PML nuclear bodies contribute to the basal expression of the mTOR inhibitor DDIT4

The promyelocytic leukemia (PML) protein is an essential component of PML nuclear bodies (PML NBs) frequently lost in cancer. PML NBs coordinate chromosomal regions via modification of nuclear proteins that in turn may regulate genes in the vicinity of these bodies. However, few PML NB-associated genes have been identified. PML and PML NBs can also regulate mTOR and cell fate decisions in response to cellular stresses. We now demonstrate that PML depletion in U2OS cells or TERT-immortalized normal human diploid fibroblasts results in decreased expression of the mTOR inhibitor DDIT4 (REDD1). DNA and RNA immuno-FISH reveal that PML NBs are closely associated with actively transcribed DDIT4 loci, implicating these bodies in regulation of basal DDIT4 expression. Although PML silencing did reduce the sensitivity of U2OS cells to metabolic stress induced by metformin, PML loss did not inhibit the upregulation of DDIT4 in response to metformin, hypoxia-like (CoCl2) or genotoxic stress. Analysis of publicly available cancer data also revealed a significant correlation between PML and DDIT4 expression in several cancer types (e.g. lung, breast, prostate). Thus, these findings uncover a novel mechanism by which PML loss may contribute to mTOR activation and cancer progression via dysregulation of basal DDIT4 gene expression.


PML loss results in lower DDIT4 expression.
In order to investigate if PML might also regulate DDIT4, we first generated two PML knock-down cell lines. PML was stably silenced in U2OS osteosarcoma cells and TERT-immortalized normal human diploid fibroblasts (NHDF) using a previously well characterized short-hairpin RNA 28 (shPML-A; Fig. 1A). In both cells lines, silencing of PML was associated with decreased levels of DDIT4 protein (Fig. 1A), with U2OS cells showing a greater decrease than NHDF cells (57% vs 47%, Fig. 1B). To control for off-target effects with the shPML hairpin, we also silenced PML in U2OS cells using a second commercially available short-hairpin RNA (shPML-B) and observed a similar decrease (45%) in DDIT4 expression by Western blotting (Fig. 1A and B). Since the U2OS shPML-A showed a greater decrease in DDIT4 expression than the shPML-B cells, we used the shPML-A U2OS cell line for all subsequent experiments unless otherwise indicated.
Activation of mTOR leads to phosphorylation of S6-Kinase, which subsequently phosphorylates and activates the ribosomal protein S6 at serine residues 235/236 29 . Given that DDIT4 is an inhibitor of mTOR activity, we next assessed whether PML knock-down affected the phosphorylation status of S6 in our cell system, and indeed we observed an increase in S6 phosphorylation in shPML cells that correlated with reduced DDIT4 expression (Fig. 1C). These results are consistent with previous work linking PML loss to increased phospho-S6 20 . However, our new data extends these findings to include reduced DDIT4 expression as a contributing mechanism to mTOR activation following the loss of PML.
The DDIT4 protein is more stable in PML-depleted cells. We next investigated possible explanations for altered DDIT4 protein expression in the shPML cells, which could include reduced transcription, translation, and/or changes in protein stability. DDIT4 is subject to post-translational regulation through ubiquitination and degradation through the 26S proteasomal pathway and has a reported half-life of ~5-10 min under normal growth conditions [30][31][32] . Using cycloheximide to block de novo protein synthesis, we observed the degradation of DDIT4 over 40 min in U2OS shControl and shPML cells and found that DDIT4 appeared to be more stable in the shPML cells ( Fig. 2A). In addition, we performed three independent experiments and calculated a decay curve for DDIT4 protein loss in the shControl and shPML cells (Fig. 2B). Despite DDIT4 protein levels being lower in PML-depleted cells (Fig. 1), the DDIT4 protein was significantly more stable at 20 and 40 min in the shPML cells with a half-life of ~16 min compared to ~10 min in shControl cells. Thus, the reduced DDIT4 levels seen in the shPML cells are not due to increased degradation of DDIT4.
Loss of PML protein is associated with decreased DDIT4 transcription rather than altered translation. Since reduced DDIT4 protein expression in PML depleted cells cannot be explained by altered protein stability, we sought to determine if PML loss affected DDIT4 transcription and/or translation. To address these questions we used whole genome microarrays to assess the gene expression changes in U2OS shControl and shPML cells. In addition, since PML is known to associate with mTOR 20 and various components of the translation initiation machinery including eIF3E 33 , eIF3K 34 and eIF4E 35 , we also assessed differences in translation by identifying the genes with altered association with the polyribosome fraction (translation state array) 36 . Using these approaches, we identified DDIT4 as significantly depleted in both the total RNA and polysome fractions of the shPML cells (p < 0.01; Fig. 2C). Since DDIT4 transcripts were reduced in the total RNA fraction in shPML U2OS cells compared to shControl cells, altered transcription rather than translation is likely responsible for the reduced DDIT4 transcripts in the polysome fraction. Consistent with these data, we confirmed that total DDIT4 gene expression was significantly impaired in the shPML U2OS cells by quantitative reverse transcription-PCR (qRT-PCR) (p < 0.01; Fig. 2D). We also assessed DDIT4 mRNA expression in NHDF cells, which have reduced Scientific RepoRts | 7:45038 | DOI: 10.1038/srep45038 DDIT4 protein expression with PML silencing (Fig. 1). Although not statistically significant, we did observe a trend towards reduced DDIT4 mRNA in the shPML NHDF cells (p = 0.19) (Fig. 2D). These results indicate that PML loss is associated with decreased DDIT4 gene expression rather than altered translation. Given the robust and reproducible changes in DDIT4 gene expression with PML loss in U2OS cells, all further investigation of the transcriptional regulation of DDIT4 by PML employed this cell line.
In order to address how transcription of the DDIT4 gene might be affected by PML loss, we cloned a 1683 bp fragment (− 1561 to + 122) of the DDIT4 promoter from U2OS cells into a luciferase expression system (Fig. 2E). This DNA fragment encodes several reported regulatory elements for DDIT4 transcription in response to various stimuli, including IR and hypoxia. For example, a p53 response element is located about 300 bp upstream of the TSS and in SAOS2 and U2OS cells this is important for p53-dependent activation of DDIT4 in response to ionizing radiation 23 . As well, DDIT4 induction during hypoxia depends on an SP1 site located at − 474 to − 446 and a hypoxia response element (HIF1α binding) located − 427 to − 407 27,37 . We also analyzed the DDIT4 promoter region (− 2500 to + 500) for potential association with PML using the ENCODE transcription factor binding data available through UCSC genome browser (https://genome.ucsc.edu/) and found two regions within our cloned DDIT4 promoter sequence that were associated with PML by ChIP-Seq (Fig. 2E). The shorter region (peak 1, − 1218 to − 990) was identified in K562 cells and had a ChIP-seq peak signal score of 778/1000. The longer region (peak 2, − 193 to + 316) was identified in both K562 and GM12878 cells ChIP-Seq signal scores of 925/1000 and 1000/1000 respectively and our cloned promoter fragment contains the sequences from − 193 to + 122 of peak 2, which overlaps the transcription start site (TSS) of DDIT4 (Fig. 2E). The expression vector containing the 1683 bp DDIT4 promoter region robustly induce luciferase expression 300-fold compared to promoterless pLuc control vector in the shControl cells (Fig. 2F), indicating that we had cloned an active promoter. In the shPML U2OS cells, the same reporter induced luciferase activity 250-fold compared to the promoterless control. While the luciferase activity in the shPML U2OS cells tended to be lower than in the shControl cells, this trend was not statistically significant (p-value = 0.15). Thus, loss of PML only modestly impacted DDIT4 promoter activity when assessed in our extra-chromosomal luciferase reporter assay. Since PML is not known to act directly as a TF, these results indicate that the significant changes in endogenous DDIT4 transcription triggered by PML loss may be occurring at the level of chromatin and/or subnuclear chromosome organization, where PML NBs may play a role in organizing chromosomal domains 14 . PML nuclear bodies are associated with DDIT4 gene loci. PML NBs are known to associate with sites of active transcription and at least one gene, TAP1, is transcriptionally regulated through association with PML NBs 11,12 . Other genes, such as BCL2, show no association with PML NBs [11][12][13] . In the Ching et al., study 11 , the relative localization of the TAP1 and BCL2 genes relative to PML NBs was determined using immuno-fluorescent in situ hybridization (FISH) in cell nuclei fixed to preserve their 3D organization. Therefore, we wanted to employ 3D immuno-FISH to study the relative localization of the DDIT4 gene loci to PML NBs as compared to TAP1 (positive control) and BCL2 (negative control). However, before beginning our 3D immuno-FISH studies, we first examined the relative expression of DDIT4, PML, BCL2 and TAP1 between shControl and shPML U2OS cells using RNA-Sequencing (Fig. S1A). The reduced PML and DDIT4 expression in shPML cells based on fold-change relative to shControl cells was very similar to the qRT-PCR data we obtained in Fig. 2D for U2OS cells. We also confirmed BCL2 and TAP1 gene expression changes between shControl and shPML U2OS cells by qRT-PCR (Fig. S1B). TAP1 and DDIT4 gene expression were, respectively, upregulated or downregulated by PML loss, consistent with a role for PML NBs in regulating these genes. In contrast, loss of PML did not affect BCL2 expression in U2OS cells.
Having characterized the effects of PML loss on the expression of TAP1 and BCL2 in U2OS cells, we next proceeded to 3D immuno-FISH analysis of wild-type U2OS cells using fluorescently labelled BAC probes to detect chromosomal regions containing the DDIT4, PML, BCL2 and the TAP1 genes relative to PML NBs (Fig. 3A). PML NBs appeared to be in closer proximity to the DDIT4 loci than any of the other gene loci; including TAP1, which was used as a positive control. In addition, we did not consistently observe PML NBs in close proximity to PML loci, indicating that, in U2OS cells, PML NBs may not regulate PML expression. Using imaging software to visualize in 3D the spatial relationship between PML NBs and each gene loci, we calculated the shortest Euclidian distances between each gene probe and the nearest PML body to provide an average gene probe-PML NB distance per nuclei. This analysis revealed that TAP1 was significantly closer (average 1.6 μ m) to PML NBs than the negative control BCL2 (average 2.0 μ m, p-value < 0.05, Fig. 3B). The DDIT4 gene locus was also significantly closer to PML NBs than BCL2 (1.4 μ m, p-value < 0.0003). Consistent with our visual assessment, the PML locus was not significantly closer to PML NBs than BCL2 (1.8 μ m, p-value = 0.22) in U2OS cells. These distances are consistent with previous studies using similar 2D and 3D immuno-FISH methods, where a positive association between gene loci and PML NBs ranged from ~1-1.6 μ m and unassociated genes were found at distances of 2 μ m or greater from a PML NB [10][11][12] . Given the previous observation that PML NBs are associated with actively transcribing genes 12 , we employed immuno-RNA FISH to determine if PML NBs are associated with DDIT4 RNA transcriptional foci (Fig. 4A). In these studies we used BCL2 RNA-FISH as a negative control since PML loss did not affect BCL2 expression (Fig. S1). Similar to the immuno-DNA-FISH studies, we calculated the distance between each DDIT4 or BCL2 RNA focus to the closest PML NB (Fig. 4A). We found that BCL2 transcriptional foci were on average 1.7 μ m from a PML NB, mirroring the DNA FISH results; whereas, DDIT4 transcriptional foci were on average 1.1 μ m from the nearest PML NB, which is significantly closer than BCL2 (p-value < 0.0001; Fig. 4B). We also noted that the RNA-FISH focus to PML NB distances for DDIT4 RNA foci were much smaller than for BCL2 foci (Fig. 4B), and were closer than observed for DDIT4 DNA FISH (Fig. 3), which suggests that actively transcribed DDIT4 loci are more closely associated with PML NBs. Together, these data support a model in which PML NBs are associated with actively transcribing DDIT4 gene loci in U2OS cells and this association supports a role for PML NBs in the regulation of DDIT4 gene expression.

Effects of PML silencing on DDIT4 induction by CoCl 2 . Having established that PML loss reduces basal
DDIT gene expression, we next explored whether PML silencing affects the ability of cells to regulate DDIT4 protein expression in response to hypoxia-like, metabolic, and genotoxic stress. DDIT4 is highly upregulated in response to hypoxic stress, and PML can inhibit mTOR activation under hypoxic conditions 20 . Therefore, we also analyzed DDIT4 response in shPML cells to hypoxia-like conditions induced by cobalt chloride (CoCl 2 ) treatment, which simulates hypoxic signaling through the stabilization of HIF1α and has been used to induce DDIT4 expression 24 . CoCl 2 treatment significantly increased DDIT4 mRNA levels in both the shControl and shPML U2OS cells (p < 0.01; Fig. 5A). This increase was correlated with an increase in DDIT4 protein expression (Fig. 5B). When these protein changes were quantified, we identified a ~6-fold increase in CoCl 2 -treated shControl cells (Fig. 5C). CoCl 2 -treated shPML U2OS cells had a similarly robust upregulation of DDIT4, reaching levels similar to CoCl 2 -treated shControl cells (Fig. 5C), suggesting that PML loss does not substantially affect DDIT4 induction in response to hypoxia-like stress. To assess the functional consequence of DDIT4 upregulation, we again analyzed S6 phosphorylation as an indicator of mTOR activity (Fig. 5D). In both shControl and shPML cells, CoCl 2 treatment resulted in decreased phosphorylation of S6, consistent with the observed increases in DDIT4 and its inhibitory activity on mTOR. However, although DDIT4 was robustly activated in both cell lines with CoCl 2 (Fig. 2B), we did observe that the shPML cells retained some S6 phosphorylation in the presence of CoCl 2 whereas phosphorylation of S6 was undetectable in the shControl cells. These results are consistent with a role of PML in directly inhibiting mTOR under hypoxic conditions 20 rather than through regulation of DDIT4.
Reduced basal DDIT4 expression with PML silencing indirectly limits DDIT4 upregulation following metformin and ionizing radiation treatment. Metformin is an inhibitor of oxidative phosphorylation and produces metabolic and oxidative and energy stress in treated cells 38 which increases DDIT4 expression in an AMPK-independent and p53-dependent manner, resulting in mTOR and growth inhibition 26 . Since PML is also implicated in p53-mediated transcriptional control 16,17,39 , we wanted to assess if PML loss could affect DDIT4 expression in response to metformin. Using quantitative real time PCR, we found that metformin treatment (5 mM) increased DDIT4 mRNA expression in both the shControl and shPML U2OS cells (Fig. 6A). This mRNA upregulation correlated with an increase in DDIT4 protein expression in U2OS cells as assessed by Western blotting (Fig. 6B). Similarly, metformin treatment also increased DDIT4 expression in the shControl and shPML NHDF cells (Fig. 6B). Quantification of the relative DDIT4 protein levels indicated significant (p < 0.05) DDIT4 upregulation with metformin-treatment in both the shControl and shPML U2OS cells (Fig. 6C). A similar trend was also observed in the NHDF cells (Fig. 6C). Thus, these data indicate that metformin-mediated increases in DDIT4 expression are not dependent on the presence of PML. However, the reduced basal levels of DDIT4 in untreated shPML cells results in reduced overall levels of DDIT4 following metformin treatment in shPML U2OS cells as compared to metformin-treated shControl cells ( Fig. 6A and C). Finally, to assess the response of DDIT4 to genotoxic damage we used ionizing radiation (IR), which can induce DDIT4 in a p53-dependent manner 23,40 . Similar to metformin treatment, 8 Gy of IR induced DDIT4 expression in both shControl and shPML U2OS cells but the final level of DDIT4 protein was reduced in PML-depleted versus control cells (Fig. S2).
The differences in total DDIT4 after metformin treatment could result in less efficient inhibition of mTOR by DDIT4 in the shPML cells. We therefore also assessed the impact of metformin treatment on the activation of mTOR by monitoring phospho-S6 in shControl or shPML U2OS cells (Fig. 6D). In shControl cells, metformin caused a 50% reduction in phospho-S6, consistent with an increase in mTOR inhibition via increased DDIT4. In contrast, the untreated shPML cells showed increased phospho-S6 relative to untreated shControl cells, consistent with decrease DDIT4 levels in the shPML cells, and metformin treatment only reduced phospho-S6 by 15%. Although metformin caused a dose-dependent decrease in viability in both the U2OS shControl and shPML cells, the shPML cells were more resistant overall (Fig. 6E). Thus, loss of PML reduces the sensitivity of U2OS cells to metformin, and our data implicates dysregulation of DDIT4 as a possible mechanism promoting cell growth in metformin-treated cells that correlates with increased mTOR activity.
Together, our data support a model in which PML loss is associated with decreased basal DDIT4 protein expression but not the direct induction of DDIT4 in response to metabolic, hypoxic and genotoxic stress. However, PML loss and the associated reduction in basal DDIT4 expression might indirectly influence the overall response to metabolic (e.g. metformin) stress.
PML and DDIT4 expression are positively correlated in several cancer cell lines and cancer types. Our data indicates that, under normal conditions, PML NBs play an important role in maintaining basal DDIT4 gene expression and that PML loss can lead to dysregulation of DDIT4 expression and therefore inappropriate mTOR activity that could drive cancer growth and progression. Therefore, we predict that PML and DDIT4 expression should be positively correlated, where cancer cells with low PML have low DDIT4 gene expression and vice versa. To test this hypothesis, we analyzed cancer genomics data available through cBioPortal (http://www.cbioportal.org) for mRNA expression correlations between PML and DDIT4, BCL2 or TAP1. Our initial analysis of 967 cancer cell lines from the Broad Institute's Cancer Cell Line Encyclopedia (http:// www.broadinstitute.org/ccle) 41 showed weak (Spearman's r = 0.199) but significant (p < 0.01) positive correlation between PML and DDIT4 gene expression (Table 1). In comparison, PML and TAP1 showed a stronger correlation (Spearman's r = 0.489, p < 0.01); and as predicted from cell lines studies, PML and BCL2 gene expression were not significantly correlated (Spearman's = 0.006, p = 0.854). These data are consistent with previous studies showing a positive correlation between PML and TAP1 and no correlation between BCL2 and PML expression 11,12 , and our immuno-RNA FISH data that indicated that PML NBs are more closely associated with actively transcribing DDIT4 loci but not BCL2 loci (Fig. 6).
We also assessed PML and DDIT4 mRNA expression correlations by cancer cell tissue type and found significant (p < 0.05) and strong correlations (Spearman's r = 0.275-0.345) in breast, ovarian and lung cancer cells (Table 1). Expression data for U2OS cells were also included in the CCLE osteosarcoma cell line data and, although the data was not significant (p = 0.130), DDIT4 showed a positive correlation with PML (Spearman's r = 0.312), consistent with our observations in this study. Analysis of other available TCGA gene expression data sets for human cancers accessible via the cBioPortal also identified cancers where DDIT4 and PML expression were significantly (p < 0.05) and positively (Spearman's r = 0.263-0.354) correlated, including breast 42 , ovarian 43 , prostate 44 , colorectal 45 , kidney 46 bladder 47 and lung 48 cancers (Fig. 7A and B).
Since the strongest correlation between DDIT4 and PML expression occurred in lung cancer derived cell lines (Table 1), and specifically lung adenocarcinoma (Spearman's 0.399, p < 0.01, Fig. 7B), we extended our analysis to lung cancer subtypes. In addition to a positive correlation between DDIT4 and PML mRNA expression in lung  = 0.365, Fig. 7B). In contrast, PML and DDIT4 showed no correlation (Spearman's r = 0.081, p = 0.281) in lung squamous cell carcinomas (Fig. 7C) 49 ; whereas, PML and TAP1 were positively correlated (Spearman's r = 0.566, p < 0.01). As expected, PML and BCL2 were not correlated (Spearman's r = 0.092, p = 0.219) in the lung squamous cell carcinoma samples. Thus, our analysis of TCGA and Broad cancer gene expression data sets revealed that PML and DDIT4 are positively correlated in several human cancer types, particularly those that are known to have lower PML (i.e. lung, breast, prostate) 1 . Thus, these analyses suggest that PML likely plays a role in regulating DDIT4 gene expression in multiple cell types and cancers of diverse tissue origin. Discussion PML NBs are associated with transcriptionally active gene loci 12 ; however, few genes are known that are specifically associated with and regulated by PML NBs. PML NBs are also involved in regulating cell fate decisions (e.g. apoptosis, senescence) following cellular stress (e.g. hypoxia, DNA damage), and loss of PML can promote cancer cell growth in part by activating mTOR 4,5 . In this study, we uncovered a novel mechanism by which PML loss can contribute to mTOR activation via the dysregulation of DDIT4, an inhibitor of mTOR. Silencing PML expression in both U2OS and NHDF cells resulted in decreased DDIT4 protein expression that correlated with increased mTOR activity (Fig. 1). The dysregulation of DDIT4 under basal conditions is, at least in part, due to altered DDIT4 transcriptional regulation as indicated by reduced mRNA levels in PML-depleted cells (Fig. 2), and by the close association of actively transcribing DDIT4 gene loci with PML NBs (i.e. within 1.1 μ m, p-value < 0.0001; Fig. 4). PML loss did not appear to affect the upregulation of DDIT4 in response to hypoxia-like (CoCl 2 , Fig. 5), metabolic (metformin, Fig. 6), or genotoxic stress (IR, Fig. S2). However, the reduced levels of basal DDIT4 in the shPML cells were correlated with reduced overall levels of DDIT4 compared to shControl cells following metformin treatment. Thus, our study has identified DDIT4 as a novel PML NB-associated gene, which like TAP1 10,50 , may be regulated through its association with PML NBs.
How PML and PML NBs might regulate DDIT4 gene expression is unclear but may involve coordination of active chromosomal domains at PML NBs and/or the post-translational modification (or reversible sequestration) of TFs and chromatin remodeling enzymes at PML NBs 14 . This is the most likely scenario given that PML is not known to act as a transcription factor or repressor directly by binding DNA. In addition, even though PML NBs make extensive contacts with chromatin at their periphery, the bodies themselves are devoid of nucleic acid 51 and thus genes regulated by PML NB-association would never appear to colocalize but rather are always juxtaposed to the body 10,11,52 . Although beyond the scope of this work, identification of these TFs and chromatin enzymes is rather daunting given that the ENCODE ChIP-Seq data reveals over 60 transcriptional regulators that can associate with the DDIT4 promoter. However, we can speculate about likely candidates, of which there are several based on known association with PML and PML NBs in the literature. These include SP1, EP300, JUN/ FOS, HIF1α and p53, to name just a few that might regulate DDIT4 basal and inducible-expression following cellular stress in a PML-or PML NB-dependent manner. For example, p53 and SP1 can associate with PML NBs and are known to regulate DDIT4 expression 15,23,27,53 and could therefore contribute to the basal regulation of DDIT4.
In our studies analyzing DDIT4 expression in response to metformin (Fig. 6) and IR (Fig. S2), the induction of DDIT4 did not appear to be PML-dependent. However, the maximal level of DDIT4 protein achieved in PML-depleted U2OS cells following these stresses was reduced relative to treated control cells. PML NBs play a role in p53 activation, and DDIT4 induction in response to metformin 26   silencing reduces basal DDIT4 expression 26 , we speculate that the basal regulation of DDIT4 expression by p53 may be impaired in PML-depleted cells, and that this then leads to an attenuated DDIT4 response to metforminand IR-induced stress. Finally, our study provides a new context in which to view changes in PML expression during cancer development and progression. PML is an important tumor suppressor whose expression is frequently reduced or lost in several cancers from diverse tissue types such as lung, prostate, colon and breast 1 . One previously uncharacterized consequence of the loss of PML, uncovered in this study, is a reduction of DDIT4 expression (Fig. 1). Our analysis of the TCGA cancer transcriptome datasets available through the cBio Portal (http://www.cbioportal. org/), revealed a positive correlation between PML and DDIT4 expression in several cancers, including breast, prostate, ovarian and lung cancers (Fig. 7). Based on our results in U2OS cells and our analysis of cancer transcriptomes, it is likely that reduced PML expression observed in these cancers is correlated with reduced DDIT4 expression, which in turn could promote mTOR activation and promote cancer cell growth. Thus, in addition to PML NB-mediated nuclear retention of mTOR and inactivation of AKT, we have uncovered a novel mechanism by which PML NB loss contributes to mTOR activation in cancer via dysregulation of DDIT4 gene expression.

Materials and Methods
Reagents and drug/chemical treatments. All chemicals were purchased from Sigma Aldrich unless otherwise indicated. Metformin and CoCl 2 were dissolved in growth media and added to cells for 24 h unless otherwise indicated in the figure legends. For Western Blotting, CoCl 2 -treated cells were washed with PBS containing 500 μ M CoCl 2 to prevent reversal of drug response prior to cell lysis. Cycloheximide was added to growth media at 50 μ g/ml.
AlamarBlue Viability Assay. Cell viability and metformin sensitivity of shControl and shPML U2OS cells was measured using the AlamarBlue assay (Life Technologies Ltd., cat#DAL1100) following the manufactures instructions. Briefly, 1500 cells were seeded in 96-well plates and grown for 24, 48, and 72 h. AlamarBlue reagent was added to 4 wells per condition at each time point, incubated for 2-4 h at 37 °C and the amount of dye reduction was quantified by measuring the absorbance at 590 nm using an Infinite 200 Pro microplate reader (Tecan US Inc., Morrisville, NC). Metformin (1 mM, 5 mM, 25 mM) was added to cells 24 h after plating and AlamarBlue was added for 4 h at 48 h post-metformin treatment.
Translation State Array Analysis. Polysome profiling was performed essentially as previously described 36 .
Cytoplasmic extracts containing polysomes from shControl and shPML U2OS cells were loaded onto a 10-50% sucrose gradient and polysomes were separated by ultracentrifugation in a Beckman SW41 rotor at 36 000 rpm for 1 h 50 min, 4 °C. The sucrose gradients were fractionated and absorbance at 254 nm was determined. RNA was harvested from fractions containing the high molecular weight polysomes and the isolated RNA was pooled and concentrated by ethanol precipitation for microarray analysis. Total cellular RNA was also harvested in parallel for microarray analysis.
The quality of RNA samples was assessed using the Experion Bioanalyzer system (Bio-Rad) and had RNA Quality Index values greater than 9. Each RNA sample (1 μ g) was amplified using the Amino Allyl MessageAmp II aRNA amplification kit and labeled with AlexaFluor 555 or 647 (Life Technologies). Quantity and quality of amplified aRNA was assessed using a Nanodrop spectrophotometer and the Experion Bioanalyzer. Samples were compared in a dye swap experiment, with 3 μ g of each labeled, fragmented aRNA (6 μ g total per slide) hybridized to proprietary human cDNA microarray slides. These arrays consist of roughly 35 000 spots, representing roughly 17 000 different 50-mer oligonucleotides spotted in duplicate on Nexterion-E epoxy microarray slides (Schott).
Hybridizations were performed in Ambion SlideHyb #2 buffer (Life Technologies) at 42 °C for 16 h using the automated TECAN-4800 Hybridization station (TECAN). Hybridized slides were scanned at 10 μ m resolution using an Axon GenePix 4200AL scanner (Molecular Devices, Sunnyvale, CA) and gridded using SpotReader (Niles Scientific). Calibration of spot rejection was done by visual inspection of gridded images and scatter plots of M = log(532/635) versus A = log(532) + log(635), with special attention being paid to outliers. A GPR file was generated and processed to flag spots with a signal to noise ratio < 5.
Analysis was performed with Acuity 4.0 (Axon Instruments). The data were normalized using Lowess and the relative mRNA abundances for DDIT4 were determined for 3 independent experiments. Student's t-tests were used to calculate p-values.
Fluorescent images were acquired on a customized Zeiss Cell Observer inverted microscope (Intelligent Imaging Innovations) equipped with a CoolSNAP HQ2 CCD camera (Photometrics) using a 1.4 NA 63 × oil-immersion lens and captured using Slidebook 5.1 software (Intelligent Imaging Innovations). Image stacks were exported to Imaris 7.2 software (Bitplane) for further analysis. Distances between PML NBs and the FISH probes were determined in 3D after minimal threshold adjustments to define the Spectrum Orange labelled probes and PML NB immuno-signal followed by semi-automated image processing, which included manual validation that detected centroid positions were accurately defined to PML NBs and most intense FISH probe signal. The Euclidian distance between the centroid of fluorescence of defined FISH probe signals and closest PML body were then calculated for each DAPI signal-defined nucleus and exported to Excel (Microsoft) for analysis. At least 50 nuclei were analyzed for each FISH probe to generate mean distances between a PML NBs and a given FISH signal for each FISH probe.

RNA-immunoFISH.
RNA FISH probe generation -PCR reactions to clone DNA fragments were generated using the following primers pairs: DDIT4 (forward 5′ -CTCCGGCACTCTGAGTTCATC-3′ and reverse 5′ -GACACCCACCCCTTCCTACTC-3′ ) and BCL2 (forward 5′ -CTTCAACGACGGTTTATTTGCTGTC-3′ and reverse 5′ -TATGCTGCAGATTGAAAAGTGCAAA-3′ ). Amplicons were gel purified and cloned using a pGEM T-easy cloning kit (Promega), sequenced and subsequently used for RNA probe generation. Briefly, 10 μ g of linearized probe DNA was used for in vitro transcription with T7 RNA polymerase (New England Biolabs) following manufacturer's instructions. To generate single stranded cDNA probes for each gene, 5 μ g of purified RNAs were mixed with 1.5 μ l of 5 μ g/μ l random hexamers. RNAs were denatured at 65 °C for 5 min and cooled immediately on ice to bind hexamers. After cooling, 4 μ l of 5X first strand buffer, 2 μ l 100 mM DDT and 7.5 μ l of dNTP mix (0.7 mM biotin-16-dUTP, 1.3 mM dTTP, 2 mM dATP, 2 mM dCTP, 2 mM dGTP) were added. The reaction was heated to 42 °C for 2 min before the addition of 400 U of Superscript III (Invitrogen). The reaction mix was incubated at 42 °C for 90 min. RNA was removed from the reaction using 2 μ l 4 M NaOH and incubation at 37 °C for 30 min. The reaction was neutralized using 5 μ l 1 M Tris-HCl, pH 7.5 Probe was precipitated using ethanol sodium acetate precipitation.

RNA-Sequencing.
For RNA sequencing and differential expression analysis, shControl and shPML U2OS cells were grown to 80% confluence in 10 cm dishes and prepared for Illumina HiSeq sequencing as previously described 58 . Sequencing reads (GEO accession #GSE94126) were mapped using TopHat (version 2.0.0) to the known transcriptome (-G option; Gencode version 7 annotations) for the human reference genome (hg19) using default parameters 59 . Cufflinks (version 2.0.2) was used to quantify gene and isoform expression for known transcripts (-G option; Gencode version 7 annotations) using the default parameters 60 . Only uniquely aligned reads were used for expression quantification. Pairwise differential gene expression analysis was performed using Cuffdiff 61 with the default parameters.