Attenuation of nonsense-mediated mRNA decay facilitates the response to chemotherapeutics

Nonsense-mediated mRNA decay (NMD) limits the production of aberrant mRNAs containing a premature termination codon and also controls the levels of endogenous transcripts. Here we show that when human cells are treated with clinically used chemotherapeutic compounds, NMD activity declines partly as a result of the proteolytic production of a dominant-interfering form of the key NMD factor UPF1. Production of cleaved UPF1 functions to upregulate genes involved in the response to apoptotic stresses. The biological consequence is the promotion of cell death. Combined exposure of cells to a small molecule inhibitor of NMD, NMDI-1, and the chemotherapeutic doxorubicin leads to enhanced cell death, while inhibiting UPF1 cleavage protects cells from doxorubicin challenge. We propose a model to explain why the expression levels of genes producing mRNAs of diverse structure that encode proteins of diverse function are under the purview of NMD.


INTRODUCTION
An estimated ~one-third of inherited diseases are the result of premature termination codon (PTC) acquisition 1 . Nonsense-mediated mRNA decay (NMD) is a conserved mRNA quality control pathway deployed by cells to eliminate mRNAs containing a PTC. Because proteins damage, regulated cell death or apoptosis ensues 28 . How cells that are exposed to apoptotic insults deal with apoptotic stresses has only been recently studied 29 . Clearly, there is dynamic resculpting of the proteome and the transcriptome: a transcriptional response (e.g. by the p53 network) precedes the loss of cellular metabolism. Concomitant with a decrease in cellular viability is a global decrease in mRNA levels 30,31 . Superimposed on these changes are proteolytic events, carried out by caspase enzymes that actively promote apoptosis and dismantle the cell 32,33 .
Here, we examine how NMD is integrated into the network of processes that define the apoptotic response. We find that NMD is inhibited during apoptosis, in part by the proteolytic production of a dominant-interfering form of UPF1. Inhibiting UPF1 cleavage protects cells from the effects of doxorubicin. Conversely, decreasing the efficiency of NMD using a small molecule inhibitor sensitizes cells to doxorubicin. We propose that the efficiency of NMD can be tuned by extracellular stimuli, and one purpose for the NMDmediated control of endogenous gene expression is to assist in the establishment of a particular state by tailoring the mRNA milieu to one that can respond to potentially diverse stimuli.

NMD activity is blunted during doxorubicin treatment
We examined the stability of a panel of known NMD target mRNAs 34 in human MCF7 breast cancer cells during doxorubicin treatment. Pre-treatment with doxorubicin (5 μM) resulted in significant increases in the half-lives of PANK2, TSTD2, and NAT9 mRNAs but not β-actin mRNA after actinomycin D-mediated transcriptional arrest (Fig. 1a), indicating a decline in NMD activity. To support this, we measured the level of each mRNA relative to the level of the pre-mRNA from which it derives as a function of time after doxorubicin treatment to control for transcriptional effects. An increase in the mRNA/pre-mRNA ratio (a metric used to distinguish a subset of direct NMD targets from those that are not in UPF1ablated HeLa cells 35,36 ) may not reliably distinguish NMD targets from those that are not during doxorubicin treatment, because global transcriptional shut-down may inflate this number. However, a decrease in this ratio would rule out the possibility that NMD activity is blunted. Consistent with our half-life data, the mRNA/pre-mRNA ratio, as assessed using RT-quantitative PCR (qPCR), increases for all three transcripts (none of which is known to be stress-regulated), in response to doxorubicin ( Supplementary Fig. 1a). The mRNA/pre-mRNA ratios for three additional known NMD-targeted transcripts [34][35][36][37][38] , CDKN1A, GADD45α, and GADD45β, were also significantly increased by 5 hours (h) of doxorubicin treatment. As in HeLa cells, the ratio of CDKN1A, GADD45α, and GADD45β mRNAs to their corresponding pre-mRNAs is elevated upon UPF1 depletion in MCF7 cells, indicating that these mRNAs are indeed NMD targets in MCF7 cells ( Supplementary Fig. 1b). Decreases in pre-mRNA levels cannot account for the increased mRNA/pre-mRNA ratio since, even in the most extreme example (CDKN1A RNA), at 5 h the mRNA/pre-mRNA ratio increased ~5.7 fold relative to 0 h, while the pre-mRNA level decreased only ~3.2 fold.
To further corroborate the inhibition of NMD during doxorubicin treatment, we transfected MCF7 cells with the previously described β-globin (β-Gl) NMD reporter plasmids 39 encoding either β-Gl Norm transcripts that lack a PTC, or β-Gl Ter transcripts that harbor a PTC at position 39. Cells were cotransfected with a plasmid encoding the mouse urinary protein (MUP) transcript to control for variations in transfection efficiency and RNA recovery and, 24 h later, were exposed to doxorubicin (5 μM). By 5 h of doxorubicin treatment, the level of β-Gl Ter mRNA increased from ~65% to ~85% the level of β-Gl Norm mRNA. These measurements occur on the backdrop of global RNA degradation at later time points 29,30 , accounting for why the normalized ratio of β-Gl Ter mRNA to β-Gl Norm mRNA is not elevated at later time points.
We performed additional mRNA decay assays using a previously described HeLa Tet-off cell system 40 to halt the synthesis of human β-Gl Norm mRNA or β-Gl Ter mRNA and subsequently measured the remaining levels of each mRNA relative to the level of MUP mRNA after doxycycline addition. We used this system because the Tet-off promoter that controls the production of β-Gl Norm mRNA or β-Gl Ter mRNA is not stress-responsive and HeLa cells, like MCF7 cells, are devoid of erythroid cell-specific β-Gl mRNA. Without doxorubicin, the level of β-Gl Ter mRNA declined to ~50% of its starting level by ~180 minutes of doxycycline addition, in agreement with reported values 40 , while the level of β-Gl Norm mRNA did not decrease during this time ( Fig. 1b; top). In contrast, pretreatment of cells with doxorubicin for 1 h before doxycycline addition eliminated the selective decay of β-Gl Ter mRNA; the half-lives of both β-Gl Norm mRNA and β-Gl Ter mRNA exceeded the chase period ( Fig. 1b; middle). Doxorubicin treatment mirrored the effect of the translational inhibitor puromycin, which is known to inhibit NMD ( Fig. 1b; bottom). From all results, we conclude that NMD activity is attenuated during doxorubicin treatment.
We next examined biochemical changes to the key NMD factor, UPF1, that correlate with doxorubicin treatment (Fig. 1c). We exposed MCF7 cells to doxorubicin (5 μM) for varying amounts of time and analyzed cell lysates using western blotting and an in-house generated polyclonal rabbit serum raised against the N-terminal 416 amino acids of human UPF1. To eliminate post-lysis proteolysis, lysates were generated in the presence of a protease inhibitor cocktail supplemented with N-ethylmaleimide at levels (50 μg/mL) known to alkylate the most active viral cysteine proteases. In addition to full-length UPF1, two additional bands of greater mobility were resolved by 5 h of doxorubicin treatment (Fig. 1c). Phosphorylation of UPF1 at both its N-and C-termini is a key feature that differentiates UPF1-bound NMD targets destined for degradation from those that are not 15,41,42 . Western blotting using a monoclonal antibody recognizing phosphorylated S1116 revealed that UPF1 phosphorylation levels diminish by 5 h. Both of these changes to UPF1 preceded maximal cleavage of poly (ADP-ribose) polymerase (PARP), a well-characterized biochemical marker for apoptosis. Increasing doxorubicin concentrations ten-fold (50 μM) to accelerate apoptotic progression generated higher-mobility UPF1 species by 2 h, i.e. well before production of the PARP cleavage product at 8 h (Fig. 1d). Thus, the generation of fastermigrating UPF1 species, which we characterize as cleavage products (CPs; see below), and the reduction of UPF1 phosphorylation occur early during apoptotic progression (see below).

Multiple apoptotic insults cause UPF1 hydrolysis
We characterized the upper UPF1 CP because we observed that it was consistently generated by an array of treatments in many cell lines (see below). We verified that this band derives from cellular UPF1 rather than a protein that fortuitously cross-reacts with our polyclonal anti-UPF1 serum by using siRNA to reduce the level of UPF1 in human cervical carcinoma HeLa cells to <10% of normal and subsequently exposing cells to cycloheximide (CHX) to induce apoptosis (Fig. 2a). In addition to halting protein synthesis, CHX causes apoptosis via incompletely understood mechanisms 43 . siRNA treatment reduced the levels of both full-length UPF1 and the UPF1 CP. Because new protein synthesis is halted by CHX, the UPF1 CP is unlikely to be a UPF1 isoform explained by the hypothetical possibility that alternative splicing of UPF1 pre-mRNA is induced during apoptosis.
Cleavage of proteins by caspases, a class of cysteine proteases, during apoptosis is a common event 32,33 . "Bystander" cuts to proteins fortuitously encoding a caspase cleavage site may occur during apoptosis, but cleavage early during apoptotic progression and cleavage conservation across species indicate functional relevance 33 .
To examine the timing of UPF1 CP generation, we treated HeLa cells with the clinically used topoisomerase inhibitor etoposide (ETP). ETP induced generation of a UPF1 CP before full induction of cleaved initiator caspase 9 (CASP9) and cleaved executioner CASP3 (Fig.  2b). Generation of the UPF1 CP prior to full CASP9 and CASP3 cleavage is recapitulated in human embryonic kidney (HEK)293T cells during CHX treatment ( Supplementary Fig. 2a). We examined the effects of other apoptotic inducers on UPF1 CP generation. Treatment of HEK293T cells with staurosporine also yielded two UPF1 CPs, prior to full cleavage of CASP9 and CASP3 ( Supplementary Fig. 2b). Exposure of the human Daudi B-lymphoblast cell line to either tumor necrosis factor-α (TNF-α) or doxorubicin led to generation of the UPF1 CP prior to maximal PARP cleavage ( Supplementary Fig. 2c). Staurosporinechallenged Jurkat T-cells also yielded a UPF1 CP prior to maximal cleavage of CASP3 or PARP ( Supplementary Fig. 2d).
To probe whether generation of the UPF1 CP is evolutionarily conserved, we exposed mouse C2C12 myoblasts to CHX or ETP, both of which generated a UPF1 CP prior to maximal CASP3 cleavage (Fig. 2c). Exposure of canine (MDCK), bovine (MDBK), and Chinese hamster (CHO) cells to staurosporine led to UPF1 CP production (Supplementary Fig. 2e-g). Likewise, exposure of African Green monkey (COS-7) cells to staurosporine or doxorubicin yielded a UPF1 CP ( Supplementary Fig. 2h). UPF1 CP levels varied drastically across cell lines, likely for three reasons: (i) in non-human cells, we cannot assess how efficiently our treatments elicited apoptosis because antibodies to human PARP, cleaved human CASP9, and cleaved human CASP3 do not cross-react; (ii) our anti-UPF1 antiserum was raised against the first 416 amino acids of human UPF1 and may exhibit reduced crossreactivity to non-human UPF1 CP; and (iii) as a result of cleavage at the N-terminus (see below), the human UPF1 CP exhibits less than one-third the immunoreactivity of full-length human UPF1 with this UPF1 antiserum. Notwithstanding this, UPF1 CP generation is an early event that is evolutionarily conserved, indicating that UPF1 cleavage may play a role in the cellular response to apoptotic induction.

Mapping UPF1 hydrolysis
To probe whether caspases are involved in UPF1 CP generation, we pre-incubated HEK293T cells with a panel of caspase inhibitors followed by exposure to CHX (Fig. 3a). Cells treated with each caspase inhibitor showed drastically reduced UPF1 CP levels, with Z-DEVD-fmk and Z-VAD-fmk lowering the level of UPF1 CP to nearly undetectable. Thus, caspases, and/or alternative proteases 44 activated downstream of caspases, are involved in UPF1 CP production. We exposed a HeLa cell line stably expressing N-terminally tagged FLAG-UPF1 45 to CHX. While anti-FLAG immunoblots failed to reveal any UPF1 CP even after long exposure, anti-UPF1 immunoblots using antiserum raised against amino acids 1-416 (Fig. 3b) yielded detectible UPF1 CP (Fig. 3c), indicating that cleavage occurs within the first 416 amino acids of UPF1 so as to eliminate the FLAG epitope but preserve partial immunoreactivity with the UPF1 antiserum.
Inventories of in vivo apoptotic cleavage events indicate that cleavage specificity in living cells is determined chiefly by an aspartic acid residue at the P1 position; P4-P2 residues contribute far less to specificity in cells than is indicated by in vitro-derived peptide-based substrate profiles 32,33 . Accordingly, we focused our attention solely on aspartic acid (D) residues in human UPF1 and interrogated residues D27, D37, and D75 near the UPF1 Nterminus by mutating each to asparagine (N). Full-length wild-type (WT) UPF1 and, separately, each variant was expressed bearing an N-terminal MYC-tag and a C-terminal FLAG-tag in HeLa cells at a level equal to endogenous UPF1, and cells were subsequently challenged with CHX. For UPF1 WT, UPF1 D27N and UPF1 D75N, the UPF1 CP was generated at ~one-third the level of uncleaved UPF1, as judged using an anti-FLAG immunoblot ( Supplementary Fig. 3a). Both the UPF1 CP and uncleaved UPF1 retained the C-terminal FLAG tag, allowing unambiguous assessment of the ratio of UPF1 CP to fulllength UPF1. UPF1 D37N yielded no UPF1 CP, indicating that the amide bond after D37 is the site of hydrolysis (i.e. D37 is the P1 residue).
We generated HeLa cells stably expressing one copy of retrovirally introduced MYC-UPF1-FLAG WT or MYC-UPF1-FLAG D37N transgene. Each protein was expressed at ~2.7 fold the level of endogenous UPF1 (Fig. 3d). In these cell lines, the D37N mutation abolished UPF1 CP generation in response to CHX and doxorubicin (Fig. 3d). MCF7 cells stably transduced with MYC-UPF1-FLAG WT also generated the UPF1 CP at ~one-third the level of uncleaved UPF1 in response to doxorubicin, and the UPF1 CP matched the molecular weight of a UPF1 fragment encompassing residues 38-1118 ( Supplementary Fig. 3b). We cannot detect the N-terminal 37 amino acid fragment released upon cleavage, likely either for technical reasons or because this fragment is unstable. Having established that one cleavage event occurs at the after position 37, we examined the conservation of surrounding amino acids by aligning UPF1 sequences from multiple species using ClustalX ( Supplementary Fig. 3c). The putative consensus cleavage site EFTD is completely conserved in human, bovine, mouse and Xenopus laevis UPF1-it deviates in chicken UPF1 at a single amino acid (where D is G) -and harbors T at the P2 residue, consistent with the high frequency of S and T residues at P4, P3, and P2 residues in cellular apoptotic protein cleavage sites 32 . Previously confirmed caspase substrates also bear similar cleavage sites: protein kinase C ζ is cleaved after EETD 46 , and the NF-kB p65/RelA subunit is cleaved after VFTD 47 .
We characterized which caspase(s) are sufficient to cleave UPF1 in vitro by treating immunoprecipitated samples of full-length MYC-UPF1-FLAG WT or the non-cleavable MYC-UPF1-FLAG D37N variant with recombinant caspases (Supplementary Fig. 3d). CASP3 and CASP7 cleaved MYC-UPF1-FLAG WT but not MYC-UPF1-FLAG D37N into a fragment with the same molecular weight as a Δ37-UPF1-FLAG variant lacking the Nterminal MYC-tag and first 37 residues of MYC-UPF1-FLAG WT (recapitulating the mapped UPF1 CP). This is consistent with our observation that Z-DEVD-fmk and Z-VADfmk blunt UPF1 CP production (Fig. 3a).

UPF1 CP is not functional in NMD
What might cleavage at D37 in human UPF1 accomplish? Both serine 10 (S10) and threonine 28 (T28) are phosphorylated by the NMD-associated kinase SMG1 18 , and phosphorylation is critical for NMD 7,18,19,41,42 . Cleavage would cause a loss of these phosphorylation sites and, indeed, experimental truncation of the first 35 amino acids in Arabadopsis thaliana UPF1 (causing loss of three phosphorylation sites) eliminates its NMD activity and causes it to act dominant negatively 48 . A previously described deletion of the N-terminal 63 amino acids of human UPF1 (dNT) causes loss of NMD activity and dominant negative behavior, as does mutation of the threonine 28 phosphorylation site to alanine 15,49 .
We assayed the NMD activity of exogenously expressed UPF1 proteins without endogenous UPF1. We depleted endogenous UPF1 levels in HEK293T cells to <10% of normal using siRNA and subsequently transiently introduced one of several siRNA-resistant UPF1 expression vectors: MYC-UPF1-FLAG WT; MYC-UPF1-FLAG D37N; Δ37-UPF1-FLAG; MYC-UPF1-FLAG TEV (described later); MYC-UPF1 dNT 15 ; or MYC-UPF1 R843C, which abolishes UPF1 helicase activity 50 . Transfections included either a "Norm" or a "Ter" plasmid set to assess NMD activity. The "Norm" set consists of the β-Gl Norm reporter plasmid, the MUP reference plasmid, and a T-cell receptor (TCR)β-based reporter plasmid. This TCRβ-based reporter plasmid contains a bidirectional promoter driving synthesis of an HA-Cerulean fluorescent protein and, in the opposite orientation, a 3X FLAG-mCherry fluorescent protein whose transcript contains a 3′ UTR composed of a TCRβ minigene lacking introns (Δ JC intron) 51 (Fig. 4a). The "Ter" plasmid set contains the β-Gl Ter reporter plasmid, the MUP reference plasmid, and a TCRβ reporter plasmid bearing (+) the JC intron >55nt downstream of the mCherry termination codon 51 rendering the mCherry transcript an EJC-mediated NMD substrate (Fig. 4a). Each variant was expressed at a level equivalent to endogenous UPF1 as assessed by comparing anti-UPF1, anti-MYC, and anti-FLAG immunoblots (Fig. 4b).

UPF1 CP is a dominant-interfering protein
Could the UPF1 CP play a dominant-interfering role in suppressing NMD even at substoichiometric levels relative to uncleaved UPF1? We challenged HeLa-cell UPF1 function by introducing increasing amounts of plasmid DNA to express increasing but substoichiometric amounts of Δ37-UPF1-FLAG or, as a control, MYC-UPF1-FLAG WT; in parallel, we introduced empty vector DNA (ϴ) as an additional control (Fig. 5a). These transfections included the "Norm" or "Ter" plasmid sets. While the flexible linker and FLAG epitope of Δ37-UPF1-FLAG limit its complete resolution from endogenous UPF1 (Fig. 5a), the level of Δ37-UPF1-FLAG can be compared to the level of MYC-UPF1-FLAG WT in anti-FLAG blots, and since MYC-UPF1-FLAG WT is cleanly resolved from endogenous UPF1 in anti-UPF1(1-416) blots, it is possible to determine the levels of Δ37-UPF1-FLAG relative to endogenous UPF1 (Fig. 5a). Levels of the β-Gl Ter NMD substrate revealed that, relative to transfections employing empty vector, increasing amounts of the Δ37-UPF1-FLAG elicited an increase in the level of β-Gl Ter mRNA (Fig. 5b). Δ37-UPF1-FLAG expression at ~one-third the level of endogenous UPF1 (Fig. 5a) yielded a ~2.4 fold higher β-Gl Ter mRNA level than in transfections employing empty vector (Fig. 5b, red arrow). Δ37-UPF1-FLAG expression increased the level of the NMD substrate mCherry-TCRβ +JC intron mRNA ~2.3 fold ( Fig. 5c; in the fourth sample, the decrease to ~1.5 fold and large error bars are likely due to experimental noise since levels of β-Gl Ter mRNA continue to increase with increasing Δ37-UPF1-FLAG levels). The amount of 3xFLAG-mCherry that derived from mCHERRY-TCRβ +JC intron mRNA increased with increasing yet substoichiometric amounts of Δ37-UPF1-FLAG (Fig. 5a). Almost no changes were observed in the level of mCherry-TCRβ +JC intron mRNA or its product protein when endogenous UPF1 was challenged with increasing but substoichiometric levels of MYC-UPF1-FLAG WT (Fig. 5a,c). We confirmed that challenge of endogenous UPF1 in HEK293T cells with increasing amounts of Δ37-UPF1-FLAG, relative to empty vector control, increased β-Gl Ter mRNA levels, whereas MYC-UPF1-FLAG WT had no such effect ( Supplementary Fig. 4a,b).
We sought to rule out the trivial explanation for the lack of Δ37-UPF1-FLAG function in NMD, i.e. that the truncated protein is misfolded, by characterizing the composition of the RNP containing either MYC-UPF1-FLAG WT or Δ37-UPF1-FLAG. HEK293T cells were depleted of endogenous UPF1 using siRNA, and either MYC-UPF1-FLAG WT or Δ37-UPF1-FLAG was expressed at a level equivalent to the normal level of endogenous UPF1 (Fig. 6). MYC-UPF1-FLAG WT or Δ37-UPF1-FLAG complexes were immunoprecipitated from lysates using anti-FLAG resin, each immunoprecipitate was divided in half, and one half was incubated with BSA while the other half was incubated with RNase ONE to identify protein-protein interactions that are stabilized by RNA.
Immunoblotting using antibodies directed against p-S1089 or p-S1116 in UPF1 revealed slightly enhanced phosphorylation of Δ37-UPF1-FLAG relative to MYC-UPF1-FLAG WT (Fig. 6). Accumulation of C-terminal phosphates is a feature of ATPase-deficient UPF1 variants that cannot support NMD 7,15,21,42 . Equivalent levels of EJC components UPF2, UPF3X and MLN51, the cap-binding protein CBP80, and the poly(A) binding protein (PABP)C1 were co-immunoprecipitated with both UPF1 variants (Fig. 6). Δ37-UPF1-FLAG retrieved slightly increased levels of SMG5 and SMG7 relative to MYC-UPF1-FLAG WT in RNase-insensitive interactions (Fig. 6). Equivalent levels of SMG6 were retrieved in a partially RNase-sensitive interaction (Fig. 6). SMG6 association with a region outside of the UPF1 N-terminus is consistent with several recent reports 7,52,53 . Our results indicate that gross misfolding of the UPF1 CP cannot explain its nonfunctional and dominant-interfering behavior.
We characterized the binding of MYC-UPF1-FLAG WT, Δ37-UPF1-FLAG, and MYC-UPF1-FLAG D37N to PTC-containing mRNAs relative to their PTC-free counterparts by transfecting cells expressing equivalent levels of each UPF1 variant with a combination of plasmids encoding β-Gl Ter mRNA and MUP mRNA, or separately, plasmids encoding the β-Gl Norm mRNA and MUP mRNA. We measured the binding of each variant to β-Gl Ter mRNA and to its PTC-free counterpart ( Supplementary Fig. 5) in immunoprecipitates. As previously reported, MYC-UPF1-FLAG WT retrieved ~26-fold higher levels of the PTCcontaining mRNA when adjusted for expression levels 7 and MYC-UPF1-FLAG D37N did likewise. Like the nonfunctional dNT UPF1 variant as well as a non-functional 4SA variant lacking four phosphorylation sites 7 , Δ37-UPF1-FLAG also retrieved β-Gl Ter mRNA relative to β-Gl Norm mRNA with an efficiency that was comparable to that of MYC-UPF1-FLAG WT and MYC-UPF1-FLAG D37N (~36-fold enrichment), despite being nonfunctional (Figs. 4,5). We conclude that, like the dNT variant 7 , Δ37-UPF1-FLAG is not misfolded-it can bind to the same complement of proteins as wild-type UPF1 and is enriched on a PTC-bearing transcript. Rather, a defect in the NMD cycle after RNA binding occurs.

UPF1 cleavage upregulates genes involved in apoptosis
What is the physiological relevance of UPF1 cleavage and the attenuation of NMD for cells exposed to chemotherapeutics that cause apoptosis? We interrogated the 91 genes upregulated upon UPF1 downregulation in Mendell et al. 3 using the online DAVID gene ontology tool to cluster genes by function. We found a cluster (11 genes) under "positive regulation of programmed cell death" (p=2.3E-4) (Supplementary Data 1) as well as a group of genes belonging to "p53 signaling pathway" (4 genes) and "regulation of cell cycle" (5 genes). Results from Viegas et al. 36 showed a cluster (20 genes) under "positive regulation of programmed cell death" (p=7.4E-8) (Supplementary Data 1). DAVID analysis of results from Cho et al. 4 also yielded genes in "positive regulation of programmed cell death" (10 genes) and regulation of cell cycle (5 genes). Indeed, NMD targets that we previously analyzed ( Supplementary Fig. 1a) include CDKN1A mRNA, which encodes the classical cell-cycle inhibitory protein p21, and GADD45α and GADD45β mRNAs, which produce proteins involved in cell-cycle arrest that also transiently upregulate CASP3 and CASP7 to promote apoptosis 54 . Thus, we hypothesized that, among the transcripts upregulated upon NMD attenuation are a group that the cell can exploit in response to apoptotic inducers.
We generated stably transduced HeLa-cell lines bearing either empty vector or a fully functional MYC-UPF1-FLAG TEV allele that harbors the tobacco etch virus (TEV) protease cleavage site substituted into the D37 position (Fig. 4). Since TEV protease has no cleavage sites in the mammalian proteome 55 , transfection with a plasmid encoding two complementary MYC-tagged TEV protease fragments expressed from a bidirectional promoter allows MYC-UPF1-FLAG TEV to be specifically cleaved in living cells in the absence of apoptotic inducers (Fig. 7a). To identify changes in cellular mRNAs that are due UPF1 CP production, we performed RNA-Seq on the empty-vector cell line and the MYC-UPF1-FLAG TEV cell line, both in the presence or absence of TEV expression, with the assumption that we would uncover direct and indirect NMD targets.
We note that abundance changes in either class of targets may have important effects on cellular physiology. To control for differences in the cellular responses to plasmid identity and transfection, we normalized changes in mRNA abundance of the MYC-UPF1-FLAG TEV cell line with and without TEV protease to changes in the empty vector cell line with and without TEV protease. We recovered upregulation of mRNAs for CDKN1A (~3-fold), GADD45α (~3.7-fold) and GADD45β (4.7-fold) (Supplementary Data 2). By expressing increasing amounts of MYC-UPF1-FLAG WT or Δ37-UPF1-FLAG in either HeLa (Fig. 7b) or MCF7 (Fig. 7d) cells and measuring the resultant changes in mRNA abundance (Fig. 7c,  Fig. 7e), we verified the upregulation of a subset of additional genes, each of which has individual literature-documented roles in promoting cell-cycle arrest or apoptosis when its expression is increased (Table 1). While we cannot explain the decrease observed for most mRNA ratios at the highest level of Δ37-UPF1-FLAG in HeLa cells (but not in MCF7 cells), we note that this expression level is greater than that observed for the CP in doxorubicin-treated cells.
To support these observations, we transfected HeLa cells with substoichiometric amounts of either MYC-UPF1-FLAG WT or an equivalent amount of Δ37-UPF1-FLAG ( Supplementary Fig. 6), treated cells 48 h later with 5,6-dichloro-1-β-D-ribofuranosyl-1Hbenzimidazole (DRB), and analyzed the half-lives of endogenous NMD targets after DRBmediated transcriptional arrest 10 . In contrast to GAPDH and β-actin mRNAs, there were noted increases in the stability of GAS5 ncRNA 56 as well as GADD45α, BAK1, and BCL3 mRNAs. Stability of ARF1 and SERPINE1 mRNAs (two SMD targets) were unaffected ( Supplementary Fig. 6). Thus, UPF1 CP can partially attenuate NMD levels at substoichiometric amounts. Each of the genes verified from the RNA-seq data can individually promote either cell cycle arrest or apoptosis (Table 1) and thus may be exploited by cells in response to chemotherapeutic treatment.

Modulation of NMD activity affects doxorubicin sensitivity
Two testable hypotheses follow from the observation that generating UPF1 CP in the absence of chemotherapy augments the expression of genes involved in apoptotic progression. First, inhibiting UPF1 CP production should slow the cell-death response to doxorubicin. Second, inhibiting NMD through exogenous introduction of UPF1 CP or small-molecule treatment should promote doxorubicin-mediated cell death.
To test the first hypothesis, we utilized HeLa cell lines stably expressing MYC-UPF1-FLAG WT or non-cleavable MYC-UPF1-FLAG D37N (Figs. 3d, 4b, 8a), both of which support NMD. Each was expressed at ~2.7 fold above the level of endogenous UPF1 and, more importantly, at levels identical to one another (Fig. 8a). We exposed these cell lines to a range of doxorubicin concentrations and assessed cell viability after 16 h using an assay that detects ATP generation by living cells. At a sub-lethal doxorubicin concentration (0.5 μM), no statistically significant difference in viability was detected. However, as doxorubicin toxicity increased, the MYC-UPF1-FLAG D37N cell line showed increased resistance to death relative to the MYC-UPF1-FLAG WT cell line, reaching a maximum of ~2.2-fold greater survival.
With the first hypothesis verified, we next transiently expressed either MYC-UPF1-FLAG WT or Δ37-UPF1-FLAG in HeLa cells ( Supplementary Fig. 7a) or MCF-7 cells (Supplementary Fig. 7c) and challenged transfectants with doxorubicin. Although we observed statistically significant increases in sensitivity for Δ37-UPF1-FLAG transfectants relative to MYC-UPF1-FLAG WT transfectants, the effect was mild in both cell types ( Supplementary Fig. 7b, d), likely because the toxic effects of lipofection obscure differences between the two transfected cell populations and limit the dynamic range of the assay.
Therefore, we utilized a small-molecule inhibitor of NMD, NMDI-1 (Fig. 8c), that interferes with the interaction between UPF1 and SMG5 57,58 . Application of NMDI-1 and the consequential attenuation of NMD may more fully replicate the complete inhibition of NMD mediated by the UPF1 CP generation explored here as well as UPF1 dephosphorylation and the generation of additional UPF1 CPs seen with doxorubicin (Fig. 1c). NMDI-1 is effective in HeLa cells, raising the levels of β-Gl Ter mRNA and another NMD target -a PTCbearing glutathione peroxidase 1(GPx1) mRNA -~3.8-fold and ~1.6 fold, respectively, at a concentration of 10 μM, which was used in subsequent experiments ( Supplementary Fig.  8a,b).
Since NMDI-1 had no effect in MCF7 cells ( Supplementary Fig. 8c, d), we focused on HeLa cells. Because of results indicating that the efficacy of combination small-molecule treatments is affected by both drug order and timing 59 , we considered three treatment regimens. First, we challenged HeLa cells with various concentrations of doxorubicin alone. Second, we continuously co-incubated cells for 16 h with doxorubicin and NMDI-1. Third, we applied a transient pulse of NMDI-1 for 8 h, washed cells to remove NMDI-1, and then applied doxorubicin. At sub-lethal doses of doxorubicin (0.5 μM), none of the treatments significantly affected viability (Fig. 8d), in accordance with previous observations 58 . Confirming our hypothesis that inhibiting NMD should increase sensitivity to doxorubicin, continuous co-treatment with NMDI-1 led to statistically significant decreases in cell viability relative to doxorubicin treatment alone (Fig. 8d, blue histograms). Transient pretreatment with NMDI-1 led to an even more pronounced effect and up to a ~2.5 fold reduction in cell viability at 50 μM doxorubicin relative to doxorubicin alone (Fig. 8d, red  histograms), despite the total time of exposure to NMDI-1 being half of that in the cotreatment regimen. Thus, inhibiting NMD promotes doxorubicin-mediated cell death, and conversely, inhibiting UPF1 CP generation obscures this effect.

DISCUSSION
Here, we observe that NMD activity is blunted during chemotherapeutic treatments (doxorubicin, staurosporine, etc.) that ultimately cause apoptosis. During treatment with doxorubicin and other clinically relevant small molecules (e.g. etoposide), one or more UPF1 CPs are produced. The UPF1 CP that we have mapped to a region encompassing UPF1 amino acids 38-1118 acts to inhibit NMD in dominant-interfering fashion, i.e. at substoichiometric levels relative to cellular UPF1 ( Fig. 5; Supplementary Fig. 4). Inhibition is tunable-the more UPF1 CP is generated, the more PTC-containing reporter mRNA is stabilized (Fig. 5b,c). Increases in PTC-reporter mRNAs (~2.4-fold; Fig. 5b) are less than those achievable using UPF1 ablation (~6-10-fold) 60 , and fold-changes in endogenous NMD targets are smaller (Supplementary Data 1), indicating that substoichiometric generation of UPF1 CP may be a way to fine-tune gene expression with physiological consequences (Fig. 8). The combined effects of this UPF1 CP, as well as additional UPF1 CPs (Fig. 1c,d), and changes to UPF1 phosphorylation status (Fig. 1c) all likely contribute to inhibition of NMD and upregulation of cell-cycle inhibitory and apoptosis-promoting transcripts seen during doxorubicin treatment ( Fig. 1a; Fig. 7). Small molecule-mediated inhibition of NMD may provide an improved therapeutic strategy when delivered in combination with cytotoxic agents already in clinical use (Fig. 8d).
Transient pre-treatment with NMDI-1 before addition of doxorubicin leads to enhanced cell death relative to either doxorubicin alone or to continuous co-treatment of NMDI-1 and doxorubicin. This suggests one model for NMD involvement in enabling the establishment of different cellular states by sculpting the mRNA milieu (Fig. 8e). Transcription produces mRNAs that are (red) or are not (blue) NMD targets, or are indirect NMD targets (black). NMD activity degrades NMD-sensitive transcripts that are either not allowed into the pool of translated mRNAs or allowed at only low levels. Here, we show that NMD activity is under the purview of the cell: NMD activity is tuned by generating a UPF1 CP (Fig. 5) that inhibits NMD, allows increased amounts of NMD-sensitive transcripts into the mRNA milieu, and likely also regulates indirect NMD targets.
Which transcripts are direct NMD targets or indirect NMD targets may be an academic distinction to the cell. Clearly, inhibiting NMD is able to change the mRNA milieu to promote physiological consequences (Fig. 8). The sum effect of these changes to the mRNA pool alters the cellular state to one that is competent to respond (via death) to the insult that elicited the inhibition of NMD (doxorubicin). Such a model explains why transient pretreatment with NMDI-1 before application of doxorubicin is a more effective treatment regimen than mere co-treatment of NMDI-1 and doxorubicin. During the pre-treatment pulse, the cell has already attenuated NMD and established an mRNA milieu that can respond to doxorubicin even before doxorubicin is applied, making the response to doxorubicin (death) more rapid. An important caveat to this model is that NMD inhibition also increases the levels of truncated aberrant proteins that have detrimental effects on cellular metabolism. However, the overall response is likely the same, namely, increased sensitivity to doxorubicin.
This model for NMD also extends to a recent report 13 showing that UPF1 levels are decreased by production of miR-128 in mouse embryonic neural stem cells. Reduction of UPF1 levels by this mechanism, and thus the inhibition of NMD, enables differentiation signals (retinoic acid) to elicit a response (neurogenesis). That many transcripts encoding proteins with diverse function are upregulated with modest magnitudes upon NMD inhibition makes this an attractive way for the cell to quickly respond to diverse stimulionly a subset of the changes made need to be exploited in response to any one stimulus. Mild upregulation of irrelevant transcripts is of little impact to cellular physiology -indeed, NMDI-1 addition has no effect at sublethal doxorubicin doses (Fig. 8d), mice tolerate NMDI-1 despite mild upregulation of endogenous NMD targets 57 , and very recently published NMD-inhibitory compounds also show little toxicity 61 .

Cell culture and transfections
All cell lines were cultivated in DMEM (Gibco) containing 10% fetal bovine serum (Gibco) with the exception of Jurkat and Daudi cells, which were grown in RPMI-1640 (Gibco) with 10% fetal bovine serum. HeLa and MCF7 cells (ATCC) were transfected with plasmid DNA using Lipofectamine LTX (Invitrogen), and HEK293T cells were transfected with plasmid DNA using Lipofectamine 2000 (Invitrogen). Transfections using siRNA employed RNAi MAX (Invitrogen) according to manufacturer's directions, with the exception of Fig. 2a, which employed Oligofectamine (Invitrogen). Cells were plated in antibiotic-free medium for a minimum of 24 h before transfection and harvested 48 h after plasmid transfections or 72 h after siRNA transfections.

Immunoprecipitation and on-bead RNase digestion
HEK293T cells were transfected as described in figure legends. Cells were lysed as described for western blotting. Input lysate protein concentrations were determined using the Bradford method (Biorad), equalized, and pre-cleared twice using protein-A conjugate agarose (Roche) for 30 min with end-over-end rotation at 4°C. Pre-cleared lysates were subjected to immunoprecipitation (IP) using anti-FLAG M2 Sepharose (Sigma) for 2 h at 4°C, washed with lysis buffer supplemented to contain 0.1% TritonX-100, and divided into two equal volumes. One-half of each IP was incubated with BSA in RNAse ONE (Promega) reaction buffer, and the other half was incubated with 1000U RNAse ONE (Promega) in reaction buffer for 30 min at 4°C. Samples were washed three times with wash buffer and eluted using 3X FLAG peptide (Sigma) according to manufacturer's directions.

In-vitro caspase cleavage assays
HEK293T cells were transfected with the indicated constructs. Cells were harvested 48 h later, and anti-FLAG immunoprecipitation was performed as above, without RNase ONE digestion. Proteins were eluted with 3X FLAG peptide. One microliter of each immunoprecipitate was incubated for 5 h at 37°C with 6U of either Caspase 3 or Caspase 7 in caspase cleavage buffer (50 mM HEPES pH 7.2, 50 mM NaCl, 0.1% CHAPS, 10 mM EDTA, 5% glycerol, and 10 mM DTT).

mRNA decay assays
For mRNA decay assays using actionmycin D, MCF7 cells were plated at 103,000 cells/well in 24-well dishes. After 24 h, cells either were or were not pre-treated for 1 h with 5 μM doxorubicin before addition of 3 μg/ml actinomycin D (Sigma). Cells were harvested at the indicated time points. Doxorubicin-treated cells received doxorubucin during the chase period. For Tet-off assays, HeLa Tet-off cells (Clontech) were plated at 40,000 cells/well in 24-well dishes. After 16 h, cells were transfected with the indicated plasmids in the presence of 1 μg/ml doxycycline to inhibit transcription. After 48 h, cells were washed three times with medium lacking doxycycline and incubated for 5h without doxycycline to induce transcription. Cells were then either treated with nothing, treated with 50μM doxorubicin for 1 h prior to transcriptional shut-off, or treated with 50 μg/ml puromycin for 3 h prior to transcriptional shutoff. At t=0, cells were cultured in medium containing 2 μg/ml doxycycline to induce shut-off and subsequently harvested at the indicated time points. For cells treated with doxorubicin or puromycin, these compounds were included in the chase. RT-qPCR was used to assess mRNA levels during the chase time. Data for actinomycin D decay assays (Fig. 1a) as well as for Tet-off decay assays (Fig. 1b), with the exception of the no treatment β-Gl 39 Ter mRNA data, were fitted to best fit linear regression lines because they clearly exhibited single component decay kinetics. In contrast, the no treatment β-Gl 39 Ter mRNA has been shown to decay with two-component kinetics 62 . For DRB-treated mRNA decay assays, HeLa cells were transfected with the indicated plasmids using Lipofectamine LTX. After 48 h, cells were treated with 100 μg/ml DRB (Sigma). Cells were harvested at the indicated time points, and RT-qPCR was used to assess transcript levels.

TEV cleavage and RNA-seq
For the TEV cleavage experiments, stably transduced HeLa cell lines were plated at 200,000 cells/well in a 6-well dish and cultured without antibiotics for 48 h. Either the bidirectional TEV protease-encoding plasmid or empty vector was introduced, and cells were harvested and flash frozen 16 h later. RNA was processed as for RT-qPCR. Biological duplicates of each sample (each duplicate consisted of 6 pooled wells) were used for sequencing. Totalcell RNA was submitted to the Whitehead Institute Genome Technology Core for RNA-seq. RNA concentrations were determined using a NanopDrop 1000 spectrophotometer (NanoDrop, Wilmington, DE), and RNA quality was assessed using a Agilent Bioanalyzer (Agilent, Santa Clara, CA). Poly(A) + libraries were prepared using the automated IntegenX Apollo system. Sequencing was performed using an Illumina HiSeq 2500 in 40-bp singleread mode. Data analysis was performed by the University of Rochester Genomics Research Center. Raw reads generated from the Illumina HiSeq2500 sequencer were demultiplexed using configurebcl2fastq.pl version 1.8.3. Low complexity reads and vector contamination were removed using sequence cleaner ("seqclean") and the NCBI univec database, respectively. The FASTX toolkit (fastq_quality_trimmer) was applied to remove bases with quality scores below Q=13 from the end of each read. Processed reads were then mapped to the UCSC Hg19 genome build using SHRiMP version 2.2.3, and differential expression analysis was performed using Cufflinks version 2.0.2; specifically, cuffdiff2 and usage of the general transfer format (GTF) annotation file for the given reference genome. Pooled duplicate values were used for analyses. Specifically, fold-change in Fragments Per Kilobase of transcript per Million (FPKM) values for cell lines stably transduced with empty vector and subsequently transiently transfected with TEV-encoding plasmid or empty vector were calculated. The same calculation was performed with FPKM values for the MYC-UPF1-FLAG TEV cell line. These values obtained for the MYC-UPF1-FLAG TEV cell line were then normalized to the fold-change value obtained for the empty vector stable cell line, and this calculation was performed for all genes represented.

Cell survival assays
Cells were plated in white opaque 96-well plates using multichannel pipettors for accuracy and treated as described in the figure legends. CellTiter-Glo luminescent assays (Promega) were performed according to manufacturer's directions. Data were collected using a SpectraMax M2 plate reader.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material. NMD is inhibited during doxorubicin treatment. (a) mRNA decay assays in MCF7 cells. MCF7 cells either were (red) or were not (black) pre-treated with 5 μM doxorubicin for 1 h before addition of 3 μg/ml actinomycin D to halt transcription. Cells were collected at the indicated times after actinomycin D addition. Levels of the indicated NMD-targeted mRNAs were assessed by RT-qPCR, normalized to 18s rRNA levels, and displayed as a percentage of the levels at t=0 h. Error bars=S.E.M., n=4 independent biological quadruplicates. (b) Human β-Gl mRNA half-life studies in HeLa Tet-off cells. HeLa Tet-off cells were transfected with plasmids encoding human β-Gl Norm mRNA and MUP mRNA or β-Gl Ter mRNA and MUP mRNA. β-Gl Norm and β-Gl 39 Ter mRNA transcription occurs under the agency of the non-stress-responsive Tet-off promoter. Cells were either pre-treated with nothing (top), 50 μM doxorubicin for 1 h (middle), or 50 μg/ml puromycin for 3 h (bottom) before transcriptional shut-off with 2 μg/ml doxycycline. Cell aliquots were removed at the indicated "chase" time points, and RT-qPCR was used to assess the remaining levels of β-Gl Norm and β-Gl Ter mRNAs, each after normalization to MUP mRNA. (c) Western blots of lysates of MCF7 cells from a (blots derive from and are representative of the three biological replicates in a) that had been exposed to doxorubicin (5 μM) for the indicated times. CP, cleavage product. GAPDH levels serve as loading controls. Three-fold serial dilutions (wedge) reveal the dynamic range of analysis. (d) As in c, but cells were exposed to a 10-fold higher concentration of doxorubicin and analyzed at earlier time points. Representative of 2 biological replicates. Popp   UPF1 CP production is an early and conserved event. (a) Western blots of lysates of HeLa cells transfected with 100 nM of either control (Ctrl) siRNA or UPF1 siRNA and, 48 h later, exposed to cycloheximide (CHX, 300 μg/mL) for 3 h. (b) Western blots of lysates of HeLa cells (Homo sapiens) exposed to CHX (100 μg/mL or 300 μg/mL) for either 3 or 5 h, or to etoposide (ETP; 44 μM) for 6 h, incubated in fresh medium, and withdrawn from ETP at the indicated times. (c) Essentially as in b, except C2C12 myoblasts (Mus musculus) were analyzed. CHX concentrations were 100 μg/mL or 300 μg/mL, and a 5 h pulse of ETP was Popp  used at 100 μM. Note that at least two apoptotic inducers were used for each cell line, and at least two cell lines were tested with each apoptotic inducer.  mCherry TCRβ ΔJC intron mRNA in the presence of each UPF1 variant is defined as 100%. Error bars=S.E.M., asterisk=p<0.05 relative to the UPF1 siRNA + MYC-UPF1-FLAG WT sample using the Student's t-test. n=3 independent biological replicates.