HIF Induces Human Embryonic Stem Cell Markers in Cancer Cells

Stem cells grow in a specialized microenvironment, the stem cell niche, which can regulate the balance among self-renewal, differentiation, and stem cell quiescence. Several factors are important within the niche, including cell interactions and environmental factors (1). Recent evidence suggests that many stem cells are also localized in areas with, and that benefit from, low oxygen, supporting the hypothesis that hypoxia might be important for the undifferentiated phenotype of stem/precursor cells (1, 2).

Many solid tumors contain poorly vascularized regions that are severely hypoxic (3) and contribute to cancer progression by activating transcription factors that promote cell survival, tumor angiogenesis, and metastasis (2, 3). Tumor hypoxia is also associated with resistance to radiation and chemotherapy (4). It is therefore not surprising that tumor hypoxia is associated with a more aggressive disease course and poor clinical outcomes (4).

A small proportion of cancer cells exhibit stem cell properties (5). These cells, which have been considered as tumor-initiating cells or cancer stem cells, show the ability of self-renewal and multipotential differentiation, and have the ability to initiate and propagate tumors (5). Recent findings suggest that these tumorigenic cells are rather common in some cancers (6), suggesting that stem-like cell state could be more dynamic than previously thought. It has been suggested that hypoxia could contribute to the formation of a cancer stem cell niche within the tumor (1).

Low oxygen levels stabilize hypoxia-inducible factor (HIF)α subunits and thereby induce the transcriptional activity of the heterodimers comprising an α subunit and a β subunit (HIFβ/ARNT) that can bind to hypoxia response elements and activate multiple genes, including cMYC and OCT4 (2, 3, 7, 8). In some cells, HIF signaling is also known to regulate cellular metabolism by upregulating the expression of the glycolytic genes and downregulating mitochondrial activity by transactivating PDK1, a repressor of pyruvate dehydrogenase, and upregulating miR-210, and a suppressor of the iron–sulfur cluster assembly protein U (ISCU; ref. 9). Under normoxic conditions, HIFα undergoes prolyl hydroxylation, binds to an ubiquitin E3 ligase, the Von Hippel–Lindau (VHL) protein, and undergoes polyubiquitination-dependent rapid proteasomal degradation. Other means of HIF regulation have also been described, including regulation through enzymes involved in the Krebs cycle and microRNA (miRNA)- or histone deacetylase-dependent regulation (10–13).

Some aggressive cancers and cancer stem cells display gene expression signatures characteristic of embryonic stem cells (ESC; ref. 14–16). However, whether OCT4 is expressed and functions in cancer cells is not yet clear (17, 18). It has been proposed that HIFs play a role in tumor aggressiveness, and the HIF target, miR-210, could be a circulating biomarker for certain tumors (19, 20). Whether hypoxia and, specifically, active HIFs are responsible for the human ESC (hESC) signature observed in aggressive tumors is, however, unknown.

Data presented here show that HIF induces hESC markers, including the critical induced pluripotent stem cell (iPSC) inducers, OCT4, SOX2, NANOG, MYC, and miR-302 in cancer cells. Furthermore, HIF, combined with the traditional iPSC inducers, is efficient in generating iPSC-like colonies that are highly tumorigenic. In prostate tumor specimens, HIF1α colocalizes with hESC markers, NANOG and OCT4, and expression of these stem cell markers positively correlates with high Gleason score, indication of prostate tumor aggressiveness. Furthermore, primary nonstem (NS) glioma cells are able to form neurospheres that upregulate hESC markers in hypoxic but not in normoxic conditions.

hESCs lines were obtained from Wicell Research Institute (Madison, WI) and cultured as previously described (21, 22).

HCT116, HT29, DLD1, and RKO (colorectal carcinoma); HeLa and ME180 (cervical carcinoma); A549 and H1299 (lung carcinoma); MCF7 (breast carcinoma); U251 (glioma); and Hep3B and HuH7 (hepatocarcinoma) cells were from the American Type Culture Collection. 786-O cells transfected either with an empty vector (EV) or wild-type VHL (23), and HCT116 Dicer hypomorph line (HCT116 Dcr-, ref. 24) were obtained from Dr. W.G. Kaelin (Dana-Farber Cancer Institute, Boston, MA) and Dr. B. Vogelstein (Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University, Baltimore, MD), respectively.

Cultures enriched for or depleted of glioma stem cells (GSC) were isolated from primary human brain tumor specimens as previously described in accordance with a Duke University Institutional Review Board (IRB)–approved protocol concurrent with the national regulatory standards with patients signing for informed consent (25). CD133⁺ cells were designated as GSCs, whereas CD133⁻ cells were used as NS glioma cells.

For hypoxia induction, cells were cultured in multigas incubators (Sanyo). Nitrogen gas was supplied to the chambers to induce a controlled reduced percentage of oxygen. For normoxia, cells were cultured in incubators containing 5% CO₂ and atmospheric concentration of O₂, approximately 20% O_2.

A549 cells were first infected overnight in 6-well plates by retroviruses with nondegradeable (nd) forms of HIF1α and HIF2α (23) or an EV control, and then infected 6 days later by lentiviruses expressing OCT4/SOX2 and NANOG/LIN28 (OSLN, ref. 26) overnight. Alternatively, cells were infected with OSLN and ndHIF viruses simultaneously. Viral transductions were performed in the presence of 4 mg/mL of polybrene (Sigma-Aldrich). Cells were trypsinized, transferred to 10-cm dishes, and cultured in Dulbecco's modified Eagle's medium (DMEM)/10% FBS. iPSC-like colonies were picked at day 12 postinfection based on hESC-like colony morphology. The selected colonies were subsequently expanded and maintained on irradiated mouse embryonic fibroblasts in hESC media. They were then stained for alkaline phosphatase (AP) activity using a Black Alkaline Phosphatase Substrate Kit II (Vector Laboratories). See Supplementary Materials and Methods for details on xenograft formation.

A 3.9-kb OCT4 promoter region-GFP fusion construct was linearized by using Apal I restriction enzyme and transfected into cells using Lipofectamine 2000, as previously described (27). Insertion events were selected by G418 (Invitrogen).

We constructed an HIF-sensor lentiviral vector expressing an enhanced yellow fluorescent protein (eYFP) under the regulation of 6 tandem repeats of HIF-binding sites (CGTGTACGTG), followed by a minimal human thymidine kinase (TK) promoter (Zhou and colleagues, in preparation).

The miR-302 cluster promoter pGL3 luciferase reporter vector contains a 3.9-kb miR-302 cluster promoter (as described in ref. 28) inserted between the KpnI and XhoI sites of pGL3-enhancer vector (Promega).

siRNA transfection was performed as described previously (24). For plasmid transfection, cells were transfected with nd forms of HIF1α and HIF2α (23) or EV control using Lipofectamine 2000. Overexpression of HIFs was confirmed by quantitative real-time PCR (qRT-PCR) and Western blot analysis 2 to 3 days after transfection. Cells containing the OCT4-GFP construct were seeded in chamber slides 24 hours before being transfected by plasmids for 6 hours, and cells were fixed 48 hours post-transfection with 4% PFA. GFP and DAPI (1 mg/mL of 4′,6-diamidino-2-phenylindole dihydrochloride, Sigma Aldrich) expressions were scored by confocal microscopy (SPE5, Leica).

Standard protocols were used. See details in the Supplementary Materials and Methods.

mRNA microarray analysis was done as described previously (22). Gene expression data analysis was done with the Rosetta Resolver gene expression analysis software (version 7.1 Rosetta Biosoftware).

Total RNA was extracted using TriZol (Invitrogen) and treated with DNase (Ambion). RNA abundance was determined using a Nanodrop ND-1000 spectrophotometer (Nanodrop Technologies). After reverse transcription reaction using Omniscript RT Kit (Qiagen), qRT-PCR was carried out by using SybrGreen Master Mix (Applied Biosystems) or, TaqMan assay (Applied Biosystems) on Applied Biosystems 7300 Real-time PCR cycler. All data were normalized to 28S or β-ACTIN transcript levels. The gene primers used are listed in Supplementary Table S6. TaqMan miRNA assays (Applied Biosystems) were used for miRNA qRT-PCR according to the manufacturer's protocol using RNU66 small nucleolar RNA as a loading control. Alternatively, the miRNA levels were determined using a quantitative primer-extension PCR assay (29). Ct values were converted to copy numbers by comparison with standard curves generated with single-stranded mature miRNAs and were expressed as copies/10 pg input RNA (approximately equivalent to copies/cell).

This study was conducted under approval by the University of Washington IRB for the research use of excess tissue from surgeries. Frozen blocks of prostate tumor tissue were obtained through the University of Washington Medical Center (UWMC) Division of Urology (Seattle, WA). Serial sections were stained with anti-HIF1α, anti-NANOG, and anti-OCT4 antibodies (Supplementary Table S7) using a protocol previously described (30). Gleason pattern of prostate tumors was given by a pathologist, and CD10, CD107b, and CD104 (BD Biosciences) antibodies were used to stain tumor specimens (30). Prostate cancer glands were scored in the presence of HIF1α and NANOG staining. See Supplementary Materials and Methods for details. For each patient, local Gleason grading was done as described previously (31) in individual microscopic areas (×100 magnification). Particularly, Gleason 5 was assigned when single round and undifferentiated epithelial cells were observed without any trace of gland formation.

Formalin-fixed and paraffin-embedded blocks of glioblastoma multiforme (GBM) were obtained through the UWMC Division of Neuropathology with IRB approval. Serial sections were processed from neuropathologically verified representative sections of grade IV glioblastoma according to a previously described protocol (32) and stained with HIF1α and NANOG antibodies (Supplementary Table S7).

P values were calculated using Student t test. As shown in Figs. 3–5, * indicates P < 0.05, and ** indicates P < 0.01. Bars show SEM for 3 or 4 separate experiments.

We analyzed the common gene expression signature of 11 different cancer cell lines under hypoxic conditions (2% O₂) and compared it with the hESC gene expression pattern (9 hESC lines; ref. 22). A significant overlap was observed between the differentially expressed mRNAs in these 2 groups (Fig. 1, cluster 5; Supplementary Fig. S1; Supplementary Table S1). Many of the overlapping mRNAs corresponded to genes involved in stemness and in the reprogramming of fibroblasts to hESC-like cells, iPSCs (Fig. 1B; refs. 26, 33). Among the hypoxia-upregulated genes were cMYC, KLF4, OCT4, and NANOG (Fig. 1B). The hypoxia-induced OCT4 upregulation was validated in multiple cancer cell lines using qRT-PCR (Fig. 1C). Multiple primers were used to confirm that the isoform of OCT4 shown to be involved in stemness (OCT4-A) was upregulated in these hypoxic cancer cell lines (Supplementary Fig. 2A and B; ref. 18). OCT4 was previously shown to be a transcriptional target of HIF2α (7). Overall, the HIF target genes and genes enriched in tumors were observed in the overlap between hypoxic cancer cell and hESC signatures (Supplementary Fig. S1A, Supplementary Tables S1 and 2; brain, prostate, and kidney tumors; Supplementary Table S3). Particularly, genes expressed in Wilms tumors were found in the hypoxic cancer cell and hESC overlap (Supplementary Fig. S1B and C). Wilms tumor and hESC similarity has previously been identified on the epigenetic level (34). The data presented suggest that HIF activity regulates these processes.

In ME180 cells, whereas SOX2, a well-established stem cell marker (26, 33), was not among the group of hypoxia-induced mRNAs at 24 hours, it was significantly induced at later time points (Fig. 1D). Moreover, OCT4, SOX2, and NANOG remained upregulated in these cells grown for multiple days in hypoxia compared with normoxia (Fig. 1D). Because multiple pseudogenes for OCT4 exist, we tested whether the true OCT4 promoter was responsive to hypoxia in cancer cells by stably integrating OCT4-GFP reporter in ME180 and U251 lines (ME180-OCT4-GFP and U251-OCT4-GFP). When these lines were incubated for 2 days in 2% O₂, 6- and 3-fold increase of GFP-positive cells was observed, respectively (Supplementary Fig. S3A and B, Fig. 2C). We also confirmed induction of OCT4 protein in ME180 cells by flow cytometry (Fig. 1E). The OCT4 antibody used in these experiments could detect a nuclear signal in tumor tissues (Fig. 4Q, Supplementary Figs. S10 and S11). Although in some situations and cell types, hypoxia is shown to affect cell division and/or cell death (3), no discernible changes in these processes were observed in hypoxia (2% O₂) for the cell lines analyzed (Supplementary Fig. S3C). Thus, the increased expression of OCT4, NANOG, and SOX2 was unlikely due to altered cell division or survival.

We also detected a convergence of miRNA signatures between hypoxic cancer cell lines and undifferentiated hESC lines. Several of the hESC-enriched miRNAs (22, 29), including miR-302a-b, miR-106a-b, miR-92, miR-372, miR-19b, miR-130a, miR-30e-5p, and miR-195, were found significantly upregulated in the cancer cell lines grown in low oxygen (Fig. 1F; Supplementary Table S4), indicating that the cancer lines move toward a stem cell–like miRNA profile when exposed to hypoxia. The hypoxia-induced upregulation of miR-302b and miR-106a in ME180 cells was validated by further qRT-PCR analysis (Supplementary Fig. S4). We tested whether miR-302 promoter was responsive to HIF using ndHIF1α and ndHIF2α (23) and a reporter construct of the miR-302 promoter linked to luciferase. Luciferase activity was upregulated in the presence of ndHIF1α and ndHIF2α, showing that the miR-302 promoter was responsive to ndHIF expression (Fig. 1G).

As reported previously, miR-210 was also consistently induced in response to hypoxia (Fig. 1F; 19, 20). miR-210 has been shown to increase MYC activity by inhibiting MNT, the antagonist of MYC (24), and to regulate metabolism by suppressing the ISCU (19). Therefore, HIF-dependent miR-210 upregulation might also contribute to the stem-like phenotypes in these hypoxic cancer cells.

To test whether the observed OCT4 upregulation was due to HIF expression, we overexpressed ndHIFs in U251 and ME180 cell lines with the OCT4-GFP construct (ref. 27; Fig. 2C). These engineered cells express GFP under the control of a 3.9-kb promoter region of OCT4 containing 3 HIF binding sites (7). Expression and activity of the ndHIF1α and ndHIF2α were first validated with Western blot analysis and HIF-binding-repeat (6×HBR)-YFP sensor construct (Fig. 2A and B). Upon transfection of U251-OCT4-GFP and ME180-OCT4-GFP with the ndHIF constructs, GFP expression was significantly increased, showing that ndHIF could activate the OCT4 promoter in U251 and ME180 cancer cell lines (Fig. 2C; Supplementary Fig. S5A–H). Endogenous hESC marker expression was also significantly increased due to ndHIF expression (Supplementary Fig. S5I).

To further test the HIF pathway function in hESC marker expression, we used 786-O and RCC4 renal carcinoma cell lines lacking VHL activity and therefore expressing constitutively active HIF (23). In normoxia, both 786-O and RCC4 cells lacking VHL function showed a significant increase in the expression of endogenous hESC markers compared with the corresponding lines with active VHL (Fig. 2D; Supplementary Fig. S6A–E). By using OCT4-GFP sensor, we showed that VHL-deficient cells expressed significantly higher levels of GFP than those with restored VHL function, suggesting activation of the OCT4 promoter by stabilized HIFs due to the lack of VHL (×30 increase; Fig. 2E). A 24-fold increase in GFP-positive cells was observed in 786+VHL cells after overexpression of ndHIFs (Fig. 2E), showing that these cells can respond to stabilized HIF. These data show that the upregulation of hESC markers in renal carcinoma lines correlated with the lack of VHL activity.

We further tested HIF dependency for hESC marker upregulation by transfecting ME180 and HCT116 cells with siRNAs directed against HIF1α, HIF2α (EPAS1), HIFβ (ARNT), and luciferase (control). The efficacy and specificity of the siRNAs were confirmed by qRT-PCR or microarray analysis (Supplementary Fig. S7). Hypoxia-induced expression of most hESC markers was dependent on HIFβ, HIF1α, and HIF2α in ME180 and HCT116 cells (Fig. 2F and G). No increase in cell death was observed in these experiments, possibly due to the short incubation period in hypoxia after HIF knockdown (Supplementary Fig. S7B). Moreover, knockdown of HIF1α in RCC4 cells upregulated genes nonrelated to apoptosis, such as TAF9B, AREG, and SGK3. These same genes were downregulated by hypoxia treatment in RCC4+VHL cells (Supplementary Fig. S7D). Thus, the observed reduction of stem cell markers was a result of HIF knockdown and not of defects in viability.

Collectively, these data were consistent with the conclusion that the hypoxia-induced hESC mRNA and miRNA signatures in cancer cells are HIF-dependent.

Because hypoxia results in an HIF-dependent upregulation of stem cell markers that are shown to be sufficient to induce a pluripotent state (iPSC; refs. 26, 33; Figs. 1 and 2), we tested whether these factors in our hands can induce iPSC colonies in cancer cells and whether such cells would have tumorigenic capacity. We introduced OCT4, SOX2, NANOG, and LIN28 with or without ndHIFs to human lung adenocarcinoma (LAC) epithelial cells (A549; Fig. 3A). Interestingly, iPSC-like colonies were observed more rapidly in the A549 cell line in these conditions than in MRC5 [a normal, primary lung fibroblast line (Fig. 3A–E; A549 10 days, MRC5 20 days)]. Furthermore, introduction of ndHIF increased the efficiency of the iPS-like cell induction in the A549 cancer cell line (7-fold increase; Fig. 3F). The iPSC-like colonies generated from A549 were positive for AP staining (Fig. 3G). The endogenous NANOG and OCT4 levels in the A549 iPS–like cells were lower than in normal H1 hESCs (Fig. 3H), and the promoter of OCT4 was only partially unmethylated (Fig. 3I), suggesting that the cells, while dedifferentiated, were not fully reprogrammed.

We hypothesized that partially reprogrammed cancer cell colonies might model the process that takes place in hypoxic regions of primary tumors and thereby show high tumor aggressiveness. We challenged this hypothesis by assessing the in vivo tumorigenic capacity of these partially reprogrammed A549 iPSC–like colonies by injecting them into the femoral muscle of immunocompromised mice. Xenograft tumors obtained were large and palpable 10 days after injection of A549 (OSNL+HIFs) iPSC–like colonies, revealing a very fast growth. No obvious tumor growth could be observed at 10 days in mice injected with A549 cells or A549 (HIF), whereas in 2 weeks, palpations began to reveal a small growth in mice injected with A549 (HIF). A hematoxylin and eosin staining on sections of A549 iPSC–like colony-induced tumors revealed highly aggressive malignant solid tumors with regions of central differentiation (Supplementary Fig. S8), high mitotic index (Fig. 3J), local invasion (Fig. 3K), necrosis, and nuclear atypia (Supplementary Fig. S8). The small and cuboidal cells with cytoplasmic droplets observed in the central portions of the xenograft were positive for both acidic and neutral mucins [magenta and blue Alcian blue (AB)-periodic acid Schiff (PAS) staining, Fig. 3L], attesting to the presence of some differentiated goblet cells. In accordance with partial dedifferentiation, the AP-positive A549 iPSC–like colonies generated aggressive invasive tumors, not typical teratomas (Supplementary Fig. S8).

These data show that expression of hESC markers in A549 cancer cells partially dedifferentiates the cells generating highly aggressive tumors.

We investigated the frequency of colocalization of HIF1α, NANOG, and OCT4 proteins in serial sections of primary prostate tumors in patient samples. Skin, seminomas, and hESCs were used as positive controls to validate the antibodies (Supplementary Fig. S9 and data not shown). In the prostate cancer specimen 09-066C (Gleason 4+3), 16.5% of tumor areas were positive for HIF1α and 7.7% were positive for NANOG (Fig. 4A, G, and K; Supplementary Fig. S10A). A significant over-lap between HIF1α and NANOG was observed: 69% of NANOG-positive cells were also HIF1α-positive (P < 0.05), whereas 32% of the HIF1α-positive cells were also NANOG-positive (Fig. 4A). In addition to morphology, previously identified markers, CD107b, CD10 and CD104 were used to distinguish between tumor and benign glands (Fig. 4O and P, Supplementary Fig. S10A; ref. 30). In specimen 07–020C, Gleason 4+3, the overlap between HIF1α and NANOG was also highly significant (91% of NANOG-positive cells were also HIF1α-positive; Fig. 4B, H–I, and L–M; Supplementary Fig. S10B). Furthermore, the 2 stem cell markers, NANOG and OCT4, were observed in the same prostate tumor regions (87% of NANOG-positive cells were also OCT4-positive; Fig. 4B, L, and Q). In the specimen 07–091LA, Gleason 3+4, of the 2,039 prostate glands counted 36.5% were positive for HIF1α, whereas 20.8% were positive for NANOG (Fig. 4C, J, and N). Highly significant overlap between NANOG and HIF1α-stained glands was observed with 99% of the NANOG-positive glands also positive for HIF1α (P < 0.01), whereas 57% of HIF1α-positive glands were also positive for NANOG (Fig. 4C). A detailed analysis revealed a high correlation in the number and location of positive cells in the overlapping regions for HIF, NANOG and OCT4 staining (Fig. 4 D–F). We performed a goodness of fit test of a linear regression between the percentage of NANOG- or OCT4-positive and that of HIF1α-positive cells and obtained an R square value ranging from 0.55 to 0.65. The results suggest that NANOG staining correlates with HIF1α and OCT4 staining (R square > 0.5) although the correlation is weakened by a few outliers and/or variation seen in different microareas of the tissue sections (Fig. 4D–F). These results are in accordance with the mRNA expression data (Fig. 1A–C), showing that a subset of hESC mRNAs is expressed in hypoxic cancer cells and the protein expression of members of the hESC key regulators, NANOG, and OCT4 correlates with HIF expression in primary prostate tumors. These data, however, also point out that not every hypoxic cancer cell can upregulate NANOG and OCT4, suggesting that other factors including the history of the cell and the level of HIF may play a role.

To assess the correlation of HIF1α, NANOG, and OCT4 expressions with clinicopathologic outcome of prostate cancer, we sampled more patients for our analysis. Since all specimens we analyzed were from recent occurrences of the disease (less than 3 years) and the follow-up clinical characterizations were not available yet, we relied on Gleason score (measures the diversion from the normal gland structure), the single most important prognostic factor in prostate cancer, for indication of disease outcome (31). When we analyzed Gleason 5 glands only (most unstructured glands), we observe highly enriched expressions of NANOG and OCT4 (56.2% and 41.7%, respectively, Fig. 4R). These values are significantly higher than the observed frequencies for NANOG and OCT4 when all diseased glands are included to the analysis (Gleason 3–5 together; Fig. 4R). We also show that the coexpression of HIF1α and NANOG in Gleason 5 glands only (20.4%) significantly increased as compared with the levels in overall diseased glands (12.3%).

Taken together, the significantly higher frequency of expressions of NANOG and OCT4 in Gleason 5 glands suggest the importance of the stem cell-specific factors in the prostate cancer progression, and the increased coexpression of HIF1α and NANOG support our hypothesis that HIF1α plays a role in the cancer progression by acquiring stem cell factors expression.

Our mRNA profiling of cancer cell lines also revealed an upregulation of known cancer stem cell markers in hypoxia (Supplementary Table S5). In hypoxic glioma cell line U251, we observe an upregulation of CD133 (PROM1), a glioma stem cell (GSC)–enriched marker that was previously shown to be induced at the cell surface by hypoxia in glioblastoma cells (refs. 35, 36; Supplementary Table S5A). To test whether similar correlations between hypoxia, GSC, and hESC markers were observed in primary tumors, we isolated and analyzed glioma cells from primary tumors from multiple patients. We generated cultures enriched for or depleted of the glioma stem–like subpopulation (25). Enrichment of the GSC population was validated by several functional assays, including fluorescence-activated cell sorting for cell surface markers such as CD133 (37). Following separation of the stem and NS populations, we subjected the cells to hypoxia and analyzed OCT4, NANOG, SOX2, OLIG2, CD133, cMYC, BMI-1, and MUSASHI by qRT-PCR. Of these genes, in addition to CD133, we found that OCT4, NANOG, and cMYC were consistently upregulated by hypoxia in populations obtained from several patient donors (Supplementary Fig. S12A). As shown previously, many primary glioma sections were found to express nuclear HIF1α (4). Some of these tumor regions also showed nuclear NANOG and OCT4 expressions (Fig. 5A, Supplementary Fig. S11). We therefore tested potential functional correlation between hypoxia, neurosphere formation, and iPSC inducer expression.

To test whether hypoxia-induced neurosphere formation capacity correlates with hESC marker expression, we first subjected glioma stem and NS cell populations to normoxia versus hypoxia to test their ability to form spheroids, a functional assay linked to self-renewal (ref. 25; Fig. 5B and C, Supplementary Fig. S12B). As described previously, only CD133+ GSCs were able to form neurospheres in this media in normoxia (Fig. 5B). In contrast, we found that culture at 2% O₂ supported neurosphere formation in both the stem and NS cell populations. Specifically, NS cells cultured at 2% O₂ formed larger neurospheres at a higher frequency (16-fold increase in hypoxia versus normoxia, P < 0.01; Fig. 5B and C, Supplementary Fig. S12C). This was also seen in the stem population (1.5-fold increase, P < 0.01; Fig. 5B and C). We analyzed on the molecular level whether the hypoxia-induced increase in neurosphere formation capacity correlates with hESC marker induction. Interestingly, many of the hESC markers were upregulated in hypoxia-induced NS–derived neurospheres (2–6-fold increase; Fig. 5D, Supplementary Fig. S12F). In particular, some of the key hESC-enriched miRNAs were upregulated in hypoxia-induced neurospheres (Fig. 5D). These data show a correlation between hypoxia-induced neurosphere formation and hESC marker expression.

Hypoxia did not significantly change the growth rate of glioma NS cells under these experimental conditions (Supplementary Fig. S12D and E), showing that observed increase in hESC markers and neurosphere number in hypoxia are not due to increased cell growth.

We analyzed expression profiles of cancer cell lines in hypoxia and show that 11 cancer cell lines (from prostate, brain, kidney, cervix, lung, colon, liver, and breast tumors) grown under hypoxic conditions share an overlapping gene expression signature with 9 hESC lines. Among the hESC-enriched genes that are upregulated in hypoxic cancer cell lines are the genes that are sufficient to induce reprogramming; OCT4, NANOG, SOX2, KLF4,cMYC, and miR-302 cluster. Moreover, HIF in combination with iPSC inducers OCT4, SOX2, NANOG, and LIN28 can induce efficiently partially reprogrammed A549 iPS–like cells that can generate highly aggressive tumors when injected in mouse muscle. Most of the hESC-enriched miRNAs are also upregulated in hypoxic cancer cells. Furthermore, we showed that hESC marker upregulations in cancer cell lines were HIF-dependent. We therefore analyzed the frequency of HIF and hESC marker colocalization in primary tumors and found a significant correlation between NANOG-, OCT4-, and HIF-positive regions in primary prostate cancer. Furthermore, the frequency of NANOG and OCT4 expressions increased significantly in higher-grade tumors. Many of the hypoxia-induced genes in cancer cells are previously identified markers for cancer stem cells. We therefore analyzed the correlation between hESC marker expression and stem cell character in hypoxic gliomas using a neurosphere assay as a functional read-out assay linked to self-renewal. Both an increase of hESC markers and neurosphere formation were observed in hypoxic glioma cells, indicating a hypoxia-induced correlation between stemness and neurosphere forming ability. Our data implicate HIF in acquisition of the dynamic state of stemness in pathologic cases.

Our large-scale profiling experiments revealed a common set in 624 genes that were expressed at higher levels in hypoxic relative to normoxic cancer cell lines and in undifferentiated relative to differentiated hESC lines. Comparison with other hESC-enriched gene sets revealed a significant overlap with the identified candidate group (Supplementary Fig. S1). Importantly, a set of 31 genes that showed an overlap between all 3 hESC profiling sets and hypoxic cancer cell lines are targets for HIF, OCT4, SOX2, and/or NANOG and are enriched in Wilms tumors that show hESC-epigenetic pattern (34). Previous analyses have shown that many prostate tumors express HIF (38, 39). Furthermore, NANOG was shown to be important for prostate tumor development (40). Our present study shows a significant overlap between HIF, NANOG, and Oct4 expressions in primary prostate tumors. Furthermore, increase in NANOG and OCT4 expressions and prostate cancer Gleason score shows the significant positive correlation. These data support the hypothesis that one of the roles of stabilized HIF transcription factor is to activate hESC marker expression in primary tumors.

We found that HIF can induce iPSC gene expression in cancer cells. Recent results have shown that hypoxia is advantageous for iPSC induction (ref. 41; J. Mathieu, W. Zhou, B. Stadler, C. Ware, H. Ruohola-Baker, unpublished); however, the mechanism is unknown. We propose that hypoxia is important in iPSC induction due to HIF transcription factor action. At least one of the iPSC inducers, OCT4, is shown to be a direct target of HIF transcription factor (7). Further studies are required to test whether the other known iPSC inducer genes NANOG, SOX2, cMYC, and KLF4 are direct targets of HIF. Although the promoters of other hESC genes, including miR-302, have potential HIF-binding sites, it is not yet clear whether they are directly regulated by HIF or if their expression is dependent on OCT4. It will also be interesting to reveal in the future whether other known HIF targets could play a role in iPSC induction.

We show that upregulation of hESC signature correlates with increased neurosphere frequency in primary glioma cells (this study; ref. 25). Increased neurosphere growth in hypoxia, even from single cells, suggests that hypoxia can increase the stem-like cell population in gliomas. More experiments are required to define whether the stem-like cell group selectively expands or whether de novo dedifferentiation takes place in the cancer cell population. However, the fact that a significant increase (16-fold) in neurosphere formation among the NS cell population is observed (Fig. 5C) suggests that the HIF pathway might have the capacity to initiate dedifferentiation by upregulating inducers of iPSC formation.

Previous studies with iPSC inducers (26, 33) have shown that reversal of differentiation to recapitulate the stem cell state seems to be inherent within each cell type under the proper conditions. Therefore, it is perhaps not surprising that a cancer cell can also revert to a cell with precursor character in response to stem cell marker expression (42, 43). However, it raises the question whether hypoxic regions of tumors have activated HIF that can upregulate stem cell marker expression on a level that could lead to induction of more potent tumor cells. Previous findings do suggest that this might be the case (14–16, 25). Furthermore, to mimic what could occur in vivo in hypoxic regions, we overexpressed iPSC inducers and HIFs in A549 lung cancer and picked colonies early in the reprogramming process. We obtained partially dedifferentiated iPSC-like colonies expressing a lower level of endogenous NANOG and OCT4 than hESCs. Postulating that these partially reprogrammed cells might mimic the HIF-induced state in tumors, we analyzed their tumor capacity in mice. Dramatically, the partially reprogrammed A549 colonies generated highly aggressive tumors that grew into large size in 10 days, whereas the control A549 line showed no tumor growth in such a short time period. These partially reprogrammed A549 colonies further mimicked aggressive lung tumors in that they did not produce teratomas but could generate some differentiated endothelial cells. An article published while our study was under consideration supports our findings by showing that OCT4 and NANOG overexpressions induce cancer stem cell–like properties in LAC (44). These authors also showed that OCT4 and NANOG are coexpressed in high-grade LAC and are associated with poor survival rate of patients.

It is critical to determine which of the iPSC inducer genes have potential to increase tumor potency. Answering this question has been initiated recently by showing that NANOG function is critical in tumor development (40, 45). It will be important to further analyze the capacity of OCT4, KLF4, and miR-302 in this process (17, 18). However, not all stem cell inducers are upregulated by hypoxia. One such exception is let-7 regulator, lin-28. Although lin-28 has been shown to promote transformation and to be associated with advanced human malignancies (46), this gene is not upregulated in hypoxia in any of the 11 cancer cell lines used in our study (however, we cannot rule out the possibility that hypoxia affects Lin-28 protein levels). These data suggest that upregulation of HIF pathway targets is important for stemness and thereby tumor potency; however, other factors also contribute to the process.

In summary, our data support the hypothesis that HIF targets are critical for stemness in malignant cells. Hypoxia is also beneficial for self-renewal of hESCs (1). One common denominator in these cell types, ESCs and cancer cells, is low mitochondrial activity (Warburg effect in cancers and low mitochondrial activity in stem cells; refs. 47, 48). Future studies will reveal whether the HIF-dependent metabolic pattern is critical for stemness.

No potential conflicts of interest were disclosed.

We thank members of the Ruohola-Baker Lab for helpful discussions; the Rosetta Gene Expression Laboratory for processing microarray experiments; M. Deng (CHDD Cellular Morphology Core), Dr. Colm Morrisey and Dr. X. Zhang (Department of Urology, University of Washington) for help and advice regarding the tumor specimens; E. Liebsekind, A. Bendoraite, D. Humes, and S. Einsele for technical help; Dr. W.G. Kaelin for providing pVHL-deficient 786-0 and RCC4 renal carcinoma cells transfected either with an empty vector or wild-type VHL and Dr. B. Vogelstein for HCT116 Dicer hypomorph cell line; and Dr. Cui for providing the OCT4-GFP construct.

This work was supported by the Brain Tumor Society, Goldhirsh Foundation and NIH grants NS047409, NS054276, CA112958, and CA116659 for J. Rich, the French Ministry of National Education/Ecole Normale Supérieure for A. Hubaud, the Robert Chang Scholarship of Sichuan University—University of Washington exchange program for A.J. Wang, the Regione Autonoma della Sardegna for C.M.A. Pinna, Jaconnette L. Tietze Young Scientist Award for B. Stadler, Breast Cancer Pilot Fund through the Fred Hutchinson/UW Cancer Consortium for J. Mathieu and H. Ruohola-Baker, R01GM083867-01 NIGMS for H. Ruohola-Baker, and 1P01GM081619-01 NIGMS and the Institute for Stem Cell and Regenerative Medicine (ISCRM) at the University of Washington for C. Ware, C.A. Blau, and H. Ruohola-Baker.

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.