PERK-mediated induction of microRNA-483 disrupts cellular ATP homeostasis during the unfolded protein response

Endoplasmic reticulum (ER) stress activates the unfolded protein response (UPR), which reduces levels of misfolded proteins. However, if ER homeostasis is not restored and the UPR remains chronically activated, cells undergo apoptosis. The UPR regulator, PKR-like endoplasmic reticulum kinase (PERK), plays an important role in promoting cell death when persistently activated; however, the underlying mechanisms are poorly understood. Here, we profiled the microRNA (miRNA) transcriptome in human cells exposed to ER stress and identified miRNAs that are selectively induced by PERK signaling. We found that expression of a PERK-induced miRNA, miR-483, promotes apoptosis in human cells. miR-483 induction was mediated by a transcription factor downstream of PERK, activating transcription factor 4 (ATF4), but not by the CHOP transcription factor. We identified the creatine kinase brain-type (CKB) gene, encoding an enzyme that maintains cellular ATP reserves through phosphocreatine production, as being repressed during the UPR and targeted by miR-483. We found that ER stress, selective PERK activation, and CKB knockdown all decrease cellular ATP levels, leading to increased vulnerability to ER stress–induced cell death. Our findings identify miR-483 as a downstream target of the PERK branch of the UPR. We propose that disruption of cellular ATP homeostasis through miR-483–mediated CKB silencing promotes ER stress–induced apoptosis.

Eukaryotic cells employ the endoplasmic reticulum (ER) 4 organelle to fold secreted and membrane proteins, synthesize hydrophobic lipids and sterols, and store free calcium (1). Physiologic, pathologic, or environmental processes that interfere with ER functions lead to a condition known as ER stress (2), which is detected by a conserved intracellular signal transduction mechanism, the unfolded protein response (UPR) (3). The UPR initiates transcriptional and translational programs that seek to restore ER homeostasis.
In mammals, the UPR is controlled by three ER-resident transmembrane sensors: inositol-requiring enzyme 1 (IRE1), PKR-like endoplasmic reticulum kinase (PERK), and activating transcription factor 6 (ATF6) (3). Each of these UPR regulators has ER stress-sensing luminal domains coupled across the ER membrane to cytosolic effector domains that initiate distinct signal transduction cascades during the UPR (3). In particular, PERK responds to ER stress by oligomerizing to activate its cytosolic kinase domain, which specifically phosphorylates a serine at position 51 of the ␣ subunit of eukaryotic initiation factor 2 (eIF2␣) (4). eIF2␣ partners with eIF2␤ and eIF2␥ to form the heterotrimeric eIF2 that is essential for canonical translational initiation at AUG start codons, but phosphorylation of eIF2␣ at serine 51 abrogates eIF2-dependent protein translation, leading to global slowdown of protein synthesis and diminished protein folding demands on the ER (5,6). However, translation of many UPR genes, including activating transcription factor 4 (Atf4), C/EBP homologous protein (Chop), Gadd34, and BiP/Grp78, persists or even increases despite phosphorylation of eIF2␣ because the 5Ј-UTRs of these transcripts contain upstream open reading frames or internal ribosome entry sites that enable them to bypass p-eIF2␣-mediated translational slowdown or employ noncanonical modes of translational initiation during the UPR (7)(8)(9)(10)(11)(12)(13). Thus, the PERK arm of the UPR dampens protein translation and also triggers transcriptional programs by production of transcription factors such as activating transcription factor 4 (ATF4) and C/EBP-homologous protein (CHOP).
The transcriptional and translational programs governed by IRE1, PERK, and ATF6 help cells to cope with and adapt to ER stress by improving the fidelity of protein folding, enhancing the elimination of irreparably damaged ER proteins, and alleviating the protein-folding demands upon the ER through translational attenuation. The requirement of the UPR in maintaining ER homeostasis is exemplified by the exquisite sensitivity of cells bearing knockouts of UPR genes to ER stress-induced damage and cell death and, conversely, the enhanced resistance to ER stress when UPR-associated proteins are artificially activated (14 -19). However, if the transcriptional and translational programs of the UPR fail to restore ER homeostasis and remove ER stress, persistent UPR activation creates a maladaptive cellular environment that can culminate in cell death. The mechanisms by which UPR signaling shifts from the restoration of ER homeostasis to the promotion of cell death are under active investigation, as this has important implications for cell survival and function in many disease conditions.
Metazoan cells contain an abundance of small noncoding microRNAs (miRs) that function in post-transcriptional gene silencing. miR genes are present throughout the genome and can be found within introns or exons of other genes. miRs are initially transcribed as a much longer primary miRNA precursor in the nucleus (20). Progressive processing by Drosha/ DGCR8 and Dicer enzymes ultimately yields the mature 18 -23-nucleotide miR in the cytosol (20). miRs typically regulate target genes by binding to cognate sequences within the 3Ј-UTR of target mRNAs, leading to translational inhibition and mRNA degradation (20). By this mechanism, miRs play an important role in regulating a wide variety of cellular signaling events through modulating the protein levels of their target genes (21). A growing number of studies have implicated a role for miRs in the UPR (22). Numerous miRs have been identified that target and down-regulate expression of UPR regulators (23)(24)(25)(26)(27)(28)(29). Conversely, UPR-induced transcription factors such as CHOP can directly up-regulate specific miR genes (30,31). These findings suggest that miRs sculpt the UPR by regulating expression of UPR genes, and the UPR also induces miRs as part of its transcriptional programs to influence cellular responses to ER stress.
In this present study, we hypothesized that the UPR may induce miRs that influence the cell life/death choice in response to ER stress. We screened a commercially assembled human microRNA gene panel for miRs up-regulated in HEK293 cells exposed to chemical forms of ER stress. We identified a small group specifically induced by the PERK arm of the UPR under experimental conditions when unabated PERK signaling causes cell death. We found that one of these, miR-483, promoted cell death when expressed in human cells. We provide evidence that miR-483 decreased cell viability by targeting expression of creatine kinase brain-type, leading to loss of cellular ATP stores.

Comprehensive analysis of ER stress-regulated human miRNAs
Cells respond to ER stress by activating potent transcriptional programs that up-regulate many mRNAs. We hypothesized that ER stress also broadly reshapes the microRNA transcriptome, similar to its effects on the mRNA transcriptome. To investigate this potential role, we profiled the expression of 754 human miRNAs tiled on printed array cards (TaqMan array human MicroRNA AϩB cards, Thermo Fisher Scientific). We probed these cards with total RNA samples collected from HEK293 cells treated for 24 h with solvent control or tunicamycin, an agent that blocks N-linked glycosylation to create potent ER stress, and performed quantitative RT-PCR to determine the expression levels of all 754 miRNAs in our experimental samples. Using this strategy, we identified numerous miRNAs whose levels were profoundly regulated positively and negatively by ER stress (Fig. 1A). We focused our subsequent analysis on miR-215 and miR-483-5p, two of the most highly ER stress-induced miRNAs, because they were previously found to influence cell growth and differentiation in experimental cancer models. In glioma and colorectal cell lines, miR-215 promoted tumor differentiation and survival (32,33). In animal tumor models, miR-483-5p prevented metastases (34). Another highly ER stress-induced microRNA was miR-616*. Interestingly, we found that the gene encoding miR-616 lay within intron 2 of DDIT3/Chop, a proapoptotic transcription factor strongly up-regulated by ER stress. This genomic pairing of miR-616* within DDIT3/Chop was unique to primates and not seen in mice or lower organisms. Based on these functional and genomic observations, we selected miR-215, miR-483, and miR-616* for additional investigation to see whether they influenced cellular survival or death in response to ER stress.
Next, we examined whether miR-215, miR-483, and miR-616* induction was evident in other human cell types and in response to other sources of ER stress. To do this, we cultured HeLa cells and treated them with tunicamycin or thapsigargin, agents that induce potent ER stress by inhibiting protein glycosylation or disrupting ER calcium stores, respectively. We collected total RNA after 24 or 48 h of drug treatment and performed qRT-PCR to measure miR-215, miR-483, and miR-616* levels. We found progressive induction of all three microRNAs after chemical ER stress induction (Fig. 1B), similar to the transcriptional up-regulation of genes induced by ER stress, such as BiP and Chop (Fig. 1C) (18). These results supported our qRT-PCR array findings that miR-215, miR-483, and miR-616* are induced by ER stress.

miRs-215, 483-5p, and 616* are induced by the PERK arm of the unfolded protein response
In human cells, ER stress-induced transcriptional programs are activated by the UPR signal transduction cascades controlled by IRE1, PERK, and ATF6. To determine whether the UPR regulated the expression of miR-215, miR-483, and miR-616*, we selectively activated the IRE1, PERK, or ATF6 pathways in HeLa cells and then measured miRNA transcript levels by qRT-PCR. To activate the transcriptional program initiated by IRE1, we expressed the spliced X-box-binding protein 1 (XBP1s) transcription factor specifically produced by IRE1's endoribonuclease function (35,36). We confirmed increased XBP1s transcriptional activity after transfection through marked increase in mRNA levels of ERdj4, a target gene of XBP1s ( Fig. 2A) (37). By contrast, mRNA levels of Chop and BiP, target genes of the PERK and ATF6 pathways (7, 37), were not statistically significantly increased with XBP1s expression ( Fig. 2A). To activate the PERK pathway, we treated HeLa cells expressing Fv2E-PERK, a chimeric fusion of the PERK kinase domain with two FK506-binding domains, with AP20187, a small-molecule ligand previously demonstrated to drive the homodimerization, trans-phosphorylation, and activation of Fv2E-PERK (10,19,38). Prior studies demonstrated that AP20187 dimerization of Fv2E-PERK selectively activated the cells were treated with tunicamycin (Tm) at 5 g/ml for 24 h, and total RNA was isolated. The expression profiles of 754 human miRNAs from Tm-treated samples (n ϭ 3) were determined using Taqman miRNA array AϩB cards and are shown relative to untreated samples in the heat map. Three miRNAs induced by Tm treatment, miR-483-5p, miR-215, and miR-616*, were selected for further analysis. B and C, HeLa cells were treated with Tm (5 g/ml) or 500 nM thapsigargin (Tg) for 0, 24, or 48 h. Total RNA was collected, and miR-483-5p, miR-215, and miR-616* levels were measured by qPCR and shown relative to levels at 0 h. In  Total RNA was collected, and ERdj4, Chop, and BiP mRNA levels were measured by qRT-PCR and shown relative to GFP-transduced cells. Values are expressed as mean Ϯ S.D. (error bars) of at least three independent experiments. *, p Ͻ 0.05, analysis of variance. B, C, and D, HeLa cells expressing Fv2E-PERK or Tet-regulated ATF6(373) were treated with AP20187 (1 nM) or doxycycline (100 g/ml) for 48 h as indicated. BiP and Chop levels were measured by qRT-PCR, and Xbp-1 mRNA splicing was determined by RT-PCR. As a positive control, in D, Xbp-1 mRNA splicing was determined in cells treated with tunicamycin (5 g/ml). E, HeLa cells were transduced with XBP1s (ϩ) or GFP (Ϫ). After 3 days, miR-483-5p, miR-215, and miR-616* levels were quantified by qPCR and shown relative to GFP-transduced levels. F, HeLa cells stably expressing a chemical dimerizer-inducible Fv2E-PERK were treated with AP20187 (1 nM). After 48 h, miRNA expression levels were quantified by qPCR from AP20187-treated cells (ϩAP) and shown relative to N,N-dimethylformamide-treated samples (ϪAP). G, HeLa cells stably expressing a Doxregulated transcriptional activator domain of ATF6 were treated with 100 g/ml Dox. After 48 h, miRNA expression levels were quantified by qPCR from untreated cells (ϪDox) and shown relative to Dox-treated cells (ϩDox). Values are expressed as mean Ϯ S.D. of at least three independent experiments. *, p Ͻ 0.05, Student's t test.

PERK-induced miR-483 disrupts cellular ATP
translational and transcriptional programs of the PERK branch of the UPR without activating IRE1 or ATF6 in cells (10, 19, 38 -40). Consistent with our prior studies, the addition of AP20187 to cells bearing Fv2E-PERK caused marked induction of Chop, a target gene robustly up-regulated by PERK activation (Fig. 2B). By contrast, BiP, a marker for ATF6 signaling, and spliced Xbp1 mRNA, a marker of IRE1 activation, were not induced with AP20187 addition (Fig. 2, B and D). To activate the ATF6 pathway, we used HeLa cells expressing the 373-amino acid cytosolic transcriptional activator fragment of ATF6 (ATF6(373)) under the control of a doxycycline-regulated promoter (40 -42). Prior studies demonstrated that the addition of doxycycline lead to expression of the ATF6 transcriptional activator fragment and potent induction of the BiP gene, without triggering Xbp1 mRNA splicing or activating the PERK pathway (39,40,42). Consistent with our prior studies, the addition of doxycycline to cells bearing the TetOn-ATF6(373) construct led to marked induction of BiP, with lesser effects on Chop and no splicing of Xbp-1 mRNA (Fig. 2, C and D).
Next, we applied these validated XBP1s, Fv2E-PERK, and doxycycline (Dox)-regulated ATF6(373) systems to determine how UPR signaling regulated the three ER stress-induced miRs selected for analysis. We found that XBP1s expression did not induce miR-215, miR-483, and miR-616* (Fig. 2E). By contrast, selective PERK activation using the Fv2E-PERK system up-regulated all three miRs (Fig. 2F). Selective activation of the ATF6 pathway using the Dox-inducible ATF6(373) system up-regulated miR-616* but had no significant effects on the levels of miR-215 or miR-483 (Fig. 2G). These results showed that the PERK signaling pathway was the most potent of the three signaling arms of the UPR for induction of miR-215, miR-483, and miR-616*.

Overexpression of miR-483 promotes cell death
Unabated ER stress or sustained signaling by the PERK arm of the UPR promotes cell death by mechanisms that are poorly understood. We investigated whether the ER stress and PERKinduced expression of miR-215, miR-483, and miR-616* influenced cell survival by overexpressing them in HeLa cells by lentiviral transduction. We used the psiCHECK-2 microRNA Biosensor luciferase reporter assay (Promega) to empirically confirm that these overexpressed miRs were functional by showing silencing of Renilla luciferase and a reduced ratio of Renilla/firefly luciferase luminescence using this assay (Fig. S1). When we examined protein lysates collected after 48 h of miR expression, we found strong production of cleaved PARP and cleaved caspase-3, markers of apoptosis, in cells expressing miR-483 (Fig. 3, A and B). By contrast, we observed little cleaved PARP production with miR-215 or miR-616* expression (Fig.  3A). These findings implicated miR-483 as a mediator of PERKinduced cell death. Because ATF4 and CHOP are also induced by PERK and promote cell death by inducing downstream target genes during the UPR (39,43,44), we next examined whether the up-regulation of miR-483 by PERK signaling operated through ATF4 or CHOP. We used adenoviruses to overexpress ATF4 or CHOP, which were previously shown to potently increase levels of functional ATF4 or CHOP proteins (39,43). We confirmed selective production of ATF4 or CHOP protein in transduced cells, compared with GFP-transduced cells (Fig. 3D). When we transduced cells with ATF4 for 48 h, we found strong induction of miR-483 (Fig. 3D). By contrast, CHOP expression failed to induce miR-483 (Fig. 3D).
To further test the link between ATF4 and miR-483, we introduced short hairpin RNAs against Atf4 mRNA sequences (shATF4) into the HeLa[Fv2E-PERK] cells. This shATF4 construct was previously demonstrated to robustly knock down ATF4 in HT1080 and DLD1 human cell lines (45). We confirmed that ATF4 production was substantially diminished in our HeLa cells expressing shATF4 compared with control scrambled mRNA hairpin sequences after tunicamycin or AP20187 exposure (Fig. 3C, top). We confirmed that ASNS, a downstream target gene of ATF4, was not normally induced when ATF4 was knocked down (Fig. 3C, bottom) (17,43). When we measured miR-483 levels under these conditions, we found significantly reduced miR-483 production when ATF4 was knocked down compared with controls. These findings placed ATF4 upstream of miR-483 induction and suggested that ATF4 may directly up-regulate miR-483 transcription.

miR-483 silences creatine kinase brain-type during the UPR
To explore the mechanism by which miR-483 caused cell death during the UPR, we identified potential mRNA targets based upon the presence of bioinformatically predicted miR-483-binding sites within the 3Ј-UTR of mRNAs using the Tar-getScan algorithm (http://www.targetscan.org/vert_72/) (81). 5 We focused on two predicted miR-483 target mRNAs that encoded for proteins with functions required for cell homeostasis and whose loss could increase cell death upon miRNA-483-mediated gene silencing. The creatine kinase brain-type (CKB) mRNA contained a miR-483 binding site in its 3Ј-UTR that was conserved across mammalian CKB mRNAs (Fig. 4A). CKB catalyzes the production of phosphocreatine to ensure that cells have sufficient ATP for cellular energy demands. Recent experiments demonstrated that miR-483 overexpression suppressed tumor liver metastasis by down-regulating CKB in the hepatic microenvironment of experimental animal models (34). Another mRNA encoding X-linked inhibitor of apoptosis (XIAP) also contained multiple putative miR-483binding sites in its 3Ј-UTR (Fig. 4A), although these were not conserved in Xiap mRNAs from other species. XIAP prevents cell death by directly inhibiting and promoting the degradation of caspase enzymes (46). We previously demonstrated that XIAP protein levels fall in response to ER stress and PERK activation (39). These observations suggested that miR-483 impaired cell viability via two different mechanisms: 1) suppression of XIAP to release the brake against caspases and 2) suppression of CKB to disrupt cellular ATP stores.
In support of this model, we saw clear drops in XIAP and CKB protein levels after miR-483 expression (Fig. 4, C and D) comparable with the drops in XIAP and CKB levels seen after tunicamycin treatment or selective PERK activation (Fig. 4B) (39). To further test miR-483's role in suppressing XIAP and CKB, we used CRISPR-Cas9 gene editing to delete miR-483 from HeLa[Fv2E-PERK] cells (Fig. 4E). Then we selectively activated the PERK pathway using AP20187. In HeLa[Fv2E-PERK] cells retaining miR-483, we found a progressive reduction in XIAP and CKB levels upon AP20187 treatment, but this was abolished in cells deleted for miR-483 (Fig. 4F). These findings provided experimental evidence linking miR-483 expression to down-regulation of XIAP and CKB protein levels.
To determine whether miR-483 directly silenced the Xiap or CKB mRNAs through putative binding sites in the 3Ј-UTR, we used Renilla luciferase mRNA reporters fused to the 3Ј-UTRs of Xiap or CKB lacking or mutated at the miR-483 binding sites (34). Next, we expressed miR-483 in cells bearing these luciferase 3Ј-UTR reporters. We found that Renilla luciferase fused to the Xiap 3Ј-UTR lacking both putative miR-483-binding sites was still suppressed by miR-483 (Fig. 4G). By contrast, mutation of the miR-483-binding site in the 3Ј-UTR of CKB rendered the Renilla luciferase-CKB-3Ј-UTR reporter construct insensitive to miR-483 suppression (Fig. 4H). These findings provided evidence that miR-483 directly targeted the CKB mRNA to silence its expression, whereas it indirectly down-regulated XIAP. Our findings indicate that miR-483 directly silences CKB and indirectly gene-silences XIAP after induction by PERK signaling.

ER stress suppresses CKB to deplete cellular ATP stores
To explore how cellular ATP levels changed during the UPR, we created HeLa cells bearing a fluorescence-based intracellular reporter of ATP concentrations, ATeam (adenosine 5Ј-triphos- Cell lysates were immunoblotted for cPARP, cleaved caspase-3, or actin (loading control). C, HeLa[Fv2E-PERK] cells stably expressing a short hairpin sequence against Atf4 (shATF4) or a scrambled control (shCont) were treated with Tm (5 g/ml) or AP20187 (1 nM) for 48 h. Cell lysates were immunoblotted for ATF4 or actin (loading control). Total RNA was collected, and miR-483 levels were analyzed by qRT-PCR and shown relative to levels in cells with no drug treatment (Ϫ). ASNS is a transcriptional target of ATF4 and was also evaluated by qRT-PCR to confirm reduced ATF4 transcriptional function in shATF4-treated cells. Representative immunoblots are shown, and values are expressed as mean Ϯ S.D. (error bars) of at least three independent experiments. *, p Ͻ 0.05 Student's t test. D, top, cells were transduced with adenovirus expressing GFP, ATF4, or Chop at 10 9 (ϩ), 10 10 (ϩϩ), or 10 11 (ϩϩϩ) pfu, and protein lysates were immunoblotted for ATF4 or CHOP. ␤-Tubulin was used as a loading control. Bottom, HeLa cells were transduced with adenovirus expressing ATF4 or CHOP at 10 11 pfu. Total RNA was collected after 48 h, and miR-483 levels were analyzed by qRT-PCR and are shown relative to untransduced controls (Ϫ). Values are expressed as mean Ϯ S.D. of at least three independent experiments. *, p Ͻ 0.05, Student's t test.

PERK-induced miR-483 disrupts cellular ATP
phate indicator based on ⑀ subunit for analytical measurements) (47). In this reporter, a cyan fluorescent protein (CFP) and a yellow fluorescent protein (YFP) are linked by the small ATP-binding ⑀ subunit of a bacterial F 0 F 1 -ATP synthase (Fig. 5A). When ATP is absent, the ⑀ subunit linker is distended, allowing for distinct CFP fluorescence (Fig. 5A). Upon ATP binding to the ⑀ subunit, the

PERK-induced miR-483 disrupts cellular ATP
linker contracts to bring the CFP and YFP in close proximity, thereby enabling excitation of acceptor YFP by donor CFP (YFP-FRET signal) (Fig. 5A). Prior studies had demonstrated that the ATeam construct enabled sensitive real-time identification of HeLa cells with depleted ATP levels by quantification of the YFP-FRET/CFP ratio in live cells (47).
We transduced HeLa cells stably expressing the ATeam reporter with miR-483 or GFP. We found a substantial decrease

PERK-induced miR-483 disrupts cellular ATP
in the YFP/CFP ratio in cells expressing miR-483 (Fig. 5A). Consistent with our live-cell measurements, we also saw a trending decrease in ATP levels using an enzymatic assay of lysates prepared from cells expressing miR-483 (Fig. 5B). Additionally, following glucose starvation, miR-483-transduced cells exhibited decreased cell viability compared with GFPtransduced control cells, whereas CKB shRNA-transduced cells were also less viable than cells expressing control shRNA sequences (Fig. 5C). These findings suggested that miR-483dependent silencing of CKB leads to cellular ATP depletion with deleterious consequences for cell viability.
To further determine how CKB protein levels impacted cell survival during ER stress, we tested whether restoration of CKB could prevent ER stress-induced cell death. We transfected cells with CKB and exposed them to increasing doses of tunicamycin. We confirmed increased CKB protein levels by immunoblotting in transfected cells compared with mock-treated cells (Fig. S2). After 72 h, we saw increased cell death morphologically and biochemically with increased production of cleaved caspase-3 on immunoblotting, and this was partially attenuated in cells overexpressing CKB (Fig. 5D). These results provide evidence that CKB played a cytoprotective role during the UPR, and its down-regulation by PERK's induction of miR-483 increased cellular vulnerability to ER stress-induced cell death.

Discussion
Unchecked ER stress leads to cell death, and dysregulated PERK activity can drive this process (18,19,43,48). In this study, we identified novel signal transduction steps by which PERK signaling impairs cell viability (Fig. 5E). First, we found that the PERK arm of the UPR selectively up-regulated miR-483 through the actions of the ATF4 transcription factor. Then we demonstrated that miR-483 silenced two genes regulating vital cellular processes: XIAP and CKB. miR-483 directly targeted the 3Ј-UTR of CKB, leading to its silencing, consistent with prior report of this interaction in metastatic cancer cells (34). Next, we provided evidence that miR-483 overexpression depleted cellular ATP stores. Based on these findings, we propose that dysregulated PERK signaling disrupts ATP homeostasis and metabolically weakens the cell. Coupled with other maladaptive changes induced by extended UPR signaling, the cell readily succumbs to cell death in the face of continued ER stress.
Our identification of miR-483 as part of the PERK signaling cascade adds to the growing number of genes induced by the PERK arm of the UPR that elicit maladaptive cellular effects. The best-characterized example is CHOP. CHOP's proapoptotic properties were initially noted because of the enhanced ability of Chop Ϫ/Ϫ cells to survive chemically induced ER stress in vitro (44) and were later reaffirmed, in vivo, by the partial amelioration of pathology and improved tissue function when Chop was deleted in some mouse models of diseases associated with ER stress (49 -53). CHOP's transcriptional program imposes protein-folding demands on the ER and disrupts the luminal ER redox environment during the UPR (43,54), and in some cell types, CHOP directly induces cell death executioners (48,55,56). However, CHOP expression does not trigger cell death, and Chop deletion does not change outcomes in many other ER stress disease models (39,43,(57)(58)(59)(60). By contrast, ATF4 can mediate ER stress-induced cell death both in vitro and in vivo (61)(62)(63)(64)(65). ATF4 can biochemically partner with CHOP via their bZIP domains and shares many transcriptional targets with CHOP (7,43,66). ATF4 also engages apoptotic pathways by stimulating TRAIL death receptors and suppressing anti-apoptotic IAP molecules (39,67). These findings implicate both ATF4 and CHOP as PERK-induced transcription factors that drive the maladaptive phase of the UPR leading to cell death. Indeed, ATF4 and CHOP co-expression caused increased protein expression, increased oxidative stress, reduced ATP/ADP ratios, and more cell death (43).
This reduction of ATP during the UPR was previously proposed to be a consequence of increased energy consumption arising from ATF4-CHOP-driven protein expression. Our current study provides a direct mechanism into how ATP is depleted during the UPR, through our identification of an ER stress-and PERK-induced miR-483 that directly silences the CKB enzyme required for maintaining cellular ATP stores. ATF4 promotes the loss of ATP via miR-483 induction and suppression of phosphocreatine generation, but ATF4 also concomitantly promotes new protein synthesis by activation of the ATF4-CHOP transcriptional program (43). Clearly, increased energy demand in the face of falling energy supplies is unsustainable for the cell. Reducing energy demands and restoring ATP levels could be accomplished if ATF4, CHOP, and miR-483 levels returned to baseline. However, in situations with persistent ER stress that lead to dysregulated PERK signaling, sustained induction of ATF4, CHOP, and miR-483 could inexorably worsen imbalances in cellular energetics and ultimately culminate in an inviable metabolic environment where apoptosis is triggered. This imbalance in cellular energetics created by dysregulated PERK signaling is an important determinant in the decision to undergo cell death in response to chronic ER stress. miR-483 joins the growing list of small RNAs regulated by PERK signaling (24 -26, 30, 31, 68). Interestingly, many of these PERK-modulated miRs also influenced cell survival in response to ER stress. For instance, miR-211 is induced rapidly after ER stress by the PERK arm of the UPR, but miR-211 levels decline with extended ER stress (24). miR-211's targets include Chop and circadian cycle regulators, and silencing of these genes promotes cell survival (24,68). Induction of miR-211 could contribute to the protective effects of PERK activation in response to acute ER stress. By contrast, the PERK-regulated miR-106b-25 cluster targets the Bim Bcl-2 family pro-apoptosis factor, and extended experimental PERK signaling (Ͼ24 h) downregulated miR-106b-25, leading to increased Bim protein (26). Loss of miR-106b-25 and ensuing increase in BIM is another maladaptive consequence of extended PERK signaling. In our studies, we identified miR-483 induction in cells exposed to extended experimental ER stress or PERK activation (Ͼ24 h). miR-483-mediated silencing of CKB operates at similar time points as the loss of miR-106b-25 silencing of Bim. Together, these distinct maladaptive effects of extended PERK signaling-loss of CKB and rise of BIM-influence the final cellular decision to undergo cell death. These studies suggest that miRs

PERK-induced miR-483 disrupts cellular ATP
are dynamically regulated in response to acute and extended PERK signaling and help shape the protective or maladaptive effects on the cell. miR-483 has been readily detected in blood, ascites, and pleural fluid samples from patients with a wide variety of cancers. miR-483-5p was significantly overexpressed in plasma from both hepatocellular carcinoma and chronic lymphocytic leukemia/lymphoma patients (69,70). miR-483-5p was also upregulated in circulating blood samples from patients with adrenocortical carcinoma, and miR-483-5p overexpression correlated with significantly worse survival and recurrence in this cancer (71)(72)(73). Multiple myeloma patients with increased circulating miR-483-5p had significantly worse progression-free survival compared with patients with low plasma levels of miR-483-5p (74). Based on these cancer epidemiologic associations, clinical tests for miR-483-5p are under development for potential use as a minimally invasive biomarker for cancer diagnosis and prognosis (75). A role for ER stress and PERK signaling in up-regulating circulating miR-483-5p in these different cancers has not been examined. However, cancers thrive under many adverse environmental conditions linked to ER stress and UPR activation, including hypoxia, starvation, and oxidative stress (76). Based on our findings, the increased miR-483-5p levels found in these patients could plausibly arise from increased ER stress and PERK activation in these cancers. If future studies of these cancers validate this conjecture, plasma testing of miR-483-5p could be used as a biomarker to quantify the amount of ER stress or the strength of UPR activation in tumors. In turn, these miR-483-5p-positive cancers could be particularly responsive to the growing number of small molecules that modulate the UPR (77)(78)(79)(80). In particular, our studies predict that pharmacologic inhibition of the PERK branch of the UPR or the integrated stress response will be potent in driving down ATF4-CHOP-miR-483 levels, whereas agents that enhance phosphorylation of eIF2␣ will have the opposite effect.

TaqMan array miRNA profiling
Total RNA from HEK293 cells treated with and without 5 g/ml tunicamycin for 24 h was extracted using the mirVana miRNA isolation kit (Ambion, Foster City, CA) according to the manufacturer's instructions. 1 g of RNA was reverse-transcribed using Megaplex RT Primers (human pool A and B version 2.1, Applied Biosystems) and the TaqMan miRNA reverse transcription kit (Applied Biosystems). Real-time qRT-PCR with human TaqMan array cards (A and B) was performed on Applied Biosystems 7900HT system. Data analysis was performed using DataAssist Software version 2.0 (Applied Biosystems).
For qPCR analysis, total RNA, including the miRNA fraction, was prepared using the miRNeasy mini kit (Qiagen), and 1 g of RNA was subjected to an RT reaction with the miScript II RT kit (Qiagen) and proprietary miR-483-5p, miR-616*, and miR-215 primers (Qiagen miScript Primer Assay). Real-time qPCR for mature miRNAs was carried out by an miScript SYBR Green PCR kit using specific miRNA assay primers obtained from Qiagen on a CFX96 thermal cycler. Expression of the U6 small RNA was used as an internal control (Qiagen miScript primer assay).

Conditional activation of individual UPR pathways
Stable transfectants of HeLa cells with chemical-genetic constructs to regulate the PERK pathway (HeLa/Fv2E-PERK) or ATF6 (HeLa/TO-ATF6(373)) were established previously (39). PERK signaling in HeLa/Fv2E-PERK cells and ATF6 signaling in HeLa/TO-ATF6(373) were specifically activated by AP20187 (ARIAD) or by withdrawing Dox from the culture medium, respectively. To selectively activate the IRE1-XBP1 pathway, enforced expression of spliced XBP1 was introduced by lentiviral transduction of human XBP1s. Lentivirus packaging was performed with the integration-defective lentiviral system to avoid random integration of the transgene (LENTI-Smart NIL, Invivogen, San Diego, CA).

PERK-induced miR-483 disrupts cellular ATP
pATeam-Blast was lentivirally transduced into HeLa/Fv2E-PERK cells, and HeLa/ATeam cells were established. Flow cytometry analysis was performed on a BD FACSCanto at the Flow Cytometry Core at the San Diego Center for AIDS Research; compensation was performed using UltraComp eBeads (Fisher, 501129038) bound to antibodies of the corresponding wavelengths. Dead cells were excluded using LIVE/ DEAD Fixable Far Red Stain (Thermo Fisher Scientific, L34973), and singlet discrimination was performed using FSC-H versus FSC-W gating. The CFP, YFP, and YFP acceptor (excited by 435 nm) fluorescent signals from the ATeam reporter were quantified from 10,000 events. Data analysis was done using Flowing Software.
ADP/ATP ratio values in cell lysates were determined using an ADP/ATP ratio assay kit (Sigma-Aldrich, MAK135) as directed.

CRISPR/Cas9 gene editing
To create miR-483-deleted cells, two sgRNAs were designed. Two sets of synthesized DNA oligonucleotides (senseleft, caccgGCATTGCTGTGGGGGAGAGG; antisense-left, aaacCCTCTCCCCCACAGCAATGCc; sense-right, caccgGG-CACCACCTAGGAGGCTGG; antisense-right, aaacCCAGC-CTCCTAGGTGGTGCCc) were obtained from Valuegene (San Diego, CA). Each set of oligonucleotides was phosphorylated with T4 PNK (New England Biolabs) by incubation for 30 min at 37°C, annealed in ligation buffer (New England Biolabs) by incubation for 5 min at 95°C, and declined in temperature to 25°C at 0.1°C/s on a thermal cycler. The oligonucleotide reaction mixture was diluted 100ϫ in water and inserted into linearized PX330 (a gift from Feng Zhang) by BbsI with a quick DNA ligation kit (New England Biolabs), and PX330-miR483-Left and PX330-miR483-Right plasmids were constructed. 1 g of each plasmid was mixed with 0.2 g of pBabe-puro-empty vector. The plasmid mixture was transfected into HeLa-Fv2E-PERK cells. At 48 h after transfection, culture medium was replaced with fresh DMEM containing 2.5 g/ml puromycin to select for cells expressing transgene. After puromycin selection, single cells were isolated in 96-well plates. Clones were expanded, and deletion of miR-483 was confirmed by PCR amplification of the genomic DNA flanking the miR-483 gene with the following primer set (forward, GGTGCCAGCCAGTCCTTG; reverse, CCACAACCAGAGGGACACC), and PCR products were resolved on 2.5% agarose gel in 1ϫ TBE buffer.

Luciferase 3-UTR reporters
The psiCheck2 luciferase reporter assays (Promega, Madison, WI) were performed as described previously (34). The WT and mutant CKB luciferase constructs were provided by Dr. Tavazoie (34). The Xiap 3Ј-UTR containing putative miR-483binding sites or truncated to omit both miR-483-binding sites was cloned into the psiCheck2 Dual-Luciferase reporter vector (34). Cells were co-transfected with miRNA expression constructs and luciferase reporter constructs, and Renilla and firefly luciferase activities were determined using a dual-luciferase assay kit per the psiCheck2 manufacturer's instructions (Promega) (34). The psiCheck2 luciferase reporter was used to functionally validate expression of transduced miRs, by cloning synthetic miR-binding sites into the 3Ј-UTR of Renilla luciferase, and measuring the ratio of Renilla and firefly luciferase activities per the manufacturer's instructions.

Western blot analysis
For Western blot analysis, 10 -20 g of protein was prepared in SDS lysis buffer (62.5 mM Tris-HCl, pH 6.8, 2% (w/v) SDS, 10% glycerol), mixed with 100 mM DTT in 1ϫ LDS buffer (Invitrogen), and incubated at 70°C for 5 min prior to loading onto a 4 -15% gradient SDS-polyacrylamide gel (Bio-Rad). After SDS-PAGE, samples were transferred to polyvinylidene difluoride membranes (Bio-Rad) in 1ϫ Tris/glycine buffer containing 20% methanol and transferred at 250 mA for 1 h. The membrane was blocked in 5% skim milk in TBS buffer containing 0.1% Tween 20 (TBST) for 1 h at room temperature and was subsequently incubated with primary antibody in 5% BSA in TBST buffer at 4°C overnight with gentle shaking. After incubation with primary antibody, the membrane was washed in TBST buffer three times at room temperature and then incubated with secondary antibody conjugated with horseradish peroxidase (HRP) for 1 h at room temperature. HRP-catalyzed chemiluminescent signal using SuperSignal West Pico or Femto solution (Pierce) was detected on ChemiDoc (Bio-Rad).