5′TRU: Identification and Analysis of Translationally Regulative 5′Untranslated Regions in Amino Acid Starved Yeast Cells*

We describe a method to identify and analyze translationally regulative 5′UTRs (5′TRU) in Saccharomyces cerevisiae. Two-dimensional analyses of 35S-methionine metabolically labeled cells revealed 13 genes and proteins, whose protein biosynthesis is post-transcriptionally up-regulated on amino acid starvation. The 5′UTRs of the respective mRNAs were further investigated. A plasmid-based reporter-testing system was developed to analyze their capability to influence translation dependent on amino acid availability. Most of the 13 candidate 5′UTRs are able to enhance translation independently of amino acids. Two 5′UTRs generally repressed translation, and the 5′UTRs of ENO1, FBA1, and TPI1 specifically up-regulated translation when cells were starved for amino acids. The TPI1-5′UTR exhibited the strongest effect in the testing system, which is consistent with elevated Tpi1p-levels in amino acid starved cells. Bioinformatical analyses support that an unstructured A-rich 5′ leader is beneficial for efficient translation when amino acids are scarce. Accordingly, the TPI1-5′UTR was shown to contain an A-rich tract in proximity to the mRNA-initiation codon, required for its amino acid dependent regulatory function.

Gene expression is efficiently controlled by several regulatory mechanisms of protein biogenesis to adapt to changing endogenous and environmental conditions (1). Translational regulation has particular significance because it enables a quick and reversible adaptation, especially needed for efficient stress response (2). Stresses often induce global control mechanisms leading to the reduction of overall protein biosynthesis. This is mediated by post-translational modifications of one or more translation initiation factors (eIFs) including phosphorylations or changes in their abundance (1,3). A well-studied example is the response to amino acid (aa) 1 starvation conditions. A lack of amino acids leads to increased phosphorylation of eIF2 and results in reduced formation of ternary complex (eIF2-GTP-Met-tRNA i Met ) needed for translation initiation (4). This global control can be overruled by mRNA-specific control mechanisms to ensure efficient translation of specific mRNAs even when overall protein biosynthesis is reduced. Those mechanisms are especially directed by elements in the 5Ј untranslated regions (5ЈUTRs) of mRNAs (1).
The paradigm of an mRNA more efficiently translated under aa-starvation conditions encodes Gcn4p, the global regulator of amino acid biosynthesis (5). This effect is mediated by four short upstream open reading frames in the 5ЈUTR of the GCN4-mRNA, repressing its translation under non-starvation conditions and derepressing it when amino acids are scarce (6). This regulative mechanism is conserved from yeast to the mammalian Gcn4p-homolog activating transcription factor 4 (ATF4), whose mRNA carries two upstream open reading frames in its 5ЈUTR (7,8). Another feature known to regulate translation efficiency in yeast and human is the presence of internal ribosome entry sites (IRES) in the 5ЈUTR of mRNAs. They enable sufficient translation for specific mRNAs under conditions when canonical cap-dependent translation is inhibited by directly recruiting the ribosome to the vicinity of the initiation codon, thus bypassing the cap-structure and its associated eIFs (2). A necessity for cap-independent translation via IRES has been shown for starvation-induced differentiation in yeast. In contrast to the structure-based viral IRES-activity the activation in yeast is mediated by unstructured A-rich elements via recruitment of the poly(A) binding protein (Pab1p) (9). In accordance, strong IRES have recently been shown to possess weak secondary structures and to be predominantly located immediately upstream of the mRNA-initiation codon (10). The formation of stronger secondary structures on the other hand can influence translation efficiency via hairpin structures, inhibiting scanning of the 43S pre-initiation complex or binding of regulative proteins (1,3).
Currently, there is only a small group of proteins for which a translational regulation has been described. Smirnova and colleagues performed polysome and microarray analyses to identify mRNAs that are translationally maintained after aadepletion (11). These data represent a genome-wide approach for detecting potential candidates that are translationally regulated. Here, we describe a proteome-based approach to identify 5ЈUTRs regulating translation in dependence of aa-availability. Bioinformatical analyses of respective 5ЈUTRs disclosed a noticeable accumulation of adenine bases and their predicted secondary structures to be specifically weak or not present. The introduction of the 5ЈUTR sequences in a reporter-testing vector revealed three 5ЈUTRs that significantly increased translation when aa-starvation was induced. The strongest effects could be monitored for the unstructured TPI1-5ЈUTR. An A-rich tract in proximity to the AUG start codon was shown to be essential for the translationally regulative function of the TPI1-5ЈUTR in response to aa-starvation conditions.

EXPERIMENTAL PROCEDURES
Yeast Strains and Growth Conditions-The Saccharomyces cerevisiae strain RH2817 is of the ⌺1278b background (MAT␣, ura3-52, trp1::hisG) (12). RH3384 and RH3385 were generated by C-terminally tagging ENO1 and FBA1 with 3xmyc, respectively, according to Janke and colleagues (13). Transformations were carried out according to the lithium acetate method (14). Cultures were grown at 30°C overnight in 10 ml liquid minimal medium (YNB) containing respective supplements (amino acids, uracil), diluted and cultivated in main cultures to midlog phase before isolation of protein extracts or total RNA. Experiment-specific growth conditions are given in the respective paragraphs.
Plasmid Construction-All plasmids used in this study are listed in Table I. The PGK1-promoter was amplified with the primers 5Ј-GAT-AGATCTGCACGTGGCCTCTTATCGAG-3Ј and 3Ј-CGAAAGAAAAA-GAGAAAAAATGTCTAGTAGTTCCTTCGGATCCATGTGGAGATCTT-C-5Ј to construct the plasmid pME3680 (testing vector). This resulted in the PGK1-promoter fragment flanked by BglII restriction sites including a BamHI restriction site and ATG start codon downstream of the promoter. The BglII restriction sites enable the introduction of the PGK1-promoter fragment into BamHI restricted YEp355. To construct plasmids 'pME3681 -pME3694Ј and 'pME3783 -pME3791 respective 5ЈUTRs were amplified by PCR, inserting BglII restriction sites on both ends, and ligated with pME3680 using the BamHI restriction site previously introduced by PCR. 5ЈUTR-lengths were determined according to David and colleagues (15) (www.ebi.ac.uk/ huber-srv/queryGene, Table II) and sequences were confirmed by sequencing (see supplemental Table S1 and Fig. 5A). Plasmids were propagated in the Escherichia coli strain DH5␣ in LB medium with 100 g/ml ampicillin.
De Novo Proteome and Two-dimensional-PAGE Analysis-Amino acid starvation conditions were induced by the histidine analog 3amino-1,2,4-triazole (3AT). 50 ml yeast cultures were grown to midlog phase (OD 600 ϭ 0.8) in minimal medium prior to the addition of 3AT to a final concentration of 10 mM (16) (17). Protein concentrations were determined via BCA Protein Assay kit from Pierce (#23227). One hundred micrograms of purified protein extracts were used in two-dimensional-PAGE analyses. For the first dimension the protein samples were applied to Immobiline Drystrips (pH 4 -7, 18 cm, #17-1233-01, GE Healthcare Europe GmbH, Freiburg, Germany) via rehydration loading. The separation was carried out in the Ettan IPGphor Isoelectric focusing system (GE Healthcare) at 20°C and a maximum of 50A/ strip with the following program: 70 V for 12 h (step-n-hold), 500V for 1h (step-n-hold), 1000V for 1h (step-n-hold), 8000V for 1h (gradient), 8000V for 4 h (step-n-hold). The strips were thereafter equilibrated in equilibration buffer (50 mM Tris-HCl, pH 8.8, 6 M urea, 30% (v/v) glycerol, 2% (w/v) SDS, 0.002% BPB) containing 325 l 1 M dithiotreitol (DTT) or 125 mg iodacetamide and incubated for 30 min, respectively. The second dimension separation was executed on 12.5% polyacrylamide gels in Protean ® II xi vertical electrophoresis cells (1 mm spacers, 20 ϫ 20 cm glass plates) (Bio-Rad Laboratories GmbH, Munich, Germany). Electrophoresis was performed at 30 mA per gel. Gels were silver-stained according to Blum et al. (18), vacuum dried, and exposed on imaging plates (Fuji, Tokyo, Japan) for two weeks. The protein-spots in the resulting autoradiographies were quantified with the analysis software PDQuest ™ (Bio-Rad, Munich, Germany). The analysis was performed for five biologically independent replicates. LC-MS/MS Protein Identification-Excised polyacrylamide gel pieces of stained protein-spots were digested with trypsin according to Shevchenko et al. (19). Tryptic peptides extracted from each gel piece were injected onto a reverse-phase liquid chromatographic column (Dionex-NAN75-15-03-C18 PM) using the ultimate HPLC system (Dionex, Idstein, Germany) to further reduce sample complexity prior to mass analyses with an LCQ DecaXP mass spectrometer (Thermo Scientific, San Jose, CA), equipped with a nano-electrospray ion source. Cycles of MS spectra with m/z ratios of peptides and four data-dependent MS2 spectra were recorded by mass spectrometry. The "peak list" was created with extractms provided by the Xcalibur software package (BioworksBrowser 3.3.1, Thermo Scientific). The MS2 spectra with a total ion current higher than 10.000 were used to search for matches against a yeast genome protein sequence database from the National Center for Biotechnology Information (NCBI) Saccharomyces Genome Database (Stanford, CA, USA, 6882 sequences, March 2005, plus 180 sequences of the most commonly appearing contaminants as e.g. keratins and proteases, provided with the BioworksBrowser package) using the TurboSEQUEST algorithm (20) of the Bioworks software. The search parameters based on the TurboSEQUEST software included: (i) precursor ion mass tolerance less than 1.4 amu, (ii) fragment ion mass tolerance less than 1.0 amu, (iii) up to three missed tryptic cleavages allowed, and (iv) fixed cysteine modifications by carboxyamidomethylation (plus 57.05 amu) and variable modifications by methionine oxidation (plus 15.99 amu), and phosphorylation of serine, threonine, or tyrosine (plus 79.97 amu). At least two matched peptide sequences of identified proteins must pass the following criteria: (i) the cross-correlation scores (XCorr) of matches must be greater than 2.0, 2.5, or 3.0 for peptide ions of charge state 1, 2, and 3, respectively, (ii) ⌬Cn values of the best peptide matches must be at least 0.4, and (iii) the primary scores (Sp) were at least 600. Peptides of identified proteins were individually blasted against the SGD database (BLASTP at http://seq. yeastgenome.org/cgi-bin/blast-sgd.pl against the data set Protein Encoding Genes) to ensure their unambiguous assignment to the TurboSEQUEST-specified protein.
␤-Galactosidase Assay-Assays were performed with extracts of cells grown in liquid minimal medium. 10 ml precultures were grown overnight at 30°C and 1 ml used to inoculate 10 ml main cultures. For non-starvation conditions cells were harvested after 6 h. To induce amino acid starvation conditions, 3AT was added to a final concentration of 10 mM and cells were incubated at 30°C for 8 h to compensate for a reduced growth rate under aa-starvation conditions. Specific ␤-galactosidase activities were normalized to the total protein amount (21) in each extract and calculated according to Rose and Botstein (22) Western Hybridization Analysis-Cell main cultures were grown in 50 ml minimal media at 30°C to midlog phase (OD 600 ϭ 0.8) before further incubation with or without 10 mM 3AT for 1.5 h. Protein extraction was performed as previously described for the two-dimensional-PAGE analysis and proteins were blotted onto nitrocellulose membrane subsequently of separation by one-dimensional or twodimensional PAGE. After incubation of membranes with polyclonal rabbit anti-eIF2␣-P (BIOSOURCE, Nivelles, Belgium), polyclonal rabbit anti-eIF2␣, rabbit anti-Tpi1, monoclonal mouse anti-myc (#sc-40, Santa Cruz Biotechnology, Heidelberg, Germany) or polyclonal rabbit anti-Cdc28 (#sc-28550) antibodies and subsequent incubation with polyclonal peroxidase-coupled goat anti-mouse (#115-035-003, Dianova, Hamburg, Germany) or goat anti-rabbit (#G21234, MoBiTec, Gö ttingen, Germany) secondary antibodies, proteins were visualized by ECL technology (Amersham Biosciences, Munich, Germany). Relative quantification was carried out via PDQuest ™ analysis software (Bio-Rad, Munich, Germany). Northern Hybridization Analysis-Yeast cultures were cultivated according to the western hybridization protocol. Total RNA from yeast was isolated following the protocol described by Cross and Tinkelenberg (23). The RNA samples were denatured and separated on a 1.4% agarose gel containing 3% formaldehyde and transferred onto nylon membranes by capillary blotting. Gene specific probes were labeled radioactively with [␣-32 P]dATP using the Prime-It® II Random Primer Labeling Kit from Stratagene (#300385, La Jolla, CA). A Fuji Film BAS-1500 Phosphor-Imaging scanner (Fuji, Tokyo, Japan) and Aida Image Analyzer software (Version 4.22.034, raytest, Straubenhardt, Germany) were used for quantification of signals.
Detection of Local, Thermodynamically Optimal RNA Secondary Structures-RNALFOLD from the VIENNA Package 1.8.2 (24) was used for the detection of local, thermodynamically optimal RNA secondary structures in 5ЈUTRs and the computation of their minimal free energy (MFE). The program was called with the options "-noLP" and "-L 100". The first option inhibits lonely base pairs and the second restrains the maximal local structure size to 100 bases.
Calculation of the z-Score-The z-score is computed for each subsequence in each 5ЈUTR with a local, thermodynamic optimal RNA secondary structure and is defined as: z ϭ (m -) : , whereby, m is the MFE of the secondary structure of the target sequence S, is the mean, and the standard deviation of the MFE-values of the RNA secondary structures of random sequences with the same length and dinucleotide composition as S. The creation of random sequences with similar properties as the target sequence is done by DISHUFFLE (25). For each target sequence 100 random sequences were computed. For each of the random sequences the secondary structure is predicted with RNAFOLD also from the VIENNA Package 1.8.2 (26). Then, for each target sequence the mean and standard deviation of the MFE-values of the random sequences and the z-score are computed.

RESULTS
In our analysis, we are aiming to find 5ЈUTRs that specifically enhance the translation of their respective mRNA under aa-starvation conditions induced by the histidine analog 3amino-1,2,4-triazole (3AT). The strategy to identify regulatory 5ЈUTRs is based on (i) the two-dimensional-analysis of 35 Smethionine labeled de novo proteomes generated under differential conditions. The next steps of the analysis consist of (ii) identification of protein-spots regulated under the chosen condition, (iii) comparison of the obtained proteome data with pre-existing transcriptome data generated under similar conditions (27), (iv) evaluation of the candidate-5ЈUTRs via reporter-testing system presented in this study and (v) sequential and structural analysis of 5ЈUTRs by bioinformatical means (overview in supplemental Fig. S1).

The De Novo Biosynthesis of Various Abundant Proteins is
Up-regulated Post-transcriptionally Upon Amino Acid Starvation-Wild type yeast cells of the ⌺1278 background were cultivated in the absence or presence of 10 mM 3AT for 30 min. After the addition of 35 S-methionine the cultures were incubated for an additional hour. This enables the visualization of the effects of aa-starvation on de novo protein biosynthesis in S. cerevisiae via autoradiographies, opposing to a conventional steady-state proteome. The effective induction of aa-starvation conditions by 10 mM 3AT was verified at the level of translational regulation by a significant increase in eIF2-phosphorylation (supplemental Fig. S3A). The consequential reduction in overall protein biosynthesis and successful incorporation of radioactively labeled methionine during translation could be monitored for the soluble and insoluble fraction, resulting from cell lysis by Y-PER ® Plus reagent, via scintillation counting (supplemental Fig. S3B). The radioactive labeling of proteins by 35 S-methionine during translation enables the generation of autoradiographies from two-dimensional gels (Fig. 1). In comparison to the silverstained gels, the autoradiographies only illustrate proteins that were synthesized when aa-starvation conditions were already induced and depict changes in protein biosynthesis mediated by the addition of 3AT.
Our two-dimensional analysis was conducted on the basis of five biological replicates. It revealed a total number of 31 proteins, which showed an up-regulation under aa-starvation conditions (Fig. 2). The corresponding proteins can be assigned to several biological functions and take part in cellular processes like amino acid biosynthesis, glycolysis, oxidative stress response, and cell biogenesis. The fold changes ( ϩ3AT / -3AT ) for the up-regulated protein-spots were determined by the analysis of the autoradiographies via PDQuest ™ and range from 1.10 for Grx1p to 9.93 for Bna1p.
The obtained proteome data was compared with transcriptome data generated under similar conditions of non-starvation and aa-starvation (27) (10 mM 3AT, 8 h) (Fig. 2). For more than half of the candidates with elevated protein-spot intensities under aa-starvation conditions an up-regulation could also be found on their mRNA-level (Fig. 2, bottom part of  chart). This comparison shows that the underlying regulation for these candidates is most likely to be found on the level of transcription even though an additional post-transcriptional regulation is not excluded. For the remaining 13 identified proteins a comparison of their proteome fold change with the fold change of their mRNA levels ( ϩ3AT / -3AT ) revealed that a post-transcriptional regulation can be assumed. Their spotintensities are elevated under aa-starvation conditions, whereas their corresponding mRNA levels either stay the same or are down-regulated when aa-starvation is induced (Fig. 2, top part of chart). Because of this discrepancy in the transcriptome and proteome changes, these candidates are of specific interest in this study and define the base for further investigations. A "confidence factor" is created as a means to facilitate the assessment of a candidate's potential to be post-transcriptionally regulated. It factors in the determined de novo proteome changes under aa-starvation conditions in context with the transcriptome changes. Furthermore the number of autoradiographies in which a protein was found to be up-regulated is accounted for.
Novel Reporter-testing System to Monitor Translational Regulation Mediated by 5ЈUTRs-Translational control of individual mRNAs often depends upon the sequential characteristics and/or structural features of the transcript itself. Primarily 5ЈUTRs are known to contain a variety of elements with a regulatory effect on the translation of their mRNAs (1). This encouraged us to develop a straightforward lacZ-reporterbased testing system to enable the identification of 5ЈUTR sequences altering translation (Fig. 3A). Transcription consequently results in a lacZ-mRNA carrying the incorporated 5ЈUTR as its own. The practicability of this system to monitor translational regulation was verified by the incorporation of the 5ЈUTR of the GCN4-mRNA, known to alter translation efficiency in dependence of aa-availability (6) (supplemental Fig. S4).
The 5ЈUTRs to be analyzed via the testing system correspond to the 13 candidates post-transcriptionally up-regulated in their biosynthetic activity upon aa-starvation conditions (Fig. 2, top part, supplemental Table S2). For most of the 5ЈUTR sequences inserted in the reporter-testing vector (TV) an increase in ␤-galactosidase activity can be monitored under nonstarvation (-3AT) as well as under aa-starvation (ϩ3AT) conditions. It reaches up to 16-fold relative to the empty TV, suggesting a general promotion of translation by the respective 5ЈUTRs (representative 5ЈUTR effects shown in Figs. 3C and 3D). However, two 5ЈUTRs investigated have the contrary effect on lacZ-mRNA expression (Fig. 3B). The insertions of the 5ЈUTRs of the PGK1-and RHR2-mRNA cause a distinct reduction of ␤-galactosidase activity compared with the empty TV. For the RHR2-5ЈUTR this effect is limited to aa-starvation. For the PGK1-5ЈUTR a drastic reduction in activity manifests under both conditions of aa-availability to as low as 30% of the empty TV-activity. A significant elevation in ␤-galactosidase activity from non-starvation to aa-starvation conditions could be monitored for three candidate-5ЈUTRs, namely ENO1-, FBA1-and TPI1-5ЈUTR, and ranges from 1.7 for the ENO1-5ЈUTR to 4.1 for the TPI1-5ЈUTR (Fig.  3D). This suggests a distinct role of those three 5ЈUTRs in enabling an enhanced translation of their respective mRNAs under aa-starvation conditions, when the translation of most mRNAs is reduced.
The mapping of the transcription start sites illustrates that within each construct one main transcription start sites is used under non-starvation as well as aa-starvation conditions and therefore identical mRNAs are entering translation under either aa-availability (supplemental Fig. S6). These findings argue for a translational regulation underlying the measured increase of ␤-galactosidase activity from non-starvation to aa-starvation for the 5ЈUTRs of ENO1, FBA1, and TPI1.
Correlation Between De Novo Biosynthesis and Total Protein Amount of Candidate Proteins Within the Cell-In this study we chose the method of metabolically labeling proteins by 35 S-methionine in the course of translation. The successive analysis of the respective autoradiographies discloses aa-dependent changes in the de novo protein biosynthesis in comparison to the steady-state total protein amount given at a certain time point. This represents a sensitive approach and enables the visualization of even subtle changes in the de novo biosynthesis for a specific protein. At this time point The numbers indicate excised protein-spots with an enhanced intensity under aa-starvation conditions. changes in total protein amount within the cell might not yet be detectable. Eno1p, Fba1p, and Tpi1p were identified as proteins to be more efficiently synthesized under aa-starvation conditions (Fig. 4A). The analysis of the respective 5ЈUTRs via testing system further suggests an underlying 5ЈUTR-mediated translational regulation for these candidates (Fig. 3D). We questioned whether these translational changes might be strong enough to be reflected in elevated steadystate protein amounts of Eno1p, Fba1p, and Tpi1p after 90min of aa-starvation, whereas it is to mention that these candidates are proteins of high abundance. In the performed two-dimensional Western blot experiments several processed forms can be detected for all three proteins. All forms seem to underlie the same regulatory effect upon aa-starvation, according to their uniform change in abundance (Fig. 4A).
The determined steady-state protein amounts of myctagged Eno1p and Fba1p are not significantly increased under aa-starvation conditions. This suggests that the existing changes in their de novo protein biosynthesis are not extensive enough to effectively influence their steady-state protein amount after 90 min of treatment with 10 mM 3AT. Another possibility might be that the turnover rate for these proteins is more rapid under aa-starvation conditions than their increase in de novo biosynthesis. In contrast, for Tpi1p a significant increase in total protein could be determined even after only 90 min of growth under aa-starvation conditions. This implies a considerable enhancement in the translational rate of the TPI1-mRNA upon induction of starvation, quickly enlarging the pool of available Tpi1p in the cell. In the proteome analysis of radioactively labeled protein extracts Tpi1p was identified with an average up-regulation of 3.8 from non-starvation to aa-starvation conditions (Fig. 2). This is similar to the upregulation of total Tpi1p amount in the cell with a factor of 3.3 determined via two-dimensional Western blot experiment (Fig. 4A). In addition, the measured effects from non-starvation to aa-starvation conditions in the ␤-galactosidase assays were the highest when the TPI1-5ЈUTR was inserted in the reporter-testing vector with a factor of 4.1 (Fig. 3D). The mRNA-levels of ENO1, FBA1, and TPI1 are not significantly up-regulated when aa-starvation is induced and confirm the FIG. 2. Comparison of proteome and transcriptome data generated to monitor effects of aa-starvation conditions. All candidates listed were found to be up-regulated upon aa-starvation in the proteome analyses (see supplemental Table S2 for corresponding protein names/functions and supplemental Table S3 for data on protein sequence identification). The transcriptome data used in this comparison has been obtained under similar conditions ( a27 ). Transcriptome as well as proteome changes induced by aa-starvation conditions are displayed as the quotient of spot-intensity under aastarvation to spot-intensity under nonstarvation conditions ( ϩ3AT / -3AT ). To clearly illustrate up-and down-regulation, transcriptome and proteome changes are visualized logarithmically in a horizontal histogram. The reproducibility for each candidate is expressed as frequency, describing the number by which a protein has been identified as up-regulated in n of five biological replicates. The last column represents the 'confidence factor' which is composed of the fold change ( ϩ3AT / -3AT ) of the proteome relative to that of the transcriptome and the frequency of each proteome candidate. microarray data used in this study (Fig. 4B). Taken together these findings strongly suggest an aa-dependent translational regulation of Tpi1p biosynthesis mediated by elements in its 5ЈUTR. The resulting effects prove to be significant enough to affect the total Tpi1p amount in the cell even after a relatively short period of time.
Regulative 5ЈUTR Sequences are A-Rich and Weakly Folded-In addition to the experimental evaluation of the candidate-5ЈUTRs, they were analyzed bioinformatically. A first comparison of the respective 13 5ЈUTRs with each other revealed no distinct structural features or obvious consensus sequences. The lengths of the analyzed 5ЈUTR sequences are slightly shorter than the average length of 89 nucleotides (nt) determined for 5ЈUTRs in yeast. They rather agree with the majority of yeast-5ЈUTRs, measuring less than 50 nt in length (28) (Table II). This rather short length of most candidate-5ЈUTRs is regarded as indicator for a facilitated translation (28). Correspondingly, elevated lacZ-expressions were achieved by all but two of the thirteen 5ЈUTR sequences evaluated via reporter-testing system. To determine the potential and characteristics of secondary structure formation the minimal free energies (MFEs) for each 5ЈUTR sequence were predicted (supplemental Table S4). Hereby, the predicted structures are considered more stable the greater their negative free energies are. The greatest negative free energy (minMFE) found within each respective 5ЈUTR is listed in Table II along with the start and length of the corresponding secondary structure. The structures predicted stretch from 21 to 63 nt in length with an average of 41 nt. The corresponding average minMFE for the predicted secondary structures lies at -4.9 kcal/mol. The weakest MFE of -0.63 kcal/mol was predicted for the ALD6-5ЈUTR and the greatest for the PGK1-5ЈUTR at -9.7 kcal/mol. Even though the greatest minimal free energy was predicted for one of the longest 5ЈUTRs, belonging to the mRNA of PGK1 (82 nt), there seems to be no obvious correlation between MFE gained from secondary structure formation and 5ЈUTR-length as far as this can be stated by this rather small group of 5ЈUTR sequences analyzed. No secondary structure was predicted for the 5ЈUTR sequences of AHP1, ENO1, FBA1, FPR1, and TPI1 (Table II). Overall these findings imply a weak secondary structure formation for the candidate-5ЈUTRs. In addition, a comparison of the stabilities of the mRNA structures formed by the candidate 5ЈUTRs and randomly computed sequences of the same length and dinucleotide composition was performed. The resulting z-score further confirms the lack of significant secondary structure formation in the candidate-5ЈUTRs (supplemental Table S4). When evaluating the obtained structural information in combination with the data generated in the ␤-galactosidase assays, it is striking that for all 5ЈUTRs, showing an aa-dependent up-regulation of translation in the reporter-assays, namely ENO1-, FBA1-, and TPI1-5ЈUTR, no FIG. 3. Reporter-testing system and ␤-galactosidase assays displaying the effects of candidate-5UTRs on lacZ-reporter activity dependent on aa-availability. A,The reporter-testing vector is a 2 m yeast-E. coli shuttle vector carrying selectable marker genes as well as the constitutive PGK1-promoter with a defined transcription start site (TSS) and a lacZ-reporter-gene. Arbitrary 5ЈUTR sequences can be inserted in between promoter and reporter gene. Possible incorporated translationally regulative elements are depicted in the lacZ-mRNA-5ЈUTR such as hairpin structures, upstream open reading frames (uORFs) and internal ribosome entry sites (IRES). The three observed effects on lacZ-expression upon introduction of candidate 5ЈUTR sequences are illustrated in: B, reduced expression w/o significant 3AT effect; C, enhanced expression w/o significant 3AT effect; and D, enhanced expression w/additional positive 3AT effect. ␤-galactosidase activities were normalized to respective plasmid copy numbers (supplemental Fig. S5A) and are displayed relative to the testing vector without 5ЈUTR under the respective condition. secondary structure could be predicted. This was only the case for two other 5ЈUTRs in the bioinformatical analysis (Table II). Thus, the share of 5ЈUTRs without predicted secondary structure is well-above average for the group of 5ЈUTRs showing an aa-specific effect.
In addition to the MFEs, the GC-content for each 5ЈUTR sequence was determined averaging at 29% (Table II). Interestingly, this is consistent with the average GC-content found for the least stable structures in a genome-wide analysis of 5ЈUTRs in comparison to 47% for the most stable structures (29). Another striking feature is the high amount of adenine bases contained in the analyzed 5ЈUTR sequences. It even further increases to over 50% for all but two 5ЈUTRs proximal to the AUG translation initiation codon. For seven of the thirteen 5ЈUTR sequences this increase amounts to at least 20% including the 5ЈUTRs of ENO1, FBA1, and TPI1 (Table II).
Because of the elevated de novo biosynthesis of Tpi1p and its increased protein levels under aa-starvation conditions, the 5ЈUTR of TPI1 was analyzed in more detail. Several altered 5ЈUTRs were constructed and evaluated via reporter-testing system with special regard to the prominent A-rich tract in proximity to the AUG start codon, which leads to an adeninecontent of 68% in the last 25 nt of the TPI1-5ЈUTR (Fig. 5A). The deletion of the last 25 nt of the TPI1-5ЈUTR, containing the A-rich tract, leads to a decreased ␤-galactosidase activity under non-starvation and aa-starvation conditions and clearly reduces the degree of activity gain in response to aa-starvation in comparison to the full length TPI1-5ЈUTR (Fig. 5B). The significance of the adenine bases within the Tpi1-5ЈUTR sequence was further examined by replacing the adenine bases with thymine, cytosine, or guanine bases, while still retaining the unstructured characteristic of the natural wild-type TPI1-5ЈUTR. The implications of a complete (c) loss of adenine bases in the TPI1-5ЈUTR were determined as well as of a partial (p) loss, limited to the last 25 nt containing the A-rich tract (Fig. 5).
The 5ЈUTR with a partial replacement of adenine with thymine bases (Tp) results in a complete loss of the aa-starvation-dependent up-regulation of ␤-galactosidase activity, characteristic for the wild type TPI1-5ЈUTR, whereas lacZexpression under non-starvation conditions is not influenced (Fig. 5B). The complete substitution of all adenine bases with thymine also disables an elevated ␤-galactosidase activity under aa-starvation conditions and, in addition, reduces overall expression levels in comparison to the wild type TPI1-5ЈUTR and the TPI1-Tp exchange. Similarly, the complete or partial exchanges by cytosine or guanine bases do not significantly promote an induced lacZ-expression under aastarvation but lead to a drastic reduction of ␤-galactosidase activity under both conditions. Hereby the complete replacement with cytosine bases and both guanine substitutions (Gp and Gc) resemble the total loss of ␤-galactosidase activity as measured for the introduction of a 52 bp stem loop structure with an MFE of -42.9 kcal/mol (30) (Fig. 5B).
As with the TPI1-5ЈUTR, the AHP1-5ЈUTR has no predicted secondary structure and displays a similar length but is distinguished by a clearly reduced A-content in its last 25 nt of FIG. 4. Steady-state protein-and mRNA-amount of candidates determined under non-starvation (-3AT) and aa-starvation (؉3AT) conditions. A, myc 3 -tagged versions of the candidate-proteins Eno1p and Fba1p were hybridized to anti-myc antibody, whereas Tpi1p was detected by anti-Tpi1 antibody. For quantification the total intensity of all spots marked by crosshairs under the respective condition was determined. The illustrated change in de novo biosynthesis was previously determined by two-dimensional-PAGE of metabolically labeled protein extracts and autoradiography analysis (see Figs. 1 and 2). B, Northern hybridizations against ENO1-, FBA1-and TPI1-mRNA were quantified and normalized against ACT1 as loading control. The adjacent graphs, respectively, illustrate the fold changes relative to the signal strengths under non-starvation conditions. Cells were grown to exponential phase before further incubation for 90 min in absence (-3AT) or presence (ϩ3AT) of 10 mM 3AT.
only 52% compared with 68% for the TPI1-5ЈUTR (Table II). The 5ЈUTR of AHP1 was modified to test whether any random 5ЈUTR sequence with a high adenine-content proximal to the AUG start codon can convey an aa-dependent increase of lacZ-expression. Several bases in the last 25 nt of the AHP1-5ЈUTR sequence were exchanged for adenine bases and arranged to resemble the terminal TPI1-5ЈUTR sequence, resulting in an AUG-proximal A-content of 68% for the AHP1-A-5ЈUTR (Fig. 5A). The evaluation of this 5ЈUTR enriched in downstream adenine bases does not result in a significant increase in ␤-galactosidase activity under aa-starvation conditions, suggesting additional specificities within the TPI1-5ЈUTR to mediate its aa-specific regulatory function than solely its increased terminal A-content and absent secondary structure formation.
Taken together, our data indicates an up-regulation in the translational rate for the ENO1-, FBA1-, and TPI1-mRNAs mediated by their respective 5ЈUTR. For TPI1 these effects are significant enough to induce substantial changes to the Tpi1p-pool within the cell already after a short period of elevated biosynthesis levels. The bioinformatic analysis suggests weakly folded 5ЈUTR sequences with an enrichment of adenine bases in proximity to the AUG initiation codon to be beneficial for translation efficiency. The analysis of artificially generated 5ЈUTRs demonstrates that the adenine bases in the TPI1-5ЈUTR are specifically required for elevated expression levels under aa-starvation condi-tions and suggests that further features might contribute to this regulatory function. DISCUSSION The presented proteome analysis of radioactively labeled protein extracts enables the identification of proteins, whose de novo biosynthesis is elevated under aa-starvation conditions. Along the same lines, Smirnova and colleagues performed polysome analyses of aa-starved yeast cells, which resulted in translation profiles disclosing mRNAs with enhanced translation efficiency upon aa-withdrawal (11). Surprisingly, no significant correlations within the identified candidates can be found when comparing both data sets. Suggested by the up-regulation of biosynthesis of several proteins involved in oxidative stress response in this study but not in the analysis by Smirnova et al., these discrepancies might at least partially derive from different means of inducing aa-starvation conditions. Whereas Smirnova and colleagues completely withdrew amino acids, in this study aa-starvation was induced by 10 mM 3AT, a compound that has previously been described to additionally inhibit catalases and might thereby contribute to oxidative stress (31). Nevertheless, both studies feature the striking similarity of the translational upregulation of many candidates that are taking part in carbohydrate metabolism and energy balance, which is speculated by Smirnova et al. to occur in preparation of an elevated aa-biosynthesis. Whereas Smirnova and colleagues only FIG. 5. Evaluation of artificially generated 5UTRs. A, DNA-fragments were constructed and expressed via reportertesting system, resulting in the schematically illustrated mRNA-5ЈUTR sequences. The corresponding minimal free energies (MFEs) were determined with RNAFOLD. For the TPI1-and AHP1-5ЈUTRs the last 25 nt and altered bases are highlighted. The respective adenine-contents are displayed for full length 5ЈUTRs and the last 25 nt. B, The ␤-galactosidase assays display the influence on lacZ-expression exhibited by the respective 5ЈUTR inserted in the reporter-testing vector under non-starvation (-3AT) and aa-starvation conditions (ϩ3AT). The data was normalized to the plasmid copy numbers for each construct (supplemental Fig. S5B) and are displayed relative to the testing vector without 5ЈUTR under the respective condition.
identify a small number of such candidates with a converse transcriptional regulation, we observe that most of our candidates belonging to this group, show a post-transcriptional up-regulation, namely Tpi1p, Fba1p, Ald6p, Pgk1p, and Eno1p. This observation is further supported by the properties of the 5ЈUTRs of ENO1, FBA1, and TPI1 in mediating elevated translation rates upon aa-starvation.
Most of the evaluated 5ЈUTRs in this study enhanced the lacZ-reporter expression and yielded weak secondary structures in the bioinformatical analysis or had no structure predicted at all like the TPI1-5ЈUTR. This is in agreement with the finding that 5ЈUTRs of highly abundant proteins generally possess weaker secondary structures resulting in an elevated translation initiation of their respective mRNAs (29).
In addition to the mainly weak secondary structures predicted, another striking feature of the analyzed 5ЈUTRs is the high content of adenine bases. The analysis of variants of the unstructured TPI1-5ЈUTR containing thymine, cytosine or guanine bases in place of adenine resulted in a reduction of ␤-galactosidase activity with more drastic effects for the replacement by guanine and cytosine. The strongest translational repression was obtained for the partial (last 25nt) and complete substitution with guanine bases. Such guanine-rich sequences can adopt non-canonical four-stranded secondary structures, so-called G-quadruplexes, which have been shown to reduce the translational rate when located in 5ЈUTR sequences (32).
In addition to the overall overrepresentation of adenine bases in the analyzed 5ЈUTR sequences, they are especially dominant in proximity to the mRNA-initiation codon, reaching up to 68% for the last 25 nt of the unstructured TPI1-5ЈUTR (Fig. 5A). The substitution of these terminal adenine bases by any other nucleotide results in a complete loss of the TPI1-5ЈUTR specific up-regulation of lacZ-expression under aastarvation conditions. This effect could be mediated by the described role of unstructured A-rich tracts in serving as binding sites for the poly(A) binding protein (Pab1p). This binding is suggested to be able to substitute for cap and eIF4E in recruiting eIF4G for translation initiation, thus displaying an IRES event, as previously shown for starvation-induced invasive growth in yeast (9). In addition, particularly strong eukaryotic IRES have been linked to weak secondary structures (10) as e.g. present for the TPI1-5ЈUTR. Accordingly, the A-rich tract within the TPI1-5ЈUTR could act as IRES element, required to mediate enhanced translational activity upon aastarvation conditions, when cap-dependent translation initiation is reduced. This is further supported by the positioning of the A-rich tract immediately upstream of the mRNA-initiation codon, which serves as predominant localization site for IRES elements (10).
Despite the absence of an apparent secondary structure and a high adenine content in the last 25 nt of the FPR1-and the AHP1-A-5ЈUTR, these sequences do not significantly induce lacZ-expression under aa-starvation conditions, sug-gesting further factors to be involved in the aa-dependent regulatory function observed for the TPI1-5ЈUTR. An indication might be given by the 5ЈUTRs of ENO1, FBA1, and TPI1, representing the three 5ЈUTR sequences enhancing lacZ-expression aa-dependently. These 5ЈUTRs show an increased adenine content in their last 25 nt relative to the full length sequence, amounting to at least 20% (Table II). This could not be observed for any other unstructured 5ЈUTR analyzed including the artificially generated AHP1-A-5ЈUTR (Fig. 5A). This leads to the assumption that not only the absolute amount of adenine bases in the last 25 nt is of importance for the regulatory function of a 5ЈUTR upon aa-starvation conditions but especially their distribution along the 5ЈUTR sequence is of relevance.
The testing system as it is described in this study displays a very strict screening tool. It aims to identify proteins regulated via translation solely mediated by their 5ЈUTR sequence. An interesting continuative approach could be to expand the current reporter-testing vector to enable the incorporation of not only the 5ЈUTR but also the 3ЈUTR of an mRNA to monitor their combined influence on lacZ-reporter expression. Apart from the known regulative function of 5ЈUTRs, roles in the regulation of translation initiation have also been described for the 3ЈUTR as well as for the ring formation of the mRNA. Thus, a comprehensive picture of the broad range of translational control mechanisms could be obtained by the described expansion of the testing system, because a synergy of 5Ј and 3Ј UTR has previously been suggested to be important for the establishment of correct translation (33)(34)(35)(36).
Taken together, this study demonstrates a well-suited approach to identify translationally regulative 5ЈUTR sequences of abundant proteins. The applied combination of proteomics, transcriptomics as well as bioinformatical and molecular genetic tools enabled the identification of an A-rich tract in the downstream region of the TPI1-5ЈUTR, required to mediate aa-starvation-induced elevated expression levels. The generated reporter-testing system is not only applicable under various growth conditions but also suitable to compare the translational capabilities of different deletion mutants or strain backgrounds. It therefore opens up the possibility to gain more insight into a proteins function within the cell.