Hypoxic Inhibition of Nonsense-Mediated RNA Decay Regulates Gene Expression and the Integrated Stress Response

Abstract

Nonsense-mediated RNA decay (NMD) rapidly degrades both mutated mRNAs and nonmutated cellular mRNAs in what is thought to be a constitutive fashion. Here we demonstrate that NMD is inhibited in hypoxic cells and that this inhibition is dependent on phosphorylation of the α subunit of eukaryotic initiation factor 2 (eIF2α). eIF2α phosphorylation is known to promote translational and transcriptional up-regulation of genes important for the cellular response to stress. We show that the mRNAs of several of these stress-induced genes are NMD targets and that the repression of NMD stabilizes these mRNAs, thus demonstrating that the inhibition of NMD augments the cellular stress response. Furthermore, hypoxia-induced formation of cytoplasmic stress granules is also dependent on eIF2α phosphorylation, and components of the NMD pathway are relocalized to these granules in hypoxic cells, providing a potential mechanism for the hypoxic inhibition of NMD. Our demonstration that NMD is inhibited in hypoxic cells reveals that the regulation of NMD can dynamically alter gene expression and also establishes a novel mechanism for hypoxic gene regulation.

Cellular hypoxia, a stress common in development, cancer, and numerous other pathological conditions, affects metabolism, proliferation, apoptosis, and differentiation in a cell-specific manner. Much of the work to determine the mechanism for hypoxia-induced phenotypes has emphasized the role of transcriptional activation mediated by the hypoxia-inducible transcription factor 1. However, multiple studies have demonstrated that hypoxic cells may also regulate gene expression through a variety of transcriptional activators and repressors as well as through translational and posttranslational mechanisms (reviewed in references 27 and 37). In addition, several mRNAs, including the mRNA for vascular endothelial growth factor, have been shown to be stabilized in hypoxic cells through AU-rich sequences in their 3′ untranslated regions (UTR) (24). The exact motifs and trans-acting elements responsible for RNA stabilization in hypoxic cells are not well-characterized, however, and the extent, mechanism, and significance of hypoxic stabilization of mRNAs are not known.

Nonsense-mediated RNA decay (NMD) is a multistep intricate pathway by which mRNAs with premature termination codons (PTCs) are targeted for rapid destruction. Recent studies have provided insight into the mechanism of NMD in mammalian cells (reviewed in references 22 and 10). After mRNA splicing, exon-exon junctions are marked by an exon junction complex. The most common characteristic of mRNAs degraded by NMD is a premature termination codon located upstream of an exon junction complex, a feature not typically found in “normal” cellular mRNAs and thus signifying a mutation. If the ribosome does not traverse the exon junction complex because of the presence of a premature termination codon, this complex is able to recruit a series of enzymes, including the Rent1/Upf1 helicase, which target the mRNA for degradation. Knockdown of decapping proteins prolongs the decay of only those messages containing premature termination codons, suggesting that NMD requires decapping of the mRNA 5′ cap prior to enzymatic degradation by 5′-3′ exonucleases, although data also support a parallel mechanism of deadenylation and 3′-5′ exonucleases (23). It is thought that much of the degradation of NMD targets occurs in cytoplasmic processing bodies, which contain concentrated decapping enzymes as well as 5′ exonucleases (19). Studies in yeast have demonstrated that Rent1/Upf1 is sufficient to target mRNAs with premature termination codons to processing bodies for degradation (44).

NMD has traditionally been thought of as a mechanism to protect an organism from deleterious dominant-negative or gain-of-function effects of truncated proteins that arise from premature termination codons. It has been estimated that 30% of the mRNAs resulting from human mutations, including those responsible for thalassemia, cystic fibrosis, and Duchenne muscular dystrophy, are degraded by NMD (13). However, expression arrays comparing control cells to cells with Rent1/Upf1 silenced by small interfering RNA (siRNA) suggest that almost 10% of “normal,” nonmutated mRNAs are also directly and/or indirectly regulated by NMD (34). These putative NMD targets include diverse transcripts, including mRNAs important for amino acid transport and metabolism and proto-oncogenes. Some of these transcripts may be rendered susceptible to NMD because of alternative mRNA splicing; it has been estimated that a third of alternative splicing events in human cells generate mRNAs containing premature termination codons upstream of exon junction complexes (25).

Among the transcripts up-regulated with the knockdown of Rent1/Upf1 are those that play an important part in mediating the unfolded protein response and integrated stress response; many of these transcripts contain alternative upstream open reading frames and undergo alternative splicing which may render them sensitive to NMD (34). The unfolded protein response is activated when the accumulation of unfolded proteins in the endoplasmic reticulum (ER) initiates a coordinated set of signaling pathways, including the degradation of ER-associated mRNAs and phosphorylation of the α subunit of eukaryotic initiation factor 2 (eIF2α). Phosphorylation of eIF2α leads to a general suppression of protein translation. However, eIF2α phosphorylation also promotes the translational induction of the transcription factor ATF-4 and the subsequent up-regulation of ATF-4 transcriptional targets, including the transcription factors CHOP and ATF-3 (17, 18, 35). Recent studies have demonstrated that the unfolded protein response is also activated in hypoxic cells, promoting both the survival of hypoxic tumors and angiogenesis and playing a role in the hypoxic down-regulation of the Id-1 protein which, in turn, may help block proliferation and differentiation in hypoxic cells (4-6, 35). eIF2α is also phosphorylated in response to a variety of stresses in addition to hypoxic stress, such as amino acid deprivation, double-stranded RNA, and reactive oxygen species. The phosphorylation of eIF2α and subsequent coordinated suppression of general protein translation and up-regulation of ATF-4 and ATF-4 targets contribute to the cell's adaptation to these stresses and have been collectively termed the integrated stress response.

Despite the wide range of transcripts degraded by NMD, the potential significance of gene regulation by NMD has not been a major focus of research because NMD has been primarily thought of as a constitutive rather than regulated pathway. We sought to determine whether NMD is inhibited in hypoxic cells and what significance this putative inhibition could have on the cellular response of these cells to their environment. Specifically, because several hypoxia-induced mRNAs which play an important role in the integrated stress response are also up-regulated with the knockdown of Rent1/Upf1, we investigated whether these mRNAs were bona fide targets of NMD and thus provide evidence for the dynamic alteration of hypoxic gene expression through regulation of NMD.

MATERIALS AND METHODS

Cell culture and treatment.

Cells were culture in Dulbecco's modified Eagle's medium, 10% fetal calf serum, and antibiotics. Just prior to experiments the medium was exchanged to include 25 mM HEPES. Cells were rendered hypoxic in a PlasLabs 855 AC environmental incubator maintained at <0.1% oxygen. Where indicated, cells were treated with tunicamycin at 2.5 μg/ml or arsenic at 500 μM (both from Sigma). eIF2α wild-type mouse embryonic fibroblasts (MEFs) and eIF2α S51A MEFs were a gift of R. Kaufman. ATF-4^−/− MEFs and PERK^−/− MEFs and their corresponding controls were the kind gift of David Ron (NYU).

Immunoblot assays and XBP analysis.

Cells were harvested and collected as previously described (35), and immunoblot assays were performed with antibodies previously described, with the addition of Rent1/Upf1 (sc-488) (35). Qualitative XBP splicing was assessed as previously described (35). Immunoblots were quantitated with ImageJ.

Constructs and transfections.

hDCP1a cDNA was generated from the reverse transcription product of U2OS mRNA and cloned into the C1-red fluorescent protein (RFP) plasmid (Clontech). hRent was digested from a pCMV Sport 6 Rent1/Upf1 IMAGE clone 5555509 with XHO and HindIII and cloned into RFP C1. The β-globin wild-type and PTC 39 transgenes (29) were the kind gift of Jens Lykke-Andersen. Rent1/Upf1 shRNA was generated by cloning the previously validated Rent1/Upf1 sequence (AAGATGCAGTTCCGCTCCATTTT) (34) into LKO.1 lentivirus plasmid. Lentivirus was generated, and cells were infected as previously described (35). Genomic ATF-4, containing either the upstream open reading frame or initiating at the start codon for ATF-4 translation, was generated by PCR from genomic U2OS DNA and cloned into an EF1 cytomegalovirus (CMV) lentivirus (kind gift of Linzhao Cheng, Johns Hopkins). Stable pools were then generated by transfecting (with Lipofectamine) into ATF-4^−/− MEFs.

mRNA expression and half-life determinations.

To assess the expression of NMD targets in cells overexpressing Rent1/Upf1, U2OS cells were transfected with RFP-Rent1/Upf1 or RFP with polyethyleneimine (Sigma) and then 48 h later transfected again with these constructs along with either wild-type β-globin plasmid or the PTC 39 mutant plasmid in a 1:4 ratio. Twenty-four h later, cells were collected, RNA was isolated with Trizol (Invitrogen), and reverse transcription was performed on 100 μg of total RNA with a High-Capacity cDNA reverse transcription kit with RNase inhibitor and with random primers (ABI). Two percent of the reverse transcription product was used for real-time PCR using Sybr green PCR master mix and a MyiQ thermal cycler (Bio-Rad), with primers designed to produce amplicons that cross introns (see the supplemental material). Results were normalized to 18S RNA and compared to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) results. A similar protocol was utilized for pre-mRNA assessment, except nuclei were isolated (2) and primers contained intronic sequence. To assess the stability of mRNAs, cells were treated with 100 μg/ml of the RNA synthesis inhibitor 5,6-dichlorobenzimidazole-1-β-d-ribofuranoside (DRB; Sigma), and RNA was collected at the time of treatment and then at subsequent time points as shown in the figures. For assessment of RNA stability in hypoxic cells, cells were rendered hypoxic for 3 hours prior to treatment with DRB. The stability of wild-type and PTC 39 β-globin mRNAs in normoxic and hypoxic U2OS cells was determined by transiently transfecting U2OS cells with these transgenes, dividing them 24 h later, and then 24 h after that rendering cells either hypoxic or normoxic for 3 h. After 3 h these cells were treated with DRB, and RNA was collected just before and 6 h after treatment. To determine the stability of β-globin wild-type and PTC 39 mRNAs in wild-type and eIF2α S51A MEFs, plasmids were transfected with Lipofectamine (Invitrogen) and cells were selected with neomycin for 2 weeks. Cells were then either rendered hypoxic or maintained as normoxic for 3 hours and treated with DRB, and RNA was collected at time zero and after 6 h.

Staining for stress granules and processing bodies and localization of Rent1/Upf1.

COS-1 cells grown on coverslips were rendered hypoxic for 3 hours, treated with 500 μM arsenic for 3 h, or maintained as normoxic and then fixed with 4% paraformaldehyde-phosphate-buffered saline and permeabilized with methanol at 4°C. Cells were then blocked with 5% normal horse serum and stained with antibodies in Phems buffer directed against components of stress granules (G3BP; BD 61126 at 1:1,250) or processing bodies (p70 S6 kinase [sc-8418] at 1:200; Santa-Cruz) (1, 45). After washes with phosphate-buffered saline, cells were stained with Jackson Labs multiple-labeling-grade secondary antibodies at 1:200. Coverslips were then washed again, nuclei were stained with Topro3 (Invitrogen), and cells were visualized on a Zeiss LSM 510 confocal microscope at 100× magnification. For experiments to examine Rent1/Upf1 or Dcp1a localization, cells were transfected with the appropriate plasmids with Superfect (Qiagen) 3 days prior to fixation, staining, and visualization.

RESULTS

NMD is inhibited in hypoxic cells.

Although NMD has traditionally been thought of as a mechanism to rapidly degrade mutated mRNAs and not to play a role in the dynamic regulation of gene expression, recent data indicate that NMD may be regulated. Specifically, studies suggests that the decay of several NMD targets is inhibited by amino acid starvation (34), and in Schizosaccharomyces pombe the up-regulation of mRNAs in response to reactive oxygen species (ROS) requires Rent1/Upf1 (40). Intriguingly, both amino acid starvation and ROS stimulate eIF2α phosphorylation (42). This, plus the observation that several of the mRNAs up-regulated with Rent1/Upf1 knockdown (34) have also frequently been reported to be up-regulated in cells rendered hypoxic (which is also known to promote eIF2α phosphorylation) led us to explore whether eIF2α phosphorylation in hypoxic cells leads to an inhibition of NMD.

A well-documented NMD target is the PTC 39 β-globin mutant, which is responsible for a high proportion of thalassemia in Mediterranean populations, in which a CAG→UAG mutation leads to a premature termination codon in the second exon of the β-globin gene (29, 46). To assess the activity of NMD in hypoxic cells, we compared the stability of wild-type and PTC 39 β-globin mRNAs in normoxic and hypoxic cells. U2OS cells transiently transfected with these transgenes were rendered hypoxic for 3 h and then treated with the RNA polymerase II inhibitor DRB at a dose sufficient to inhibit RNA transcription (34), as confirmed by documenting that the pre-mRNAs of several genes transcriptionally activated in hypoxic cells were not induced when these cells were pretreated with DRB (data not shown). RNA was collected at the time of DRB treatment and 4 h later during uninterrupted hypoxia, and β-globin mRNA stability over this time period was then compared to β-globin stability in similarly treated normoxic cells. As expected, in normoxic cells the decay of the PTC 39 β-globin mutant mRNA was greater than the decay of the wild-type 39 β-globin mRNA (Fig. 1A, left panel). The stability of the wild-type mRNA was not significantly increased in hypoxic cells compared to normoxic cells. However, the β-globin PTC 39 mRNA was stabilized in hypoxic cells (Fig. 1A, left panel). When expressed as a ratio of hypoxic stability to normoxic stability, the β-globin PTC 39 mRNA stability doubled in hypoxic cells (Fig. 1A, right panel).

FIG. 1.

NMD is inhibited in hypoxic cells. (A) The stability of mRNA expressed from either wild-type β-globin (WT) or a β-globin cDNA with a premature termination codon (PTC 39) was determined in normoxic and hypoxic transiently transfected U2OS cells. Cells were rendered hypoxic or normoxic for 3 h, and then DRB was added during uninterrupted hypoxia. RNA was collected for analysis at time zero and then 4 h later. Normoxic and hypoxic β-globin expression at time zero was normalized to 1. The percentage of mRNA at each time point compared to time zero was determined, and the ratio of these percentages in hypoxic versus normoxic cells is depicted in the right panel; a ratio with a value of >1 indicates increased stability in hypoxic cells compared to normoxic cells for that time point. Real-time PCR was preformed in triplicate, and the graphs depict the averages of four separate experiments ± standard errors. (B) The mRNA and pre-mRNA expression levels of several NMD targets (DNAJB2, ARFRP1, MAP3K14, and UPP1) and control mRNA (ORC3L) were assessed in normoxic and in 3-h hypoxic U2OS cells by real-time PCR. Normoxic values were normalized to 1. Experiments were preformed in triplicate, and the graph depicts the averages ± standard errors. (C) U2OS cells were rendered hypoxic or normoxic for 3 h and then an inhibitor of RNA synthesis was added. RNA was collected for analysis at time zero and then 2, 4, and 6 h after RNA synthesis was inhibited, and the messages of NMD targets were determined by reverse transcription-PCR. Normoxic and hypoxic mRNA expression levels at time zero were normalized to 1. The hypoxic/normoxic ratio of mRNA levels at each time point is shown in the right panel, with values greater than 1 reflecting stability in hypoxic cells. Real-time PCR was preformed in triplicate, and the graphs depict the averages of four separate experiments + standard errors.

We next examined the stability of several mRNAs endogenously expressed in U2OS cells that have been reported to be NMD targets (DNAJB2, ARFRP1, MAP3K14, and UPP1) and a control mRNA not degraded by NMD (ORC3L) (34). The expression levels of these NMD targets were up-regulated in hypoxic cells (Fig. 1B). The steady-state expression of mRNA, however, is a reflection not only of its rate of degradation (e.g., by NMD) but also of its rate of synthesis, which could be suppressed or induced in hypoxic cells. When nuclei were isolated from hypoxic and normoxic cells and the pre-mRNA expression level was determined for several NMD targets, in general there was a more modest increase in pre-mRNA for each NMD target compared to its processed mRNA (Fig. 1B), suggesting that transcriptional up-regulation has only a minor role in the steady-state increase for most of these mRNAs. To directly assess the degradation rates of NMD targets in hypoxic cells, we compared the rates of mRNA decay (following inhibition of transcription with DRB) in hypoxic and nonhypoxic U2OS cells. We found that the decay of several reported NMD targets was dramatically decreased in hypoxic cells compared to normoxic cells (Fig. 1C, left panel). When expressed as the hypoxic decay/normoxic decay ratio, NMD targets were stabilized at least twofold at both 2 and 4 hours after DRB treatment (Fig. 1C, right panel). The stabilities of several other messages not degraded by NMD were unaltered in hypoxic cells. These mRNAs include those with long half-lives (e.g., GAPDH), short half-lives (e.g., p27), and intermediate half-lives (e.g., ORC3L and β-globin) (Fig. 1A and B and data not shown) (14). Similar results were also seen in HeLa cells (data not shown). Taken together, these data demonstrate that NMD is inhibited in hypoxic cells.

Inhibition of NMD in hypoxic cells is dependent on phosphorylation of eIF2α.

As discussed earlier, evidence suggests that NMD is inhibited by amino acid starvation and ROS (34, 40), two conditions that result in the phosphorylation of eIF2α, a modification that has previously been implicated only in regulating protein translation. Because NMD is also inhibited in hypoxic cells (when eIF2α is phosphorylated), we next investigated whether the hypoxic inhibition of NMD is dependent on the phosphorylation of eIF2α.

We examined the activity of NMD in eIF2α S51A knock-in MEFs in which the serine 51 of both eIF2α alleles was replaced with alanine, thus preventing the phosphorylation of eIF2α (41). As previously seen in transiently transfected U2OS cells (Fig. 1A), the stability of the PTC 39 β-globin mutant mRNA degraded by NMD, but not the stability of wild-type β-globin mRNA, was increased in hypoxic stably transfected wild-type MEFs (Fig. 2A). However, the mutant β-globin mRNA was not stabilized in hypoxic eIF2α S51A MEFs (Fig. 2A). We then assessed the stability of a number of endogenous NMD targets in hypoxic wild-type and eIF2α S51A MEFs and compared these to their normoxic stabilities. As might be expected from studies demonstrating that the efficiency of RNA decay, and NMD specifically, varies among cell lines (26), there were differences in the stabilities of NMD targets between normoxic U2OS cells (Fig. 1C) and normoxic MEF cells (Fig. 2B). However, consistent with our findings with the β-globin reporter, endogenous NMD targets were stabilized in hypoxic wild-type MEFs but not in hypoxic eIF2α S51A MEFs (Fig. 2B). Thus, eIF2α phosphorylation is necessary to inhibit NMD in hypoxic cells.

FIG. 2.

NMD inhibition in hypoxic cells is dependent on eIF2α phosphorylation. (A) The half-life of mRNA expressed from either wild-type β-globin (WT) or a β-globin cDNA with a premature termination codon (PTC 39) was determined in either normoxic or hypoxic stably transfected WT or eIF2α S51A MEFs. Cells were rendered hypoxic or maintained as normoxic for 3 h, at which time RNA was collected, and then cells were treated with DRB for 5 h and RNA was again collected. mRNA expression was normalized to 1. In the right panel mRNA stability is presented as a ratio, with a value of >1 indicating increased stability in hypoxic cells. Experiments were repeated in triplicate, and real-time PCR was performed in triplicate; graphs depict the averages ± standard errors. (B) The mRNA stabilities of several NMD targets were determined in normoxic or hypoxic wild-type MEFs or MEFs in which eIF2α cannot be phosphorylated (eIF2α S51A). Experiments were repeated in triplicate, and real-time PCR was performed in triplicate; graphs depict the averages + standard errors.

Components of the integrated stress response are degraded by NMD.

One of the consequences of hypoxic eIF2α phosphorylation is the translational up-regulation of ATF-4 and subsequent transcriptional up-regulation of ATF-3 and CHOP (6, 35). Although these proteins protect against cellular stress, high levels of ATF-4 and CHOP are deleterious to unstressed cells (31). Previous data have demonstrated that inhibition of NMD through Rent1/Upf1 knockdown increases the steady-state expression of ATF-4, ATF-3, and CHOP mRNAs, and many of the mRNAs for components of the integrated stress response have characteristics that could render them sensitive to NMD (34). Based on our data demonstrating that NMD is inhibited in hypoxic cells, we hypothesized that hypoxia-induced eIF2α phosphorylation not only increases ATF-4 translation and the transcription of ATF-4 targets but also that the phosphorylation of eIF2α, through the inhibition of NMD, stabilizes many of the mRNAs important for the cellular stress response. This dual, synergistic activity of eIF2α phosphorylation would, in effect, augment the integrated stress response.

To test this hypothesis we first needed to confirm that ATF-4, ATF-3, and CHOP were bona fide direct targets of NMD, since one interpretation of the previously described up-regulation of these mRNAs with Rent1/Upf1 knockdown is that inhibition of NMD leads to the accumulation of mRNAs with premature termination codons, synthesis of truncated proteins which do not fold correctly, and thus the induction of ER stress. To rule out that NMD inhibition leads to increased cellular stress, phosphorylation of eIF2α, and subsequent activation of stress response genes, we depleted cells of Rent1/Upf1, an established and validated method of NMD inactivation (34, 39), with siRNA or shRNA in two cell lines. Upon knockdown of Rent1/Upf1 we did not observe either eIF2α phosphorylation or cleavage of XBP-1 mRNA, two events that occur during the unfolded protein response (e.g., in hypoxic cells or cells treated with the chemical stressor tunicamycin) (Fig. 3A), demonstrating that inhibition of NMD does not itself increase levels of unfolded proteins and cellular stress.

FIG. 3.

Stress response mRNAs are substrates of NMD. (A) Rent1/Upf1 was diminished with either siRNA pools or an shRNA lentivirus in U2OS and HeLa cells. Cellular stress was determined by assessing eIF2α phosphorylation by immunoblotting or XBP mRNA splicing by reverse transcription-PCR. Hypoxic stress (3 h) and tunicamycin (0.5 μg/ml for 3 h) served as positive controls. (B) U2OS cells were transiently transfected with either RFP or RFP-Rent1/Upf1 and harvested 48 h later for immunoblotting. eIF2α is shown as a loading control. (C) U2OS cells were transiently transfected with either RFP or RFP-Rent and 48 h later transfected again with 25% of the same plasmid and 75% of either a wild-type or PTC β-globin transgene. At 24 h, cells were harvested, RNA was isolated, and expression of β-globin was assessed by reverse transcription-PCR. Expression for each construct transfected with RFP only was normalized to 1. Real-time PCR was performed in triplicate, and the graphs depict the averages of two separate experiments ± standard errors. (D) Cells prepared as for panel B had total RNA extracted, and real-time PCR was performed to quantitate messages for ATF-4, ATF-3, and CHOP, transcripts degraded by NMD (DNAJB2, ARFRP1, and MAP3K14), and control messages not degraded by NMD (ORC3L, ID-1, and GAPDH). RNA expression at time zero was normalized to 1. Real-time PCR was preformed in triplicate, and the graphs depict the averages of two separate experiments ± standard errors. (E) mRNA stabilities of three cellular stress mRNAs were assessed in U2OS cells expressing either scrambled shRNA (scr) or Rent1/Upf1 shRNA (shRent) by inhibiting RNA synthesis with DRB and assessing mRNA expression at the time of inhibition and 2 and 8 hours later by real-time PCR. Expression for each message at time zero was normalized to 1. Real-time PCR was done in triplicate with two biological replicates.

To determine if transcripts important for a cell's integrated stress response are degraded by NMD, we utilized both Rent1/Upf1-overexpressing and Rent1/Upf1 knockdown cells. Previous studies have demonstrated that tethering Rent1/Upf1 to mRNAs is sufficient to target mRNAs to processing bodies in yeast (44), and we hypothesized that simple overexpression of Rent1/Upf1 would also be sufficient to promote the degradation of bona fide NMD targets in mammalian cells. Transiently overexpressing a monomeric RFP-Rent1/Upf1 construct in U2OS cells led to increased levels of RFP-Rent1/Upf1 fusion protein as well as Rent1/Upf1 (perhaps because of translation initiation at the internal Rent1/Upf1 AUG of the N-terminal-tagged Rent1/Upf1 construct) (Fig. 3B). Overexpression of tagged Rent1/Upf1 did not affect the mRNA level of the wild-type β-globin transgene but significantly decreased the expression of the NMD-degraded PTC 39 β-globin mRNA (Fig. 3C), demonstrating that NMD activity could be increased with the overexpression of Rent1/Upf1. Rent1/Upf1 overexpression also led to significantly decreased steady-state levels of multiple mRNAs previously reported to be degraded by NMD (DNAJB2, ARFRP1, and MAP3K14) as well as the mRNAs of ATF-4, ATF-3, and CHOP (Fig. 3D). In contrast, mRNAs not thought or predicted to be degraded by NMD (e.g., ORC3L, ID-1, and GAPDH) were unchanged by overexpression of Rent1/Upf1.

Knockdown of Rent1/Upf1 increased the abundance of the mRNAs for ATF-3, CHOP, and to a lesser degree ATF-4 (data not shown). To assess if the stabilities of these mRNAs were dependent on Rent1/Upf1, RNA synthesis was inhibited with DRB in U2OS cells with Rent1/Upf1 knockdown and in U2OS control cells. mRNA was isolated at the time of DRB addition and at several subsequent time points, and RNA decay was assessed. The stabilities of ATF-4, ATF-3, and CHOP mRNAs were significantly increased with Rent1/Upf1 knockdown (e.g., after 8 hours of DRB treatment, ATF-4 mRNA was 4.5-fold increased in Rent1/Upf1 knockdown cells compared to scramble cells), demonstrating that these messages are indeed degraded in an NMD-dependent manner (Fig. 3E). Taken together, these data not only demonstrate that Rent1/Upf1 is rate limiting for NMD, but also that mRNAs for important components of the integrated stress response that help cells cope with stress but have a deleterious effect in nonstressed cells are rapidly degraded by NMD.

eIF2α-dependent hypoxic inhibition of NMD stabilizes mRNAs of the integrated stress response.

Based on our previous findings, the identification of mRNAs for components of the integrated stress response as true NMD targets predicts that these mRNAs are stabilized in hypoxic cells and that this stabilization is dependent on the phosphorylation of eIF2α. We first assessed the stability of ATF-4, ATF-3, and CHOP mRNAs in hypoxic U2OS cells and compared this to the stabilities of these messages in normoxic U2OS cells. Similar to what we noted for other NMD messages, the stabilities of ATF-4, ATF-3, and CHOP mRNAs were all significantly increased in hypoxic cells compared to normoxic cells (Fig. 4A). Although there was still decay of these mRNAs (particularly the unstable ATF-4 mRNA) in hypoxic cells over 2 hours of RNA synthesis inhibition. When expressed as a ratio of hypoxic stabilization to normoxic stabilization, the hypoxic messages were stabilized at least twofold at each time point examined. For example, after 2 hours of DRB treatment, ATF-4 mRNA had decayed to 13% of baseline levels in normoxic cells but to only 33% of baseline levels in hypoxic cells. Of note, this two- to threefold stabilization of ATF-4, ATF-3, and CHOP mRNAs in hypoxic cells is similar to the degree of stabilization of these mRNAs we observed with the knockdown of Rent1/Upf1 (Fig. 3E) and what has been reported for other NMD targets with genetic suppression of NMD (34).

FIG. 4.

eIF2α-dependent hypoxic inhibition of NMD stabilizes ATF-4, ATF-3, and CHOP to augment the stress response. (A) U2OS cells were rendered hypoxic or normoxic for 3 h prior to treatment with DRB. RNA was collected for analysis at time zero and then at 2, 4, and 6 h after addition of DRB (because of the short half-life of ATF-4, only time zero and 2 h were assessed), and the messages of NMD targets were determined by reverse transcription-PCR. Normoxic and hypoxic mRNA expression levels at time zero were normalized to 1. Real-time PCR was preformed in triplicate, and the graphs depict the averages of four separate experiments ± standard errors. The hypoxic/normoxic ratios of mRNA levels at each time point are shown in the right panel, with values greater than 1 reflecting stability in hypoxic cells. (B) Wild-type MEFs (eIF2α +/+) or MEFs which cannot phosphorylate eIF2α (eIF2α S51A) were treated as described for panel A, except the stability of ATF-4 mRNA was assessed after 1 h. (C) ATF-4^−/− MEFs were transfected with ATF-4 genomic constructs expressing either the coding region of ATF-4 or an ATF-4 cDNA containing the uORFs. The alternative uORFs are depicted by circles and hexagons. The region for ATF-4 coding is shown as a black rectangle, and introns are depicted as lines. Cells were rendered hypoxic for 3 h and then treated with an inhibitor of RNA synthesis. One hour after treatment, RNA was harvested and quantitated by real-time PCR. Normoxic cells were similarly treated. The endogenous ATF-4 half-life was determined with wild-type MEFs. mRNA expression in normoxic and hypoxic cells was normalized to 1 at time zero. Real-time PCR was preformed in triplicate, and the graphs depict the averages of four separate experiments ± standard errors. (D) U2OS cells were either infected with a lentivirus with a scrambled shRNA (scr) or a lentivirus with a Rent1/Upf1 sh RNA (shRent1/Upf1) as in Fig. 1A. Cells were then treated with 2.5 μg/ml of tunicamycin for short (A) or longer (B) times, protein was collected, and immunoblot assays were performed for proteins induced during the unfolded protein response (ATF-4, ATF-3, CHOP, and gadd34). Both actin and total eIF2α served as loading controls.

We then confirmed that the stabilities of these NMD-degraded components of the integrated stress response were increased in hypoxic cells in a manner dependent on the phosphorylation of eIF2α. In contrast to wild-type MEFs, in which the stabilities of ATF-4, ATF-3, and CHOP mRNAs were significantly stabilized when cells were rendered hypoxic, the stabilities of these mRNAs were not altered in hypoxic eIF2α S51A MEFs compared to normoxic eIF2α S51A MEFs (Fig. 4B), demonstrating that, as we found for other NMD targets, the hypoxic stabilization of these mRNAs is dependent on eIF2α phosphorylation.

ATF-4 mRNA degradation by NMD is dependent on its 5′ uORFs.

It is possible, though unlikely, that the increased stability of NMD targets in hypoxic cells occurs via a pathway that is only indirectly regulated by NMD. For example, recent evidence suggests that the degradation of some AU-rich element-containing mRNAs by AUF1 is impacted by NMD regulation of AUF1 variants (3). If the hypoxic inhibition of NMD directly increased the stabilities of NMD targets, we reasoned that not only would this hypoxia-induced stabilization be independent of the 3′ UTR of the mRNAs, the region typically affected by AUF1, but also that this stabilization might be due to alternative open reading frames which result in messages with novel premature termination codons (Fig. 4D). The ATF-4 mRNA has two alternative upstream open reading frames (uORFs) in addition to the downstream reading frame that encodes the ATF-4 protein. eIF2α phosphorylation-dependent ATF-4 up-regulation requires these two uORFs (16, 28). When eIF2α is phosphorylated, translation of the ATF-4 genomic locus occurs from a downstream open reading frame, which results in full-length ATF-4 protein. In the absence of stress, when eIF2α is not phosphorylated, only the uORFs are translated, which results in ribosome pausing at a termination codon upstream of an exon junction complex, a feature which could account for its susceptibility to NMD (34).

We constructed human genomic ATF-4 constructs that contained either the coding region of ATF-4 or the coding region along with ATF-4's uORFs. As predicted from previous studies using mouse-derived constructs (28), when transfected into ATF-4^−/− MEFs, the human ATF-4 construct that contained the 5′ uORFs was only translated when cells were rendered hypoxic or treated with tunicamycin; the construct containing only the coding region of ATF-4 was constitutively highly expressed (data not shown). In addition, the hypoxia-induced translation of the construct containing the uORFs was also dependent on eIF2α phosphorylation, as previously noted for endogenous murine ATF-4 (data not shown) (35).

We then assessed the stabilities of ATF-4 transgenes stably expressed in ATF-4 knockout MEFs. We chose to assess ATF-4 mRNA expression after only 1 hour of transcriptional inhibition, as opposed to 2 hours as previously done (Fig. 3E and 4A), because of the short half-life of the ATF-4 mRNA. In normoxic cells, the mRNA including ATF-4's uORFs was less stable than the mRNA of the ATF-4 construct that only contains its coding region, consistent with the model that increased degradation of ATF-4 mRNA by NMD is dependent on its upstream open reading frame (Fig. 4C). Similar to what we noted in human cells (Fig. 4A) and wild-type MEFs (Fig. 4B), the stability of the ATF-4 mRNA construct that contains the uORFs was increased in hypoxic MEFs compared to normoxic cells (Fig. 4C). This construct does not contain ATF-4's 3′ UTR, demonstrating that cellular hypoxia does not increase the stability of ATF-4 mRNA through AU-rich elements or other motifs in its 3′ UTR, the region which is responsible for the degradation of many short half-lived transcripts, although we do note that the stability of this transgene even in normoxic cells is slightly increased over the stability of endogenous ATF-4 (compare Fig. 4C to B). Hypoxia did not increase the mRNA stability of an ATF-4 construct that did not contain uORFs (Fig. 4C). These data demonstrate that hypoxic stabilization of ATF-4 mRNA is not dependent on its 3′UTR and suggest that translation through ATF-4's upstream open reading frame is in part responsible for this mRNA decay, as well as its increased stability in hypoxic cells.

Suppression of NMD augments the integrated stress response.

As expected, the steady-state expression levels of ATF-3 and CHOP mRNAs were dramatically increased in hypoxic cells compared to normoxic cells (9.8-fold ± 0.44-fold and 3.8-fold ± 0.11-fold, respectively [means + standard errors]), consistent with their transcriptional induction in hypoxic cells by ATF-4 as well as their stabilization by the hypoxic inhibition of NMD. Although the steady-state expression of ATF-4 mRNA was modestly, though significantly, increased in hypoxic cells versus normoxic cells (1.25-fold ± 0.019-fold; P = 0.02) (Fig. 4D), mRNA expression is a function of both decay and synthesis. It is therefore possible that without the hypoxic inhibition of NMD a hypoxic suppression of ATF-4 mRNA synthesis would significantly decrease ATF-4 expression in hypoxic cells compared to normoxic cells. To investigate this possibility we assessed the steady-state expression of the ATF-4 mRNA encoded by the two ATF-4 constructs previously described (Fig. 4C) in normoxic cells and in cells rendered hypoxic for 3 hours. Both the ATF-4 coding region construct and the ATF-4 construct containing uORFs are cloned downstream of the EF-1 promoter, and there is no reason to suspect that transcriptional activity of this promoter would differ between these two constructs. Similar to what we noted with endogenous ATF-4, steady-state mRNA expression of the construct containing the uORFs was only minimally increased in hypoxic cells compared to normoxic cells (Fig. 4D, right). However, the mRNA from the construct containing only the ATF-4 coding region was decreased in hypoxic cells compared to normoxic cells (Fig. 4D, middle). Therefore, in hypoxic cells there was a significant twofold increase in the steady-state mRNA expression of the ATF-4 construct containing the uORFs that is regulated by NMD compared to steady-state ATF-4 mRNA level from the unregulated ATF-4 construct (Fig. 4D). This is consistent with the model that without the hypoxic suppression of NMD, endogenous ATF-4 mRNA expression would be significantly lower than what we have observed in hypoxic cells.

Since our data indicated that many of the mRNAs that mediate the cellular response to stress are degraded by NMD, we asked whether the suppression of NMD would augment this stress response at the level of protein expression. Because the hypoxic activation of the integrated stress response also leads to the inhibition of NMD, we chose to assess activation of the integrated stress response in Rent1/Upf1 knockdown and control cells treated with the ER stress-inducing agent tunicamycin, which has been well-described to induce eIF2α phosphorylation, translational up-regulation of ATF-4, and the subsequent up-regulation of transcriptional targets of ATF-4. When we treated control and Rent1/Upf1 knockdown U2OS cells (as in Fig. 3A) with tunicamycin and assessed activation of the stress response by immunoblotting, as noted previously (Fig. 3A), knockdown of Rent1/Upf1 did not induce eIF2α phosphorylation or alter the kinetics of eIF2α phosphorylation in response to tunicamycin (Fig. 4E). Despite this, however, basal expression of the ATF-4 protein as well as ATF-4 protein induction with tunicamycin was increased in Rent1/Upf1 knockdown cells compared to control cells. With longer exposure to tunicamycin similar results were seen at the protein level for the ATF-4 targets ATF-3 and CHOP (Fig. 4E). In addition to being ATF-4 targets, ATF-3 and CHOP mRNAs are also degraded by NMD. gadd34, an ATF-4 transcriptional target whose mRNA we have determined is not degraded by NMD (data not shown), was also up-regulated when Rent1/Upf1 was knocked down, presumably due to the higher expression of ATF-4 (Fig. 4E). The augmentation of the integrated stress response with Rent1/Upf1 knockdown is diminished with prolonged exposure to tunicamycin. However, as reviewed in the Discussion section, below, if eIF2α phosphorylation were sufficient to inhibit NMD then longer treatment with tunicamycin would itself inhibit NMD and thus diminish the contribution of Rent1/Upf1 knockdown in augmenting the integrated stress response.

Taken together, these data validate ATF-4, ATF-3, and CHOP as true NMD targets and are consistent with the finding that NMD is inhibited in hypoxic cells and that this inhibition is dependent on the phosphorylation of eIF2α. Furthermore, the hypoxia-induced stabilization of these mRNAs suggests that an important phenotypic consequence of hypoxic suppression of NMD is the augmentation of the integrated stress response.

Stress granule formation in hypoxic cells is dependent on eIF2α phosphorylation.

We next explored potential mechanisms for the hypoxic inhibition of NMD. Although we demonstrated that NMD activity is sensitive to manipulations of total cellular Rent1/Upf1 (Fig. 3B to E), no difference in Rent1/Upf1 expression was seen in hypoxic versus normoxic cells at the protein (Fig. 3A) or mRNA (data not shown) level. We thus chose to focus on our finding that eIF2α phosphorylation was required for the hypoxic inhibition of NMD, as there are several mechanisms by which eIF2α phosphorylation could regulate NMD. As discussed earlier, an important step in NMD is a pioneering round of translation which leads to the pausing of a ribosome at a premature termination codon near an exon junction complex. Inhibition of protein translation inhibits ribosome binding to mRNA and suppresses NMD (8). Since eIF2α phosphorylation globally suppresses protein translation, one mechanism by which NMD is inhibited in hypoxic cells in an eIF2α-dependent manner could be through the suppression of protein translation. However, we do not favor this hypothesis for several reasons. Only a small amount of translation is required for NMD to occur, and the pioneer round of translation occurs during periods of decreased translation, including serum starvation and heat shock (7, 32, 36). In addition, several of the mRNAs increased by NMD inhibition in hypoxic cells, including ATF-4, ATF-3, and CHOP, are associated with polysomes and are robustly translated in hypoxic cells despite phosphorylation of eIF2α (6, 35). Hence, it is unlikely that the inhibition of protein translation mediated by eIF2α phosphorylation is the mechanism by which NMD is inhibited in hypoxic cells.

We thus pursued an alternative hypothesis for eIF2α-regulated inhibition of NMD. Recent evidence suggests that the actual degradation step of NMD occurs in cytoplasmic processing bodies which contain the decapping and exonuclease enzymes necessary for NMD degradation. This theory is supported by the finding in yeast that Rent1/Upf1 targets mRNAs with premature termination codons to processing bodies and that interference of mRNA decapping and degradation leads to an increase of these mRNAs in enlarged processing bodies (44). Cytoplasmic stress granules, in contrast, do not contain these enzymes. Because mRNAs may be found in both stress granules and processing bodies, it is thought that stress granules may serve to hold mRNAs that are later reused or transported to processing bodies for degradation (19). Phosphorylated eIF2α is found in stress granules, and eIF2α phosphorylation is both sufficient for stress granule formation and necessary for stress granule formation in response to arsenic and tunicamycin (20, 21, 33). We formed a working hypothesis that in hypoxic cells phosphorylation of eIF2α leads to stress body formation and that these stress bodies sequester mRNAs and exclude these mRNAs from processing bodies, thereby protecting them from NMD. Targets typically degraded by NMD might be directed to stress granules by Rent1/Upf1, phosphorylated eIF2α, or some combination of the two. We therefore investigated the formation of stress granules in hypoxic cells and the role these stress granules may play in the hypoxic inhibition of NMD.

Using a variety of fluorescent fusion proteins as well as indirect immunofluorescence for components found specifically in either stress granules (e.g., eIF3 and G3BP) or processing bodies (e.g., Ge-1/Hedls, which is recognized by a p70 S6 kinase antibody [45]), we examined the components necessary for stress granules and processing body formation in hypoxic cells. As noted, previous studies have shown that stress granule formation in response to arsenic is dependent on eIF2α phosphorylation (11). When we stressed wild-type MEFs, eIF2α S51A MEFs, or MEFs deficient for PERK (the eIF2α kinase predominantly activated in hypoxic cells [35]), we observed that stress granules form in hypoxic cells, and their formation is dependent on eIF2α phosphorylation by the PERK kinase (Fig. 5A and data not shown). Hence, stress granules do not form in hypoxic PERK^−/− and eIF2α S51A cells. Interestingly, arsenic induces dramatically large stress granules even in cells without PERK, but not in cells that cannot phosphorylate eIF2α, demonstrating that arsenic-induced stress granule formation is dependent on eIF2α phosphorylation that occurs through a kinase(s) other than PERK. In contrast, processing body formation is not dependent on eIF2α phosphorylation and occurs in normoxic, hypoxic, and arsenic-treated cells in roughly the same proportion (Fig. 5B, top panel).

FIG. 5.

Hypoxic induction of stress granules is dependent on eIF2α phosphorylation, and Rent1/Upf1 protein is concentrated in these granules. A. WT MEFs, MEFs deficient for the PERK kinase, and MEFs which have knock-in alleles for eIF2α with a serine-to-alanine mutation so that eIF2α cannot be phosphorylated (S51A) were either rendered hypoxic for 3 h or treated for 3 h with 500 μM arsenic. Cells were then fixed and stained for a marker for stress granules (G3BP). B. Cos-1 cells were transiently transfected with a Dcp1a-RFP construct and 24 h later plated on coverslips. Then, 24 h later they were either treated with 500 μM arsenic or rendered hypoxic for 3 h or untreated. Cells were then fixed and stained for antibodies directed against components of either stress granules (G3BP) or processing bodies (Ge-1/Hedls, which cross-reacts with the p70 S6 kinase antibody). C. Cos-1 cells were transiently transfected with RFP-Rent1/Upf1. Cells were then rendered hypoxic for 3 hours or treated with arsenic, and stress granules were visualized with indirect immunofluorescence.

In hypoxic cells, decapping proteins are concentrated in processing bodies and Rent1/Upf1 is targeted to stress bodies.

Removal of the 5′ 7-methylguanosine “cap” of mRNA by the decapping enzymes Dcp1a and Dcp2 is the first step necessary for 5′-3′ exonuclease degradation, a major path of NMD (23). These decapping enzymes are primarily concentrated in processing bodies in unstressed cells (19), and we next determined if they are also concentrated in processing bodies in hypoxic cells. When COS-1 cells, chosen for their optical properties, were transfected with a monomeric RFP-Dcp1a fusion protein, we determined that Dcp1a is concentrated solely in processing bodies in normoxic, hypoxic, and arsenic-treated cells (Fig. 5B). Dcp1a is excluded from stress granules in arsenic-treated and hypoxic cells, although stress granules and processing bodies are often in close proximity to each other, as previously reported (19). These studies, confirming similar studies in yeast and mammalian cells (19, 44), suggest that NMD-degraded mRNAs must be transported to processing bodies prior to decapping and degradation, even in hypoxic cells.

Since Rent1/Upf1 is required for NMD and recent studies in yeast have demonstrated that Rent1/Upf1 targets NMD-degraded transcripts to processing bodies, we next examined the localization of Rent1/Upf1 in mammalian cells under a variety of stresses. As might be expected for a protein that binds to cytoplasmic mRNAs and guides these mRNAs to processing bodies, the localization of Rent1/Upf1 is dynamic. Studies in yeast have revealed that Rent1/Upf1 is diffusely cytoplasmically located unless RNA degradation is inhibited (44). The lack of Rent1/Upf1 concentrated in processing bodies is thought to be due to the rapid degradation of Rent1/Upf1-delivered mRNAs in processing bodies and subsequent rapid release of Rent1/Upf1 from processing bodies; Rent1/Upf1 becomes concentrated in processing bodies only in yeast strains deleted for enzymes required for RNA degradation (44).

We also noted diffuse cytoplasmic staining of Rent1/Upf1 in unstressed mammalian cells (Fig. 5C). Of note, the number of processing bodies per cell was diminished in those cells overexpressing Rent1/Upf1 (data not shown), suggesting that rapid degradation of NMD targets by Rent1/Upf1 (as demonstrated in Fig. 3C to E) diminishes the formation of processing bodies. We then stressed cells and determined where Rent1/Upf1 localized. In hypoxic and arsenic-treated cells, Rent1/Upf1 was again primarily cytoplasmic, but Rent1/Upf1 foci colocalized not in processing bodies but in stress granules (Fig. 5C). Together, these data suggest that in while in hypoxic cells de-capping occurs primarily in processing bodies, Rent1/Upf1 is aberrantly localized to stress granules.

DISCUSSION

We have determined that NMD is inhibited in hypoxic cells, promoting the stability of many of the mRNAs important for the response to cellular stress as well as other well-characterized NMD targets. Our studies indicate that the phosphorylation of eIF2α, in addition to being a central component of the integrated stress response, is also necessary for the hypoxic inhibition of NMD. Finally, we have observed that eIF2α phosphorylation is also required for the hypoxic induction of stress granules and have generated a model, supported by our data, in which eIF2α phosphorylation-dependent stress granules that form in hypoxic cells sequester NMD targets, thus making these NMD targets unavailable for degradation in processing bodies.

Our data demonstrate that Rent1/Upf1 is concentrated in stress granules in hypoxic cells. These data, as well as data from a variety of mammalian and yeast studies, are consistent with our hypothesis in which the Rent1/Upf1 targeting of NMD substrates to processing bodies is aberrant in hypoxic cells (Fig. 6). This model proposes that with overexpression of Rent1/Upf1, processing bodies are not readily apparent as NMD substrates are quickly degraded. When enzymes necessary for the late stages of NMD are deleted in yeast strains (44), Rent1/Upf1 continues to deliver NMD substrates to processing bodies; because these mRNAs cannot be degraded, Rent1/Upf1 accumulates in processing bodies. When NMD is inhibited in hypoxic cells, however, our working model theorizes that Rent1/Upf1-tethered mRNAs are not targeted to processing bodies but instead to stress granules where these mRNAs cannot be degraded. The targeting of these mRNAs may be mediated by Rent1/Upf1, phosphorylated eIF2α, or some combination of the two. The fate of these putatively sequestered mRNAs is unknown. Phosphorylated eIF2α is localized to stress granules (21), and one possibility is that with the cessation of stress these mRNAs are rapidly available for translation, which may then help the cell to recover from stress. It is also clear, however, that at least some of the NMD targets stabilized in hypoxic cells are translated even while cells are hypoxic and eIF2α is predominantly phosphorylated (e.g., ATF-4, ATF-3, and CHOP) (6, 35). Thus, we cannot rule out the possibility that the identification of an mRNA as a potential NMD target may be one of the mechanisms used in the mRNA triage decision that has been proposed to occur in stress granules (1) and that the transit of NMD targets through stress granules subsequently leads to their translation.

FIG. 6.

Hypothesis for hypoxic suppression of NMD via sequestration of NMD targets in stress granules. In this hypothetical model, cellular hypoxia induces stress granule formation, which is dependent on eIF2α phosphorylation, and leads to the sequestration of Rent1/Upf1 and Rent1/Upf1-tethered NMD targets in stress granules. Stress granules, as opposed to processing bodies, do not contain the enzymes necessary to degrade NMD. Therefore, these Rent1/Upf1-tethered mRNAs are stabilized.

While we hypothesize that Rent1/Upf is tethered to NMD targets which then leads to the accumulation of these mRNAs in stress granules in hypoxic cells, and perhaps under other conditions of stress that lead to the phosphorylation of eIF2α (e.g., reactive oxygen species), it is possible that the sequestration of Rent1/Upf1 in stress granules is sufficient by itself to suppress NMD. This alternative hypothesis posits that Rent1/Upf1 is sequestered in stress granules independently of NMD-targeted mRNAs, allowing the latter to escape NMD and remain in the translated pool in hypoxic cells. Future studies, beyond the scope of this present work, will be needed to test the actual localization of NMD-targeted mRNAs in hypoxic cells in order to determine which model best describes the mechanism for hypoxia-induced suppression of NMD.

The cellular adaptation to stress encompasses diverse mechanisms of gene regulation, including degradation of ER-associated mRNAs, mRNA splicing of the transcription factor XBP-1, general suppression of translation, increased translation of ATF-4, and transcriptional up-regulation of ATF-4 targets. We have now demonstrated that an additional mechanism that aids in the cellular adaptation to hypoxic stress is the posttranscriptional stabilization of NMD targets involved in the integrated stress response. It is perhaps not surprising that eIF2α phosphorylation, which is already known to integrate many aspects of the cellular response to stress, is also necessary to inhibit NMD in hypoxic cells. Indeed, if an important role of NMD regulation were to control the expression of stress-related mRNAs, and translation and transcription of these stress genes are/ regulated by eIF2α phosphorylation, then eIF2α phosphorylation would be an efficient mechanism to inhibit NMD and stabilize these mRNAs.

eIF2α is phosphorylated by a variety of stresses in addition to hypoxia, including the accumulation of misfolded proteins in the ER, the rapid synthesis of proteins, amino acid deprivation, double-stranded RNA, reactive oxygen species, and cytokines (6, 15, 47). Further studies will be needed to demonstrate if eIF2α phosphorylation is sufficient to inhibit NMD. However, we note that eIF2α phosphorylation is sufficient to promote the formation of stress granules as well as to suppress protein translation, two of the more likely mechanisms responsible for the hypoxia-induced, eIF2α phosphorylation-dependent inhibition of NMD. If eIF2α phosphorylation were sufficient to inhibit NMD, then this would explain why the increased up-regulation of integrated stress response proteins seen in tunicamycin-treated cells when Rent1/Upf1 is depleted is more impressive at earlier time points (Fig. 4D), since at later time points complete phosphorylation of eIF2α by tunicamycin would fully inhibit NMD, and Rent1/Upf1 knockdown adds little to the inhibition of NMD and stabilization of integrated stress response targets.

A variety of conditions are marked by tissue hypoxia and/or other stresses that lead to the phosphorylation of eIF2α, and regulation of eIF2α phosphorylation and/or induction of ATF-4 and ATF-4 targets has been shown to play an important role in a number of diseases, including thalassemia and cancer (reviewed in references 9 and 30). While the well-established role for eIF2α phosphorylation in regulating protein synthesis is obviously an important mechanism by which eIF2α affects cellular phenotype, our data suggest that the regulation of NMD by eIF2α phosphorylation may also be a contributing factor. Many of the mutated mRNAs responsible for human disease are degraded by NMD, including mutations responsible for β-thalassemia and mutations of the tumor suppressors BRCA1 and APC (12, 38, 46). The environmental milieu, therefore, has the potential to regulate these mutated mRNAs, leading to their increased expression under hypoxia and perhaps other stress conditions.

Both hypoxia and regulation of eIF2α phosphorylation also play an important role in a number of physiological conditions, including differentiation, and our findings not only have significance for the expression of truncated proteins arising from mutated mRNA but also for the large number of cellular transcripts degraded by NMD. The characteristics of many of these cellular transcripts that render them sensitive to NMD are unknown, as is the case with CHOP. While our data suggest that an alternative upstream open reading frame is responsible for the sensitivity of ATF-4 mRNA to degradation by NMD, a common feature of NMD targets, the most frequent characteristic for nonmutated mRNAs that are degraded by NMD is thought to be alternative splicing events that lead to a premature termination codon upstream of the exon junction complex. Indeed, the alternative splicing of ATF-3 has been hypothesized to render this mRNA susceptible to NMD (34). Thus, in addition to stabilizing mutant mRNAs with a premature termination codon, the regulation of NMD also has the potential to stabilize alternatively spliced mRNAs and dramatically alter the expression profiles of stressed cells. The regulation of NMD suggests a novel mechanism for the dynamic regulation of gene expression in hypoxic cells, and the dynamic regulation of gene expression via hypoxia inhibition of NMD may play an important role in hypoxia-induced phenotypes.

Because the hypoxic induction of genes promotes angiogenesis as well as the metabolic adaptation of hypoxic cells to the stressful environment, a developing therapeutic approach has been to either augment this gene induction for the treatment of diseases that result from ischemia (e.g., myocardial infarctions and cerebral vascular accidents) or to suppress hypoxic gene induction for the treatment of hypoxic tumors. Much of this work has revolved around altering the activity of hypoxia-inducible factor 1 (reviewed in reference 43). Preclinical studies have identified a variety of antibiotics and other small molecules which allow read-through of premature termination codons and thus inhibit NMD, and some of these drugs have already demonstrated clinical efficacy in diseases caused by premature termination codons (48). Our data suggest that the hypoxic inhibition of NMD and augmentation of the integrated stress response likely serve as a protective mechanism. Therefore, augmenting the inhibition of NMD in hypoxic cells is likely to be cytoprotective and might be an effective strategy for ischemic diseases; conversely, diminishing the hypoxic inhibition of NMD would be predicted to cause cell death and could be an effective approach against hypoxic tumors.