoma cohort of TCGA research network (30) revealed that tumors with high HSP90 expression had increased levels of the HIF target gene VEGF, as well as of the glioma stem cells markers CD133 and nestin (Fig. 7A), supporting the notion that HSP90 is an important factor activated by the CSC microenvironment. To confirm that HSP90 is specifically upregulated in glioblastoma in the hypoxic CSC niche, we examined its localization in glioblastoma biopsies. The presence of necrotic areas is a diagnostic hallmark of glioblastoma and is associated with regions of insufficient blood supply and severe hypoxia. HIF levels were elevated in the perinecrotic regions (Fig. 7B), indicating that these areas are hypoxic and thus presumably more acidic as a result of the hypoxia-induced metabolic response. Importantly, HSP90 levels in the same region were also increased in 10 of 10 glioblastoma biopsies examined, with 93.0% ± 4.0% of perinecrotic/hypoxic areas showing HIF and HSP90 coexpression. Moreover, CD133 was also prominently enriched in this area (Fig. 7B), supporting our hypothesis of synergistic stimulation of HIF function by hypoxia and acidosis in an HSP90-dependent manner, leading to an expansion of the CSC pool in human glioblastomas.

Tumor cells are involved in an intimate crosstalk with their microenvironment, which critically controls major aspects of cancer cell biology. In this study, we show that two key characteristics of the tumor microenvironment, hypoxia and acidosis, synergize to potentiate HIF activity and HIF-dependent functions through complementary and additive mechanisms. We identify HSP90 as a key factor responsible for the capacity of the acidic microenvironment to promote the hypoxic response and its downstream functions, including CSC maintenance and tumorigenicity, pointing to potential strategies for the targeting of this critical cancer cell population.

Hypoxia and acidosis are characteristic features of many tumor types. Similarly to hypoxia, the acidic tumor microenvironment regulates key aspects of tumor pathophysiology. For example acidosis can enhance the resistance of tumor cells to various chemotherapeutics (15, 31, 32). Furthermore, a number of studies have linked acidosis to an enhanced invasive and metastatic capacity of tumor cells (9, 33–36). In gliomas, the acidic microenvironment has been shown to induce VEGF expression and tumor angiogenesis (16, 37). Importantly, most of the properties listed above are also regulated by hypoxia and the HIF pathway. Indeed, our results uncover an interesting link between acidosis and HIF showing that acidosis increases HIF levels and function in a synergistic manner with hypoxia. An increase of both HIF1α and HIF2α levels under acidosis has been previously reported in primary glioblastoma cells and other cancer cell types (14, 38). In contrast, other groups have observed a downregulation of HIF in acidic conditions (39, 40). By systematically assessing the effects of different levels of acidosis on HIFs we show that acidosis enhances HIF levels down to a pH of 6.6–6.8, while lower pH values decrease HIF expression, pointing to a narrow pH window for acidotic activation of the HIF pathway. Importantly, these values correspond to the extracellular pH that has been measured in the majority of tumors, including glioblastoma (14, 23), underlining the physiologic relevance of our findings.

The levels and activity of HIFs are regulated by multiple mechanisms. A central place among those is taken by the PHD/VHL-dependent HIF hydroxylation, ubiquitination and degradation. A previous report has shown that a highly acidic pH (5.8-6.2) can lead to sequestration of VHL in the nucleolus and to HIF stabilization (24). However, several lines of evidence that we present here strongly argue against a role for the VHL/PHD degradation pathway in mediating the effect of acidosis on HIF stability in the pH range that is typically encountered in tumors. For example, renal carcinoma cells lacking VHL still responded to acidosis with a striking increase in HIF levels. Furthermore, HIF induction under acidic pH was not affected by inhibition of PHDs. Instead, we show that acidosis stabilizes HIF1α and HIF2α through the upregulation of HSP90, independently of PHD and VHL. Importantly, inhibition of HSP90 function by geldanamycin or a dominant-negative mutant abolished the stabilization of HIF1/2α at low pH. Among alternative, PHD-VHL–independent mechanisms that regulate HIF stability HSP90 has been well characterized (reviewed in ref. 41). The multifunctional scaffold protein RACK1 binds the PAS-A domain of HIF with subsequent recruitment of the ubiquitin ligase complex that mediates its degradation, in a manner analogous to the mechanism activated by VHL, but independent of O₂/PHD-mediated HIF hydroxylation (28). HSP90 competes with RACK1 by binding to the HIF PAS-A domain (42), thereby stabilizing HIF (28). Taken together, our results highlight the central importance of acidosis as a key microenvironmental factor that regulates the HIF response synergistically with decreased oxygen tension through an HSP90-dependent, but PHD/VHL and oxygen-independent mechanism.

HSP90 forms a chaperone complex that can stabilize and activate a number of cellular proteins. In cancer cells HSP90 plays an additional important role by protecting various mutated or overexpressed proteins against misfolding and degradation, thereby facilitating oncogene addiction, counteracting proteotoxic stress and enabling cancer cell survival (43). HSP90 upregulation has been found in several types of cancer and has been linked to poor prognosis and increased tumor aggressiveness (44–46). This has led to the development of a number of small-molecule inhibitors, including geldanamycin and its derivatives that are currently in clinical trials for various tumor types (47). Our results highlight a general function of HSP90 in tumor physiology, revealing a novel role of HSP90 as the key mediator of the acidosis-dependent potentiation of HIF function. The clinical relevance of our findings is supported by the fact that HSP90 is prominently upregulated within the hypoxic niche of human glioblastomas and a high level of HSP90 is linked to the upregulation of HIF targets and CSC marker genes. Interestingly, a higher sensitivity of CSCs to HSP90 inhibitors compared with neural stem cells or other untransformed cells has recently been demonstrated (48, 49). Therefore, blockade of HSP90 function in glioblastoma, where HSP90 inhibitors have not been tested in clinical trials so far, could represent a viable therapeutic strategy. An alternative possibility is to aim at neutralizing tumor acidosis itself, and indeed recent studies have shown that this can be achieved by simple interventions such as bicarbonate administration, resulting in curtailed tumor growth, invasion, and metastasis (33, 50).

The synergistic increase of HIF levels and function by hypoxia and acidosis likely affects multiple HIF-dependent processes that collectively can promote tumor aggressiveness. Among those, the maintenance and expansion of the CSC pool likely plays an important role, as it is linked to numerous aspects of tumor progression that are stimulated by both hypoxia and acidosis. We and others have previously shown that CSCs are located and controlled within a hypoxic niche through HIF (12, 13). Our data support a model in which the hypoxic tumor niche induces an acidic environment, which synergizes with decreased oxygen availability to potentiate the hypoxic response and enhance HIF-dependent downstream functions in a positive feedback loop (Fig. 7C). Given the crucial role of HIFs in tumor progression, the additive induction of HIF by hypoxia and acidic pH via PHD- and HSP90-dependent mechanisms, respectively, may provide a means to fully activate HIF signaling within the hypoxic niche. Importantly, these two, synergistically acting mechanisms may allow tumor cells to fine-tune and flexibly induce HIF activity in response to different microenvironmental parameters (O₂ level, pH) to activate key hallmarks of cancer (Fig. 7C). Thus, by providing important mechanistic insight into the synergistic control of HIF function and CSC maintenance through central physiologic parameters of the tumor microenvironment, our work uncovers potential avenues for developing novel therapeutic approaches targeted against hypoxia/acidosis-driven tumor progression.

T. Acker was a consultant/advisory board member for Merck Serono (2014/2015; minor relationship). No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Acker

Development of methodology: A. Filatova, S. Seidel, N. Böğürcü, B.K. Garvalov, T. Acker

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Filatova, S. Seidel, N. Böğürcü, S. Gräf, B.K. Garvalov, T. Acker

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Filatova, S. Seidel, N. Böğürcü, B.K. Garvalov, T. Acker

Writing, review, and/or revision of the manuscript: A. Filatova, S. Seidel, N. Böğürcü, B.K. Garvalov, T. Acker

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Filatova, S. Seidel, N. Böğürcü, B.K. Garvalov, T. Acker

Study supervision: B.K. Garvalov, T. Acker

We would like to thank Barbara Lafferton, Gudrun Schmidt, Kerstin Leib, Carmen Selignow, and Tanja Diem for excellent technical assistance.

This work was supported by grants from the Deutsche Krebshilfe (T. Acker, B.K. Garvalov), the German Ministry of Education and Research (BMBF) within the National Genome Network (NGFN^plus) and Brain Tumor Network (BTN) (T. Acker), the DFGKFO210 (T. Acker, B.K. Garvalov), DFG SPP1069, DFG SPP1190 (T. Acker), the DFG Clusters of Excellence Cardio-Pulmonary System (ECCPS; T. Acker), LOEWE-OSF, UKGM Kooperationsvertrag §2, 3 (T. Acker, B.K. Garvalov), and the von Behring-Röntgen Foundation (T. Acker, B.K. Garvalov).

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