Nonsense mutations in close proximity to the initiation codon fail to trigger full nonsense-mediated mRNA decay.

Nonsense-mediated mRNA decay (NMD) is a surveillance mechanism that degrades mRNAs containing premature translation termination codons. In mammalian cells, a termination codon is ordinarily recognized as "premature" if it is located greater than 50-54 nucleotides 5' to the final exon-exon junction. We have described a set of naturally occurring human beta-globin gene mutations that apparently contradict this rule. The corresponding beta-thalassemia genes contain nonsense mutations within exon 1, and yet their encoded mRNAs accumulate to levels approaching wild-type beta-globin (beta(WT)) mRNA. In the present report we demonstrate that the stabilities of these mRNAs with nonsense mutations in exon 1 are intermediate between beta(WT) mRNA and beta-globin mRNA carrying a prototype NMD-sensitive mutation in exon 2 (codon 39 nonsense; beta 39). Functional analyses of these mRNAs with 5'-proximal nonsense mutations demonstrate that their relative resistance to NMD does not reflect abnormal RNA splicing or translation re-initiation and is independent of promoter identity and erythroid specificity. Instead, the proximity of the nonsense codon to the translation initiation AUG constitutes a major determinant of NMD. Positioning a termination mutation at the 5' terminus of the coding region blunts mRNA destabilization, and this effect is dominant to the "50-54 nt boundary rule." These observations impact on current models of NMD.

Nonsense-mediated mRNA decay (NMD) 1 is an mRNA surveillance mechanism that rapidly degrades mRNAs carrying premature translation termination codons (1). Nonsense-containing mRNAs targeted by NMD can be generated by naturally occurring frameshift and nonsense mutations, splicing errors, leaky 40 S scanning, and utilization of minor AUG initiation sites (2,3). A major function of the NMD pathway is to block the synthesis of truncated proteins that could have dominant negative effects on cell function (2,4).
Recent studies have shown that the NMD pathway in mammalian cells is linked to splicing-dependent deposition of a protein complex 20 -24 nucleotides (nt) 5Ј of each exon-exon junction (exon-junction complex; EJC). The EJC contains the general splicing activator RNPS1, the RNA export factor Aly/ REF, the shuttling protein Y14, the nuclear matrix-localized serine-arginine-containing protein SRm160, the oncoprotein DEK, and the Y14 binding protein magoh. The interaction of magoh with Y14 may have a role in cytoplasmic localization of mRNAs and in anchoring the NMD-specific factors Upf3 and Upf2 to the mRNA (5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18). Previous published data have shown that Upf3 and Upf2 join the EJC in different subcellular compartments: Upf3 (Upf3a and Upf3b) is loaded onto mRNAs in the nucleus during splicing via interactions with components of the EJC. In contrast, Upf2 joins the complex soon after cytoplasmic export is initiated (14,19,20). According to the present models, translating ribosomes displace EJCs from the open reading frame (ORF) during the "pioneer" round of cytoplasmic translation (21). If, however, the mRNA contains a premature termination codon located more than 50 -54 nt 5Ј of at least one EJC, complex components 3Ј to the termination mutation will remain on the mRNA. The retention of one or more EJCs on the mRNA triggers the NMD response by an as yet undefined mechanism. According to current models, recognition of the nonsense codon as "premature" involves a direct interaction of the phosphorylated Upf1 RNA helicase with EJC-bound Upf2 (20,22,23). Interactions between Upf1 and other Upf factors may also be involved in this process (14,19,20,(23)(24)(25)(26). The interactions of the translation termination factors eRF1 and eRF3 with Upf1 and of eRF3 with Upf2 and Upf3 appear to provide a mechanistic link between NMD and translation termination at the premature stop codon (reviewed in Refs. 1 and 27).
Studies supporting the "50 -54 nt boundary rule" suggest that an exon-exon junction serves as the major cis-acting NMDregulatory element (28 -30). Although EJC-regulated NMD is supported by multiple lines of evidence, exceptions to the 50 -54 nt boundary rule have been reported. These exceptions suggest that additional determinants may be involved in determining the net stability of nonsense-containing mRNAs. For example, nonsense codons positioned close to the initiation AUG of the TPI, immunoglobulin heavy chain, and BRCA1 mRNAs, fail to specify NMD (31)(32)(33). For the TPI and immunoglobulin heavy chain mRNAs, NMD appears to be circumvented by re-initiation of translation 3Ј to the nonsense codon. This mechanism of NMD avoidance has also been speculated for nonsense codons located early in the BRCA1 transcript (33).
Such re-initiation would allow the ribosome to disrupt EJCs located 3Ј to the nonsense mutation (31,32). Nonsense codons in the T-cell receptor-␤ transcript also fail to elicit canonical NMD; nonsense codons located more 5Ј in the mRNA trigger robust NMD, whereas nonsense codons closer to the terminal exon-exon junction induce a much weaker NMD response (34). Finally, premature termination codons located in different exons of the fibrinogen A␣-chain gene or at exon 2 of the human ALG3 gene can avoid NMD (35,36). The basis for NMD avoidance by the fibrinogen A␣-chain and ALG3 mRNAs remains undefined. Thus, present evidence indicates that determinants of NMD, at least in some systems, are more complex than would be predicted on the basis of the single 50 -54 nt boundary rule.
We have previously reported that human ␤-globin mRNAs containing naturally occurring nonsense mutations in the 5Јpart of exon 1 accumulate to levels similar to those of normal ␤-globin transcripts (37). The aim of the present study is to investigate the mechanism by which these mutant ␤-globin transcripts circumvent the full impact of NMD. Our results reveal that this unusual NMD behavior reflects the proximity of the nonsense mutations to the AUG initiation codon. Remarkably, the impact of this 5Ј localization is dominant to the 50 -54 nt boundary rule.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-The plasmids containing the normal (␤ WT ), ␤15 [CD 15 (TGG3 TGA)], or ␤39 [CD 39 (C3 T)] human ␤-globin genes were previously described (37). All variants were created within the 428-bp NcoI-BamHI fragment of the ␤-globin gene template by overlapextension PCR (37). PCR reactions were performed by using Invitrogen primers (Barcelona, Spain; see Table I). To create variants carrying DelA (deletion of 69 bp between codons 3 and 27, exclusively), the PCR reaction 1 was performed by using L5Ј␤-S and DelA-AS primers, and DNA template from the previously cloned ␤ WT gene. PCR reaction 2 was performed by using DelA-S and R5Ј␤-AS primers and the ␤ WT or ␤39 genes as template. To clone variant ␤ WT seqhet and ␤39seqhet constructs, which contain a different sequence between codons 3 and 27, from the second exon of the human ␣2-globin gene (from codon 34 through codon 56), PCR reaction 1 was performed by using Seqhet-S and Seqhet-AS primers, and the h␣2-globin gene as the DNA template (pTet-Wt detailed in Ref. 38). PCR reaction 2 was performed by using 3Ј exon1-S and 3Ј exon2-AS primers and the ␤ WT or ␤39 gene DNAs. To obtain ␤62 and ␤62DelA variants carrying the nonsense mutation (GCT 3 TAG) at codon position 62, PCR reaction 1 was performed with ATG-S and 62-AS primers and the previously cloned ␤ WT or DelA gene DNAs. PCR reaction 2 was performed with 62-S and 3Ј exon2-AS primers and the ␤ WT gene as DNA template. To clone the double nonsense mutated gene ␤15non␤39, carrying both nonsense mutations at codons 15 and 39, PCR reaction 1 was performed by using L5Ј␤-S and 15non39-AS primers and the ␤39 gene DNA. PCR reaction 2 was performed by using 15non39-S and R5Ј␤-AS primers and the same DNA   (37). To obtain plasmids used to transiently transfect HeLa cells, a human ␤-globin gene fragment extending from the transcription initiation site to 200 bp 3Ј to the polyadenylation site was amplified by PCR from the p158.2 plasmid, using LinkClaI-S and LinkNotI-AS primers ( Table I). The 1806-bp ClaI-NotI human ␤-globin fragment was sub-cloned into pTet2 plasmid, creating the pTet2-␤ WT plasmid (pTet2 plasmid is derived from pTet-Splice vector (Invitrogen) by deleting a 126-bp fragment upstream of the ClaI site; the deletion was introduced by substituting the native 500-bp XhoI/ClaI fragment by the 374-bp DNA fragment that was amplified using pTet-XhoI1/15-S and pTetClaI351/370-AS primers (Table I)). The human ␤-globin variants ␤15, ␤17, ␤39, ␤62, DelA, ␤39DelA, ␤62DelA, ␤ WT seqhet, ␤39seqhet, or ␤15non␤39 were cloned into the pTet2-␤ WT vector previously obtained, by substituting normal NcoI-BamHI fragment for the corresponding mutated fragment, which had been previously cloned into the p158.2 vector (see above). To obtain genes ␤ WT (15: 39) and ␤15 (15:39), which present the original codon 15 at position 39, initially the restriction sites HindIII-ApaI were introduced between codons 14 and 15 into the NcoI-BamHI fragment of the ␤ WT and ␤15 genes, using the QuikChange TM site-directed mutagenesis kit as before.
Reactions were performed with the following mutagenic primers: WT15:39-S and WT15:39-AS (for ␤ WT ) or 15:39-S and 15:39-AS (for ␤15) ( Table I). After to obtain the correct mutagenized ␤ WT and ␤15 clones, a 60-bp DNA sequence of the ampicillin resistance gene (Amp r ; 5Ј-TA-CATCGAACTGGATCTCAACAGCGGTAAGATCCTTGAGAGTTTTCG-CCCCGAAGAACGT-3Ј) was introduced in both clones (␤ WT and ␤15) between HindIII/ApaI sites. The sequence was previously obtained by PCR using HindIIIAmp r -S and ApaIAmp r -AS primers (Table I), using the pGEM-3 plasmid (Promega) as DNA template. The cloned NcoI-BamHI fragments were transferred to the pTet2-␤ WT vector by substituting the normal NcoI-BamHI fragment. Because the introduced DNA Amp r spacer contained one splicing donor site with high consensus value (sequence CGGTAAGA), an additional mutagenesis (T 3 A) was performed to inhibit the alternative splicing. The mutagenesis was carried out in both constructs ␤ WT (15:39) and ␤15 (15:39), with the QuikChange TM site-directed mutagenesis kit using the GA-S and GA-AS oligonucleotides ( Table I). The pTet plasmid containing the human ␣-globin gene (pTet-␣wt) used as a control for transient transfection efficiency was previously described (39). Cell Culture and Transfections-Mouse erythroleukemia (MEL) cells were cultured in RPMI medium with glutamax-1 (Invitrogen), supplemented with 10% fetal bovine serum. Stable transfections were carried out with 40 g of linearized plasmid DNA mixed with 2 g of linearized pGKpuro (plasmid containing the puromycin resistance gene). Electroporations were performed using 3 ϫ 10 7 cells per transfection, as previously described (37). Each electroporated cell pool was expanded in selective medium during ϳ10 -12 days, by adding puromycin (Invitrogen) to 2.5 g/ml. Erythroid cell differentiation was induced by adding 2% (v/v) Me 2 SO to the media of each transfection cell pool. Following 5 days, cell pools were harvested for RNA extraction. MEL cells stably expressing the tet transactivator (MEL/tTA; described in Ref. 39) were used for conditional expression of h␤-globin genes (previously cloned into the pTet vector). For transient transfections, MEL/tTA cells were split 1 day before transfection and cultured in minimal essential medium supplemented with 10% fetal bovine serum and 100 ng/ml tetracycline. Cells were washed three times in cold Dulbecco's phosphatebuffered saline and resuspended in cold serum-free minimal essential medium at a concentration of 3 ϫ 10 7 cells per ml. Two micrograms of pTet2-␤ WT or each pTet2-␤ variant were added to the cell suspension and mixed thoroughly. Electroporation was performed at 250 V, 1.180 microfarads, and low resistance in a Invitrogen Cell-Porator system. For steady-state mRNA quantification, cells were harvested after a 12-h transcription pulse. For mRNA stability analysis, cells were split into four 60-mm diameter dishes. Cells were pulsed with ␤-globin mRNA for 12 h by growth in Tet Ϫ media. Following this period, transcription from the plasmid was then blocked by the addition of Tet to the media. Cells from each culture dish were harvested in different time points for further analysis.
The adherent HeLa cells, stably expressing the tet transactivator (HeLa/tTA; described in Ref. 39), were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. Transient co-transfections were performed with the pTet-␣wt (to control for transfection efficiency) and the appropriate pTet2-␤ plasmid, using Lipofectin Plus reagent (Invitrogen), following the manufacturer's instructions, using 1 ϫ 10 5 cells per transfection and 1 g of each plasmid. Cells were harvested after 1 day transcription pulse.
RNA Isolation-Total RNA from transfected cells was prepared using the RNeasy total kit (Qiagen) following the manufacturer's instructions. Before analysis, RNA samples were treated with RNase-free DNase I (Ambion) and purified by phenol:chloroform extraction.
Primer Extension Analysis-Primer extension assays were performed using 5Ј-end-labeled primers: H␤exon3-AS or H␤3ЈUTR-AS and M␣-AS (Table I). Both primers were mixed, hybridized with 10 g of MEL RNA at 50°C during 5 h, and extended with reverse transcriptase Superscript II (Invitrogen), as described previously (37). Primer extension products were analyzed by phosphorimaging, using a Typhoon® Imager 8600 (Molecular Dynamics) with a 6-h exposure period. One, ten, and fifteen micrograms of RNA extracted from cells transfected with the normal ␤-globin gene were also transcribed to show that the experiment was carried out in the presence of primer in excess. The intensity of each band was quantified using the Image Master Phoretix® Software (Amersham Biosciences).
Ribonuclease Protection Assays-Probes used were generated by in vitro transcription of plasmids containing DNA fragments from h␤globin, h␣-globin (40), m␣-globin (40), or mGAPDH (Ambion). The h␤globin probe is a 170-bp fragment encompassing the last 20 bp of intron 2, the entire exon 3 coding region, and first 21 bp of 3ЈUTR, which was generated by PCR and inserted into the polylinker region of pGEM3. Amplification primers contained restriction sites that facilitated insertion into the vector polylinker. To generate RNA probes, each transcription vector was linearized and transcribed in the presence of [␣-32 P]CTP (400 Ci/mmol, 10 mCi/ml; Amersham Biosciences) using a Maxiscript SP6 kit, under conditions recommended by the manufacturer (Ambion). Final concentrations of ATP, GTP, and UTP were 0.5 mM and that of CTP was 0.06 mM. One microgram of RNA was added to 20 l of hybridization buffer (40 mM PIPES, pH 6.4, 1 mM EDTA, pH 8.0, 0.4 M NaCl, 80% formamide) supplemented with probes. Samples were heated at 80°C for 10 min, incubated overnight at 50°C, and digested for 15 min at room temperature in 200 l of RNase assay buffer (300 mM NaCl, 10 mM Tris-HCl, pH 7.5, 5 mM EDTA, pH 8.0) containing 1 l of RNase mixture (500 mg of RNase A/ml, 20,000 units of RNase T1/ml; Ambion). Digestions were terminated by addition of 18 l of an SDSprotease K (10%:2 mg/ml) mixture to each sample followed by incubation 20 min at 37°C. RNA was extracted, precipitated, dissolved in loading buffer, and resolved onto a 6 or 8% acrylamide 8 M urea gel. Radioactivity in bands of interest was quantified by phosphorimaging, using a Typhoon® Imager 8600.
RT-PCR Analysis-Two micrograms of each total RNA sample was mixed with 2 pmol of H␤3ЈRNA-AS primer ( Table I) Table I), and 0.3 l of enzyme. 10-l aliquots from each RT-PCR sample were analyzed by electrophoresis on 1% agarose gels.
Sucrose Gradient Polysome Fractionation-Polysome gradients were carried out as described previously (38). In brief, transiently transfected MEL cells were centrifuged at 1,000 ϫ g for 5 min at 4°C. The cell pellet was washed with ice-cold phosphate-buffered saline, lysed in TMK100 buffer (10 mM Tris-HCl (at pH 7.4), 5 mM MgCl 2 , 100 mM KCl, 2 mM DTT, 1% Triton X-100, and RNase inhibitor 100 units/ml; Promega), the nuclei were cleared at 10,000 ϫ g for 10 min at 4°C, and the supernatants (S10) were removed and layered onto the prepared 10 -50% linear sucrose gradients and centrifuged at 40,000 rpm for 85 min at 4°C. The gradients were collected in 15 fractions from the top to the bottom by displacing them upwards with 60% sucrose, and the gradient profile was monitored via UV absorbance at 254 nm with an ISCO UA-5 detector (ISCO, Lincoln, NE). RNA was extracted from sucrose fractions and ethanol-precipitated. mRNA content was determined by RPA, using probes generated by in vitro transcription of plasmids containing genomic DNA inserts for h␤-globin (38) and mGAPDH (Ambion). The final protected fragments were resolved on a 6% acrylamide-8 M urea gel. Radioactivity in bands of interest was quantified by phosphorimaging analysis (Storm 840, Amersham Biosciences).

RESULTS
␤15 and ␤17 mRNAs Are More Stable Than the NMD-targeted ␤39 mRNA-We have previously reported that a set of thalassemic h␤-globin mRNAs bearing nonsense mutations in exon 1 (␤5, ␤15, and ␤17, respectively) accumulate to steadystate levels in vivo that are significantly higher than the ␤-globin mRNA with a prototype NMD-sensitive nonsense mutation (␤39) (37,41). This observation suggested that these nonsensecontaining ␤-globin mRNAs avoid the full impact of NMD. To firmly establish that this is the case, the half-lives of two of the ␤-globin mRNAs with nonsense mutations in exon 1, ␤15 and ␤17, were determined and compared with ␤ WT and ␤39 mRNAs. Each ␤-globin gene (normal or mutant) was cloned behind a tet-controlled promoter and transfected into a MEL cell line that stably expresses the tet-transcriptional transactivator (MEL/tTA cells) (39). The cells were transcriptionally pulsed for 12 h with each ␤-globin mRNA, and the corresponding mRNA decay rate was subsequently established. An example of this study is shown in Fig. 1A, and a compilation of four independent experiments is shown in Fig. 1B. The half-life (t1 ⁄2 ) of normal ␤-globin mRNA (␤ WT ) in this system is 10 h, and that of the NMD-sensitive ␤39 mRNA is 3.5 h. These half-lives are in agreement with a recent study by Couttet and Grange (42). The half-lives of the ␤15 and ␤17 mRNA were similar to each other and were intermediate to the ␤ WT and ␤39 mRNAs; t1 ⁄2 of ␤15 mRNA is ϳ6.3 h, and t1 ⁄2 of ␤17 is ϳ5.8 h (Fig. 1B). These half-lives are consistent with the steady-state levels of ␤15 and ␤17 mRNAs in erythroid cells (37) and suggest that h␤-globin mRNAs with nonsense-mutations located in close proximity to the initiation AUG can escape the full impact of NMD.
Close Proximity of Nonsense Mutations to the Initiation AUG Minimizes NMD of h␤-Globin mRNA-The above data confirm that the ␤15 and ␤17 mRNAs are relatively resistant to NMD. We hypothesized that resistance might reflect the positioning of the nonsense codon relative to the initiation AUG. To test this hypothesis, a deletion was introduced into the ␤39 mRNA that removed 69 bp between codons 3 and 27 (␤39DelA). This deletion relocates the ␤39 nonsense mutation to a position 16 codons from the AUG while leaving the mutation within exon 2 and preserving the distance between the ␤39 nonsense codon and the terminal (exons 2-3) splice junction (195 nt) ( Fig. 2A). MEL cells were stably co-transfected with equimolar amounts of ␤39DelA plus ␤39 or ␤ WT DelA plus ␤ WT , and the levels of the corresponding mRNAs were determined after the MEL cells were stimulated to terminally differentiate (37) (Fig. 2B). The results from three independent experiments revealed that the DelA deletion increased the level of ␤39 mRNA by 7.3-fold. This increase was significantly greater than the 1.8-fold increase in the level of ␤ WT mRNA. Thus, shortening the distance between the ␤39 mutation and the AUG resulted in a significant and preferential mRNA stabilization.
The stabilizing effect of DelA on the ␤39 mRNA was confirmed by a second approach. MEL cells were stably transfected with each individual construct, the cells were induced to erythroid differentiation, and the steady-state level of each mRNA was assessed relative to an internal control (mouse (m) ␣-globin mRNA) (Fig. 2C). Results from four independent experiments indicate that ␤39DelA transcripts accumulate at about 84% of ␤ WT DelA transcript levels, whereas ␤39 transcripts show levels of accumulation at ϳ9% of ␤ WT (Fig. 2C). The studies support the conclusion that NMD is significantly abrogated when ␤-globin nonsense mutations are in close proximity to the initiation AUG.
The half-life of ␤39DelA mRNA Is Similar to That of ␤15 or ␤17 mRNAs-Based on steady-state mRNA levels (Fig. 2), we hypothesized that DelA stabilizes the NMD-sensitive ␤39 mRNA. To directly test this prediction, ␤39 and ␤39DelA genes were placed under control of the tetracycline-regulated promoter. MEL/tTA cells transfected with the ␤39 and ␤39DelA genes were transcriptionally pulsed for 12 h and the decay rates of the ␤-globin mRNAs were monitored over time (Fig.  3A). Results from three independent experiments revealed a t1 ⁄2 of 5.2 h for ␤39DelA mRNA (Fig. 3B). This half-life is equivalent to that determined for the ␤15 or ␤17 mRNAs and significantly longer than that determined for the ␤39 mRNA. These data confirm that DelA stabilizes the ␤39 mRNA.
The Relative Stability of ␤-Globin mRNAs with 5Ј-Proximal Nonsense Mutations Is Independent of Promoter and Tissue Specificity-Work from others suggests that the decay rate of nonsense-containing ␤-globin mRNAs can be impacted by an erythroid environment (43). For this reason, we next compared the NMD profiles established in MEL cells (Figs. 2 and 3) with those in HeLa cells. HeLa cells were transiently co-transfected with equimolar amounts of h␣-and ␤39DelA-globin genes under control of the CMV promoter. Twenty-four hours posttransfection, RNA was quantified by RPA using probes specific for the human ␣and ␤-globin mRNAs. The mRNA levels of ␤ WT , ␤15, ␤39, and ␤ WT DelA genes were assessed in parallel (Fig. 4). Results from three independent experiments revealed accelerated decay of ␤39 mRNA in non-erythroid cells; levels of ␤39 mRNA were 37% of ␤ WT mRNA. In contrast, mRNA with a 5Ј proximal nonsense mutation (␤15) accumulated to the same level as ␤ WT . Furthermore, deletion of sequences between codons 3 and 27 (DelA) of ␤39 mRNA increased its accumulation in HeLa cells from 37% to 86% of ␤ WT . Thus ␤15 mRNA evades NMD in the non-erythroid (HeLa) as well as erythroid (MEL) cells, whereas ␤39 is NMD-sensitive in both settings and is stabilized by DelA. Because the stabilizing impact of DelA on ␤39 mRNA is equivalent in MEL and HeLa cells, using either the homologous ␤-globin promoter or the heterologous CMV promoter in erythroid and non-erythroid cells, the rela-  Table I) were hybridized with RNA, and extended with reverse transcriptase (see "Experimental Procedures"). Human reticulocyte RNA (hRet) and untransfected differentiated MEL mRNA (t-) were also reverse-transcribed to mark the positions of h␤and m␣-globin cDNAs, respectively (indicated on the right). The mean phosphorimaging quantification of the h␤globin mRNA (h␤), normalized to the level of endogenous m␣-globin mRNA (m␣) and calculated as a percentage of the corresponding nonsense-free transcript (from four independent experiments), is shown at the bottom of the respective lanes. The standard deviations (S.D.s) are indicated.
tive impact of NMD on the various mutant mRNAs appears to be independent of promoter as well as tissue specificity. Taken together, these data substantiate the stability of the 5Ј nonsense mutations and the stabilizing effect of deleting codons 3-27 from ␤39 mRNA.
Stabilities of ␤39DelA and 5Ј-Proximal Nonsense mRNAs Do Not Reflect Activation of Abnormal Splicing Pathways-Mechanisms that might contribute to stabilization of the ␤-globin mRNAs with 5Ј-proximal nonsense mutations were explored. We initially tested the possibility that these mutations activated cryptic splicing pathway(s) with consequent alteration in mRNA sequence and circumvention of the premature termination codon. This possibility was of particular interest in light of the numerous cryptic splice-donor sites in exon 1 that can be activated in vivo (44 -47). Representative h␤-globin genes were expressed in transfected MEL cells, and the encoded mRNAs were analyzed by RT-PCR, using a set of primers that encompass the full-length ␤-globin mRNA (Fig. 5A) (see "Experimental Procedures"). A single full-length product of 625 bp was amplified from all non-deletional transcripts and a 556-bp fragment was amplified from cDNAs carrying DelA. In all cases, the cDNAs that were amplified correspond to the expected sizes of normally spliced ␤-globin mRNA. These results were confirmed by amplifying 5Ј and 3Ј halves of the h␤-globin cDNA from mRNA transiently expressed in HeLa cells ( Fig. 5B; see "Experimental Procedures"). As expected, the 5Ј half of the cDNA generated a product of 351 bp for ␤ WT , ␤39, and ␤15 transcripts and a fragment of 282 bp corresponding to the ␤ WT DelA and ␤39DelA mRNA samples. Analysis of the 3ЈcDNA fragment revealed the amplification of a 359-bp product in all samples (Fig. 5B). These results demonstrate normal splicing patterns for the transcripts containing the various nonsense and deletion mutations.
␤39DelA and 5Ј-Proximal Nonsense mRNAs Fail to Undergo Appreciable Levels of Translation Re-initiation-NMD can be abrogated if translation re-initiates downstream of the nonsense codon (31). With this in mind, we investigated whether translational re-initiation was occurring on the ␤-globin mRNAs with 5Ј-proximal nonsense mutations. Polysome distributions were determined for ␤ WT , ␤15, ␤39, ␤ WT DelA, and ␤39DelA mRNAs expressed in transiently transfected MEL cells (Fig. 6, A and B). Data revealed a normal polysome profile for the ␤ WT mRNA (ORF 146 codons) with a mean polysome size of four ribosomes. The mean polysome size for the ␤ WT DelA mRNA was three ribosomes, a size consistent with its slightly shortened ORF (123 codons). In contrast, ␤15 and ␤39 mRNAs were both limited to a mean polysome size of 1-1.5 ribosomes and ␤39DelA mRNA peaked at a mean polysome size of one ribosome (Fig. 6C). These results support the conclusion that ␤15, ␤39, and ␤39DelA mRNAs terminate translation at their respective nonsense codons without appreciable levels of downstream re-initiation.
The question of translational re-initiation was further addressed by a second approach. If residual re-initiation, even at low levels, did occur and was able to circumvent NMD, this effect should be inhibited by inactivating putative initiating AUG codons located 3Ј to the nonsense codon. The most proximal h␤-globin AUGs in a favored Kozak consensus (48) are at positions 55 (in-frame) and 73/74 (out of frame). A third AUG codon at 63/64 is in a less optimal sequence context (49). All three putative re-initiation AUGs were mutated in cis within the ␤ WT , ␤15, ␤39, and ␤39DelA genes; Met-55 was converted to isoleucine (AUG 3 AUA), in cis to AUG 3 ACG substitutions at positions 63/64 and 73/74. HeLa cells were transiently transfected with each of these constructs or with the parental ␤ WT , ␤39DelA, ␤15, and ␤39 constructs. The expression of each ␤-globin mRNA was normalized to the expression of a cotransfected h␣-globin gene (Fig. 6D). The results demonstrate that the combined array of missense mutations introduced at codons 55, 63/64, and 73/74 failed to destabilize the ␤15 or the ␤39DelA mRNAs. These data, along with the polysome profiles, lead us to conclude that the stability of the ␤15 mRNA and the stabilization of the ␤39 mRNA by the DelA deletion, do not reflect translation re-initiation.
The Proximity of Nonsense Codons to the Translation Initiation AUG Is a Major Determinant of NMD-The stabilization of the ␤39 mRNA by DelA could be explained by two models. The first model reflects the repositioning the ␤39 mutation closer to the initiation codon. The second model reflects deletion of a cis element within the DelA segment that is important to the NMD pathway. The second model might also account for the relative stabilities of ␤5, ␤15, and ␤17 mRNAs if each of the corresponding base substitutions altered the structure of the putative NMD determinant. The merit of the second model was tested by substituting the entire sequence between codons 3 and 27 in the ␤39 with a "neutral" sequence corresponding to exon 2 of the h␣2-globin gene. The resulting ␤39seqhet gene and a control ␤ WT seqhet gene were separately stably transfected in MEL cells. Following erythroid induction, h␤-globin mRNA levels were quantified relatively to the endogenous m␣-globin (Fig.  7A). Results from four independent experiments revealed that ␤39seqhet and ␤39 mRNAs accumulate to approximately the same levels when compared with the corresponding ␤ WT (31 and 28%, respectively) (Fig. 7A). These data suggest that sequences between codons 3 and 27 in the ␤-globin mRNA do not contain an NMD-relevant determinant.
The lack of support for a NMD determinant in the first exon led us to consider in further detail a model that proposes that the strength of NMD is impacted by the proximity of the nonsense mutation to the initiation AUG. This proximity effect was tested by introducing DelA in a second NMD-committed ␤-globin mRNA, ␤62 (29). In this case DelA moves the ␤62 nonsense codon to a position 39 codons from the AUG; i.e. the same distancing as in NMD-sensitive ␤39 mRNA. The ␤62 and ␤62DelA constructs were stably transfected into MEL cells in parallel with ␤ WT , ␤15, or ␤39 genes. Transfected cell pools were induced to differentiation, and RNA levels were quantified in four experiments (Fig. 7A). The ␤62 transcripts are expressed at about 18% of normal and ␤62DelA mRNA accumulates at about 15% of normal, and both of these transcripts have NMD sensitivity similar to ␤39 (Fig. 7A). These data support the conclusion that DelA stabilizes the ␤39 mRNA by bringing the nonsense codon into critical proximity with the AUG rather than by removing an NMD determinant.
The relationship between the AUG proximity of nonsense mutations to the strength of NMD was confirmed in non-erythroid cells. HeLa cells expressing the tet transactivator (HeLa/ tTA (39)) were transiently co-transfected with each of seven ␤-globin constructs (␤ WT , ␤ WT seqhet, ␤15, ␤39, ␤39seqhet, ␤62, and ␤62DelA) under control of the tetracycline-regulated promoter. Cells were cultured in the absence of tetracycline for 24 h to induce expression, and RNA levels were subsequently quantified. The expression of each ␤-globin gene was normalized to a co-transfected h␣-globin gene. The results of three studies are summarized in Fig. 7B. When compared with the level of ␤ WT mRNA, the ␤15 mRNA is expressed at 105%; ␤39 mRNA at 38%; and ␤62 and ␤62DelA mRNAs at 44 -49% of normal. The ␤39seqhet is expressed at 40% of ␤ WT seqhet. These data parallel those in MEL cells and reinforce the conclusion that DelA stabilizes a nonsense-containing mRNA by repositioning a nonsense codon close to the initiation AUG.
As an additional test of the relationship of AUG proximity to NMD we increased the distance between the ␤15 nonsense codon and the AUG. A heterologous 24-codon spacer was inserted between codons 14 and 15 of ␤15 gene. As a control, the same insertion was made in the ␤ WT gene. This insertion moves the ␤15 nonsense mutation to a position 39 codons from the containing the human ␤ WT -globin gene. The lower gel shows the analysis between exon 2 and exon 3; a single 359-bp product is observed for each construct. The 1209-bp fragment was amplified from plasmid DNA containing the human ␤ WT -globin gene. Plasmid DNA (pDNA) and water comprise the positive and negative RT-PCR controls. M represents the 100-bp DNA ladder marker. Primers localization (Table I) (Table I)  AUG. HeLa/tTA cells were co-transfected with each of these hybrid constructs and the h␣-globin gene. Cells were grown for 24 h in the absence of tetracycline, and RNA levels were subsequently quantified (Fig. 7B). These studies reveal that ␤ WT (15:39) mRNA accumulates at ϳ112% of normal (␤ WT ) and ␤15(15:39) mRNA accumulates at ϳ37% of ␤ WT . These results further support the role of AUG proximity in NMD by demonstrating that increasing the distance between the ␤15 mutation and the AUG results in mRNA destabilization.
Finally, we analyzed if the impact of AUG proximity on NMD is dominant over the general 50 -54 nt boundary rule. Here, we introduced the ␤15 mutation in cis to the ␤39 (construct ␤15non␤39). This gene was stably transfected in MEL cells, and the level of accumulation was determined after induction of differentiation. We find that the ␤15non␤39 mRNA accumulates to approximately the same level as the ␤15 mRNA (respectively, 63 and 60% of ␤ WT ; Fig. 7A). Concordant results were also obtained in HeLa cells, where ␤15non␤39 mRNA is at about 97% of ␤ WT (Fig. 7B). These data unequivocally demonstrate that the effect of the proximity of the nonsense mutation to the AUG is a dominant determinant in the inhibition of the NMD pathway. DISCUSSION Termination codons are recognized by the NMD apparatus as premature if they are located more than 50 -54 nt 5Ј to the 3Ј-most exon-exon junction (28,29). We have previously reported that human ␤-globin mRNAs bearing nonsense mutations in the 5Ј-region of exon 1 accumulate to levels similar to those of wild type ␤-globin mRNA (37). This finding is in apparent contradiction to the 50 -54 nt boundary rule, because these mutations are substantially 5Ј to the terminal exon 2-3 junction. In the present report we demonstrate that these mRNAs with AUG-proximal nonsense mutations are in fact significantly more stable than prototype NMD-sensitive ␤-globin mRNAs and that the proximity of nonsense mutation to the AUG initiation codon comprises the basis of their relative NMD resistance. Thus, AUG proximity can override the 50 -54 nt boundary rule in establishing the overall efficiency of NMD for a particular mutant mRNA.
The AUG proximity effect studied in the current report is based on the h␤-globin mRNA model. The study stems from observations originally made in patients with specific ␤-thalassemic mutations. The cell culture studies currently reported appear to accurately recapitulate our prior in vivo observations. The effect of AUG proximity on NMD mechanism in ␤-globin mRNA is independent of promoter sequence and is not restricted to erythroid cells. Thus, it is likely that these findings are not restricted to the globin genes or to the erythroid cell environment and may in fact constitute general determinants of the NMD pathway. FIG. 7. NMD resistance of human ␤-globin mRNAs carrying a 5-proximal nonsense mutation reflects the proximity of the nonsense codon to the initiating AUG. A, analysis of mRNA expression in terminally differentiated MEL cells. MEL cells were stably transfected with the h␤-globin construct specified above each lane. After erythroid induction, the normalized steady-state levels of ␤-globin mRNA were determined from either transfected or untransfected (t-) MEL cell pools by RPA using specific probes for h␤-and m␣-globin (see "Experimental Procedures"). The protected bands corresponding to the h␤-and m␣globin mRNA are indicated. The level of mRNA from each ␤-globin gene was normalized to the level of endogenous m␣globin to control for RNA recovery and erythroid induction. HeLa cells were cotransfected with the ␤-globin constructs specified above each lane and the h␣-globin gene to normalize the expression levels for transfection efficiency. The ␤-globin and ␣-globin genes are both driven by the tetracycline-regulated promoter. Twenty-four hours after transfection, steady-state total RNA from either transfected or untransfected (t-) HeLa cells was isolated, analyzed by RPA using specific probes for h␤and ␣-globin mRNAs, and quantified as indicated above. The percentage mRNA values (indicated on the bottom) were plotted for each construct, and standard deviations from three independent experiments are shown.
To our knowledge, the AUG proximity effect that we report represents a newly recognized parameter of the NMD response. However, a number of apparent exceptions to the 50 -54 nt boundary rule have been previously reported by others. The mechanisms involved in these individual cases of NMD resistance include translation re-initiation downstream of the nonsense codon (TPI and immunoglobulin heavy chain mRNAs (31,32)) and the presence of a sequence cis-acting element that confers immunity to the 50 -54 nt boundary rule (T-cell receptor-␤ mRNA (34)). In additional cases the rule is apparently circumvented, although the underlying mechanism responsible for inhibition of the NMD response remains undefined (BRCA1, fibrinogen A␣-chain, and ALG3 mRNAs (33,35,36)). We propose, based on the likelihood that the current findings are not limited to the ␤-globin mRNA, that the NMD resistance in some of these model systems reflects the AUG proximity effect.
The observed lack of cell-type specificity for NMD phenotypes in the ␤-globin mRNAs is consistent with the understanding that NMD constitutes an essential surveillance process operating in all tissues. Nonetheless, a comparison of results shown in Figs. 2C and 4 suggests that NMD of human ␤-globin mRNA may be somewhat more efficient in terminally differentiated erythroid cells when compared with HeLa cells. This may be contributed by the observation of Stevens et al. (43) that the degradation of nonsense-mutated ␤-globin mRNAs in erythroid cells is accomplished by a tissue-specific endonuclease with preference for UG dinucleotides (43). Thus, although specialized decay pathways may exist in the erythroid compartment, the general conservation of the NMD phenotypes that we observe in MEL and HeLa cells suggests that the AUG proximity effect reflects a general attribute of the NMD pathway.
The relevance of the AUG proximity of a nonsense codon to NMD is consistent with prior studies on 5Ј ORF structure in NMD in other systems. 2 For example, thrombopoietin mRNA is normally insensitive to NMD, but can be rendered NMD-sensitive by extending the length of its upstream ORF. These data led to the suggestion that mammalian mRNAs with naturally occurring short upstream ORFs may have a general resistance to NMD. 2 These data, in conjunction with our present study, would support a model in which termination of translation in close proximity to the initiation codon results in a resistance to NMD. This notion, combined with the knowledge that the NMD pathway is translation-dependent, may lead to a further proposal that the observed NMD insensitivity of mRNAs with a short 5Ј ORF, whether natural or resulting from a nonsense mutation, can mechanistically reflect ineffective and/or brief translation. In this regard, it is interesting to note that the smallest naturally occurring eukaryotic ORF that produces significant levels of protein product comprises 24 codons (51). Based on our results and those published by Zhang et al. (28), this boundary for human ␤-globin mRNAs appears to map between codons 17 and 21/22 (␤17 is the 3Ј-most mutation that fails to trigger full NMD, and the ␤21/22 mutation is the 5Ј-most mutation able to fully commit mRNA to the NMD pathway (28,37)). Thus, the reported minimal size of an ORF for effective translation roughly correlates with the boundary between NMD-resistant and NMD-sensitive mutations.
The apparent concordance between the minimal length of an ORF for effective protein synthesis and the minimal length for effective NMD may be relevant to disease pathophysiology. In the case of thalassemia, the production of truncated proteins via nonsense mutations can have a dominant-negative effect on red cell structure and function (52,53). C-terminal truncated proteins, if produced at significant levels, can block proteolytic pathways in differentiating erythroblast and/or can interfere with critical aspects of the assembly and function of hemoglobin tetramers (53). The NMD pathway is generally considered to clear cellular mRNAs that can encode such "toxic" protein fragments. However, mutant ␤-globin mRNAs with AUG-proximal nonsense mutations are relatively stable. This apparent defect in NMD surveillance would theoretically place the cell in jeopardy. Nevertheless, we note that the thalassemic phenotype of heterozygotes and homozygotes carrying AUG-proximal nonsense mutations do not appear to be any more severe than in patients carrying more distal nonsense mutations that are effectively targeted by NMD. The lack of a detectable dominant-negative effect in these patients may reflect the inefficiency with which mutant mRNAs with short 5Ј ORFs are translated. Thus, the inability of the NMD pathway to effectively clear mRNAs with 5Ј-proximal nonsense mutations may be mitigated, at least in part, by the inability of these mRNAs with very short ORFs to be effectively translated.
The relative NMD resistance of mRNAs with AUG-proximal nonsense mutations can be related to critical steps in translational biochemistry. The short ORF may result in a temporal overlap between translational initiation and termination. The ribosome may arrive at the premature termination (nonsense) codon before it has discharged its initiation factors and/or stabilized interactions with elongation/termination factors. This effect would result in interference with elongation/termination reactions that may be central to the NMD pathway. Present evidence indicates that the translation termination process is intimately linked with the NMD pathway. Analyses in other systems have shown that, during the process of translation of short upstream ORFs, interactions involving ribosomal or ribosome-associated factors, including but not limited to releasing factors, can inhibit translation elongation as well as termination (50, 54 -56). In the case of NMD, Upf1 must interact with eRF1 and eRF3 to terminate translation and, in the proper setting, to interact with additional factors to trigger NMD (reviewed in Refs. 1 and 27). Therefore, it is possible that the dynamics of the mRNP remodeling necessary for NMD cannot occur or is defective in this compressed environment of the AUG-proximal nonsense-mutated mRNAs. The present study further supports the notion that NMD is a multifaceted process that reflects the overall complexity of eukaryotic translational biochemistry.