DNMT3a epigenetic program regulates the HIF-2α oxygen-sensing pathway and the cellular response to hypoxia

Significance

DNA methyltransferase 3a (DNMT3a) mediates the de novo methylation of DNA to regulate gene expression and maintain cellular homeostasis. Mutations in DNMT3a in primary tumors suggest that the DNMT3a epigenetic program is modified during early tumorigenesis. We show that a major consequence of DNMT3a defects is the epigenetic deregulation and unscheduled activation of the EPAS1 (hypoxia-inducible factor 2α) gene that facilitates growth and viability under conditions of low oxygen availability. This represents a critical step during tumorigenesis, because cancer cells must adapt to hypoxia during the formation of the earliest multicellular foci. The data identify the DNMT3a epigenetic program as a gatekeeper of the hypoxic cancer cell phenotype.

Abstract

Epigenetic regulation of gene expression by DNA methylation plays a central role in the maintenance of cellular homeostasis. Here we present evidence implicating the DNA methylation program in the regulation of hypoxia-inducible factor (HIF) oxygen-sensing machinery and hypoxic cell metabolism. We show that DNA methyltransferase 3a (DNMT3a) methylates and silences the HIF-2α gene (EPAS1) in differentiated cells. Epigenetic silencing of EPAS1 prevents activation of the HIF-2α gene program associated with hypoxic cell growth, thereby limiting the proliferative capacity of adult cells under low oxygen tension. Naturally occurring defects in DNMT3a, observed in primary tumors and malignant cells, cause the unscheduled activation of EPAS1 in early dysplastic foci. This enables incipient cancer cells to exploit the HIF-2α pathway in the hypoxic tumor microenvironment necessary for the formation of cellular masses larger than the oxygen diffusion limit. Reintroduction of DNMT3a in DNMT3a-defective cells restores EPAS1 epigenetic silencing, prevents hypoxic cell growth, and suppresses tumorigenesis. These data support a tumor-suppressive role for DNMT3a as an epigenetic regulator of the HIF-2α oxygen-sensing pathway and the cellular response to hypoxia.

Metazoan life is dependent upon the use of molecular oxygen for an array of metabolic processes. Tissue hypoxia occurs during periods of imbalance between oxygen supply and consumption. One of the primary cellular responses to hypoxia is the activation of the hypoxia-inducible factor (HIF) program (1–4). HIF consists of oxygen-regulated α-subunits HIF-1α and HIF-2α and a constitutively expressed β-subunit (HIF-β). In the presence of oxygen, a series of nonheme Fe(II)- and 2-oxoglutarate–dependent dioxygenase oxygen sensors, referred to as HIF prolylhydroxylases (HIF PHDs), promote the hydroxylation of key proline residues on the HIF-α subunits (5, 6). This serves as a recognition site for the von Hippel-Lindau (VHL) tumor-suppressor protein, which mediates ubiquitination and proteasomal degradation of HIF-1α and HIF-2α (7–9). Hypoxia inhibits HIF PHDs, allowing HIF-1α and HIF-2α to evade VHL recognition and assemble with HIF-β to produce the active heterodimeric HIF factor. Once activated, HIF-1α and HIF-2α cooperate through common and distinct pathways to regulate hypoxic gene expression and cellular adaptation to hypoxia (10).

A notable feature of the HIF response is the differential expression pattern of HIF-1α and HIF-2α in normal tissues. HIF-1α mRNA is ubiquitous and constitutively expressed in adult cells. In stark contrast, HIF-2α mRNA is detected in a few cell types of adult tissues and is typically not expressed by epithelia (11). This suggests a physiological necessity to fine-tune the HIF program depending upon the cellular settings by negatively regulating the HIF-2α gene (EPAS1) upstream of the HIF oxygen-sensing enzymes. The negative regulation of EPAS1 is often compromised in cancers, as HIF-2α mRNA is observed in the vast majority of overt tumors (11–13). This is particularly evident in renal cancer. Elegant studies by the Maxwell group (13) and others (14) revealed that HIF-2α mRNA is absent in human kidney tubule epithelia but present in dysplastic foci of the nephron. In these incipient renal tumor cells, HIF-2α may function as an oncoprotein (15), collaborating with, or activating, multiple growth-promoting pathways including cancer stewards c-myc (16), ras (17), and EGFR (18, 19). Silencing of HIF-2α suppresses tumorigenesis of various genetically diverse cancers, further highlighting its central role in malignancy (16, 17, 20, 21), although this depends on the experimental context (22). Therefore, EPAS1 is silent in adult epithelia but undergoes unscheduled activation in several malignancies, driving proliferation in the hypoxic tumor microenvironment (23).

A clue to the mechanisms involved in the unscheduled activation of EPAS1 during early tumorigenesis may reside in its promoter, which harbors an enrichment of cytosine and guanine bases that often serve as sites of DNA methylation and epigenetic gene silencing (24–27). Cytosine methylation is catalyzed by a family of DNA methyltransferases (DNMTs) including DNMT1, DNMT3a, and DNMT3b. DNMT1 maintains the methylation pattern from the template strand to the newly synthesized strand during DNA replication (28). DNMT3a and DNMT3b are de novo methyltransferases that establish postreplicative methylation patterns (29). Alterations in DNA methylation patterns are common in tumors and likely play a central role in aberrant gene expression that characterizes the malignant phenotype (26, 30, 31). This is particularly evident for DNMT3a, as recent studies have identified mutations in DNMT3a in patients with acute myeloid leukemia (32, 33) or down-regulation of DNMT3a mRNA in a variety of solid tumors (34). It is suggested that DNMT3a is a tumor-suppressor gene and that its mutation, or mRNA down-regulation, contributes to reducing global DNMT3a methyltransferase activity (35, 36). Currently, a key challenge is to link aberrant methylation profiles commonly observed in malignant lesions, including alterations in the DNMT3a epigenetic program, to genes that directly promote the tumorigenic phenotype.

Here we show that DNMT3a methylates and silences EPAS1 in normal cells. Loss of DNMT3a observed in primary tumors and malignant cells causes unscheduled EPAS1 activation. This allows emerging cancer cells to exploit the HIF-2α program that facilitates cancer cell traverse of the hypoxic barrier and formation of tumors larger than the diffusion limit of oxygen. We suggest that the DNMT3a epigenetic program is a gatekeeper of the hypoxic cancer cell phenotype.

Results

EPAS1 Undergoes Epigenetic Silencing by DNA Methylation.

One of the salient features of the EPAS1 gene is the presence of a large CpG island surrounding the promoter (Fig. 1A) often associated with DNA methylation and epigenetic gene silencing (24–27). Because kidney tubule epithelial cells usually do not express HIF-2α (13), we hypothesized that EPAS1 may undergo epigenetic silencing by DNA methylation. Methylation-sensitive restriction digest PCR (MSRPCR) analysis of DNA isolated from human kidney samples showed cleavage resistance to the methylation-sensitive restriction enzyme HpaII within the EPAS1 promoter region (Fig. 1B, Left). This indicates the presence of substantial DNA methylation at the EPAS1 locus as each amplicon possessed at least two HpaII cleavage sites, which were digested efficiently by the methylation-insensitive isoschizomer MspI (Fig. 1B). As expected, the H19 gene was insensitive to HpaII digestion, indicating methylation at this imprinted locus (37), whereas the constitutively expressed GAPDH was digested, implying minimal methylation at this site (Fig. 1B). MSRPCR analysis also indicated the presence of methylated DNA at the EPAS1 locus in primary cultures of human renal tubule epithelial cells (RECs) (Fig. 1B, Right) and normal human astrocytes (NHAs) (Fig. 1C). The presence of methylcytosine in the EPAS1 promoter was further validated using the bisulfide deamination (Fig. S1A) and methylated-DNA immunoprecipitation (MeDIP) (Fig. 1D) assays. To determine whether these DNA methylation events are involved in the regulation of EPAS1 expression, RECs were treated with the general DNA methylation inhibitor 5-azacytidine (5-aza). Upon exposure to 5-aza, the EPAS1 promoter became significantly more sensitive to HpaII digestion (Fig. 1E and Fig. S1B), suggesting a loss in DNA methylation at this locus. This effect was correlated with an increase in HIF-2α mRNA (Fig. 1F), which was not caused by alterations in HIF-2α mRNA stability (Fig. S1C). The increase in HIF-2α mRNA in 5-aza–treated RECs was sufficient to produce detectable HIF-2α protein under hypoxic conditions (Fig. S1D). HIF-1α mRNA and protein levels were essentially unaltered by 5-aza in RECs (Fig. 1F and Fig. S1D), as expected. In situ hybridization demonstrated that 5-aza–induced expression of HIF-2α occurred in the vast majority of cells (Fig. 1G and Fig. S1E). This suggests that epigenetic silencing of EPAS1 by DNA methylation is a widespread phenomenon that explains the lack of HIF-2α mRNA in most cells composing the REC population.

Fig. 1.

Epigenetic silencing of EPAS1 by DNA methylation in normal human cells. (A) Schematic diagram of the EPAS1 promoter region and CpG island (dashed green) surrounding the transcription start site defined as DNA with at least 50% GC content over a minimum of 200 bp. The locations of the PCR amplicons (1–4) are indicated. (B) MSRPCR analysis of human kidney and REC samples. EPAS1 and the controls H19 and GAPDH are assessed for HpaII and MspI sensitivity before PCR amplification. (C) MSRPCR analysis of genomic DNA extracted from NHAs shows a similar pattern of DNA methylation at the EPAS1 locus. (D) MeDIP of the EPAS1, H19, and GAPDH loci from primary RECs. Inputs represents 1% of the lysates that were precipitated with a methylcytosine (^mC) antibody or IgG as a negative control. (E) MSRPCR of the EPAS1 locus, using primer set 4 and quantitative (q)PCR amplification, in RECs left untreated (Ctrl) or exposed to 5-azacytidine for 72 h. Values were calculated by assessing the percentage of DNA uncleaved following HpaII digestion relative to the input. (F) The expression of HIF-2α and HIF-1α mRNA was quantified by qPCR in 5-aza–treated RECs. Values are displayed relative to untreated RECs. (G) Global up-regulation of HIF-2α mRNA levels in 5-aza–treated RECs. In situ hybridization of HIF-2α and GAPDH mRNA on low-passage RECs untreated or exposed to 10 μM 5-aza for 72 h. (Insets) Images are lower magnification. Columns represent the mean of at least three experiments (n ≥ 3). Error bars represent SEM. Significance was measured by Student t test; **P < 0.01, ***P < 0.001.

DNMT3a Methylates and Silences EPAS1 in Epithelia.

We sought to determine which DNMT enzymes were responsible for the DNA methylation observed on the EPAS1 promoter. Chromatin immunoprecipitation revealed a strong and specific binding of DNMT3a, but not DNMT3b, at the EPAS1 locus (Fig. 2A). DNMT3b was efficiently pulled down during the experiment, as it bound the H19 locus as previously reported (38). Renal epithelial cells express the first (DNMT3a1) but not the second form of DNMT3a (DNMT3a2) (Fig. S2A; compare RECs with induced pluripotent stem cell RT-PCR and Western blot). Two different lentiviral-expressed shRNA constructs were produced that effectively silenced DNMT3a protein and mRNA (Fig. S2B) in RECs. The eye absent homolog 4 (EYA4) transcript, a known DNMT3a-regulated gene (39), was up-regulated following DNMT3a knockdown, demonstrating shRNA-mediated loss of DNMT3a function (Fig. S2C). DNMT3a-depleted RECs displayed increased sensitivity to HpaII cleavage at the EPAS1 locus, but not the imprinted H19 locus (Fig. 2B), and elevated levels of HIF-2α mRNA compared with control lines (Fig. 2C). In contrast, silencing of DNMT3b (Fig. S2D) had no significant effect on the methylation status of EPAS1 (Fig. 2B) or HIF-2α mRNA expression (Fig. 2C). Importantly, HIF-2α protein accumulated in DNMT3a-depleted RECs incubated under hypoxic conditions, but not in controls (Fig. 2D). These data highlight the role of DNMT3a in the control of HIF-2α upstream of the oxygen sensors by methylating and epigenetically silencing EPAS1.

Fig. 2.

DNMT3a methylates and silences EPAS1 in normal human RECs. (A) Endogenous DNMT3a, DNMT3b, and RNA polymerase II were immunoprecipitated with the associated chromatin from REC lysates. EPAS1, H19, and GAPDH loci were amplified by PCR. (B) MSRPCR analysis of genomic DNA extracted from RECs expressing control shRNA, shRNA targeting DNMT3b, or two independent DNMT3a shRNAs. EPAS1 (primer sets 2 and 4), H19, and GAPDH were amplified following MspI/HpaII restriction digestions. (C) RECs infected with lentiviruses expressing two shRNAs targeting DNMT3a and an shRNA targeting DNMT3b were assessed for HIF-2α mRNA expression levels by qPCR. Values are relative to unaltered (parental) cells. (D) Total protein extracts of RECs exposed or not to 1% oxygen atmosphere were immunoblotted for the expression of HIF-2α, HIF-1α, DNMT3a, DNMT3b, and an actin control. Columns represent the mean of at least three independent experimental repeats (n ≥ 3). Error bars represent SEM. Significance was measured by Student t test; *P < 0.05. NS, not significant.

DNMT3a Loss Activates EPAS1.

We next asked whether naturally occurring defects in DNMT3a could explain the unscheduled activation of EPAS1 commonly observed in primary tumors and overt cancer cells. Strikingly, methylation of the EPAS1 promoter detected in normal kidneys was consistently lost in pair-matched early (stage I) and overt (stage III) renal tumors (Fig. 3A and Fig. S3A), correlating with an increased expression of HIF-2α mRNA (Fig. 3B and Fig. S3B) and protein levels (Fig. S3C) (13, 14), whereas the control H19 and GAPDH loci maintained their methylation status (Fig. 3A and Fig. S3A). Consistent with the in vivo data, increased HIF-2α mRNA expression in renal cell carcinoma (RCC) and glioblastoma tumorigenic cell lines (Fig. S3 D and E) was associated with loss of EPAS1 DNA methylation observed in RECs and NHAs (Fig. 3C and Figs. S1A and S3F). Interestingly, loss of methylation in stage I and III tumors was correlated with a systemic reduction in DNMT3a mRNA expression (Fig. 3B and Fig. S3G) and protein levels (Fig. S3C), but not DNMT3b or DNMT1 (Fig. 3B and Fig. S3H), compared with adjacent kidney tissues. In addition, several primary tumors harbored gross truncations of DNMT3a mRNA, heralding loss of function (Fig. S3I). DNMT3a protein (Fig. 3D) and mRNA (Fig. S3J) levels were considerably lower in cancer cells compared with their normal counterparts, whereas DNMT3b (Fig. 3D and Fig. S3K) and DNMT1 (Fig. S3K) did not show consistent differences. Neither the DNMT3a1 nor DNMT3a2 isoform was detected in RCC and glioblastoma tumorigenic cell lines (Figs. S2A and S3L). 5-Aza treatment had no significant effect on HIF-2α expression in 786-0 RCC, as mRNA and protein levels in this cell line were already similar to those observed in treated RECs (Fig. 3E and Fig. S3M). This suggests that loss of DNA methylation observed in these cancer cells accounted for most of the EPAS1 expression, likely as a consequence of reduced DNMT3a activity. To test this, stable RCC 786-0 cell lines were produced that expressed exogenous DNMT3a, which bound to the EPAS1 and EYA4 genes (Fig. S3N). Reintroduction of DNMT3a in DNMT3a-defective RCC 786-0 was sufficient to induce HpaII digestion resistance at the EPAS1 locus (Fig. 3F) and reduce HIF-2α mRNA to levels similar to those observed in primary RECs (Fig. 3G). Expression of DNMT3a, but not DNMT3b, also reduced HIF-2α mRNA levels in RCC4 cells (Fig. S3O). Ectopic expression of DNMT3a restored methylation (Fig. 3F) and silenced EPAS1 (Fig. 3G) in glioblastoma, preventing HIF-2α protein expression even under hypoxic conditions (Fig. S3P). These data suggest that the unscheduled activation of EPAS1 observed in cancer cells is a direct consequence of naturally occurring DNMT3a defects.

Fig. 3.

Naturally occurring defects in DNMT3a trigger the unscheduled activation of EPAS1. (A) MSRPCR analysis of genomic DNA-isolated RCC tumors and normal adjacent tissue. EPAS1 (Upper) and H19 (Lower) loci were assessed. (B) HIF-2α, DNMT3a, and DNMT3b mRNA levels were quantified in eight stage I RCC tumors relative to normal adjacent tissue. Real-time PCR analysis of mRNA expression was standardized with GAPDH. (C) MSRPCR for EPAS1, H19, and GAPDH in RECs and RCC (786-0) or NHAs and glioblastoma (U87mg) cultured cells. (D) DNMT3a and DNMT3b protein expression levels were assessed by Western blot analysis in renal and glioblastoma cells. (E) HIF-2α mRNA levels were determined in untreated and 5-aza–treated RECs and 786-0 cells by qPCR. (F) RCC 786-0 cells stably expressing GFP or GFP-DNMT3a or U87mg cells expressing FLAG or FLAG-DNMT3a were analyzed by MSRPCR for methylation at the EPAS1, H19, and GAPDH loci. (G) RECs or RCC 786-0 stably expressing GFP or GFP-DNMT3a and NHAs or U87mg expressing FLAG or FLAG-DNMT3a were assessed for HIF-2α mRNA expression by qPCR. Values are relative to the parental 786-0 cell line. Columns represent the mean of at least three independent experimental repeats (n ≥ 3). Error bars represent SEM. Significance was measured by Student t test; *P < 0.05, ***P < 0.001.

DNMT3a Is a Regulator of the Hypoxic Cell Phenotype.

We predicted that the DNMT3a epigenetic program is intimately involved in hypoxic cellular life, in part by negatively regulating the HIF cascade of gene expression. To test this, the role of DNMT3a in cellular adaptation to hypoxia was examined under various settings (Fig. 4). Because DNMT3a regulates an array of different genes, the relative contribution of EPAS1 silencing in DNMT3a-mediated regulation of the hypoxic cell phenotype was addressed (Fig. 5). Silencing DNMT3a (Fig. S2B) had no discernible effects on REC proliferation and viability in normoxia (Fig. 4A). Intriguingly, DNMT3a-depleted RECs were significantly more proficient at proliferating and remaining viable under prolonged hypoxic conditions (Fig. 4A and Fig. S4 A and B). Likewise, reintroduction of DNMT3a had no measurable consequence on normoxic cancer cells, but significantly impaired their ability to proliferate and remain viable in hypoxia (Fig. 4B and Fig. S4 C and D). Next, an in vitro spheroid assay was used as a model system to study the normoxic–hypoxic transition that occurs in the core of early tumors. To do this, 50 cells per well of control U87mg cells were plated and growth was monitored over time. Initially, U87mg cells formed small spheroids (∼50 μm in diameter) that grew to ∼300 μm after 8 d (Fig. 4C, Upper and Fig. 4D, black line), becoming hypoxic, as HIF-2α protein could be detected by Western blot (Fig. 4E). In contrast, U87mg spheroids expressing DNMT3a initially formed small spheroids on day 1, but were unable to grow to a diameter of more than ∼150 μm (Fig. 4C, Lower and Fig. 4D, red line), coinciding with the diffusion limit of oxygen in spheroid masses (40). RT-PCR analysis demonstrated that HIF-2α mRNA expression was reduced in DNMT3a-expressing spheroids (Fig. 4E), confirming that epigenetic silencing of EPAS1 was maintained. Next, 10⁵ cells per well were plated in the spheroid assay to induce the formation of large masses that promptly became hypoxic in their cores. Control cells formed large and highly condensed spheroids that expanded over time (Fig. S4E). Staining with Hypoxyprobe confirmed that the cores of these control spheroids were hypoxic (Fig. 4F). DNMT3a-expressing U87mg cells formed more loosely packed masses that remained oxygenated at their core (Fig. 4F) and did not change size (Fig. S4E). Expression levels of exogenous DNMT3a in U87mg cells were similar to (clone 2) or higher than (clone 1) those of unaltered human astrocytes (Fig. S4F). Similar data were obtained for the RCC 786-0 cell lines (Fig. S4G), altogether supporting the notion that DNMT3a specifically modulates the proliferation and viability of hypoxic cells. Because reintroduction of DNMT3a prevents cancer cells from forming hypoxic cores, it would be predicted that it would also suppress their ability to form tumors in nude mice. Reintroduction of DNMT3a prevented RCC 786-0 and astrocytoma U87mg from forming tumors, even after a long period following injection (Fig. 4 G and H). Together, these data suggest a tumor-suppressor role for DNMT3a by in part preventing growth and viability of hypoxic cancer cells.

Fig. 4.

DNMT3a regulates cellular adaptation to hypoxia. (A) RECs transfected with control (Ctrl) or DNMT3a-specific shRNA were exposed to normoxic (21% O₂) or hypoxic (1% O₂) conditions for 96 h before a 1-h BrdU labeling or trypan blue staining. (B) Glioblastoma U87mg cells stably expressing FLAG or FLAG-DNMT3a were exposed to normoxic or hypoxic BrdU labeling or trypan blue staining. (C) Representative spheroids from FLAG- and FLAG-DNMT3a–expressing U87mg cells (50 cells per well) grown for the indicated times. (D) Quantification of spheroid diameter from FLAG- or FLAG-DNMT3a–expressing U87mg cells (50 cells per well) grown for the indicated times. (E) FLAG- and FLAG-DNMT3a–expressing 8-d-old U87mg spheroids were lysed and HIF-2α or DNMT3a mRNA/protein levels were assessed by RT-PCR or Western blotting (WB). (F) U87mg cells (10⁵) with or without DNMT3a stably expressed were plated and grown for 96 h before the addition of Hypoxyprobe to the media for 1 h. Spheroids were then prepared for cryosectioning, Hoechst staining (blue), and immunofluorescence detection of Hypoxyprobe (green)-positive cells. (G and H) Nude mouse xenograft assays performed with RCC 786-0 (G) stably expressing GFP or GFP-DNMT3a or glioblastoma U87mg (H) stably expressing FLAG or FLAG-DNMT3a. Tumor diameters were measured at the 9-wk end point and means were determined. Columns represent the mean of at least three independent experimental repeats (n ≥ 3). Scale bars, 100 μm. Error bars represent SEM. Significance was measured by Student t test; **P < 0.01, ***P < 0.001.

Fig. 5.

DNMT3a regulates cellular adaptation to hypoxia by preventing HIF-2α expression. (A) Silencing of HIF-2α in DNMT3a-depleted RECs suppresses adaptation to hypoxia. RECs were transduced with lentiviruses expressing shRNAs against DNMT3a and HIF-2α mRNA or control shRNA. Infected cells were grown under normoxic or hypoxic conditions before proliferation analysis by BrdU incorporation and trypan blue viability analysis. (B) Glioblastoma U87mg expressing FLAG or FLAG-DNMT3a infected with an empty vector (Ctrl) or a HIF-2α cDNA–containing lentivirus were grown in a normoxic or hypoxic atmosphere. Cell proliferation was measured by BrdU incorporation and cell viability by trypan blue staining. (C and D) Fifty cells per well of U87mg with FLAG-DNMT3a and stably coexpressing control empty vector or a HIF-2α lentivirus construct were plated, their growth was quantified (C), and representative (D) spheroids are presented from the indicated times. (E) U87mg cells (10⁵) with FLAG-DNMT3a and empty vector or HIF-2α cDNA were Hoechst-stained (blue) and stained for Hypoxyprobe (green)-positive cells. (F) U87mg cells stably expressing FLAG-DNMT3a and stably coexpressing control empty vector (Ctrl) or HIF-2α lentivirus constructs were injected into nude mice. Xenograft diameters were measured at end point. Scale bars, 100 μm. Column mean values are indicated, and error bars represent SEM. Significance was measured by Student t test; ***P < 0.001.

DNMT3a Regulates Cellular Adaptation to Hypoxia by Preventing HIF-2α Expression.

Finally, we wanted to look at the specific role of EPAS1 silencing in DNMT3a-mediated prevention of growth and viability under low oxygen tension. Silencing of HIF-2α in DNMT3a-depleted RECs prevented their proliferation and viability under prolonged hypoxic conditions to levels similar to those of unaltered controls (Fig. 5A and Fig. S5 A–C). This suggests that the growth and viability advantage observed in hypoxic RECs following silencing of DNMT3a is mediated, at least in part, by EPAS1 activation and the HIF-2α hypoxic program. In addition, DNMT3a-competent U87mg cells were transduced to express exogenous HIF-2α protein and mRNA to levels similar to those observed for endogenous HIF-2α (Fig. S5 D and E). This allowed methylation of the endogenous EPAS1 promoter to be maintained (Fig. S5F), whereas exogenous HIF-2α expression occurred in a DNMT3a-positive background (Fig. S5D). Ectopic expression of HIF-2α restored the ability of DNMT3a-expressing U87mg cells to proliferate and maintain viability under hypoxic conditions (Fig. 5B and Fig. S5G). Likewise, transient expression of HIF-2α was sufficient to restore DNMT3a-competent 786-0 cell survival in hypoxia to levels similar to those of unaltered cells (Fig. S5 H and I). Importantly, restoration of HIF-2α activity enabled cancer cells to grow past the point of oxygen limitation in spheroid assays (Fig. 5 C and D), producing a hypoxic core (Fig. 5E and Fig. S5J) and masses similar in size to DNMT3a-deficient controls (Fig. 4 C and D). Exogenous expression of HIF-2α was sufficient to restore the ability of DNMT3a-competent cells to form tumors in nude mice (Fig. 5F and Fig. S5K). These data suggest that DNMT3a-mediated silencing of EPAS1 plays a key role in its ability to restrict cellular adaptation to a hypoxic environment and promote tumor suppression.

Discussion

In this report, we present evidence that the unscheduled activation of EPAS1 observed in dysplastic foci stems from the deregulation of the DNMT3a epigenetic program, representing a key event that confers growth and survival properties to malignant cells under hypoxia. Incipient cancer cells encounter regions of hypoxia in the earliest multicellular foci as oxygen diffuses through only a few cellular layers. This represents a key challenge, as these cells must passage through the hypoxic barrier to pursue their tumorigenic endeavor. Loss of DNMT3a enables cells to exploit the HIF-2α hypoxic program, facilitating their adaptation to low oxygen tension. Reintroduction of DNMT3a restores epigenetic silencing of EPAS1, preventing cellular proliferation and viability under hypoxia, thereby inhibiting tumorigenesis. Ectopic expression of HIF-2α is sufficient to override DNMT3a tumor-suppressive capacity, underlining the role of the DNMT3a epigenetic program in HIF-2α–dependent tumorigenesis. These results suggest that naturally occurring alterations in DNMT3a provide the epigenetic landscape for enhanced cancer cell adaptation to the hypoxic tumor microenvironment.

The mechanism involved in EPAS1 regulation has remained an unanswered question in the field of tumor hypoxia. Whereas HIF-1α is constitutively and widely expressed in normal tissues, HIF-2α mRNA is only detected in a select subset of adult cells. Paradoxically, HIF-2α mRNA is detected in a majority of tumors in vivo and in overt cancer cell lines. The data shown here indicate that defects in DNMT3a-mediated epigenetic silencing of EPAS1 may explain the unscheduled expression of HIF-2α mRNA in early cancer. Interestingly, the appearance of HIF-2α mRNA in renal cancers is often associated with a decrease or outright absence of HIF-1α mRNA, stemming from alterations in the HIF-1α locus (41). One possible model to explain this apparent switch between HIF-1α and HIF-2α would be a concomitant inactivation of HIF-1α and a loss of DNMT3a activity, which would relieve EPAS1 from its epigenetic restraint. Mutations in upstream regulators, such as VHL in kidney, its breast-specific functional homolog SHARP1 (42), or alterations in other epigenetic programs (43), may serve to further amplify the HIF-2α response following EPAS1 activation. This provides an interesting example of how genetics interacts with epigenetics to modify cellular metabolism and confer the cancer hallmarks (24, 26, 27, 30, 31).

Recent studies have provided evidence that DNMT3a epigenetic reprograming occurs in human cancer. DNMT3a is one of the few epigenetic regulators recurrently mutated, or down-regulated, in primary tumors (32–34). Our data are consistent with these reports, as we have observed either a significant down-regulation or gross truncations of DNMT3a mRNA in stage I/III RCC samples and various overt cancer cells compared with their normal tissue and cell counterparts, respectively. Because DNMT3a undergoes posttranslational modifications, it is possible that the screen has missed other cancer-causing events associated with DNMT3a loss. Perhaps more importantly, reintroduction of DNMT3a in cancer cell lines is sufficient to suppress their ability to adapt to hypoxia and form tumors in xenograft assays. DNMT3a loss imparts these phenotypic properties, at least in part, by causing the unscheduled activation of EPAS1, which represents a key epigenetic driver event associated with cancer cell growth and survival in hypoxia. It remains unknown whether DNMT3a loss is an oncogenic event; however, it likely plays a central role in genetically diverse cancer cells to proliferate and remain viable under hypoxic conditions. Based on these observations and those of others, we propose that DNMT3a functions as a gatekeeper tumor suppressor (35, 36) by controlling hypoxic growth and viability. Obviously, further genetic studies will be necessary to clearly demonstrate biallelic-inactivating mutations of DNMT3a in primary tumors. Nonetheless, the data shown here underlie the fundamental role of epigenetic reprogramming in the attainment of the hypoxic cancer cell phenotype.

Experimental Procedures

Cell Culture and Reagents.

Human renal epithelial primary cells were kindly provided by Christopher Kennedy (Kidney Center, University of Ottawa) and maintained in epithelial cell medium (ScienCell). Normal human astrocytes were a kind gift of Alexandre Prat (Université de Montréal, Montréal, Canada) and grown in astrocyte growth media. Induced pluripotent stem cells were provided by William Stanford (Faculty of Medicine, University of Ottawa). Human renal carcinoma cell lines 786-0, KTCL, and RCC4 were obtained from the American Type Culture Collection. Glioblastomas U118mg and U87mg were generously provided by Ian Lorimer (Ottawa Regional Cancer Center, Ottawa, Canada). All nonprimary cell lines were maintained at 37 °C in a 5% CO₂ environment with high-glucose DMEM supplemented with 5% (vol/vol) FBS, 100 U/mL penicillin, and 100 µg/mL streptomycin. Cells were incubated at 37 °C in 21% or 1% O₂, 5% CO₂, and N₂-balanced atmosphere. 5-Azacytidine (10 μM) was added to the cell media fresh every day. Hypoxyprobe (100 μM) (Hypoxyprobe, Inc.) was added for 1 h before analysis.

Human Tissues and Sample Preparation.

Frozen primary tumors and normal adjacent kidney biopsies were obtained from the Ontario Tumour Bank (Ontario Institute for Cancer Research). Experiments were approved by the Ontario Cancer Research Ethics Board (TEC 018-10) and samples were selected randomly from individuals under 18 y old, with informed consent obtained prior to specimen deposition. Samples were crushed in liquid nitrogen and DNA/RNA was isolated with TriPure reagent (Roche) following the manufacturer’s instructions.

MSRPCR.

The MSRPCR assay amplified genomic DNA (1 μg) that was previously restriction endonuclease-digested for 48 h with the isoschizomers MspI or HpaII. Genomic DNA was extracted by lysing cultured cells or tissue in 1% SDS followed by proteinase K digestion, ethanol precipitation, and phenol-chloroform purification.

Multicellular Spheroid Assays.

Fifty or 1 × 10⁵ cells were plated in multiwell plates precoated with 1% SeaPlaque agarose (Cambrex) to prevent anchorage-dependent growth. To promote cell–cell adhesion and induce formation of the multicellular sphere, plates were rotated for 1 h and grown under standard conditions for the indicated number of days. Images were taken daily to measure cell growth. For hypoxia analysis, spheroids were washed with PBS, flash-frozen in OCT compound, sliced, and fixed in paraformaldehyde before immunohistochemistry analysis of pimonidazole adducts.

Xenograft Tumors.

Animal experimentation was performed in accordance with University of Ottawa policy and approved by the Animal Care Committee (CMM-181). Exponentially growing cells (10⁷) were resuspended in 200 μL PBS and injected s.c. into the flanks of CD-1 nude female mice (Charles River). Tumor growth was recorded weekly and animals were killed when the end point was reached or 8–9 wk postinjection.

Statistical Analysis.

For normal and tumor samples, P values were determined via paired t test and calculated using Statistica software (StatSoft). Significant differences between groups are shown as *P < 0.05, **P < 0.01, ***P < 0.001.