DEAD Box Protein Family Member DDX28 Is a Negative Regulator of Hypoxia-Inducible Factor 2α- and Eukaryotic Initiation Factor 4E2-Directed Hypoxic Translation

Hypoxia is a deficiency in oxygen delivery to tissues and is connected to physiological and pathophysiological processes such as embryonic development and cancer. The master regulators of oxygen homeostasis in mammalian cells are the heterodimeric hypoxia-inducible transcription factors 1 and 2 (HIF-1 and HIF-2, respectively). The oxygen-labile HIF-2α subunit has been implicated not only in transcription but also as a regulator of eukaryotic initiation factor 4E2 (eIF4E2)-directed hypoxic translation.

ABSTRACT

Hypoxia is a deficiency in oxygen delivery to tissues and is connected to physiological and pathophysiological processes such as embryonic development and cancer. The master regulators of oxygen homeostasis in mammalian cells are the heterodimeric hypoxia-inducible transcription factors 1 and 2 (HIF-1 and HIF-2, respectively). The oxygen-labile HIF-2α subunit has been implicated not only in transcription but also as a regulator of eukaryotic initiation factor 4E2 (eIF4E2)-directed hypoxic translation. Here, we have identified the DEAD box protein family member DDX28 as an interactor and negative regulator of HIF-2α that suppresses HIF-2α’s ability to activate eIF4E2-directed translation. Stable silencing of DDX28 via short hairpin RNA (shRNA) in hypoxic human U87MG glioblastoma cells caused an increase of eIF4E2 binding to the m⁷GTP cap structure and the translation of eIF4E2 target mRNAs (including the HIF-2α mRNA itself). DDX28 depletion elevated nuclear and cytoplasmic HIF-2α protein, but HIF-2α transcriptional activity did not increase, possibly due to its already high nuclear abundance in hypoxic control cells. Depletion of DDX28 conferred a proliferative advantage to hypoxic, but not normoxic, cells. DDX28 protein levels are reduced in several cancers, including gliomas, relative to levels in normal tissue. Therefore, we uncover a regulatory mechanism for this potential tumor suppressor in the repression of HIF-2α- and eIF4E2-mediated translation activation of oncogenic mRNAs.

INTRODUCTION

The procurement of oxygen is a fundamental aspect of survival for aerobic organisms in all domains of life. Mammals have evolved complex circulatory, respiratory, and neuroendocrine systems to satisfy the need for molecular oxygen as the primary electron acceptor in oxidative phosphorylation, which supplies energy in the form of ATP (1). The ability of a cell to acclimate to low oxygen (hypoxia or ≤1% O₂), which usually arises due to an imbalance in supply and demand, is essential from the earliest stages of life (2, 3). Hypoxia plays a role in several physiological and pathophysiological conditions, such as embryonic development, muscle exercise, wound healing, cancer, heart disease, and stroke (4). Prior to the establishment of uteroplacental circulation, embryonic cells receive only as much as 2% O₂, and following oxygenation by maternal blood, the embryo still contains discrete regions of hypoxia (2, 5). This form of physiological hypoxia helps govern the process of development through cell fate determination, angiogenesis, placentation, cardiogenesis, bone formation, and adipogenesis (3, 6–10). Hypoxia is also a feature of the tumor microenvironment and plays a key role in several cancer hallmarks toward tumor progression (11, 12). The major cellular response to hypoxia is mediated by hypoxia-inducible transcription factors 1 and 2 (HIF-1 and HIF-2, respectively). The HIFs are heterodimeric, composed of an oxygen-labile α subunit, HIF-1α or HIF-2α, and a constitutively expressed HIF-1β subunit (11). HIF-1 and HIF-2 activate the transcription of hundreds of genes (some shared and some unique), including those involved in metabolism and erythropoiesis, in order to simultaneously reduce the activity of energy-expensive processes and promote the increased uptake of nutrients and oxygen (13). Further investigation into the unique roles of HIF-1α and HIF-2α and how they might be differentially regulated will reveal novel insights into hypoxic gene expression.

Unlike HIF-1α, which is more involved in acute hypoxia (<24 h), HIF-2α is linked to chronic hypoxia (14). Further, HIF-2α accumulates in both the cytoplasm and the nucleus (15). Indeed, HIF-2α, but not HIF-1α, participates in the recruitment of select hypoxia-responsive mRNAs to the translation apparatus (16), not only in hypoxia but in the low range of physiological oxygen (<8% O₂) (17). Hypoxia is a potent inhibitor of mammalian target of rapamycin complex 1, which suppresses canonical cap-dependent translation by sequestering the cap-binding protein eukaryotic initiation factor 4E (eIF4E) (18–21). Alternative modes of translation initiation, including cap-independent mechanisms as well as noncanonical cap-dependent translation mediated by the eIF4E2 cap-binding protein, are utilized in hypoxia. HIF-2α and RBM4 recognize select mRNAs by binding to RNA hypoxia response elements (rHREs) in their 3′ untranslated regions (UTRs) and initiating their translation via the 5′ cap through eIF4E2, eIF4G3, and eIF4A (16, 22). HIF-2α and eIF4E2 are essential for embryonic development (3, 23) and are important contributors to tumor progression (24, 25), both of which are hypoxia-driven processes. The protein levels of eIF4E2 do not change between normoxia and hypoxia (16, 17). HIF-2α, on the other hand, is constantly degraded in normoxia via a family of prolyl hydroxylases that are inhibited by hypoxia (11). It is mostly unclear how the activities of eIF4E2 or HIF-2α in hypoxia are regulated with respect to translation initiation. A greater understanding of these mechanisms will shed light on how gene expression is coordinated in physiological and pathophysiological processes that are linked to hypoxia.

Here, we demonstrate that the DEAD box protein DDX28 is a negative regulator of HIF-2α in a human glioblastoma cell line. DDX28 has been detected in mitochondria, cytoplasm, and nuclei (26). However, known functions of DDX28 are limited to its RNA interference (RNAi)-mediated silencing, which impairs mitoribosome assembly and, subsequently, oxidative phosphorylation (27, 28), which is also an outcome of hypoxia (29). While it is known that hypoxia increases HIF-2α levels, our data show that DDX28 protein levels concurrently decrease. We also show that DDX28 interacts with HIF-2α but not with HIF-1α or the m⁷GTP cap structure. We demonstrate a link between these two observations by depleting DDX28 levels during hypoxia, causing eIF4E2 and its mRNA targets (including the HIF-2α mRNA) to associate more with the m⁷GTP cap structure and polysomes, respectively. Furthermore, hypoxic depletion of DDX28 caused a significant increase in cell proliferation only during hypoxia. While HIF-2α levels increased in both the nucleus and the cytoplasm upon DDX28 depletion, there was no increase in HIF-2α’s transcriptional activity. We propose a model where hypoxia-reduced DDX28 binds to and sequesters a pool of HIF-2α, suppressing HIF-2α’s ability to activate eIF4E2 cap binding and the translation of target mRNAs, including HIF-2α (EPAS1 gene) itself. Our data suggest that the reduced amount of DDX28 is still useful during hypoxia to restrain the HIF-2α/eIF4E2 translational axis, which can be oncogenic (24, 25).

RESULTS

DDX28 interacts with HIF-2α but not HIF-1α or eIF4E2.

U87MG human glioblastoma cells were used in this study because they have previously been characterized as models for hypoxia research and the interaction between HIF-2α and eIF4E2 in noncanonical cap-dependent translation (16, 17, 22, 24, 25). When U87MG cells were exposed to hypoxia (1% O₂) for 24 h, we observed not only an increase in HIF-2α but a concurrent decrease in DDX28 levels (Fig. 1A). Furthermore, when DDX28 was stably depleted via two independent short hairpin RNA (shRNA) sequences to produce two unique cell lines, the hypoxia-dependent increase in HIF-2α was further increased relative to levels in controls expressing a nontargeting shRNA (Fig. 1B). Since the HIF-α subunits can be stabilized by an increase in reactive oxygen species (ROS) (30) and some DDX28 is mitochondrial (26) and involved in mitoribosome biogenesis (28), we tested whether shRNA-mediated depletion of DDX28 in hypoxia produced more ROS or disrupted mitochondrial networks, morphology, and membrane potential. Fragmentation of mitochondrial networks occurs in response to ROS-induced stress (31, 32), and stress-damaged mitochondria are likely to be more round than rod shaped (31, 33, 34). We did not observe significant changes to mitochondrial fusion, morphology, or membrane potential between hypoxic DDX28-depleted cells and control cells (see Fig. S1A and B in the supplemental material). Cells induce the transcription of an array of genes in response to ROS. However, we did not observe significant changes between hypoxic DDX28-depleted cells and control cells (all were less than 2-fold) in the mRNA abundances of four genes encoding antioxidant proteins: Cu/Zn-superoxide dismutase (35), NAD(P)H:quinone oxidoreductase (36), sulfiredoxin-1 (37), and thioredoxin reductase 1 (38) (Fig. S1C).

FIG 1.

DDX28 interacts with HIF-2α but not with HIF-1α or eIF4E2. (A) Western blot of total HIF-2α and DDX28 protein levels in normoxia (21% O₂) and hypoxia (1% O₂). GAPDH was used as a loading control. (B) Western blot of DDX28 and HIF-2α (arrow, hypoxia-inducible lower band) normoxic and hypoxic protein levels in control (Ctrl) cells stably expressing a nontargeting shRNA or in cells stably expressing one of two shRNAs targeting DDX28 mRNA: knockdown 1 (KD1) and KD2 cells. Actin was used as a loading control. (C) GFP coimmunoprecipitation in hypoxic cells stably expressing FLAG-DDX28 and transfected with either recombinant GFP–HIF-2α or an empty GFP vector as a control. (D) FLAG coimmunoprecipitation of recombinant FLAG-eIF4E2 from hypoxic cells cotransfected with HA–HIF-2α. Cells transfected with the empty FLAG vector were used as a control. (E) FLAG coimmunoprecipitation in hypoxic cells stably expressing FLAG-DDX28 and transfected with recombinant HA–HIF-1α. Cells transfected with the empty FLAG vector were used as a control. The HA–HIF-α vectors produce stable nondegradable proteins that can be expressed in normoxia and retain their function. SE, short exposure (15 s); LE, long exposure (60 s). Twenty-five micrograms of whole-cell lysate was used as the input. Experiments are representative of three biological replicates and were performed with U87MG glioblastoma cells.

We next investigated whether DDX28 was directly involved in HIF-2α regulation by performing a coimmunoprecipitation. Exogenously tagged proteins were used due to the lack of specificity of the antibodies for the endogenous protein or its low abundance. In hypoxic U87MG cells, exogenous green fluorescent protein (GFP)–HIF-2α, but not GFP alone, coimmunoprecipitated with DDX28 (Fig. 1C). Since HIF-2α is a known interactor of eIF4E2 at the 5′ mRNA cap in hypoxia (16), we tested whether DDX28 was part of this complex. However, exogenous eIF4E2 coimmunoprecipitated with HIF-2α but not with DDX28 (Fig. 1D). Further, to demonstrate specificity for the HIF-2α homolog, HIF-1α did not coimmunoprecipitate with DDX28 (Fig. 1E). These data suggest that DDX28 interacts with a distinct pool of HIF-2α that is not associated with eIF4E2. However, this interaction did not alter HIF-2α protein stability (Fig. S2A) nor the transcription of the HIF-2α gene EPAS1 (Fig. S2B).

Depletion of DDX28 enhances the association of eIF4E2 with m⁷GTP and polysomes.

We performed m⁷GTP cap-binding assays to test whether the increased HIF-2α levels in DDX28-depleted cells had an effect on the translation initiation potential of eIF4E2. In hypoxia, eIF4E2 bound more to m⁷GTP (by 1.6- ± 0.1-fold and 1.5- ± 0.2-fold) in two DDX28-depleted cell lines than it did in the control (Fig. 2A and B). Surprisingly, depletion of DDX28 in normoxia had an even greater effect on eIF4E2 binding to m⁷GTP (with 2.9- ± 0.4-fold and 2.5- ± 0.1-fold increases) than it did in the control (Fig. 2C and D). These data are also in support of the findings illustrated in Fig. 1D, in which DDX28 does not bind the m⁷GTP cap.m⁷GTP association of a cap-binding protein like eIF4E2 does not necessarily imply an increase in translation initiation (39). Therefore, we performed polysome fractionation to test whether eIF4E2 had a greater association with polysomes isolated from DDX28-depleted cells than with those isolated from control cells in normoxia and hypoxia. Using densitometry to quantify total eIF4E2 associated with monosomes (low translation, fractions 1 to 3) and polysomes (medium to high translation, fractions 4 to 9), we show that 5% ± 1.2% of total eIF4E2 in normoxic control cells was associated with polysomes (Fig. 3A). In DDX28-depleted cells, the proportion of polysome-associated eIF4E2 increased to 23% ± 3.2% (Fig. 3B). Hypoxic control cells displayed 22% ± 5.1% of the eIF4E2 associated with polysomes (Fig. 3C), an increase relative to levels in normoxic control cells (5% ± 1.2%) (Fig. 3A). Hypoxic DDX28-depleted cells had an even greater proportion of eIF4E2 associated with polysomes (41% ± 4.1%) than did hypoxic control cells (Fig. 3D). The polysome association of the canonical cap-binding protein eIF4E did not change under any condition (Fig. 3E). Our data show that hypoxia or knockdown of DDX28 in normoxia increased the proportion of polysome-associated eIF4E2 relative to that of normoxic control cells (Fig. 3F). However, DDX28 depletion and hypoxia together significantly increased the polysome association of eIF4E2 relative to that of normoxic control cells (Fig. 3F).

FIG 2.

Knockdown of DDX28 increases the cap-binding affinity of eIF4E2. Shown are Western blots and quantifications of eIF4E2 capture with m⁷GTP-bound agarose beads in cells stably expressing one of two distinct shRNA sequences targeting DDX28 in 1% O₂ hypoxia (A and B) and 21% O₂ normoxia (C and D). Thirty-five micrograms of whole-cell lysate was used as the input. Ctrl, control cells stably expressing nontargeting shRNA; KD1 and KD2, knockdown cells stably expressing one of two distinct shRNA sequences targeting DDX28; LE, long exposure (30 s for panels A and B and 60 s for panels C and D). Data (n ≥ 3) are means ± SEM normalized to the input. *, P < 0.05 by a one-sample t test against a hypothetical mean (μ = 1). Experiments were performed with U87MG glioblastoma cells.

FIG 3.

Knockdown of DDX28 increases polysome-associated eIF4E2 only in hypoxia. (A to D) Polysomal distribution of DDX28, eIF4E2, and eIF4E protein measured by Western blotting in control cells stably expressing nontargeting shRNA in 21% O₂ normoxia (A) and 1% O₂ hypoxia (C) and in knockdown (KD1) cells stably expressing an shRNA targeting DDX28 in normoxia (B) and hypoxia (D). Ribosomal protein L5 (rpL5) was used as a marker of protein integrity in each fraction. (E and F) The eIF4E (E) or eIF4E2 (F) protein associated with polysomes (fractions 4 to 9) as a percentage of total protein (fractions 1 to 9) was quantified by densitometry. Data (n = 3) are means ± SEM. *, P < 0.05 using one-way ANOVA and Tukey’s HSD post hoc test. Experiments were performed with U87MG glioblastoma cells.

Depletion of DDX28 increases the translation of eIF4E2 target transcripts in hypoxia.

We performed quantitative reverse transcription-PCR (qRT-PCR) on monosome and polysome fractions in normoxia and hypoxia to measure the DDX28-dependent association of eIF4E2 and eIF4E target transcripts. We chose epidermal growth factor receptor (EGFR), insulin-like growth factor 1 receptor (IGF1R), and HIF-2α (EPAS1 gene) mRNAs, previously identified as eIF4E2-dependent transcripts due to the presence of an rHRE in their 3′ UTRs (16, 17). Eukaryotic translation elongation factor 2 (EEF2) and heat shock protein 90 alpha family class B member 1 (HSP90ab1) mRNAs were chosen as previously characterized eIF4E-dependent transcripts due to the presence of a 5′-terminal oligopyrimidine motif (17, 40). We observed a significant increase in the association of EGFR mRNA with polysomes relative to that with monosomes from 4.5-fold ± 0.9-fold in controls to 11.3-fold ± 1.6-fold in DDX28-depleted cells in hypoxia (Fig. 4A). Similarly, the polysome association of IGF1R and EPAS1 mRNAs significantly increased from 3.2-fold ± 0.3-fold and 2.6-fold ± 0.3-fold in controls, respectively, to 5.4-fold ± 0.7-fold and 6.2-fold ± 0.8-fold in hypoxic DDX28-depleted cells (Fig. 4A). Under normoxia, there was no difference between the associations of EGFR mRNA with polysomes and monosomes between controls (3.5-fold ± 0.5-fold) and DDX28-depleted cells (3.8-fold ± 0.5-fold) (Fig. 4B). Similarly, there was no statistical difference between the polysome and monosome associations of IGF1R and EPAS1 mRNAs in controls (5.6-fold ± 0.8-fold and 1.7-fold ± 0.2-fold) and DDX28-depleted cells (3.5-fold ± 0.5-fold and 1.8-fold ± 0.2-fold) (Fig. 4B). Moreover, EGFR protein levels (Fig. 4C), but not total EGFR mRNA levels (Fig. 4D), increased in hypoxic DDX28-depleted cells relative to those in the control. Neither of the eIF4E-dependent transcripts displayed significant changes in polysome association in normoxia or hypoxia between DDX28-depleted cells and controls. The one exception was EEF2 mRNA, which displayed a small significant increase in normoxic polysome association relative to monosome association in DDX28-depleted cells (1.52-fold ± 0.04-fold) compared to that in controls (1.05-fold ± 0.06-fold) (Fig. 4E). These data suggest that DDX28 depletion significantly increases the translation of eIF4E2-dependent, but not eIF4E-dependent, transcripts, including the HIF-2α mRNA (EPAS1) itself, in hypoxia.

FIG 4.

Depletion of DDX28 increases the polysome association of eIF4E2-dependent transcripts in hypoxia. (A and B) The associations of the eIF4E2-dependent transcripts epidermal growth factor receptor (EGFR), insulin like growth factor 1 receptor (IGF1R), and HIF-2α (EPAS1 gene) with polysomes (fractions 4 to 9 from Fig. 3) were measured relative to those of monosomes (fractions 1 to 3) by qRT-PCR in control cells expressing a nontargeting shRNA or knockdown (KD1) cells stably expressing an shRNA targeting DDX28 in 1% O₂ hypoxia (A) and in 21% O₂ normoxia (B). (C and D) Western blots of total EGFR, DDX28, and actin protein levels (C) and qRT-PCR of total EGFR mRNA levels normalized to those of the endogenous control genes RPLP0 and RPL13A (D) in hypoxic control cells or DDX28 KD cells. (E) The associations of the eIF4E-dependent transcripts eukaryotic translation elongation factor 2 (EEF2) and heat shock protein 90 alpha family class B member 1 (HSP90ab1) with polysomes were measured relative to those of monosomes by qRT-PCR in control cells or DDX28 KD cells in hypoxia and normoxia. Data (n ≥ 3) are means ± SEM. *, P < 0.05 using an unpaired two-sample t test except in panel D, for which a one-sample t test against a hypothetical mean (μ = 1) was used.

Depletion of DDX28 in hypoxia increases cytoplasmic and nuclear HIF-2α levels but not its nuclear activity.

Hypoxic cells were fractionated into cytoplasm and nuclei to measure the effects of DDX28 depletion on HIF-2α levels in both compartments. In accordance with the above-mentioned observations that DDX28 depletion increases total HIF-2α protein levels and its cytoplasmic activity (translation), we show that DDX28 depletion increased cytoplasmic HIF-2α levels compared to levels in the control (Fig. 5A). However, we also observed an increase in the nuclear levels of HIF-2α. Therefore, we investigated whether DDX28 depletion in hypoxia also increased the nuclear activity of HIF-2α (transcription) by measuring the mRNA abundance from its gene targets relative to that in the control. We chose six genes that contain in their promoters hypoxia response elements that are more dependent on HIF-2α than on HIF-1α (41–44). None of the six genes displayed a significant increase in mRNA abundance in DDX28-depleted cells relative to that of control cells (Fig. 5B). In fact, two genes displayed significant decreases in mRNA abundance, albeit by 2-fold at most. These data suggest that while both nuclear and cytoplasmic HIF-2α levels increase in response to DDX28 depletion, the effect on HIF-2α transcriptional activity is likely minimal.

FIG 5.

Depletion of DDX28 in hypoxia increases cytoplasmic and nuclear HIF-2α levels but not its nuclear activity. (A) Representative Western blot (n = 3) of HIF-2α protein levels in cytoplasmic and nuclear fractions of control cells expressing a nontargeting shRNA or knockdown (KD) cells stably expressing an shRNA targeting DDX28 in 1% O₂ hypoxia. Lamin A/C was used as a nuclear marker, and α-tubulin was used as a cytoplasmic marker. (B) The mRNA abundance of HIF-2α gene targets in hypoxia was measured via qRT-PCR. Data (n ≥ 3) are means ± SEM for DDX28 KD cells relative to those for control cells and normalized to values for endogenous control genes RPLP0 and RPL13A. *, P < 0.05 using a one-sample t test against a hypothetical mean (μ = 0). CITED2, Cbp/P300-interacting transactivator 2; EPO, erythropoietin; IGFBP3, insulin-like growth factor binding protein 3; ITPR1, inositol 1,4,5-trisphosphate receptor type 1; LOXL2, lysyl oxidase-like 2; OCT4, octamer-binding transcription factor 4. Experiments were performed with U87MG glioblastoma cells.

DDX28 depletion causes an increase in cell viability and proliferation in hypoxia but not normoxia.

We next investigated whether the increase in eIF4E2-directed translation in DDX28-depleted cells provided a benefit to cells by measuring their viability and proliferation in normoxia and hypoxia. To assess this, we monitored the number of viable DDX28-depleted and control cells over 72 h at 24-h intervals using crystal violet staining. We observed no significant differences in viability at 24, 48, or 72 h between normoxic DDX28-depleted and control cells (Fig. 6A). However, both hypoxic DDX28-depleted cell lines had significantly increased viability compared to that of the control at each time interval (Fig. 6B). To measure proliferation, we monitored the incorporation of bromodeoxyuridine (BrdU) into the DNA of actively dividing cells via immunofluorescence. In normoxia, one DDX28-depleted cell line displayed a significant increase in BrdU incorporation relative to that of control cells (Fig. 6C). However, following 24 h of hypoxia, both DDX28-depleted cell lines displayed significant increases in BrdU incorporation relative to that of the control (Fig. 6D). To test whether overexpressing exogenous DDX28 would have the opposite effect (decreased viability, proliferation, and HIF-2α levels) in hypoxia, we generated two stable clonal U87MG cell lines expressing FLAG-DDX28 and a control expressing FLAG alone. We did not observe any significant differences in viability, proliferation, or HIF-2α levels between the overexpressing cell lines and the control (Fig. S3). The one exception was a decrease in viability after 24 h of hypoxia in one of the overexpressing cell lines relative to that of the control, but this difference ceased at 48 h and 72 h. Since hypoxia appears to reduce the levels of exogenous DDX28 (Fig. S3E), it is possible that the overexpressing cell lines do not overexpress DDX28 enough in hypoxia to suppress HIF-2α or that DDX28 is already suprastoichiometric to HIF-2α. Our data suggest that depletion of DDX28 increases cell viability and proliferation in hypoxia, likely through an increase in the translation of select mRNAs and perhaps other unidentified pathways.

FIG 6.

DDX28 depletion causes an increase in cell viability and proliferation only in hypoxia. (A and B) Viable cell counts were measured with crystal violet staining after 24 h, 48 h, and 72 h in 21% O₂ normoxia (A) and in 1% O₂ hypoxia (B) for control cells expressing a nontargeting shRNA or knockdown cells stably expressing one of two shRNAs targeting DDX28 (KD1 and KD2). All absolute cell count values were normalized to the number of cells present on day 0 for each individual independent experiment, representing the fold change in the number of cells at each time point relative to the value on day 0. (C and D) Proliferation was measured as a percentage of the value for the BrdU-positive control and DDX28 KD cells via immunofluorescence after 24 h in normoxia (C) or hypoxia (D). Data (n ≥ 3) are means ± SEM. *, P < 0.05 using an unpaired two-sample t test. Experiments were performed with U87MG glioblastoma cells.

DISCUSSION

HIF-2α and eIF4E2 contribute to the ability of a cell to adapt to hypoxic conditions through selective gene expression by transcriptional and translation regulation. We have previously shown that eIF4E2 knockdown represses hypoxic cell proliferation and survival, migration and invasion, and tumor growth (24, 25). Furthermore, eIF4E2-directed translation is active in the low range of physiological oxygen, where HIF-2α, but not yet HIF-1α, is stabilized (3 to 8% O₂) (17). Total levels of eIF4E2 protein are minimally altered upon hypoxic exposure (16, 17, 25, 45), so how is it regulated? Posttranslational modifications of eIF4E2, such as ISGylation, have been identified (46), as have protein interactors, such as eIF4G3 and eIF4A (16, 22). However, the hypoxic stabilization of HIF-2α is essential for eIF4E2 hypoxic activity (16) and is likely a more upstream regulatory step of eIF4E2-directed translation.

We have uncovered a new mode of HIF-2α regulation that affects the activity of hypoxic translation via eIF4E2. HIF-2α coimmunoprecipitates with DDX28 (Fig. 1C) and eIF4E2 (16), but eIF4E2 does not coimmunoprecipitate with DDX28 (Fig. 1D). This suggests that DDX28 interacts with a different pool of HIF-2α than the one that interacts with eIF4E2. In agreement, DDX28 did not interact with the m⁷GTP cap structure (Fig. 2). However, the depletion of DDX28 did significantly increase the ability of eIF4E2 to associate with m⁷GTP, suggesting that its effect on HIF-2α influences eIF4E2 activity. Indeed, eIF4E2 did associate more with polysomes upon DDX28 depletion (Fig. 3F). Mechanisms by which HIF-2α acted at the 3′ UTRs of the rHREs of select mRNAs along with RBM4 to mediate joining of the 5′ end and eIF4E2 were initially proposed (16, 22, 47). It is important to note that cap-binding assays are performed with m⁷GTP bound to agarose beads and not to an mRNA (48). Therefore, these data suggest that HIF-2α may act directly on eIF4E2 at the 5′ UTR to regulate its cap-binding potential (Fig. 7). Next steps might examine how posttranslational modifications, such as phosphorylation, regulate the interaction between HIF-2α and DDX28.

FIG 7.

Model of HIF-2α regulation via DDX28. (A) DDX28 binds to and sequesters a pool of HIF-2α, suppressing HIF-2α’s ability to activate eIF4E2 5′ cap binding to transcripts harboring RNA hypoxia response elements (rHREs) in their 3′ UTRs, such as the HIF-2α mRNA itself. The remaining free cytoplasmic pool of HIF-2α, along with the binding of known interactors (eIF4G3, eIF4A, and RBM4), activates eIF4E2-dependent translation. (B) Very low levels of DDX28 as observed in many cancers or RNAi-mediated silencing in hypoxia frees up more HIF-2α to hyperactivate eIF4E2-directed translation above the levels observed in hypoxia. We propose that DDX28 may be a hypoxic tumor suppressor that restrains the HIF-2α/eIF4E2 oncogenic pathway.

Since DDX28 is involved in mitoribosome biogenesis (28), a potential concern regarding DDX28 depletion may be indirect HIF-2α stabilization through disruption of mitochondria and subsequent ROS production (30). However, most mitochondrial proteins are synthesized by cytoplasmic ribosomes, while mitoribosomes synthesize only 13 proteins involved in oxidative phosphorylation (OXPHOS). Indeed, previous work demonstrated that silencing DDX28 to only 5% of control levels resulted in a similar 5% decrease in OXPHOS (28). We demonstrate here that mitochondrial networks, morphologies, membrane potentials, and cellular ROS are similar between hypoxic DDX28-depleted and control cells. The reduction of DDX28 observed in hypoxia likely contributes to the shift away from OXPHOS and toward the glycolysis typically observed in cancer cells.

Depletion of DDX28 significantly increased the ability of eIF4E2 to bind m⁷GTP in both normoxia and hypoxia, but it was unexpected that this increase was greater in normoxia (Fig. 2). The polysome-associated eIF4E2 increased in DDX28-depleted normoxic cells, but the increase was not statistically significant (Fig. 3F). Further, the polysome association of the eIF4E2 mRNA targets EGFR, IGF1R, and EPAS1 did not increase in normoxic DDX28-depleted cells relative to levels in the control (Fig. 4B). With cells under hypoxic conditions, not only did we observe the effects of DDX28 depletion on eIF4E2 m⁷GTP and polysome association, but here EGFR, IGF1R, and EPAS1 mRNAs were significantly more associated with polysomes than in the control (Fig. 4A). We speculate that depleting DDX28 under normoxic conditions may increase the very low levels of HIF-2α to levels still undetectable via Western blotting but enough to significantly increase the cap-binding potential of eIF4E2. The relative increase in HIF-2α in normoxic DDX28-depleted cells relative to the level in the control may be greater than that in hypoxic cells, but there may be an unmet requirement in normoxia for a threshold amount of total HIF-2α protein to efficiently activate eIF4E2 and to translate its mRNA targets. Unexpectedly, the increase in nuclear HIF-2α upon DDX28 depletion in hypoxia produced minimal changes to the transcription of its target genes (Fig. 5). Perhaps in hypoxic control cells, the HIF-2α DNA binding sites are saturated. Conversely, the cytoplasmic HIF-2α protein binding partners (i.e., eIF4E2 and DDX28) are likely not saturated due to the much lower levels of HIF-2α in this compartment than in the nucleus. Therefore, our data suggest a 2-fold effect of DDX28 on the negative regulation of HIF-2α: (i) direct binding sequesters a pool of HIF-2α and suppresses its ability to enhance eIF4E2 cap binding and the translation of eIF4E2 target mRNAs and (ii) because the HIF-2α gene EPAS1 is a target mRNA of eIF4E2-directed translation (16), more HIF-2α protein is observed under conditions of less DDX28, such as hypoxia or shRNA-mediated silencing.

Hypoxia decreases total DDX28 protein levels, but our data suggest that the remaining DDX28 is important to restrain the HIF-2α/eIF4E2 translational axis. This brings into question why a negative regulator of this pathway would be in place, given that the expression of rHRE-containing mRNAs contributes to hypoxic survival (25). Dozens of eIF4E2 mRNA targets, such as EGFR, IGF1R, PDGFRA, EPAS1, and CDH22, have been identified (16, 17, 22, 24), and these have all been characterized as oncogenes (24, 49–51). Therefore, while important for hypoxic adaptation, tight regulation of this pathway is likely required to prevent neoplastic transformation. Indeed, mining the data in The Pathology Atlas (https://v18.proteinatlas.org/humanproteome/pathology) within The Human Protein Atlas (https://v18.proteinatlas.org/), we found that DDX28 is expressed in most tissues but lost in the majority of cancers (≥3 patients per cancer type [available from https://www.proteinatlas.org/ENSG00000182810-DDX28/pathology]) (52). Conversely, the presence of DDX28 is listed as a favorable prognostic marker in cases of renal cell carcinoma. More than 80% of all clear-cell renal cell carcinomas (CCRCC), the most common form of renal cancer, contain inactivating mutations in the VHL gene that stabilize the HIF-α subunits in a hypoxia-independent manner (53). The normoxic stabilization of HIF-2α and activation of its oncogenic pathways, which occurs in CCRCC, might be antagonized by the presence of DDX28. This study was performed in U87MG glioblastoma cells, but since DDX28 appears to be expressed ubiquitously, it may function similarly in other tissues. We provide mechanistic insight into the regulation of HIF-2α and eIF4E2-directed hypoxic translation and support for DDX28 as a tumor suppressor and prognostic marker to deepen our understanding of cancer progression.

MATERIALS AND METHODS

Cell culture.

U87MG human glioblastoma cells (HTB-14) were obtained from the American Type Culture Collection and maintained as suggested by them. Normoxic cells were maintained at 37°C in ambient O₂ levels (21%) and 5% CO₂ in a humidified incubator. Hypoxia was induced by culturing the cells at 1% O₂ and 5% CO₂ at 37°C for 24 h, unless otherwise stated, using an N₂-balanced Whitley H35 HypOxystation.

Generation of stable cell lines.

Two unique OmicsLink shRNA expression vectors (Genecopoeia) were used to target the coding sequence of human DDX28 (HSH014712-3-nU6 sequence 5′-GGTGGACTACATCTTAGAG-3′, HSH014712-3-nU6 sequence 5′-ACGCTGCAAGATTACATCC-3′). A nontargeting shRNA was used as a control. U87MG cells stably expressing C-terminal 3× FLAG-tagged DDX28 were generated by transfecting cells with the OmicsLink pEZ-M14 EX-A3144-M14 expression vector, encoding the human DDX28 coding sequence (Genecopoeia). Selection was initiated 48 h posttransfection using 1 μg/ml puromycin (HSH014712-3-nU6 sequence 5′-GGTGGACTACATCTTAGAG-3′) or 400 μg/ml G418 (HSH014712-3-nU6 sequence 5′-ACGCTGCAAGATTACATCC-3′), and single colonies were picked after 7 days.

Western blot analysis.

Standard Western blot protocols were used. Primary antibodies were as follows: anti-eIF4E2 (GeneTex; GTX82524), anti-DDX28 (Abcam; ab70821), anti-eIF4E (Cell Signaling; C46H6), anti-glyceraldehyde-3-phosphate dehydrogenase (anti-GAPDH; Cell Signaling; D16H11), anti-RPL5 (Abcam; ab137617), anti-HIF-2α (Novus; NB100-122), anti-FLAG (Sigma; F1804), anti-α-tubulin (GeneTex; GT114), anti-lamin A/C (Cell Signaling; 2032), anti-GFP (Abcam; ab290), antihemagglutinin (anti-HA; Santa Cruz; Y-11), anti-β-actin (GeneTex; GT5512), and anti-epidermal growth factor receptor (anti-EGFR; Proteintech; 18986-1-AP).

Polysome profiling and analysis.

Polysome profiling and analysis were performed as described previously (17). The total eIF4E2 or eIF4E signal was quantified by densitometry using Bio-Rad Image Lab software, and the percentages of eIF4E2 or eIF4E present in monosome and polysome fractions relative to the total signal were calculated.

RNA isolation and quantitative RT-PCR.

RNA was extracted from polysome fractions and qRT-PCR performed for the results shown in Fig. 4A, B, and E as previously described (17). RNA was extracted from cells (Fig. 4D and 5B; see also Fig. S1C in the supplemental material) using RiboZol per the manufacturer’s instructions. RNA (4 μg) was reverse transcribed using the high-capacity cDNA reverse transcription kit (Applied Biosystems). Primers and their sequences (5′ to 3′) were as follows: CITED2, CCTAATGGGCGAGCACATACA (forward) and CGTTCGTGGCATTCATGTT (reverse); CuZnSOD, CGAGCAGAAGGAAAGTAATG (forward) and TAGCAGGATAACAGATGAGT (reverse); EEF2, TTCAAGTCATTCTCCGAGA (forward) and AGACACGCTTCACTGATA (reverse); EGFR, GGAGAACTGCCAGAAACTGAC (forward) and GGGGTTCACATCCATCTG (reverse); EPAS1, TCCTCTCCTCAGTTTGCTCT (forward) and GTCCCATGAACTTGCTGATG (reverse); EPO, TGGAAGAGGATGGAGGTCGG (forward) and AGAGTGGTGAGGCTGCGAA (reverse); GAPDH, GTCAAGGCTGAGAACGGGA (forward) and CAAATGAGCCCCAGCCTTC (reverse); HSP90AB1, TGTCCCTCATCATCAATACC (forward) and TCTTTACCACTGTCCAACTT (reverse); IGFBP3, GCGCCAGGAAATGCTAGTG (forward) and AACTTGGGATCAGACACCCG (reverse); IGF1R, CCATTCTCATGCCTTGGTCT (forward) and TGCAAGTTCTGGTTGTCGAG (reverse); ITPR1, CGGAGCAGGGTATTGGAACA (forward) and GGTCCACTGAGGGCTGAAAC (reverse); LOXL2, CCCCCTGGAGACTACCTGTT (forward) and GGAACCACCTATGTGGCAGT (reverse); NQO1, AACTTGTTTGGTAGTTAGCC (forward) and TCAAGGTATCTGTGTACTGT (reverse); OCT4, GATGTGGTCCGAGTGTGGTTC (forward) and TTGATCGCTTGCCCTTCTG (reverse); RPLP0, AACATCTCCCCCTTCTCC (forward) and CCAGGAAGCGAGAATGC (reverse); RPL13A, CATAGGAAGCTGGGAGCAAG (forward) and GCCCTCCAATCAGTCTTCTG (reverse); SRXN1, CTCCACGAAGGTAGGGGTCA (forward) and CCTCCTTCCTTGAACGCAGA (reverse); and TXNDR1, GGAGTCATCAAGCATTTGAGG (forward) and GTAGGAGAATCCGGTGTCCA (reverse). Relative fold change in expression was calculated using the ΔΔC_T method (where C_T is threshold cycle), and transcript levels were normalized to those of RPLP0 and either RPL13A or GAPDH.

Immunoprecipitation and vectors.

Exogenous expression vectors used were as follows. FLAG–GFP–HIF-2α in a pAdlox backbone was a gift from Stephen Lee (Miami, FL), FLAG-eIF4E2 was a gift from Dong-Er Zhang (Addgene plasmid 17342, RRID:Addgene_17342 [http://n2t.net/addgene:17342]), and HA–HIF-1α and HA–HIF-2α were gifts from William Kaelin (Addgene plasmid 18955, RRID:Addgene_18955 [http://n2t.net/addgene:18955], and Addgene plasmid 18956, RRID:Addgene_18956 [http://n2t.net/addgene:18956]). The HA–HIF-α vectors produce stable nondegradable proteins that can be expressed in normoxia and retain their function (54). Control vectors were of the same backbone and tag without a gene insert. Cells were transfected with 4 μg DNA complexed with 20 μg polyethylenimine (PEI) diluted in 600 μl of lactate-buffered saline (20 mM sodium lactate and 150 mM NaCl, pH 4.0). The DNA-PEI complexes were diluted in 2.4 ml Dulbecco’s modified Eagle’s medium (DMEM) without fetal bovine serum (FBS) or antibiotics and added to cells at 37°C for 8 h followed by replenishment with complete medium. Cells were lysed after 48 h in 200 μl lysis buffer (10 mM Tris-HCl [pH 7.4], 150 mM NaCl, 0.5 mM EDTA, 0.5% NP-40, 1× protease inhibitor cocktail). Lysates were centrifuged at 12,000 rpm for 10 min at 4°C and diluted with 500 μl of dilution buffer (10 mM Tris-HCl [pH 7.4], 150 mM NaCl, 0.5 mM EDTA, 1× protease inhibitor cocktail). Protein (1 mg/ml) was incubated with 25 μl of GFP-Trap magnetic microbeads (ChromoTek), anti-FLAG M2 magnetic beads, or anti-HA magnetic beads. Only GFP-Trap required preblocking with 3% bovine serum albumin (BSA) in TBS [50 mM Tris, 150 mM NaCl, pH 7.4] and was washed per the manufacturer’s instructions. Immunoprecipitation was carried out for 1 h at 4°C with rotation. The GFP beads were washed four times with a more stringent wash buffer (10 mM Tris-HCl [pH 7.4], 500 mM NaCl, 0.5 mM EDTA, 0.1% NP-40), while FLAG and HA beads were washed four times with TBS. Proteins were eluted at 95°C in 1× Laemmli sample buffer. Whole-cell lysate (25 μg) was used as the input. When two proteins did not appear to interact, cross-linking with 4% paraformaldehyde for 20 min prior to lysis was performed to verify that transient interactions were not missed.

Analysis of cap-binding proteins.

An analysis of cap-binding proteins was performed as previously described (48), but the elution step was performed at 70°C to avoid melting the agarose beads. The eIF4E2 signal in the input and cap elution lanes was quantified by densitometry using Bio-Rad Image Lab software. The eIF4E2 signal in the DDX28 knockdown cap elution lane relative to that in the control lane was normalized to the eIF4E2 input signal ratio.

Cellular fractionation.

After 24 h of hypoxia, cells were lysed in 400 μl harvest buffer (10 mM HEPES, 50 mM NaCl, 500 mM sucrose, 0.1 mM EDTA, 10 mM dithiothreitol [DTT], 2 mM NaF, 0.5% Triton X-100, 1× protease inhibitor cocktail). Lysates were centrifuged at 8,000 rpm for 10 min at 4°C to pellet nuclei. The supernatant was collected as the cytoplasmic fraction, and the nuclear pellet was washed twice with 800 μl nuclear wash buffer (10 mM HEPES, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 10 mM DTT, 2 mM NaF, 1× protease inhibitor cocktail), with centrifugation at 13,000 rpm at 4°C for 5 and 10 min following the washes. The nuclear pellet was resuspended in 400 μl RIPA buffer (20 mM Tris-HCl [pH 7.5], 10 mM NaCl, 1% NP-40, 0.1% SDS, 0.5% sodium deoxycholate, 1 mM EDTA, 10 mM DTT, 2 mM NaF, 1× protease inhibitor cocktail) and rotated at 4°C for 15 min. Insoluble proteins were pelleted by centrifugation at 13,000 rpm for 10 min at 4°C, and the supernatant was collected as the nuclear fraction. Equal volumes of cytoplasmic and nuclear samples were mixed with 1× Laemmli sample buffer and boiled at 95°C for 90 s for Western blot analysis.

Viability assay.

For each indicated time point, 10,000 cells per well were plated in triplicate in a 24-well plate. The following day (day 0), cells were incubated at the oxygen concentrations indicated in the figures, and following each 24-h increment, cells were washed once with PBS and stained with 400 μl of 1% crystal violet solution prepared in 20% methanol, with gentle rocking for 20 min at room temperature. Cells were gently washed with water to remove excess stain. Plates were air dried overnight, and 400 μl of 10% acetic acid was added to each well and incubated on a shaker for 20 min at room temperature for destaining. The absorbance at 595 nm was measured using a microplate reader.

Cell proliferation assay.

Cells (250,000) were seeded on coverslips and incubated at their indicated oxygen concentrations for 24 h prior to treatment with 10 μmol/liter bromodeoxyuridine (BrdU) cell proliferation labeling reagent (Sigma) for 1 h. Cells were washed with PBS and fixed in cold methanol for 10 min. Excess methanol was removed by washing the slides for 5 min with PBS, and coverslips were incubated with 1:100 primary antibromodeoxyuridine antibody (RPN202; GE Healthcare) in the dark for 1 h. Coverslips were washed three times for 5 min with PBS and incubated with 2 μg/ml goat anti-mouse Alexa Fluor 555 secondary antibody (Invitrogen) in the dark for 1 h at 37°C. Cells were counterstained with Hoechst dye (1 μg/ml) for 5 min, and coverslips were mounted on microscope slides using ProLong Gold antifade reagent. Cells were imaged with the Nikon eclipse Ti-S inverted microscope. An average of 200 cells was assessed for positive BrdU labeling per biological replicate.

Immunofluorescence and microscopy.

Immunofluorescence and microscopy were performed as previously described (55). For mitochondrial shape and fusion, images were analyzed using Volocity (PerkinElmer). Percentages of fused mitochondria were determined by dividing the sum of the surface area of a single mitochondrion (≤5 μM) by the sum of the total surface area of all mitochondria. The Shape factor was calculated using Shape factor in Volocity. A minimum of 30 cells per replicate were measured.

TMRE mitochondrial membrane potential assay.

Cells were seeded (5,000) into 96-well plates and incubated for 24 h in normoxia (21% O₂) followed by 24 h in hypoxia (1% O₂). Following 24 h in hypoxia, positive controls received 10 μM carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP) (dimethyl sulfoxide [DMSO] was used only for DDX28 knockdown [KD] cells and the control) for 20 min to depolarize mitochondria. All cells were then given 200 nM tetramethylrhodamine ethyl ester (TMRE) for 30 min in the dark. Wells were aspirated and washed with 1× PBS-0.2% BSA, and then 100 μl 1× PBS-0.2% BSA was added to each well. Fluorescence was read on a plate reader at an excitation/emission of 544/584 nm.

Cycloheximide chase experiment.

Cells were grown to ∼80% confluence in 10-cm plates and incubated in 1% O₂ for 24 h. Cycloheximide (100 μg/ml) was added to each plate, and whole-cell lysates in RIPA buffer were prepared at 0, 0.5, 1, and 2 h of cycloheximide treatment. After incubation on ice for 10 min, lysates were centrifuged and supernatants were retained for protein quantification via a bicinchoninic acid (BCA) assay. SDS-PAGE and immunoblotting of HIF-2α and actin proteins were performed. The intensities of the protein bands were quantified using ImageJ software (NIH). HIF-2α protein degradation was plotted after the band intensity was normalized to that of the corresponding actin, and the values were expressed as proportions of the initial levels of HIF-2α.

Statistical analyses.

Results are expressed as means ± standard errors of the means (SEM) from at least three independent experiments. Experimental data were tested using an unpaired two-tailed Student t test when only two means were compared or a one-way analysis of variance (ANOVA) followed by Tukey’s honestly significant difference (HSD) test when three or more means were compared. A P of <0.05 was considered statistically significant using GraphPad Prism.