Functional mapping of the translation-dependent instability element of yeast MATalpha1 mRNA

The determinants of mRNA stability include specific cis-acting destabilizing sequences located within mRNA coding and noncoding regions. We have developed an approach for mapping coding-region instability sequences in unstable yeast mRNAs that exploits the link between mRNA translation and turnover and the dependence of nonsense-mediated mRNA decay on the activity of the UPF1 gene product. This approach, which involves the systematic insertion of in-frame translational termination codons into the coding sequence of a gene of interest in a upf1delta strain, differs significantly from conventional methods for mapping cis-acting elements in that it causes minimal perturbations to overall mRNA structure. Using the previously characterized MATalpha1 mRNA as a model, we have accurately localized its 65-nucleotide instability element (IE) within the protein coding region. Termination of translation 5' to this element stabilized the MATalpha1 mRNA two- to threefold relative to wild-type transcripts. Translation through the element was sufficient to restore an unstable decay phenotype, while internal termination resulted in different extents of mRNA stabilization dependent on the precise location of ribosome stalling. Detailed mutagenesis of the element's rare-codon/AU-rich sequence boundary revealed that the destabilizing activity of the MATalpha1 IE is observed when the terminal codon of the element's rare-codon interval is translated. This region of stability transition corresponds precisely to a MATalpha1 IE sequence previously shown to be complementary to 18S rRNA. Deletion of three nucleotides 3' to this sequence shifted the stability boundary one codon 5' to its wild-type location. Conversely, constructs containing an additional three nucleotides at this same location shifted the transition downstream by an equivalent sequence distance. Our results suggest a model in which the triggering of MATalpha1 mRNA destabilization results from establishment of an interaction between translating ribosomes and a downstream sequence element. Furthermore, our data provide direct molecular evidence for a relationship between mRNA turnover and mRNA translation.

mRNA turnover is a regulated process that is essential to the course of gene expression and dependent on specific cis-acting sequences and trans-acting factors (28,40,44,45). In Saccharomyces cerevisiae, as in mammalian cells, one major class of sequences that regulates mRNA decay rates also promotes poly(A) shortening, a rate-limiting event for the turnover of many mRNAs (for reviews, see references 7, 10, 28, 49). Conventional mapping of such instability elements (IEs) involves the construction of chimeric genes, composed of segments encoding both stable and unstable mRNAs, and the analysis of in vivo decay rates of the resulting chimeric mRNAs. This approach in S. cerevisiae, combined with deletion and mutational analyses, has successfully localized instability determinants to the coding regions of the MAT␣1 (6,39), HIS3 (22), STE3 (19), SPO13 (56), and RPL2 (48) mRNAs, the 3Ј untranslated regions of the STE3 (19) and MFA2 (38) mRNAs, and the 5Ј untranslated regions of the PPR1 (47) and SDH2 (9) mRNAs.
Here, we report a new method for mapping coding region IEs in inherently unstable mRNAs. Development of the new approach was made possible by the identification of gene products required for nonsense-mediated mRNA decay (11,17,31,32,33,41,43) and prior demonstration of an intimate link between mRNA decay and translation. The latter is exempli-fied by the location of some IEs to mRNA coding regions (6,19,22,54), the accelerated degradation of mRNAs promoted by premature translational termination (16,34,41), the association of trans-acting degradation factors with ribosomes (2,4,8,46), and the stabilization of mRNAs that occurs when translation is inhibited (3,5,23,28,42,49,57). Of particular relevance to the present study were earlier experiments which showed that normally unstable chimeric PGK1-MAT␣1 and ACT1-MAT␣1 mRNAs were stabilized when an in-frame translational termination codon was inserted at the junction of the sequences from the respective stable and unstable mRNA components of the chimeras (39). Since MAT␣1 IE activity required its translation, we hypothesized that translation-dependent IEs could, in general, be mapped by insertion of in-frame nonsense codons. Those inserted 5Ј to the element should stabilize the transcript, while downstream codons should be inconsequential to mRNA half-life. To circumvent activation of the nonsense-mediated mRNA decay pathway, half-lives of allelic transcripts could be measured in a strain deficient for Upf1p, a trans-acting factor essential for activity of this decay pathway (32,33,41).
We have used the MAT␣1 gene as a model to test the feasibility of nonsense codon mapping since the precise coding sequence location of its 65-nucleotide (nt) IE has been defined (6,39). In addition to mapping the element, this method has allowed us to identify a boundary for the translation dependence of element function, thus providing insight into the role of the IE as a destabilizer of the MAT␣1 transcript. Application of the new mapping protocol to other genes should facilitate localization of potential coding-region IEs and thereby provide further understanding of the cis-acting determinants of mRNA stability.

MATERIALS AND METHODS
Yeast strains and plasmids. The yeast strains used in this study were yRP582 (MATa rpb1-1 ura3-52 leu2- 3,112; provided by C. Decker and R. Parker, University of Arizona, Tucson) and yAH01, which is isogenic to yRP582 except that it contains a disruption of UPF1 (upf1⌬). Disruption of the chromosomal UPF1 gene in yAH01 was accomplished by using a plasmid-borne upf1::LEU2 allele constructed by standard techniques (35) from pRS314UPF1. The latter contained the UPF1 gene on a 4.2-kb EcoRI-BamHI DNA fragment (33) subcloned into the yeast shuttle vector pRS314 (55). pRS314UPF1 was digested with BglII to remove a 1.6-kb sequence from the UPF1 gene, and a LEU2 gene from plasmid pJJ250 was cloned into the BglII sites of the resulting plasmid (29). To facilitate cloning of the LEU2 marker, pJJ250 was digested with HindIII, which cuts once, 3Ј of LEU2. The 5Ј overhangs were filled in with Klenow enzyme (Boehringer Mannheim Biochemicals), and a BglII linker (New England Biolabs) was ligated. This DNA was digested with BglII and BamHI, and the LEU2 containing fragment was then ligated to the plasmid carrying the partial deletion of the UPF1 gene. Transformation of yRP582 with a 4.4-kb XhoI-BamHI fragment from pRS314upf1::LEU2 was followed by Southern analysis of genomic DNA to check for UPF1 disruption. Disruption was also monitored by Northern (RNA) blotting to detect stabilization of the nonsense-containing CYH2 pre-mRNA and hence the presence of a nonsense decay phenotype (reference 18 and data not shown). Table 1 lists plasmids containing wild-type and allelic MAT␣1 genes which differ by the presence or absence of a single in-frame translational termination codon and/or the deletion or insertion of an AAT triplet at codon 82. Plasmids containing an in-frame translational termination codon in the MAT␣1 sequence are numbered according to the codon changed. The plasmids are all derived from pGALMAT␣1 (provided by G. Caponigro and R. Parker), in which a GAL1 promoter fused to a MAT␣1 gene was ligated to EcoRI-HindIII-cut pSEH.BX (a derivative of pRIP1H [39] in which XhoI, HindIII, and BglII sites were replaced by PvuI, NheI, and ClaI sites, respectively). They were transformed into strains yRP582 and yAH01 by using a modification of the lithium acetate method of yeast transformation (53), and synthetic medium lacking uracil was used for their selection and maintenance.
Site-directed mutagenesis. The insertion of in-frame nonsense codons into the MAT␣1 coding sequence was performed by using a modification of the procedure described by Kunkel et al. (30). A BamHI-HindIII fragment harboring the MAT␣1 gene from pGALMAT␣1 was cloned into the polylinker of pBluescriptII KSϩ phagemid (Stratagene), and this construct transformed into Escherichia coli CJ236 (dut-1 ung-1). R408 helper phage (Promega) was used for synthesis of single-stranded phagemid DNA and was infected into cells at a ratio of five phage per cell. Approximately 400 ng of uracil-containing single-stranded DNA isolated from infected cells was used per in vitro mutagenesis reaction performed with a commercially available mutagenesis kit (Bio-Rad). Second-strand DNA synthesis was primed by a mutagenic oligonucleotide complementary to the sequence being mutated except for a 1-to 3-nt mismatch, depending on the oligonucleotide used (Table 2). All oligonucleotides were either purchased from  Operon Technologies, Inc., or synthesized on an Oligo 1000 DNA synthesizer (Beckman). Replicative-form DNAs were transformed into E. coli MV1190, which selects for the mutated, non-uracil-containing DNA strand. Plasmids were sequenced over the region being mutated by the dideoxy-chain terminator method (52), using a Sequenase version 2.0 DNA sequencing kit (USB/Amersham) and the sequencing primers MATAH01, MATAH02, and RP-18 (Table 2). Those with desired termination codons were digested to remove the allelic MAT␣1 DNAs, and the BamHI-HindIII fragments were subcloned into the original yeast vector downstream of the GAL1 promoter. For construction of MAT␣1 alleles having a deletion or extra copy of codon 82 (AAT), the MAT␣1 sequences of the nonsense-containing alleles MATUAG78, MATUGA79, MATUAG80, and MATUAG81 were subcloned into the vector M13mp19 (36). Recombinant phage were plaqued on E. coli JM109, and after verification of the presence of an insert, phage stocks were prepared. These phage were used to infect E. coli CJ236, and the uracil-containing single-stranded DNA was isolated. Mutagenesis was performed with the appropriate oligonucleotides (Table 2) essentially as described for the nonsense codon insertion experiments.
mRNA synthesis and decay measurements. Transcriptional pulse-chase experiments were performed essentially as described by Decker and Parker (12) except that galactose inductions were performed for 30 min and medium containing 4% glucose was preheated to 59ЊC. Cells harvested both prior to induction and at various times with respect to inhibition of transcription were concentrated, frozen rapidly on dry ice, and stored at Ϫ70ЊC. Total cellular RNA was extracted by the hot phenol method, and mRNA decay rates were measured as described elsewhere (18). RNA fractionated by electrophoresis was transferred to Zeta probe membranes (Bio-Rad), and blots were probed to detect MAT␣1 and STE2 transcripts. Fragments of the MAT␣1 (0.5-kb BamHI-HincII fragment) and STE2 (2.3-kb SalI fragment) genes were labeled by a random priming procedure (15), using a kit from Boehringer Mannheim Biochemicals. Northern blots were quantitated by using a Betascope blot analyzer. Half-life values were obtained by expressing data as the log 10 of the percentage of each mRNA remaining versus time at 36ЊC. All values represent the average Ϯ1 standard deviation of at least two half-life determinations, using two independently isolated transformants.
Analysis of a second MAT␣1 transcript of lower molecular weight. To determine the nature of a potential MAT␣1 decay intermediate, DNA fragments with specificity for the 5Ј and 3Ј termini of the full-length transcript were generated. A 0.5-kb BamHI-HincII fragment spanning the MAT␣1 coding sequence, with the exception of the 3Ј-terminal 70 nt, was digested with HaeIII, yielding a 161-bp 5Ј complementary fragment. Digestion of the 0.5-kb DNA with MspI generated a 3Ј complementary sequence of 136 bp. Digests were electrophoresed in 1.5% agarose, and DNA fragments of interest were purified by using ultrafree-MC filter units (Millipore Corp., Bedford, Mass.) and labeled by the random priming protocol.

RESULTS
In-frame translational termination codons map the MAT␣1 IE to its coding sequence location. The MAT␣1 mRNA is inherently unstable, having a half-life of approximately 5 min (23,39,45). Previous analyses of chimeric genes identified MAT␣1 coding sequences conferring mRNA-destabilizing activity (39), and subsequent studies defined a 65-nt IE spanning nt 201 to 266 as the minimal sequence capable of destabilizing an mRNA (6). The element, whose location within the MAT␣1 coding region is depicted in Fig. 1A, is bipartite. The first 33 nt are highly enriched for rare codons, while the following 32 nt are predominantly AU rich. Both sequences are necessary for element activity (6) and are shown in Fig. 1B. To establish the feasibility of our new approach to mapping coding region instability elements, we first determined the effect that inserting in-frame nonsense codons at positions 18, 52, and 139 had on MAT␣1 mRNA half-life ( Fig. 2A). The first two nonsense codons precede the IE, whereas that at position 139 follows the IE. Transcription of allelic MAT␣1 mRNAs was regulated by fusing all constructs to a GAL1 promoter. Following a 30-min induction in medium containing galactose, transcription was terminated by the combined effects of glucose repression and thermal inactivation of RNA polymerase II, and mRNA halflives were measured in both UPF1 and upf1⌬ isogenic strains. In all experiments, inhibition of transcription was monitored by following decay of the unstable STE2 mRNA, whose half-life of 5 to 6 min is comparable to that of the wild-type MAT␣1 mRNA (23).
In the upf1⌬ strain, in which nonsense-mediated mRNA decay is blocked, nonsense codons at positions 18 and 52 stabilized the MAT␣1 transcript to half-lives of ϳ13.8 and ϳ12.5 min, respectively ( Fig. 2A and B). However, if ribosomes were allowed to translate to codon 139, the mRNA remained unstable, with a wild-type half-life of ϳ5.2 min ( Fig. 2A and B). In all strains, the STE2 mRNA had a half-life of ϳ5 to 6 min ( Fig. 2A and C), hence ruling out the possibility that extended half-lives for the MATUAG18 and MATUAG52 alleles were a consequence of incomplete inhibition of transcription. These initial results demonstrated the feasibility of nonsense codon insertion as a method to localize an IE within the coding sequence and were in agreement with a previous observation that an in-frame translational termination codon 5Ј to the IE stabilizes the encoded mRNA (39).
In the UPF1 background, the allelic MAT␣1 transcripts were very unstable, with half-lives of ϳ1.0 min, except for  176)]. The two parts of the element, a 5Ј 33-nt sequence enriched for rare codons and a 3Ј 32-nt AU-rich sequence (6), are indicated. UAG 50 depicts the location, relative to the IE, of the in-frame translational termination codon inserted at PGK1-MAT␣1 and ACT1-MAT␣1 hybrid fusion junctions generating stabilized chimeric transcripts (39). (B) Primary sequence of the 65-nt element. Codons 78 and 79 define the 3Ј-and 5Ј-most codons of the rare codon and AU-rich sequences, respectively (6). Rare codons are indicated by a single underline, and double underlines mark the 19-nt sequence within the AU-rich span postulated to be a specific protein binding site (6). Sequences with complementarity to 18S rRNA (45) are indicated in boldface. All numerals refer to codon numbers. VOL. 16,1996 MAPPING OF INSTABILITY ELEMENT OF YEAST MAT␣1 mRNA MATUAG139 whose half-life of ϳ4.0 min more closely resembled the wild-type mRNA half-life ( Fig. 2A and B). The highly unstable nature of the mRNAs with early nonsense codons is indicative of activation of the nonsense-mediated mRNA decay pathway (41,43,45,46). Termination of translation within the IE yields mRNAs of different stabilities dependent on the extent of ribosome translocation. To develop nonsense codon insertion as a general approach for mapping coding region IEs, we sought to understand the consequences on mRNA stability of inhibiting translation within an IE. We therefore inserted in-frame translational terminators at codons 67, 79, and 85 of the MAT␣1 gene (Fig. 3A) and measured the resulting mRNA half-lives.
MATUAG67 terminates translation immediately 5Ј to the IE; MATUGA79 changes the first codon in the 3Ј 32-nt AU-rich sequence to a terminator, such that ribosomes translate only the rare-codon segment of the IE; and MATUAG85 terminates translation after ribosomes progress through 60% of the AUrich sequence, equivalent to 80% of the entire element. Ribosome stalling at codons 67 and 79 yielded transcripts with half-lives of 9.8 and 9.4 min, respectively, in the upf1⌬ strain, indicating an approximately twofold stabilization relative to wild-type mRNA decay (Fig. 3). However, translation to codon 85 yielded an mRNA half-life of ϳ4 min, indicating that ribosome progression had been sufficient to promote normal decay of the MAT␣1 transcript (Fig. 3). mRNA stabilization arising from partial translation of the element, as in the case of translating only the rare-codon sequence in MATUGA79, is consistent with a previous report in which deletion of the 3Ј AU-rich portion of the element stabilized the transcript (6) and indicates that both segments of the 65-nt element are necessary for mRNA instability.
Half-life values for these three transcripts in the UPF1 genetic background reflected the overall conclusion gained from the MATUAG18 and MATUAG52 alleles. While the transcripts of the current alleles were found to be slightly more stable (mRNA half-lives of ϳ2 to 3 min) than those with more 5Ј-proximal nonsense codons (compare Fig. 3A with Fig. 2A), they were still significantly less stable than the wild-type transcript (Fig. 3A), reflecting at least partial activation of the nonsense-mediated mRNA decay pathway. Moreover, the half-life of the STE2 mRNA, measured in all strains, did not fluctuate significantly (data not shown), indicating that the differences observed in the decay of the respective MAT␣1 mRNAs were not attributable to variations in the degree of transcriptional inhibition.
Destabilization of the MAT␣1 transcript is mediated by translation of a two-codon interval of the IE's AU-rich sequence. Half-lives of the six MAT␣1 transcripts analyzed in the experiments of Fig. 2 and 3 are summarized in Fig. 4. Most striking is the abrupt nature of the transition to slower mRNA decay rates in the UPF1 strain (Fig. 4A) and the transition to more rapid mRNA decay in the upf1⌬ strain (Fig. 4B). The former phenomenon has been observed previously for the PGK1, CYC1, and HIS4 mRNAs (16,32,41,59) and may reflect the existence of cis-acting sequences capable of inactivating nonsense-mediated mRNA decay (41,43). The sudden, twofold decrease in mRNA stability that occurred as ribosomes traversed the IE in the upf1⌬ strain (Fig. 4B) was unanticipated, however, and we were interested in mapping more precisely this stability transition. To this end, constructs with single, in-frame UAG translational termination codons at positions 76, 77, 78, 80, 81, 82, 83, and 84 were constructed by oligonucleotide site-directed mutagenesis (Fig. 5A). This set of constructs, together with MATUGA79 and MATUAG85 (Fig.  3A), allows ribosomes translating the respective mRNAs to progress through the IE in increments of one codon. The effects of such ribosome progression on mRNA decay rates were determined in the UPF1 and upf1⌬ isogenic strains.
In the upf1⌬ strain, half-lives of ϳ9 to 10 min were obtained for the allelic MATUAG76, MATUAG77, and MATUAG78 mRNAs ( Fig. 5 and 6), similar to the previously determined value for the MATUGA79 transcript ( Fig. 3A and 4B), and an intermediate half-life (ϳ7.5 min) was obtained for the MATUAG80 transcript ( Fig. 5 and 6). In contrast, the MATUAG81, MATUAG82, MATUAG83, and MATUAG84 mRNAs were unstable, with half-lives of 4 to 5 min ( Fig. 5 and  6), similar to that determined for the MATUAG85 allele ( Fig.  3 and 4B). These results are summarized in a bar plot of the half-lives obtained (Fig. 6), which clearly indicates a transition in mRNA stability as translation proceeds over a two-codon interval of the IE, codons 80 and 81. The transition region is bordered by an upstream segment wherein translation termination stabilizes the mRNA and by a downstream segment in which translation termination has no effect on mRNA decay. The results demonstrate at the molecular level a clear relationship between mRNA stability and mRNA translation, supporting previous studies that have linked the two processes (6,16,19,22,23,33,40,41,44,53). Interestingly, a second inducible MAT␣1 transcript of ϳ370 nt was detected in RNA isolated from cells harboring the MATUAG84 allele (indicated with an asterisk in Fig. 5B). Its characterization is described below.
Decay measurements of these allelic MAT␣1 mRNAs in the UPF1 strain yielded half-lives of ϳ2 to 3 min (Fig. 5A), comparable to the values obtained for the alleles MATUAG67, MATUGA79, and MATUAG85 (Fig. 3A), whose nonsense codons are also located in the vicinity of the stability transition sequence. Half-life values on the general order of ϳ6 min were obtained for the STE2 transcript in all strains in which decay of the allelic MAT␣1 mRNAs was measured (data not shown), again confirming that transcription was inhibited efficiently.
Deletion or insertion of a single codon shifts the position of the MAT␣1 mRNA stability transition region. The experiments of Fig. 5 and 6 demonstrated a transition in MAT␣1 mRNA half-life as ribosomes translated codons 80 and 81 of the IE. One possibility was that this transition reflected a requirement for the ribosome (or a ribosome-borne factor) to interact with a downstream sequence or bound factor and that this interaction provided the signal for rapid decay of the transcript. Since it had previously been demonstrated that the 65-nt IE contains all sequence information required for the promotion of mRNA instability (6), a corollary of the previous hypothesis was that the sequences with which the ribosome interacted were probably within the 3Ј portion of the IE, i.e., the AU-rich region. If so, deletion or insertion of 3 nucleotides VOL. 16,1996 MAPPING 3Ј to the transition should shift the stability profile in a predictable manner. To test this model, we constructed a set of MAT␣1 alleles in which nonsense mutations before, within, and after the transition were accompanied by deletion or duplication of a downstream codon. These alleles, depicted in Fig. 7A, are identical to the nonsense-containing alleles MATUAG78, MATUGA79, MATUAG80, and MATUAG81 except that they also contain either a deletion or a duplication of codon 82. The choice of codon 82 for deletion or duplication was dictated by several considerations, including the following: (i) we sought to minimize disruption of the 19-nt sequence that begins at codon 82 because it is reiterated immediately downstream (with 14 of 19 nt being identical; see reference 6) and postulated to serve as a protein recognition site, possibly linked to transcript decay (6); and (ii) deletion/insertion at codon 82 also avoids those sequences specifically defining the stability transition region (codons 79 to 81) and those sequences of the IE having complementarity to 18S rRNA (Fig. 1B). It is conceivable that interference with any one of these nucleotide stretches could alter the destabilizing mechanism mediated by ribosome translocation.
Half-lives of the STE2 mRNA and the transcripts of the MAT␣1 deletion/insertion alleles were determined, as before, in both UPF1 and upf1⌬ strains. The short half-life of the STE2 mRNA in all experiments (ϳ4 to 7 min) indicated that transcription was inhibited efficiently (data not shown). Control constructs containing either a deletion (MAT⌬82) or an insertion (MATI82) of codon 82 in the wild-type MAT␣1 gene (Fig.  7A) were tested to establish the effects of these changes on transcript half-life. In UPF1 and upf1⌬ strains, both constructs produced mRNAs with a half-life of approximately 5 min, equivalent to that of wild-type MAT␣1 mRNA (Fig. 7A). This result eliminates a role for this nucleotide triplet in IE-destabilizing activity.
In the UPF1 strain, transcripts of all eight MAT␣1 deletion/ insertion mutants also containing a nonsense codon had halflives of ϳ2.5 to 4 min (Fig. 7A), again consistent with activation of nonsense-mediated mRNA decay triggered by the respective in-frame translational termination codons. The halflives obtained for these mutant transcripts in the upf1⌬ strain are listed in Fig. 7A and compared in Fig. 7B to D. To establish the relationship with the parent constructs, half-lives of the four nonsense-containing mRNAs that do not have an insertion or deletion of codon 82 are illustrated in Fig. 7B. In the deletion mutants (Fig. 7C), the stability transition shifted in the 5Ј direction. Compared with the original nonsense-containing transcripts, deletion of codon 82 reduced the half-lives of the MATUGA79 and MATUAG80 transcripts by 2.7 and 1.9 min, respectively, and had only minor effects on the half-lives of the MATUAG78 and MATUAG81 mRNAs (compare Fig. 7B with Fig. 7C; see also Fig. 3A, 5A, and 7A). These results support a model in which the destabilizing effects of a ribosome interaction with a downstream element have occurred three nucleo- tides "earlier" when codon 82 is absent. Such a model of ribosome interaction with a downstream element is also supported by the results of the experiments with the codon 82 insertion mutants (Fig. 7D). The insertion of an extra codon 82 resulted in a complementary shift; i.e., the stability transition was shifted by one codon 3Ј to its original location. Thus, the half-lives of the MATUAG78 and MATUGA79 mRNAs were largely unaffected by codon 82 duplication, but the MATUAG80 and MATUAG81 mRNAs increased in half-life by 1.5 and 1.9 min, respectively (compare Fig. 7B with Fig. 7D).
Characterization of the MATUAG84 370-nt transcript. A second inducible MAT␣1 transcript was produced from the MATUAG84 construct in both UPF1 and upf1⌬ strains ( Fig. 5B  and 8A). This mRNA, 5Ј MAT␣1, has an estimated size of 370 nt and is detectable with a 161-nt probe complementary to MAT␣1 5Ј sequences but not with a 136-nt probe specific for 3Ј sequences (Fig. 8A). Oligo(dT)-cellulose fractionation of RNA extracted from the upf1⌬ strain harboring pGALMATUAG84 resulted in retention of approximately 50% of the 5Ј MAT␣1 molecules, 80% of the wild-type MAT␣1 mRNA, and 65% of the STE2 mRNA (Fig. 8B). The implied presence of a poly(A) tail on the 5Ј MAT␣1 transcript eliminates the possibility that this mRNA is a decay intermediate produced by endonucleolytic cleavage of full-length molecules.
Interestingly, the 5Ј transcript has a half-life of ϳ5 min in both UPF1 and upf1⌬ genetic backgrounds (Fig. 8A and data not shown). This contrasts to the ϳ2.7-min half-life of the full-length MATUAG84 mRNA in the UPF1 strain ( Fig. 5A and 8A) and is suggestive of the 5Ј transcript's resistance to nonsense-mediated decay, with decay proceeding exclusively via the inherent pathway. Closer examination of Northern blots collected during this study revealed the presence of this second MAT␣1 transcript in both UPF genetic backgrounds of strains harboring four additional alleles, MATUAG76, MATUAG77, MATUAG78, and MATUAG80 ( Fig. 5B and 8C). As judged from coelectrophoresis of steady-state RNAs, the transcript has similar molecular weights in all strains (Fig. 8C).
However, intracellular levels of the 5Ј transcript vary for the different alleles. This is apparent from a comparison of the ratios of the 5Ј transcript to full-length mRNA, which are 0.09, 0.35, 0.14, 0.28, and 1.47 in the UPF1 background and 0.04, 0.1, 0.07, 0.08, and 0.4 in the upf1⌬ background for the alleles MATUAG76, MATUAG77, MATUAG78, MATUAG80, and MATUAG84, respectively. We suspect that these 5Ј transcripts arise as a consequence of premature 3Ј processing events and that the variation in the levels of the respective transcripts reflects the efficiency with which such processing events occur (see Discussion).

Nonsense codon mapping of mRNA instability elements.
The identification of the cis-acting determinants of mRNA stability has been facilitated to date by the use of chimeric genes and the analysis of deletions and other mutations. Here, we report a new method that allows localization of cis-acting coding-region IEs in yeast genes encoding inherently unstable mRNAs. Previous studies identified the coding-sequence location of a 65-nt IE within the MAT␣1 mRNA (6,39), and so we tested the ability of in-frame translational termination codons to map this IE in a strain inactive for nonsense-mediated mRNA decay. The localization experiments, summarized in Fig. 4B, demonstrated that insertion of nonsense codons 5Ј to the IE stabilized the transcript two-to threefold, while location of a translation termination codon downstream of the IE, by allowing translation to proceed through the element, had no effect on transcript half-life. The successful localization of the MAT␣1 IE by nonsense codon insertion provides a valuable tool in the search for coding-region IEs in other genes specifying inherently unstable mRNAs.
This new mapping procedure offers certain advantages over the chimeric gene approach in that it causes minimal perturbations to overall mRNA structure, since constructs differ from VOL. 16,1996 MAPPING OF INSTABILITY ELEMENT OF YEAST MAT␣1 mRNA the wild-type transcript by only 1 to 3 nt. In addition, the approach is less tedious since construction of chimeric genes is not required. Our procedure does not, however, delimit the IE and requires further deletion analyses to map the 5Ј and 3Ј boundaries of a localized element. A potential difficulty with nonsense codon mapping may be its inability to distinguish different IEs in a gene containing multiple destabilizing sequences. Furthermore, because of the translational dependency of the mapping protocol, only those elements whose destabilizing activities require ongoing translation can be characterized.
The mRNA half-lives obtained in our studies are in agreement with those of previously reported MAT␣1 deletion experiments (6). Termination of translation 5Ј to the IE was found to stabilize the MAT␣1 transcript two-to threefold ( Fig.  2A and B, 3, and 4B), and in experiments by Caponigro et al., in which the 65-nt element was removed from the gene, the mRNA was stabilized twofold (6). In the latter studies, a three-fold stabilization occurred if the AU-rich sequences immediately 3Ј to the IE were also deleted (6). Furthermore, the extents of message stabilization were similar irrespective of whether the 3Ј 32-nt AU-rich sequence of the element was physically removed or nonsense codons were used to block its translation (6) (Fig. 3 and 4B). Since the 65-nt IE is the only sequence element within the MAT␣1 mRNA with apparent destabilizing activity (39), and since bona fide stable mRNAs in S. cerevisiae have half-lives as long as 60 min (23,45), it might have been expected that mRNAs lacking this element, or its function, would have half-lives in excess of the ϳ14-min maximum observed here and previously (6). However, mutation of the IE of another inherently unstable mRNA, the MFA2 mRNA, also yielded mRNAs whose maximal half-life was 14.5 min (38). This finding suggests that unstable mRNAs may have other, nondiscrete sequence features that enhance their decay rates or that stable mRNAs may contain specific sequences that promote their stability. Experiments supporting the latter possibility, at least for the PGK1 mRNA, have been reported previously (19,41). cis-acting determinants of 3 end formation and mRNA stability. Several MAT␣1 alleles produced a novel transcript of approximately 370 nt (Fig. 5B and 8A and C). The presence of poly(A) tracts on these mRNAs (Fig. 8B) strongly suggests that they are not decay intermediates of the full-length transcript; rather, they appear to be products of a premature RNA processing event. The variation in the ratios of the 5Ј fragments to full-length mRNA (Fig. 8C) suggests that the efficiency of such a processing event is allele specific. A variety of predominantly AT-rich sequence motifs have been proposed to constitute mRNA 3Ј end formation signals in S. cerevisiae (1,21,24,50,51). While precise sequences vary with each gene, all require multiple sequence elements and/or specific sequence reiterations for activity (14,20,25,26). A previous comparison of the 3Ј UTRs of 15 yeast genes, including MAT␣1, identified a tripartite consensus sequence for mRNA 3Ј end processing, TAG. . .TAGT/TATGT. . .TTT (60). Analysis of the MAT␣1 coding region 3Ј to the 65-nt IE revealed the presence of a TATG. . .TTT stretch spanning codons 96 to 106. We hypothesize that this sequence acts as an internal 3Ј end formation signal in the MAT␣1 nonsense-containing alleles, since the introduction of an upstream TAG translation termination codon creates a complete 3Ј end formation element. Consistent with this interpretation is the observation that the MATUGA79 allele does not give rise to a detectable 5Ј transcript. The nonsense codon in this allele, unlike those at codons 76, 77, 78, 80, and 84, is a TGA, not a TAG. Termination of transcription in the vicinity of the proposed element would produce a transcript of approximately 370 nt, in agreement with our estimated size of the 5Ј MAT␣1 mRNA. The subtle differences between the newly created 3Ј end formation elements, i.e., the distances between the inserted TAG codons and the downstream TATG, may account for the observed allele-specific variation in 5Ј MAT␣1 mRNA synthesis. Previous studies have pointed to a spatial requirement for the sequences within such elements that dictates overall element efficiency (14).
In the UPF1 strain, the half-life of the 5Ј MAT␣1 transcript produced from the MATUAG84 allele is almost twice that of the full-length transcript (Fig. 8A). Although this truncated mRNA is of sufficient length to contain its in-frame translational termination codon, it appears to be resistant to the nonsense decay pathway. This may reflect the loss of specific cis-acting sequences 3Ј to the nonsense codon that are required for nonsense-mediated mRNA decay (16,41,61). Another sequence element, designated a stabilizer, appears to regulate this pathway and comprises the region of an mRNA in which nonsense codons lose the ability to promote rapid mRNA decay (41). The increase in the half-lives, in the UPF1 strain, of MAT␣1 mRNAs containing nonsense mutations at or beyond codon 67 suggests that like the PGK1, CYC1, and HIS4 mRNAs (16,32,41,59), this mRNA may contain such a stabilizer region (Fig. 4A).
Interrelationship of mRNA decay and translation. Numerous studies have pointed to an intimate link between the processes of mRNA translation and mRNA decay (6,16,19,22,23,28,34,41,42,43,46,54), and results reported here support this relationship. Through a detailed mutagenesis of the MAT␣1 IE's rare-codon/AU-rich sequence boundary, we have shown that ribosome progression over a two-codon region maximizes the instability phenotype of this mRNA (Fig. 5 and  6). Termination of translation 5Ј to this region stabilized the MAT␣1 mRNA two-to threefold, and termination within this region had intermediate stabilizing effects. We interpret these results in terms of a model in which the crucial destabilizing signal entails a physical interaction between a translating ribosome and an outlying site within the same mRNA. Experiments using one-codon deletions or insertions (Fig. 7B to D) have shown that the translation-dependent mRNA stability transition can be predictably shifted one codon 5Ј or 3Ј, strongly favoring an interaction between a ribosome and a downstream site. The exact nature of the downstream site with which the ribosome interacts is unknown, but it is likely that it is confined to the IE since previous studies have shown that the 65-nt element is sufficient to destabilize a heterologous gene (6). We therefore anticipate ribosome interaction with the AU-rich sequence located immediately downstream of the stability transition region. Ribosomes may recognize this sequence per se, a secondary structure within the sequence, or an RNA-protein complex.
Previous experiments point to a requirement for both the rare-codon and AU-rich sequences of the MAT␣1 IE for message destabilization (6,39). A stimulatory role for rare codons in IE activity was speculated to involve facilitation of ribosome pausing since there was no sequence specificity to the rarecodon requirement (6). The stability transition that we observe occurs at codons 80 and 81, a region of the mRNA previously assigned to the AU-rich portion of the element (6). However, codon 80 is actually the last rare codon of the IE (Fig. 1B). Hence, destabilization of the MAT␣1 transcript coincides with translation of the IE's rare-codon/AU-rich sequence boundary. Interestingly, the stability transition sequence also coincides with the 3Ј end of a 15-nt sequence of which 14 nt are complementary to 18S rRNA (Fig. 1B). A hypothetical mRNA-rRNA base-pairing interaction was previously proposed to induce a translational pause on the MAT␣1 transcript and potentially play a role in transcript turnover (45). Our current data, linking a twofold change in MAT␣1 mRNA half-life to translation of this putative 18S rRNA binding sequence, may support this conclusion.
A model for decay of the MAT␣1 transcript must, therefore, account for the involvement of the entire bipartite element, the role of translating ribosomes, the observed transition in stability which is dependent on the ribosome's position on the mRNA, and our ability to move this stability boundary through mutations. We propose that during translation, ribosomes experience an appreciable decrease in their elongation rate over the IE's rare-codon sequence. This event, perhaps in conjunction with ribosome stalling mediated by mRNA-18S rRNA base-pairing interactions, may provide a sufficient time frame for interaction of the ribosome or a ribosome-bound protein with the AU-rich mRNA sequence or a protein bound to it. The possibility that the IE's AU-rich sequence can serve as a recognition site for binding of a protein involved in transcript decay has been suggested previously (6). The decay-initiating signal that results from this ribosome-mRNA interaction is likely to enhance the transcript's deadenylation rate, hence activating a deadenylation-dependent pathway of mRNA turnover in which 3Ј poly(A) shortening leads to 5Ј decapping and ultimately 5Ј-to-3Ј exonucleolytic digestion of the mRNA coding sequence (7,12,13,37). Removal of the rare-codon segment, by interfering with the kinetics of ribosome translocation, would prevent these ribosome-mRNA interactions and so disrupt the destabilizing activity of the IE. Similarly, nonsense codons which prevent ribosomes from reaching the IE would be expected to stabilize the transcript. What, however, is the role of nonsense codons (at position 82 and beyond) that do not lead to alterations in mRNA decay rate? One possibility is that the translational pause induced by nonsense codons is sufficient to promote decay, provided that it occurs at a site that is the appropriate distance from the AU-rich region. Alternatively, nonsense codons that allow normal decay are located sufficiently far downstream that the rare codon cluster is exposed and, hence, active. The latter possibility is consistent with a 27-to 29-nt footprint for eukaryotic ribosomes (58). A comparison of the current results with those of a previous study (6) provides some insight into these possibilities. Caponigro et al. (6) demonstrated that in the context of a MAT␣1 transcript with an IE deletion, clustered rare codons 11 to 14 codons upstream of the second AU-rich region were sufficient to promote rapid decay. In the current study, however, a UAG 12 codons 5Ј to the normal AU-rich region (UAG 67 ) promoted mRNA stabilization. These results suggest that nonsense codon position simply determines whether the rare codon segment is available for execution of its normal function. It is also possible that the two types of pausing event have different effects on ribosome conformation or ribosome-associated factors and thus demand different positionings of the ribosome for its role in mRNA destabilization. These alternatives are currently being tested.