A Competition between Stimulators and Antagonists of Upf Complex Recruitment Governs Human Nonsense-Mediated mRNA Decay

The nonsense-mediated decay (NMD) pathway subjects mRNAs with premature termination codons (PTCs) to rapid decay. The conserved Upf1–3 complex interacts with the eukaryotic translation release factors, eRF3 and eRF1, and triggers NMD when translation termination takes place at a PTC. Contrasting models postulate central roles in PTC-recognition for the exon junction complex in mammals versus the cytoplasmic poly(A)-binding protein (PABP) in other eukaryotes. Here we present evidence for a unified model for NMD, in which PTC recognition in human cells is mediated by a competition between 3′ UTR–associated factors that stimulate or antagonize recruitment of the Upf complex to the terminating ribosome. We identify cytoplasmic PABP as a human NMD antagonizing factor, which inhibits the interaction between eRF3 and Upf1 in vitro and prevents NMD in cells when positioned in proximity to the termination codon. Surprisingly, only when an extended 3′ UTR places cytoplasmic PABP distally to the termination codon does a downstream exon junction complex enhance NMD, likely through increasing the affinity of Upf proteins for the 3′ UTR. Interestingly, while an artificial 3′ UTR of >420 nucleotides triggers NMD, a large subset of human mRNAs contain longer 3′ UTRs but evade NMD. We speculate that these have evolved to concentrate NMD-inhibiting factors, such as PABP, in spatial proximity of the termination codon.


Introduction
The process of nonsense-mediated decay (NMD) subjects mRNAs with premature termination codons (PTCs) to rapid decay. This helps rid the cell of aberrant mRNAs that have acquired PTCs through mutation or faulty processing [1][2][3]. Moreover, several lines of evidence suggest that NMD is also used as a posttranscriptional mechanism of normal gene regulation [4]. The NMD pathway employs a set of factors that are conserved amongst eukaryotes. Central to the NMD pathway is the Upf complex, which consists of the proteins Upf1, Upf2, and Upf3 [1][2][3]. The Upf complex interacts with the eukaryotic translation release factors, eRF3 and eRF1, and triggers NMD when translation termination takes place at a PTC [1][2][3]. In addition, the Smg proteins, which are conserved in metazoans, regulate Upf1 function by phosphorylation and dephosphorylation [2,3].
A fundamental question is how mRNAs with PTCs are distinguished from those with normal termination codons. Despite the conservation of core NMD factors, contrasting models have been proposed in mammalian cells as opposed to other eukaryotes. Evidence in Saccharomyces cerevisiae and in cell lines from Drosophila melanogaster suggests that termination codons are recognized as PTCs when positioned too far upstream of the poly(A) tail [5][6][7]. This is thought to be a consequence of an impaired interaction between eRF3 at the terminating ribosome and factors associated with the normal 39 UTR, including cytoplasmic poly(A)-binding protein (PABP) [1,5,7], which on mRNAs with regular stop codons (proximal to the poly(A) tail) stimulates normal translation termination [8]. Consistent with this model for NMD, termed the ''faux 39 UTR'' model [1,7], 39 UTRs of S. cerevisiae and D.
melanogaster mRNAs are generally short, on average ;100 and ;330 nucleotides in length, respectively [9,10]. Interestingly, recent observations show evidence that cytoplasmic PABP is not required for the discrimination of normal termination codons from PTCs in S. cerevisiae [11]. Thus, cytoplasmic PABP may function redundantly with other 39 UTR-associated factors to antagonize NMD. 39 UTRs of human mRNAs are on average longer (;750-800 nucleotides [12]) than those of S. cerevisiae and D. melanogaster, and current models for NMD in mammalian cells do not involve the length of the 39 UTR. Rather, the exon junction complex (EJC), which is deposited 20-25 nucleotides upstream of mRNA exon-exon junctions after pre-mRNA splicing [13], is thought to play a central role. A termination event more than ;30 nucleotides upstream of one or more EJCs is thought to trigger NMD through EJC-mediated recruitment of the Upf complex [2,3]. This is consistent with observed direct interactions between EJC components and Upf3 proteins [14][15][16][17][18]. However, the EJC plays no apparent role in NMD in D. melanogaster [19] or in Caenorhabditis elegans [20] and no evidence for the existence of an EJC has been reported in yeast. Nevertheless, a conceptually similar model to the EJC model was proposed earlier for NMD of the PGK1 mRNA in yeast, in which a ''downstream sequence element'' (DSE), when present downstream of a termination codon, promotes NMD through recruitment of the protein Hrp1p, which interacts with Upf proteins [21,22].
A fundamental difference between the faux 39 UTR and the EJC/DSE models for NMD is that the EJC/DSE models propose that NMD-stimulating factors (the EJC and Hrp1p, respectively) trigger NMD when positioned downstream of a termination codon, whereas the faux 39 UTR model postulates that NMD is caused instead by the absence of NMDantagonizing factors, such as cytoplasmic PABP, which normally positively influence translation termination and mRNA stability. Here, we present evidence for a merged model for NMD in human cells, which likely can be extended to other eukaryotes. According to this model, PTC recognition is determined by a competition between 39 UTRassociated factors, which stimulate (including the EJC) or antagonize (including cytoplasmic PABP) the recruitment of the Upf complex to the terminating ribosome. Our observations suggest that the fundamental principles of the NMD pathway are much more conserved between mammals and other eukaryotes than previously anticipated.

UTR Introns Are Not Sufficient for Triggering Human NMD
The EJC model for human NMD postulates that any translation termination event taking place .50-55 nucleotides upstream of an exon-exon junction should result in NMD. However, during our studies of the human NMD pathway, we observed that a b-globin mRNA, in which the adenovirus major late (AdML) intron was inserted into the 39 UTR 175 nucleotides downstream of the normal b-globin mRNA translation termination codon, did not show enhanced mRNA decay as compared to the wild-type b-globin mRNA in human HeLa Tet-off cells (compare Figure 1A and 1B). Moreover, in contrast to a well-characterized b-globin NMD substrate, which contains a premature termination codon at position 39 (b39; Figure 1C), the AdML intron containing bglobin mRNA is not stabilized when the central NMD factor hUpf1 is knocked down ( Figure 1A, middle panel and Figure  S1) or when a point mutation causes translation termination to take place downstream of the inserted intron ( Figure 1A, bottom panel). Sequencing of cDNAs derived from the mRNAs in Figure 1A revealed the expected splicing patterns (unpublished data). Thus, in contrast to the prediction from the EJC model for human NMD, the AdML intron is not sufficient for triggering NMD when positioned in the 39 UTR of b-globin mRNA, even though the AdML intron has been previously demonstrated to recruit an EJC [13].
The observation in Figure 1A was surprising, because a bglobin mRNA in which the MINX-intron had been placed in the 39 UTR was previously found to cause reduced mRNA steady-state levels [23]. Therefore, to rule out the possibility that the observations in Figure 1A represent an unusual property of the specific mRNA reporter, we tested two other substrates. As seen in Figure 1D and 1E, insertion of the triosephosphate isomerase (TPI) mRNA intron 6 or the AdML intron, 140 or 149 nucleotides downstream of the termination codons of b-globin or TPI mRNAs, respectively, failed to cause hUpf1-dependent mRNA decay, despite the previously demonstrated ability of each of these introns to recruit an EJC [13]. Sequencing of cDNAs derived from the tested mRNAs revealed the expected splicing patterns (unpublished data), although a minor fraction of the bTPIi6 mRNA fails to remove the TPI intron (see asterisk in Figure 1D). We attempted to test the b-globin mRNA with the MINX-intron in the 39 UTR, which was previously found to accumulate at reduced steady-state levels as compared to wild-type b-globin mRNA [23] (construct generously provided by A. Kulozik and M. Hentze). However, the MINX intron (as well as a number of other introns tested in this study) failed to be spliced out of the b-globin mRNA 39 UTR in the HeLa Tet-off cells used here (unpublished data). We conclude that a 39 UTR intron is not sufficient to trigger NMD in human cells. This contradicts an EJC-centric model for human NMD.

Extended 39 UTRs Trigger Human NMD
Our observation that 39 UTR introns are not sufficient for triggering human NMD spurred us to test whether cytoplasmic PABP may antagonize NMD in human cells, as it does in S. cerevisiae and D. melanogaster. We therefore first manipulated the position of the poly(A) tail relative to the termination codons of b-globin and TPI reporter mRNAs and tested the effect on mRNA decay. As seen in Figure 2A and 2B and Figure S2, artificial extension of the 39 UTRs of b-globin or TPI mRNAs (from 292 and 447 nucleotides, respectively, to 846-1,112 nucleotides) through insertion of fragments of glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (Figures 2A and S2) or green fluorescent protein (GFP) ( Figure  2B) mRNAs results in mRNA destabilization (compare with Figures 1B and S2, top panel). This is due to NMD because depletion of the NMD factors hUpf1 or hUpf2 stabilizes the mRNAs (Figure 2A and 2B and Figures S1 and S2). Moreover, the introduction of single point mutations in the termination

Author Summary
The nonsense-mediated mRNA decay pathway is responsible for rapidly degrading mRNAs with premature termination codons. This is important because it prevents the production of potentially deleterious truncated proteins from aberrant mRNAs, such as those that have undergone erroneous processing. How does the cell discriminate aberrant mRNAs from those that are normal? Here we present evidence that in human cells, the targeting of an mRNA to nonsense-mediated mRNA decay depends on a competition between proteins associated with the mRNA 39 UTR that stimulate or antagonize mRNA decay. We show that cytoplasmic poly(A)binding protein, a protein associated with the mRNA 39 end poly(A) tail, antagonizes mRNA decay. By contrast, a protein complex deposited onto mRNAs upon pre-mRNA splicing, called the exon junction complex, stimulates mRNA decay. Our observations suggest that the competition between these proteins, and probably other unknown proteins with similar activities, determines whether a key protein complex in the pathway, the Upf complex, is recruited to the mRNA upon translation termination, which leads to mRNA decay.
codons that results in termination on the same mRNAs in proximity (180-357 nucleotides upstream) of the poly(A) tail, results in mRNA stabilization (Figure 2A and 2B and S2, bottom panels). Even though sequencing of cDNAs derived from the tested mRNAs revealed no cryptic splicing in the extended 39 UTRs (unpublished data), depletion of the central EJC component eIF4AIII results in stabilization of the bGAP mRNA ( Figure S3), possibly reflecting the ability of the EJCs in the b-globin mRNA open reading frame to stimulate translation as has been previously observed [24,25]. Successive shortening of the 39 UTR of the bGAP mRNA revealed that a 39 UTR as short as 422 nucleotides can trigger NMD ( Figure S4). This is surprising because a large fraction of human mRNAs contain 39 UTRs longer than 422 nucleotides [12]. We conclude that artificially extended 39 UTRs trigger NMD in human cells. This is consistent with recent reports in which steady-state levels of PTCs containing TPI, b-globin, and Ig-l reporter mRNAs lacking 39 UTR introns were measured [26][27][28], and with observations using unspliced Rous sarcoma virus RNAs in chicken cells [29]. Thus 39 UTR introns are neither necessary ( Figure 2) nor sufficient ( Figure  1) for human NMD.

An Intron-Less mRNA Can Undergo NMD
Having observed that 39 UTR introns are not required for NMD, we asked whether a completely intron-less mRNA can undergo NMD. It was observed previously that introduction of PTCs in the naturally intron-less Hsp70 and histone H2A mRNAs does not result in their decreased steady-state levels, which led to speculations that intron-less mRNAs are immune to NMD [30]. However, it has been pointed out that wild-type  Figure S1). For each panel, schematics are shown for each tested mRNA below the panels; b-globin or TPI exons are indicated as light-gray bars, introns as lines, AdML exons are shown in black, and TPIi6 exons are patterned. Numbers above the panels indicate time after transcriptional repression. mRNA half-lives were calculated by comparison with the constitutively expressed internal control mRNAs (bUAC-GAP or bwt; top panels in each assay) and are given on the right. doi:10.1371/journal.pbio.0060111.g001 Hsp70 and histone H2A mRNAs are both highly unstable and may thus not be further destabilized by a PTC [5]. We therefore tested the stability of three naturally occurring intron-less mRNAs (encoding eRF3b, SFN, and TBCC) and found that both wild-type and PTC containing versions of these mRNAs were unstable (;100-to 150-min half-lives, unpublished data). Thus, mRNA instability may be a general feature of natural intron-less mRNAs. However, when the only intron in the Glutathione Peroxidase 1 (GPx1) mRNA is removed, introduction of a PTC triggers NMD, although not as efficiently as in the presence of the intron ( Figure 2C, compare lower and upper panels). Thus, neither 39 UTR introns nor internal introns are essential for human NMD. However, similarly to a previous report [30] we have not been able to identify a natural human intron-less mRNA for which NMD could be observed, perhaps due to the observed inherent instabilities of the tested mRNAs.

Cytoplasmic PABP Antagonizes Human NMD
To more directly test whether cytoplasmic PABP antagonizes NMD in human cells, we examined the effect of positioning cytoplasmic PABP in proximity of a PTC using two different approaches. First, as seen in Figure 3A and 3B, when an MS2-PABPC1 fusion protein (PABPC1 is one of five human cytoplasmic PABPs [31]) is artificially tethered downstream of a PTC in two different b-globin NMD reporter mRNAs, a partial rescue of NMD is observed. This rescue is due to tethered PABPC1, because similar levels of unfused PABPC1 ( Figure 3A) or MS2 coat protein ( Figure 3A and 3B) do not stabilize the mRNAs. Moreover, tethering of the nuclear poly(A)-binding protein PABPN1 does not rescue NMD ( Figure 3A) even though it is expressed at levels similar to MS2-PABPC1 ( Figure S5). The efficiency of the rescue from NMD by tethered PABPC1 decreases as the MS2 binding sites are moved more distal to the PTC in the b-globin PTC-39 mRNA ( Figure S6).
As a second independent approach to ask whether PABP can antagonize human NMD, we tested the effect of inserting a binding site for PABP downstream of the PTCs. As seen in Figure 3C and 3D, inclusion of a poly-A 30 -stretch, but not that of a random 30-nucleotide stretch, 182 or 230  Figure S1). Constitutively expressed bwt mRNA was used as internal controls for quantification. Numbers above the panels indicate time after transcriptional repression. Schematics on the bottom show the used constructs with b-globin exons indicated as light-gray bars, introns as lines, and GAPDH and GFP sequences as dark-gray and dotted bars, respectively. Numbers on the right indicate mRNA half-lives (t 1/2 ; in minutes) calculated from the shown experiment. Numbers in parentheses indicate the fold stabilization with standard deviation (n 3) upon hUpf knockdown as compared to the Luc control. (C) mRNA decay assays for intron-containing GPx1 mRNA with a PTC (GPx1-UAA) or a GPx1 mRNA with a PTC expressed from an intron-less construct (GPx1Di-UAA) (see schematics below) in the presence of Luc or hUpf1 siRNAs as indicated on the left. mRNA decay rates for the shown experiments are given on the right and numbers in parentheses indicate the fold stabilization with standard deviation (n 3) upon hUpf knockdown as compared to the Luc control. doi:10.1371/journal.pbio.0060111.g002 nucleotides downstream of the PTC of two different NMD reporter mRNAs, results in partial rescue of NMD. Thus, similarly to S. cerevisiae [7] and D. melanogaster [5], cytoplasmic PABP can antagonize NMD in human cells when placed in proximity of a PTC. Recent observations suggest that while cytoplasmic PABP can antagonize NMD in S. cerevisiae, it is not required for discriminating a normal mRNA from an NMD substrate [11]. Attempts at testing whether PABPC1 is required for preventing NMD in human cells failed because HeLa Tet-off cells became detached from plates upon short interfering RNA (siRNA)-mediated PABPC1 depletion (unpublished data).

A Subset of Naturally Occurring Long 39 UTRs Can Antagonize NMD
Our observations raise the question of whether naturally occurring mammalian mRNAs with long 39 UTRs, which can be several kilobases in length, are normal targets of NMD or whether they have evolved mechanisms to evade the NMD pathway. We noted that mRNAs identified by microarray assays to be upregulated upon hUpf1 knockdown in HeLa cells [32] contain on average significantly longer 39 UTRs than those mRNAs unaffected by hUpf1 knockdown ( Figure S7). Moreover, the majority of these 39 UTRs (75%) are longer than the ;420 nucleotides observed here to trigger NMD in the bGAP reporter mRNA (Figures S4 and S7). It is possible that at least a subset of these transcripts undergo NMD due to an increased distance between the termination codon and the poly(A) tail. Indeed, when the 1,342-nucleotide 39 UTR of one of these mRNAs, encoding hSmg5, is replaced for the bglobin 39 UTR (bSmg5), the chimeric mRNA undergoes NMD ( Figure 4, top two panels). Thus, the Smg5 39 UTR stimulates NMD, and a subset of mRNAs may have evolved long 39 UTRs to be regulated by the NMD pathway. However, numerous mRNAs with long 39 UTRs are not upregulated upon hUpf1 knockdown [32] ( Figure S7). When the 39 UTRs from two such mRNAs, Cript1 and Tram1, were inserted into the b-globin mRNA, no NMD was observed ( Figure 4, bottom panels). This is in sharp contrast to the observations using artificial long 39 UTRs (compare to Figures 1 and S4) and suggests that the ability of a subset of endogenous long 39 UTRs to evade NMD is an acquired property (see Discussion).

A 39 UTR Intron Can Enhance Human NMD
Our observations that 39 UTR introns are neither necessary ( Figure 2) nor sufficient ( Figure 1) for human NMD raises the question of whether introns play any role in human NMD. We therefore tested the effect of inserting the AdML intron into the 39 UTR of an mRNA, which already undergoes NMD due to an extended 39 UTR. Interestingly, insertion of the AdML intron downstream of the termination codon of bGAP mRNA results in enhanced mRNA decay ( Figure 5; bGAP-AdML mRNA). This effect is only observed when the intron is positioned in the 39 UTR, as insertion of the same intron upstream of the termination codon (without disrupting the open reading frame) does not enhance mRNA decay ( Figure  5; bAdML-GAP mRNA). Thus, while a downstream intron is neither sufficient nor necessary for triggering NMD in human HeLa cells (Figures 1 and 2) [33], it can enhance the degradation of an mRNA that is already a target of NMD  due to an extended 39 UTR ( Figure 5A and 5B). Consistent with this, the presence of an intron appears to also stimulate NMD of a PTC containing GPx1 mRNA ( Figure 2C).

PABPC1 Can Out-Compete the Interaction between hUpf1 and eRF3 In Vitro
How does cytoplasmic PABP antagonize NMD when positioned in proximity of the termination codon? Both cytoplasmic PABP and Upf1 have been previously observed to stimulate translation termination in yeast cells [8,34] and to associate with translation release factor eRF3 [35][36][37][38][39][40]. This raised the possibility that cytoplasmic PABP inhibits NMD by preventing Upf1 from interacting with eRF3 and the terminating ribosome. As seen in the co-immunoprecipitation (co-IP) assays in Figure 6A, endogenous hUpf1 and PABPC1 can both be observed in complex with eRF3 in RNase-treated HeLa cell extracts. However, PABPC1 co-IPs much more efficiently than hUpf1 with eRF3 ( Figure 6A), in spite of comparable estimated number of molecules of cytoplasmic PABP (;8 3 10 6 /cell) and hUpf1 (;3 3 10 6 /cell) in HeLa cells [41,42]. Consistent with this, bacterially expressed GST-tagged eRF3 was found to associate much more efficiently with rabbit reticulocyte-lysate-translated PABPC1 (K d ; 5 nM) than hUpf1 (K d . 1 lM) (unpublished data).
To test whether PABPC1 can antagonize the interaction between eRF3 and hUpf1 in vitro, we immunopurified transiently expressed epitope-tagged eRF3, PABPC1, and hUpf1 proteins from HEK 293T cells and tested the ability of hUpf1 to associate with eRF3 in the presence of increasing amounts of PABPC1. As seen in Figure 6B, in contrast to the negative control protein hnRNP A1, increasing amounts of PABPC1 efficiently prevent the interaction between hUpf1 and eRF3, even when hUpf1 is present in 10-to 40-fold excess over PABPC1 ( Figure 6B, compare lanes 5 and 6 with lanes 4 and 7). Thus, PABPC1 can antagonize the interaction between hUpf1 and eRF3 in vitro. However, no reduction in the co-IP efficiency between hUpf1 and eRF3 was observed upon transient over-expression of FLAG-tagged PABPC1 in HeLa or HEK 293T cells (unpublished data). Thus, either exogenous PABPC1 failed to express at adequate levels to observe a competition in cells, or the relation between hUpf1, PABPC1, and eRF3 is more complex in cells than it is in vitro.
To test whether amino acid residues of eRF3, which are important for cytoplasmic PABP interaction, are also  1-3), immobilized on an anti-myc antibody resin, was incubated with various amounts of FLAG-tagged hUpf1, PABPC1, or hnRNP A1 as indicated. Estimated amounts of FLAG-tagged proteins in each reaction are given in lM. Bottom panels show 5% of input protein for each reaction. The asterisk (*) and dagger ( ) on the right indicates likely degradation products of PABPC1 and cross-reacting Myc-eRF3, respectively. (C) Lanes 1-6: Western blot for exogenously expressed Myc-tagged eRF3 (wt, lanes 1-3) or eRF3 KAKA mutant protein (KAKA, lanes 4-6) that co-IP with FLAG-tagged PABPC1, hUpf1, or as a negative control, MS2, as indicated above the lanes. Lanes 7-9: Western blot for exogenously expressed Myctagged eRF1 that co-IP with FLAG-tagged eRF3, eRF3 KAKA mutant protein, or as a negative control, MS2, as indicated above the lanes. For all lanes, 5% of total input extracts are shown in the bottom panels. doi:10.1371/journal.pbio.0060111.g006 important for the interaction with hUpf1, we constructed a eRF3 protein (eRF3 KAKA) mutated in four N-terminal residues that are conserved between cytoplasmic PABPbinding proteins [38,43]. As seen in the co-IP assays in Figure  6C, the exogenously expressed eRF3 KAKA mutant protein is equally impaired in interaction with PABPC1 and hUpf1 ( Figure 6C, compare lanes 5 and 6 with lanes 2 and 3). As a control, the mutant eRF3 KAKA protein associates with eRF1 with similar affinity as wild-type eRF3, suggesting that these mutations do not cause gross structural alterations, although local changes cannot be ruled out. Thus, hUpf1 and PABPC1 interact with a similar, though not necessarily overlapping region of eRF3. The ability of PABPC1 to antagonize the association between hUpf1 and eRF3 in vitro could therefore be a result of a direct competition for eRF3 binding, or of a local structural alteration of eRF3 upon PABPC1 binding, which prevents hUpf1 association.

Discussion
Previous contrasting models for PTC-recognition in NMD invoke either 39 UTR-associated factors that stimulate NMD, the EJC in human cells [2,44], and DSE-binding proteins in yeast [21], or factors that stimulate normal translation termination and antagonize NMD [1,45]. Our observations, together with the observations in the paper by Eberle et al. [46], are consistent with a unified model for human NMD, in which the balance between NMD-antagonizing (such as PABPC1) and NMD-stimulating (such as the EJC) factor(s) that are associated with the mRNA 39 UTR, determines whether termination is considered normal or premature ( Figure 7A). According to this model, a translation termination event proximal to cytoplasmic PABP (Figure 3), or other unknown NMD-antagonizing factors, precludes the interaction of hUpf1 with eRF3 ( Figure 6C) and thus prohibits NMD ( Figure 7A, top). By contrast, if hUpf1 associates with eRF3, NMD ensues ( Figure 7A, bottom). This occurs when cytoplasmic PABP, or other inhibitory factors, are spatially distant from the termination event ( Figure 2) and is enhanced when a splicing event downstream of a termination codon results in deposition of an EJC ( Figure 5), which provides higher affinity for the hUpf complex ( Figure  7A, bottom). However, an exon-exon junction in the 39 UTR is not sufficient for NMD (Figure 1). This suggests that a proximal cytoplasmic PABP is dominant over 39 UTR exonexon junctions, which is consistent with the observation that the affinity of PABPC1 for eRF3 appears to be several orders of magnitude higher than that of hUpf1 ( Figure 6 and unpublished data). However, while introns are observed to only stimulate NMD of the substrates tested in this study, it cannot be ruled out that a subset of human mRNAs requires downstream introns for NMD. Previous experiments, in which EJC or hUpf proteins tethered to an mRNA 39 UTR were observed to trigger NMD, may have been assisted by the extended 39 UTRs resulting from insertion of multiple tethering sites and/or by the recruitment of multiple NMDpromoting factors [15,17,[47][48][49]. The model depicted in Figure 7A may be extended to eukaryotes other than mammals and is consistent with the observation in Drosophila S2 cells that the decay of an NMD reporter mRNA is inhibited upon cytoplasmic PABP depletion [5]. In this case it is predicted that a large subset of normally stable endogenous mRNAs become NMD substrates, thus out-titrating the NMD pathway.
How does cytoplasmic PABP antagonize NMD? While PABPC1 can out-compete the association of hUpf1 with eRF3 in vitro ( Figure 6B), a more complex relationship may exist between these proteins in the cell. For example, we failed to observe exogenously expressed PABPC1 out-compete the co-IP of endogenous hUpf1 with eRF3 (unpublished data). Moreover, in S. cerevisiae, cytoplasmic PABP truncated of its C-terminal eRF3-interaction region was capable of suppressing NMD when tethered in proximity of a PTC [7]. However, we found no loss of eRF3-association of a similarly truncated PABPC1 in co-IP assays between exogenously expressed human proteins (unpublished data), suggesting that eRF3 may form a complex with PABPC1 through additional regions. Understanding the specific mechanism by which NMD is antagonized by cytoplasmic PABP, and likely other 39 UTR-associated factors, is an important goal for future studies and could involve both direct competition with the Upf complex as well as modulation of the translation termination event that excludes Upf complex recruitment in a more indirect manner. Another open question is how the interplay between eRF3, PABP, and the Upf complex influences events downstream of translation termination. Interestingly, it was previously observed that the interaction between eRF3 and cytoplasmic PABP stimulates mRNA deadenylation in yeast [50], and that deadenylation can be an early step in NMD [51][52][53]. Clearly, a great deal remains to be learned about the relationship between eRF3, the Upf complex, and cytoplasmic PABP and how it controls the fates of mRNAs after translation termination.
It is likely that 39 UTR-associated factors (indicated by a question mark in Figure 7A) other than cytoplasmic PABP can antagonize NMD. This hypothesis is consistent with the observation that in yeast cells, cytoplasmic PABP is not required for discriminating tested NMD substrates from their normal counterparts [11]. An excellent candidate for such an activity is the yeast protein Pub1p, which has been identified as a factor that binds downstream of upstream open reading frames (uORFs) in GCN4 and YAP1 mRNAs to prevent NMD [54]. It is possible that Pub1p and factors with similar activities are found in a subset of normal 39 UTRs. It remains to be tested whether Pub1p acts on the terminating ribosome in a manner similar to cytoplasmic PABP. Similarly, factors other than the EJC could provide an enhanced affinity for the Upf complex and stimulate NMD. For example, the protein Hrp1p appears to serve such a role in the yeast PGK1 NMD substrate [21]. Moreover, human Staufen1 and histone mRNA stem loop binding protein have been shown to recruit hUpf1 to the 39 UTR of specific mRNAs to trigger NMD-like mRNA decay [55,56]. Thus, our observations suggest that the NMD pathway is much more conserved between mammals and other eukaryotes than previously appreciated. Nevertheless, there is evidence that differences exist between yeast and mammalian cells as to which round of translation can stimulate NMD [28,[57][58][59].
Our observations suggest that while artificial long 39 UTRs trigger NMD (Figure 2), a subset of mRNAs containing long 39 UTRs have evolved mechanisms to evade NMD (Figure 4). Future studies should reveal the mechanism by which this is accomplished. This could conceivably be achieved by (i) induced looping of the 39 UTR, thus placing the poly(A) tail and cytoplasmic PABP in close spatial proximity to the translation termination event ( Figure 7B, top), or (ii) by recruitment of factors that antagonize NMD (such as PABPC1 or Pub1p) to the 39 UTR in proximity to the termination codon ( Figure 7B, bottom). The observation that cytoplasmic PABP alleviates NMD when placed in the vicinity of a PTC (Figure 3) [5,7,46] and the finding in the paper by Eberle et al. that artificially induced 39 UTR looping rescues reporter mRNAs with extended 39 UTRs from NMD [46], provides proof-of-principle evidence for each of these models. The mechanism by which specific mRNAs evade the NMD pathway is an important subject for future investigation and is likely to vary between individual mRNAs.
After the submission of this paper, we have become aware of two other studies reporting that cytoplasmic PABP antagonizes human NMD when placed in proximity to a PTC [60,61].

Materials and Methods
Plasmid constructs. All plasmid sequences are available upon request. Plasmids expressing different b-globin reporter mRNAs were derived from the pcTET2-bwt plasmid that was constructed by inserting the human b-globin gene between HindIII and ApaI sites of a pcDNA3-based plasmid containing six copies of the Tet-operator sequences upstream of the TATA box. For extended 39 UTR constructs, parts of the GAPDH mRNA coding sequence and the entire GAPDH 39 UTR (pcTET2-bGAP) or the GFP ORF (pcTET2-bGFP) were inserted between NotI and XbaI sites of the pcTET2-bwt plasmid, thus replacing the b-globin 39 UTR. The b-globin stop codon was mutated to UAC by site-directed mutagenesis to generate pcTET2-bGAP-UAC and pcTET2-bGFP-UAC. Plasmids expressing bGAP-UAC-696, bGAP-UAC-485, and bGAP-UAC-422 mRNAs were generated by site-directed mutagenesis of pcTET2-bGAP-UAC to introduce a stop codon (UAA) in the GAPDH sequence, respectively, 696, 485, or 422 nucleotides upstream of the polyadenylation site. The plasmid expressing bwt mRNA was described earlier [47]. To construct plasmids expressing b39-2xMS2-Ex2, b39-2xMS2-Ex3, and b39-2xMS2-3UTR, the 2xMS2 cassette from the previously described plasmid pcb-2bs [47] was inserted into the BamHI, EcoRI, or NotI sites, respectively, of the pPC-b39 plasmid described earlier [62]. A stretch of A 30 (pPC-b39-A 30 ) or N 30 (pPC-b39-N 30 ) was inserted into the BamHI site of pPC-b39 plasmid using annealed DNA oligos. Similarly, A 30 (pcTET2-bGAP-A 30 ) or N 30 (pcTET2-bGAP-N 30 ) was inserted into the XbaI site of the pcTET2-bGAP plasmid. To construct the pcTET2-bGAP-4xMS2 plasmid, four MS2 binding sites were amplified from a previously described plasmid pcb-4bs [47] and inserted into the XbaI site of pcTET2-bGAP. Plasmids expressing bAdML, bAdML-UAC, and bTPIi6 mRNAs were constructed by cloning the AdML intron or TPI intron 6 (TPIi6) and flanking exon sequences into the XbaI site in pcTET2-bwt or pcTET2-bwt-UAC plasmids. bGAP-AdML and bAdML-GAP mRNA-expressing plasmids were constructed by inserting the same AdML intron into XbaI and EcoRI sites, respectively, in the pcTET2-bwtGAP plasmid. Plasmids expressing chimeric b-globin mRNAs with 39 UTRs from Smg5, Cript1, and Tram1 genes, the respective 39 UTRs, were cloned into the NotI-XbaI sites of pcTET2-bwt.
Plasmids expressing TPI reporter mRNAs were constructed by inserting the entire human TPI gene between HindIII and XbaI sites of the pcTET2 plasmid. A NotI site was inserted into exon 6 (in a manner that preserved the encoded protein) by site-directed mutagenesis. Codon 189 was mutated to TGA using site-directed mutagenesis to generate pcTET2-TPI-189. To remove intron 6, a NotI-XbaI fragment containing exon6-intron6-exon7 was replaced by the same region amplified from TPI cDNA, to generate pcTET2-TPIDi6-189. To extend the TPI 39 UTR, a fragment containing part of the GAPDH coding region and 39 UTR was inserted into the NotI site of pcTET2-TPIDi6-189 to generate pcTET2-TPIDi6-189-GAP, or into the NotI site of pcTET2-TPIDi6 to give rise to pcTET2-TPIDi6-GAP. TPI-AdML mRNA-expressing plasmid was constructed by inserting the AdML intron and flanking exonic sequences into the XbaI site of pcTET2-TPI.
The constructs for knockdowns were based on the pSHAG plasmid (a gift from Dr. G. Hannon) and contained inserts expressing precursors to hUpf1, hUpf2, or eIF4AIII siRNAs described earlier [63,64].
NMD factor knockdown. NMD factor knockdowns were performed by co-transfecting cells with reporter mRNA plasmids and plasmids encoding small hairpin (sh)RNAs targeting hUpf1, hUp2, or eIF4AIII, 60 h before pulse-chase mRNA decay assays were carried out.
mRNA decay assays and Northern blots. mRNA decay assays were performed in HeLa Tet-off cells in DMEM/10% FBS/tetracycline (50 ng/ml) transfected with b-globin mRNA expression plasmids. For each 2-cm well of HeLa Tet-off cells, 10 ng of pcbG or pcbwt (as an internal control) and 0.2 lg of tetracycline-regulated reporter mRNA expression plasmids were co-transfected using TransIT HeLa Monster reagent (Mirus). For knockdowns, 0.5 lg of pSHAG plasmids were co-transfected. In each transfection, empty pcDNA3 vector was added to 1 lg of total plasmid. 36-40 h after transfection, or approximately 60 h in the case of knockdowns, transcription of reporter mRNAs was induced by removal of tetracycline through washing cells with 1 ml of phosphate-buffered saline (PBS) and adding DMEM/10% FBS. 6 h later, transcription was shut off by adding tetracycline to a final concentration of 1 lg/ml. Cells were washed with 1 ml PBS and taken up in 500 ll of TRIzol (Invitrogen) starting 30 min after tetracycline addition (0 min time point), and subsequently at time points indicated in each figure. For analysis of knockdown of endogenous hUpf1, hUpf2, and eIF4AIII, 0.2 lg of the plasmid pSUPERpuro was co-transfected instead of the plasmids expressing b-globin mRNA, and cells were treated and harvested as described earlier [66]. Total cellular RNA was isolated and analyzed by Northern blots as described earlier [47]. The anti-sense RNA probe used for b-globin mRNA detection was described earlier [47]. Northern blots for exogenously expressed TPI mRNAs were probed using UltraHyb reagent following the manufacturer's protocol (Ambion), with a short anti-sense RNA probe complementary to the bovine growth hormone 39 UTR sequence encoded from the pcDNA3 plasmid. GPx1 mRNAs were probed as described earlier [18].
Co-IPs between wild-type or KAKA-mutant eRF3 and eRF1 were performed as described above from the cells co-transfected with 0.5 lg of plasmids expressing FLAG-tagged proteins (eRF3, eRF3-KAKA, or MS2 as control), 0.5 lg of plasmids expressing Myc-eRF1, and 0.1 lg of Myc-hnRNP A1 expressing plasmid. Endogenous eRF3 IPs ( Figure 6A) were performed as described above except that ;2.5 3 10 7 HeLa cells were lysed in 1 ml hypotonic gentle lysis buffer, and the lysates were incubated with 10 lg of anti-eRF3 rabbit polyclonal antibody (Abcam), or rabbit pre-immune serum as control, preconjugated to 5 mg of protein-A sepharose beads (GE Healthcare).
In vitro competition assay. Approximately 10 7 HEK293T cells from a 10-cm plate expressing Myc-eRF3, or Myc-peptide as a negative control, were lysed in 1 ml hypotonic gentle lysis buffer as described above. The RNase A-treated, cleared extracts were subsequently incubated with 40 ll anti-Myc resin (Sigma) at 4 8C for 2-3 h, following which the beads were washed eight times with 1 ml of NET-2 buffer. The beads were divided into eight equal parts, and indicated amounts of FLAG-hUpf1, FLAG-PABP1, or FLAG-hnRNP A1 proteins, which had each been affinity-purified from RNase Atreated HEK293T cell extracts (protein concentrations estimated by comparison in anti-FLAG Western blot to a GST-FLAG fusion protein of known concentration), were incubated in 50 ll of NET-2 supplemented with 0.1 mg/ml BSA and 0.2 mg/ml FLAG peptide. The reactions were gently shaken at 4 8C for 2-3 h following which the beads were washed eight times with 1 ml of NET-2 buffer. The beads were resuspended in 25 ll of SDS-loading buffer (10 mM Tris-HCl [pH 6.8], 2% SDS, 10% glycerol, 0.5% bromophenol blue, and 50 mM DTT), and 10 ll of the protein sample was resolved on SDS-PAGE followed by Western blot analysis using anti-FLAG M2 antibody (Sigma, 1:1,000 dilution). Figure S1. Knockdown Efficiencies of NMD and EJC Factors Western blots showing the efficiency of knockdown of hUpf1 (lanes 5 and 11), hUpf2 (lane 6), and eIF4AIII (lane 12). Protein levels are compared to 100%, 50%, 25%, and 10% of cell extract from cells expressing an siRNA against F-Luciferase (lanes 1-4 and 7-10). hUpf3b levels served as a loading control. Found at doi:10.1371/journal.pbio.0060111.sg001 (1.4 MB TIF). Figure S2. An Extended 39 UTR in TPI mRNA Triggers NMD mRNA decay assays showing decay rates of TPI mRNAs with different length 39 UTRs, due to insertion of a fragment of GAPDH mRNA (see schematics below), in human HeLa Tet-off cells co-expressing siRNAs targeting hUpf1 or Luciferase (Luc; as a control) as indicated. Constitutively expressed bwt mRNA was used as an internal control for quantification. Numbers indicated above the panels indicate time after transcriptional repression. Schematics on the bottom show the used construct with TPI exons indicated as light-gray bars (not to scale), introns as lines, and GAPDH sequences as dark-gray bars. PTC(189) refers to a PTC at codon 189. Numbers on the right indicate mRNA half-lives (t 1/2 ; in minutes) calculated from the shown experiment with the average fold increase and standard deviation over the half-life of TPI-189Di6-GAP mRNA in the presence of Luc siRNA calculated from three or more experiments given in parentheses below. Found at doi:10.1371/journal.pbio.0060111.sg002 (5.1 MB TIF). Figure S3. bGAP mRNA Is Stabilized by Knockdown of the EJC Factor eIF4AIII mRNA decay assays showing decay rates of the b-globin-derived bGAP mRNA with an extended 39 UTR in human HeLa Tet-off cells knocked down (using RNAi) for eIF4AIII, or as a negative control, Luciferase (Luc), as indicated on the left of each panel (the knockdown efficiency for eIF4AIII is shown in Figure S1). Constitutively expressed bwt mRNA was used as an internal control for quantification. The mRNA half-lives are shown on the right, and the average fold increase in comparison to the Luc control is given with standard deviation in parentheses below. Found at doi:10.1371/journal.pbio.0060111.sg003 (1.6 MB TIF).