The Escherichia coli threonyl-tRNA synthetase gene contains a split ribosomal binding site interrupted by a hairpin structure that is essential for autoregulation


Pascale Romby E-mail; Tel. (3) 88 41 70 51; Fax (3) 88 60 22 18.


The expression of the gene encoding Escherichia coli threonyl-tRNA synthetase (ThrRS) is negatively autoregulated at the translational level. ThrRS binds to its own mRNA leader, which consists of four structural and functional domains: the Shine–Dalgarno (SD) sequence and the initiation codon region (domain 1); two upstream hairpins (domains 2 and 4) connected by a single-stranded region (domain 3). Using a combination of in vivo and in vitro approaches, we show here that the ribosome binds to thrS mRNA at two non-contiguous sites: region −12 to +16 comprising the SD sequence and the AUG codon and, unexpectedly, an upstream single-stranded sequence in domain 3. These two regions are brought into close proximity by a 38-nucleotide-long hairpin structure (domain 2). This domain, although adjacent to the 5′ edge of the SD sequence, does not inhibit ribosome binding as long as the single-stranded region of domain 3 is present. A stretch of unpaired nucleotides in domain 3, but not a specific sequence, is required for efficient translation. As the repressor and the ribosome bind to interspersed domains, the competition between ThrRS and ribosome for thrS mRNA binding can be explained by steric hindrance.


The formation of a translation initiation complex in bacteria requires a limited number of well-characterized signals on the mRNA, which are clustered around the translation initiation site. The first important signal is the Shine–Dalgarno (SD) sequence, a stretch of 3–7 nucleotides that pairs to the 3′ end of the 16S rRNA (Shine and Dalgarno, 1974; Hui and de Boer, 1987). A second signal, located 4–7 nucleotides downstream, is the initiation codon, which can be AUG, GUG or UUG in E. coli. The interactions between the SD sequence and its complement on the 16S RNA and between the initiation codon and the initiator tRNA require that the mRNA region carrying these signals is not sequestered in a stable secondary structure (De Smit and van Duin, 1990). Footprinting experiments delimited the length and position of the ribosomal binding site (RBS), showing that the initiating ribosome protects about 40–50 nucleotides (Murukawa and Nierlich, 1989; Huttenhofer and Noller, 1994). Statistical analysis indicates that, in addition to the SD sequence and initiation codon, the whole RBS is rich in A and U nucleotides, which are probably not able to form stable secondary structures (Ganoza and Louis, 1994). However, numerous examples of secondary structures in RBSs with an inhibitory effect on translation initiation have accumulated, and a thermodynamical framework has been proposed that quantitatively predicts the effect of secondary structure on translation initiation (Draper, 1993; De Smit and van Duin, 1994; De Smit, 1998).

Surprisingly, stable basepairing in the RBS is tolerated in some cases and suggests that secondary structure in mRNA can exist within the ribosomal track (for a review, see De Smit, 1998). For example, the spacing between the SD sequence and the AUG codon in T4 gene 38, which is greater than optimal, has been proposed to be shortened by the presence of a 8-bp-long hairpin (Hartz et al., 1991). Similarly, it has been shown that the ribosome is able to accommodate the 38-nucleotide stem–loop of the selenocysteine insertion sequence (from E. coli fdhF gene) if cloned 3 nucleotides downstream from the start codon, and also the structured unspliced intron from the T4 g60 gene placed between the SD and the start codon (Ringquist et al., 1993). Another example is a 5-bp-long hairpin, located 3 nucleotides upstream of the SD sequence of the T4 IIB gene, that can be increased to 9 bp without inhibitory effect (Shinedling et al., 1987). Longer hairpin structures can be tolerated at a minimal distance of 11 nucleotides upstream of the SD sequence of the repB gene in plasmid pMU720 (Wilson et al., 1994) and of the repA gene in plasmid R1 (Malmgren et al., 1996).

In translational control, the mRNA structure plays an active role by providing specific recognition sites for repressors (for a review, see Draper et al., 1998). The thrS gene, which encodes threonyl-tRNA synthetase (ThrRS), is regulated at the translational level by a feedback mechanism (Springer et al., 1985). ThrRS binds to a site located in the leader mRNA upstream of the SD sequence and inhibits translational initiation by competing with ribosome binding (Moine et al., 1990). The cis-acting region of the thrS leader mRNA essential for the control is called the operator and is composed of four structural domains (Fig. 1). Domain 1 contains the initiation codon and the SD sequence, domains 2 and 4 are two stable hairpin structures that bind ThrRS and domain 3 is a single-stranded region, which acts as a linker between the two structured hairpins. The hairpin loops of domains 2 and 4 both mimic the anticodon region of E. coli tRNAThr (Graffe et al., 1992; Romby et al., 1992). Symmetrical interactions between domains 2 and 4 of the mRNA and the two tRNAThr anticodon recognition sites (one per subunit) of the homodimeric ThrRS have been shown to be responsible for regulation (Romby et al., 1996). Although the interaction between the enzyme and its operator region has been studied in detail, little is known about the regions of thrS mRNA that are recognized by the ribosome, and how the repressor and the ribosome binding sites might interfere.

Figure 1.

. The effect of stabilization of the upper stem of domain 2 on the in vivo expression of thrS (A) and on ribosome binding in vitro (B). The location of the mutations are shown on the secondary structure of thrS operator region. The four domains of the operator without the 5′ end are indicated, and the SD sequence is underlined. The initiation codon is indicated by Met. A. β-Galactosidase levels (given in Miller units) are synthesized from IBPC5421 lysogenized with λMΔ20-10 and its mutated derivatives, λMΔ20-10L6, λMΔ20-10M1 and λMΔ20-10N1. The empty vector and the thrS-carrying plasmid are pBRDAR1SuUAG and pBRUB4SuUAG respectively. The transformed strains were grown in MOPS medium supplemented with glucose, tetracycline and amino acids as indicated in Experimental procedures. The repression for each construct is the ratio of β-galactosidase in the presence of the empty plasmid to that of the thrS plasmid. The standard deviations are given below the numbers. B. Quantitative determination of the toeprint (%) with wild-type and GC-d2 RNAs at 37°C as a function of 30S subunit concentration.

Domain 2 located at the 5′ edge of the SD sequence (Fig. 1) may be suspected to be inhibitory. Another possibility is that the ribosome is able to accommodate the stem–loop structure of domain 2 in the mRNA track. The fact that it is difficult to predict why certain mRNA structures are tolerated and others are inhibitory prompted us to study the role of domain 2 in translation efficiency. The combination of in vitro and in vivo data allows the definition of the mRNA regions involved in the ribosome binding and provides an explanation for the thrS translational control mechanism.


Mutations that increase the stability of domain 2 do not affect thrS expression in vivo and ribosome binding in vitro

At high growth rate, about 600 molecules of ThrRS are synthesized per genome, indicating that thrS is a reasonably well-expressed gene. Therefore, its mRNA is efficiently translated, and the stem of domain 2 just upstream of the SD sequence does not appear to be strongly inhibitory. This could be explained either by the low stability of the stem closing domain 2, which could be in an open conformation often enough to permit efficient ribosome binding, or by the fact that domain 2 does not interfere with ribosome binding. Stabilization of the stem of domain 2 should decrease initiation in the first case, but not in the second. We therefore introduced mutations stabilizing domain 2 and investigated their effect on thrS expression and control (Fig. 1A). The mutations were introduced by site-directed mutagenesis of a thrS–lacZ translational fusion cloned in λ. The mutated bacteriophage was then integrated as a single copy in the E. coli chromosome. Monolysogens were transformed with an empty vector and with a derivative plasmid overproducing ThrRS. The wild-type fusion synthesizes 1013 Miller units of β-galactosidase with the empty vector and only 41 in the presence of the ThrRS-overproducing plasmid, i.e. the repression factor is 1013/41 = 24. Interestingly, changing the three upper U-A basepairs of the stem of domain 2 one by one to more stable G-C basepairs does not significantly affect thrS expression and control (Fig. 1A). This result shows that a strong increase in the stability (−5.8 kcal) of the stem of domain 2 is without effect on thrS mRNA expression.

To prove directly that ribosome binding is unchanged in a mutant with an increased stability of domain 2, toeprinting experiments were performed with wild-type and a mutant RNA having all three U-A pairs changed to G-C pairs (GC-d2 in Fig. 1). Toeprinting is based on the inhibition of reverse transcription of the mRNA by the ternary complex formed with initiator tRNAfMet, the 30S subunit and mRNA (Hartz et al., 1988). We have shown previously that the 30S/tRNAfMet complex bound to thrS mRNA produces a toeprint at position +16 (+1 being the A of the initiation codon) (Moine et al., 1990). With the GC-d2 mutant, the intensity of the toeprint is similar to that of the wild-type mRNA at any ribosome concentration (Fig. 1B), showing that ribosome binding in vitro is not affected. Structure probing experiments indicate that the conformation of GC-d2 RNA and ThrRS footprint remain unchanged. This is supported by the fact that GC-d2 RNA acts as a competitive inhibitor of tRNAThr aminoacylation and that the deduced apparent dissociation constant of the mutant operator for ThrRS (around 10−8 M) remains identical to the wild-type mRNA–ThrRS complex (Romby et al., 1996).

Destabilization or deletion of domain 2 does not cause an increase in expression in the absence of repression

As the above data indicate that domain 2 is not detected as inhibitory by the ribosome, mutations that weaken the helix of domain 2 should have no effect on expression. However, we have shown previously that changes in this stem can abolish control, because they may lead to a complete structural rearrangement of the whole domain (Moine et al., 1988). Therefore, in order to discriminate between direct effects on translational initiation and on control, we tested the effect of mutations that destabilize domain 2 in the presence of a mutation that abolishes regulation.

The mutation G-32 to A (BS4-9) is known to abolish control by ThrRS without affecting the mRNA structure (Brunel et al., 1993). This mutation causes the expression of the thrS–lacZ fusion to increase from about 1250 to 2900 β-galactosidase units (Fig. 2). As this mutation does not affect ribosome binding (Brunel et al., 1992), the increased expression results from the fact that ThrRS is no longer able to compete with the ribosome for mRNA binding. The mutation G-46 to A (CS29) introduced in the BS4-9 construct destroys the unique G-C basepair in the stem close to the SD sequence. This mutation does not cause any increase in expression compared with BS4-9 alone (Fig. 2), thus indicating that the destabilization of domain 2 has no effect on expression once control is abolished. This is further supported by the fact that mutations that abolish the first (L19 and L7) basepair do not significantly affect expression and regulation in the absence of the BS4-9 mutation (Fig. 2). Finally, a deletion of the whole hairpin domain (CS30Δ2 in Fig. 2) does not cause an increase in expression over the BS4-9 mutation, supporting the hypothesis that domain 2 does not interfere with translation initiation in its native context.

Figure 2.

. The effect of the destabilization of the upper stem of domain 2 or its entire deletion on the in vivo expression of thrS. The mutations are shown on the secondary structure of thrS mRNA. The nucleotides deleted in CS30Δ2 are squared. IBPC5421 was lysogenized with λMΔ20-10 and its mutated derivatives, λMΔ20-10L19, λMΔ20-10L7, λMΔ20-10BS4-9, λMΔ20-10BS4-9/CS29 and λMΔ20-10CS30Δ2. Same legend as in Fig. 1.

Domain 3 is necessary for efficient translation in vivo

Footprinting experiments indicate that the RBS covers 54–57 nucleotides in gene 32 mRNA (Huttenhofer and Noller, 1994), the 5′ boundary being around 20 nucleotides upstream of the SD region. As domain 2 is apparently tolerated by the initiating ribosome, we suspected that the RBS of thrS might extend upstream of domain 2. To test this hypothesis, we constructed several thrS–lacZ fusions in which transcription initiates from the lac promoter at different positions upstream of domain 2 (ILOΔ1–5 in Fig. 3). Primer extension experiments were performed for all the constructs to verify that the transcription initiation point was precisely as expected from the position of the cloned lac promoter (data not shown).

Figure 3.

. The effect of changing the location of the transcriptional start on in vivo expression of thrS. The sites of mutations are shown on the secondary structure of thrS mRNA. The transcriptional starts are shown with arrows and are labelled ILOΔ1, ILOΔ2, ILOΔ4 and ILOΔ5. IBPC5421 was lysogenized with λMΔ20-10 and its derivatives, λMΔ20-10ILOΔ1, λMΔ20-10ILOΔ2, λMΔ20-10ILOΔ4 and λMΔ20-10ILOΔ5. Same legend as in Fig. 1.

In the first construct (ILOΔ1 in Fig. 3), transcription starts at G-159. This construct synthesizes about 1550 β-galactosidase units, and the repression factor is close to the value for the wild-type thrS–lacZ fusion. This result confirms previous work showing that sequences upstream of domain 4 are not essential for translational feedback (Brunel et al., 1992). A second construct, ILOΔ2, initiates transcription at G-68 and synthesizes about 3100 β-galactosidase units. This elevated value is caused by decreased control, as the repression factor drops from about 20 to 4 (Fig. 3). This is consistent with former data indicating that a deletion of domain 4 significantly decreases the binding affinity for ThrRS, whereas ribosome binding is not affected (Brunel et al., 1993). With ILOΔ4, transcription starts at the 5′ edge of domain 2 at position −49, where U has been changed to G. To maintain basepairing at the edge of the stem, A-13 was changed to C. Interestingly, the expression drops drastically as the ILOΔ4 construct synthesizes only 20 β-galactosidase units. These data indicate that domain 2 inhibits translation strongly when domain 3 is missing. An additional mutation, G-46 to A, was introduced in the ILOΔ4 construct to give ILOΔ5 (Fig. 3). The same mutation was shown to destabilize the bottom part of the stem of domain 2 (Moine et al., 1988). This last construct synthesizes about 3760 β-galactosidase units compared with 20 for ILOΔ4. Therefore, sequences in domain 2, if in an open conformation, are able to restore translation initiation compared with ILOΔ4, suggesting that a stretch of unpaired nucleotides and not a specific sequence is required upstream of the SD. For these last two constructs, regulation is strongly reduced.

Northern analysis was used to show that RNA synthesis was equivalent for all the constructs. For this purpose, we measured the steady-state level of a tRNA5Arg mutant placed downstream of thrS–lacZ fusion and synthesized from the same transcript. As this reporter tRNA is stable, its level of accumulation reflects the synthesis of the entire transcript. Northern analysis (Fig. 4A) clearly indicates that the reporter tRNA concentration, and thus that of the hybrid mRNA, is almost the same for all constructs with the exception of ILOΔ4, which is much lower. For ILOΔ4, the low reporter tRNA concentration reflects a defect in transcription elongation caused by transcriptional polarity. This effect results from the low translation efficiency of ILOΔ4 (Fig. 3). This polarity effect was shown using a strain mutated for the rho factor, which is responsible for polar effects. This mutant and its wild-type isogenic strain were lysogenized with the two λ phages carrying the ILOΔ4 and ILOΔ5 constructs. The difference between the reporter tRNA concentration in ILOΔ4 and ILOΔ5 is large in a rho+ strain, but much smaller in a rho strain (Fig. 4B). Thus, polarity explains the low reporter tRNA concentration with the ILOΔ4 construct. Altogether, the Northern experiments indicate that transcription initiation varies by less than a factor of two between the ILO constructs. Thus, the β-galactosidase levels synthesized from the different fusions reflect translational effects.

Figure 4.

. Northern analysis of the level of transcription from the thrSlacZ fusions starting at different locations in the thrS operator. The upper band is 5S RNA, and the lower band is tRNA5Arg. A. RNAs extracted from IBPC5421 lysogenized with λMΔ20-10, which does not carry the gene for the tRNA5Arg mutant, with λMΔ20-10ILOΔ1, λMΔ20-10ILOΔ2, λMΔ20-10ILOΔ4 and with λMΔ20-10ILOΔ5. B. RNAs extracted from IBPC5751 (rho) lysogenized with λMΔ20-10 ILOΔ5 and with λMΔ20-10ILOΔ4, or from IBPC5741 (rho+) lysogenized with λMΔ20-10ILOΔ5 and with λMΔ20-10ILOΔ4. The RNA extracted from IBPC5421 lysogenized with λMΔ20-10 does not carry the gene for the tRNA5Arg mutant. The 5S probe was more isotopically diluted than in (A), which explains the weaker bands.

Domain 3 is necessary for ribosome binding in vitro

Toeprinting experiments were performed on several truncated thrS mRNAs in order to study more directly the inhibitory effect of domain 2 on ribosome binding when domain 3 is missing. Ternary complexes were formed with the wild-type and variant mRNAs, the initiator tRNAfMet and increasing concentrations of 30 subunits (Fig. 5). With ILOΔ4 RNA, which starts at the 5′ edge of domain 2, the toeprint at position + 16 is significantly weaker than the wild-type thrS operator, especially at the low concentration of 30S subunits. Therefore, in the absence of domain 3, the hairpin domain 2 is inhibitory for ribosome binding. Conversely, an additional mutation introduced into ILOΔ4, which changes G-40 to A and thus destabilizes domain 2, binds to the ribosome as efficiently as wild-type thrS mRNA (L7-3 in Fig. 5). Chemical probing on this double mutant revealed that most of the paired nucleotides in domain 2 become reactive towards chemical probes at their Watson–Crick positions (data not shown), indicating that the whole domain is destabilized. The mutation X-18, which reduces the complementarity between the SD sequence and the 16S RNA from 6 to 3 nucleotides, was also tested in ILOΔ4 (Fig. 5). This mutation strongly decreases in vivo translation and ribosome binding in vitro to the full-length thrS mRNA (Brunel et al., 1992; 1993). The ribosome binding to X-18-ILOΔ4 RNA is even worse compared with X-18 RNA (Fig. 5). Thus, the in vitro data on ribosome binding are well correlated with the in vivo results (Fig. 3) and show that two regions of thrS mRNA are important for efficient ribosome binding: the SD sequence and unpaired nucleotides upstream of the SD.

Figure 5.

. Toeprint with the truncated (ILOΔ4, X-18 and L7-3) thrS operators at different 30S subunit concentrations. A. Secondary structure model of the truncated thrS mRNAs for which the 5′ end starts at position −49. The 3′ end is at position +61 (+1 being the A of the AUG codon). B. Quantitative determination of the toeprint with the different mutant RNAs at 37°C as a function of 30S subunit concentration. The names of the mutants and the corresponding symbols are shown in the insert.

We have also shown that the truncated ILOΔ4 RNA behaves as a competitive inhibitor of tRNAThr aminoacylation (data not shown). The deduced ThrRS binding affinity is about 50-fold lower than that for the wild-type mRNA but is equivalent to that of a fragment consisting only of domain 2 (Romby et al., 1996). These results are consistent with the in vivo data showing that regulation is strongly affected if transcription starts at the 5′ edge of domain 2 (Fig. 3). It also shows that domain 1, which is part of the RBS, does not contribute to ThrRS binding affinity.

The ribosome protects two non-contiguous regions of the thrS leader

In order to define the regions of thrS mRNA bound to the ribosome, the footprint of the 30S subunit on the mRNA in the presence of the initiator tRNAfMet was studied using several structure-specific probes (Ehresmann et al., 1987). RNase T2 (unpaired nucleotides) and RNase V1 (paired nucleotides) were used as structure-specific enzymes. The four bases were each probed at one of their Watson–Crick positions using dimethylsulphate (DMS), kethoxal and 1-cyclohexyl-2-morpholino-carbodiimide-metho-p-toluene sulphonate (CMCT), and riboses were tested using hydroxyl radical hydrolysis generated by Fe2+-EDTA (Wu et al., 1983). Footprinting experiments were performed with thrS mRNA, which carries a mutation extending the SD sequence from 6 to 8 nucleotides (XI-26), in order to increase ribosome binding affinity (see Fig. 5). This mutation does not affect the conformation of the mRNA and ThrRS footprint (Brunel et al., 1993). The experimental results are shown in 6Fig. 6A, and the reactivity changes are reported on the secondary structure model of thrS mRNA in 6Fig. 6B.

Figure 6.

. Enzymatic and chemical probing on the ternary XI-26 mRNA–30S subunit–tRNAfMet complex. A. Autoradiogram of experiments performed at 25°C on XI-26 RNA with RNase V1 (V1), RNase T2 (T2), DMS and Fe-EDTA. Lanes 1 and 2, incubation controls of RNA in the absence (lane 1) or in the presence of 30S subunits (lane 2); lanes 3 and 4, free RNA treated with 0.05 units and 0.1 units of RNase V1, respectively; lanes 5 and 6, RNA bound to 30S subunit (250 nM) in the presence of initiator tRNAMet (1 μM) treated with 0.05 units and 0.1 units of RNase V1, respectively; lane 7, RNase T2 (0.05 units), DMS and Fe-EDTA reactions performed on free RNA; lanes 8 and 9, reactions performed on mRNA in the presence of 250 nM (lane 8) or 500 nM (lane 9) 30S subunit and initiator tRNAMet (1 μM); lanes A, C and G, sequencing reactions. Experimental details are given in Experimental procedures. B. Chemical and enzymatic probing and 30S footprint on XI-26 mRNA. Reactivity of Watson–Crick position towards DMS, and CMCT modifications: reactive under native conditions, •G strong; G moderate and G weak reactivity; no symbols, unreactive. RNase T2 cleavages are represented by arrows, and RNase V1 cuts by triangles. Reactivity changes induced by 30S subunit binding in the presence of initiator tRNAMet: protected bases and RNase cuts are noted by circles and protected riboses by squares, (▪, •) strong and (□, ○) moderate protections. Enhancements are represented by stars. Nucleotides in italics are not determined. The protected regions are framed.

A region of around 35 nucleotides in thrS mRNA was found to be protected from enzymatic and chemical attack by the 30S subunits in the presence of initiator tRNA. The strongest protections occurred in the region comprising nucleotides −12 to +16 (Fig. 6). The 3′ boundary is in good agreement with the position of the toeprint at + 16 and also with previous studies performed on natural mRNAs, where the 3′ ends of the protected regions were localized between +16 and +21 (Philippe et al., 1993; Huttenhofer and Noller, 1994). The only base-specific protections are located in the SD sequence (A-10, A-9, G-8 and G-7) and at the initiation codon (A + 1, U + 2), reflecting the formation of stable basepairings with the 3′ end of 16S RNA and the anticodon sequence of the tRNAfMet respectively. In addition, extensive protection of riboses was observed in the ternary complex (Fig. 6A), indicating that the 30S subunit makes contact with the ribose–phosphate backbone, as was previously described for gene 32 mRNA (Huttenhofer and Noller, 1994). A stimulatory interaction between a ‘downstream box’ in mRNA and a specific sequence in 16S rRNA (nucleotides 1471–1482) has been proposed (Sprengart et al., 1990). Some complementarity exists with nucleotides U + 8 to U + 18 of thrS mRNA. However, we failed to observe protection of bases in this region whereas moderate enhanced reactivities occur at position N3 of C + 14, U + 17 and U + 18 (Fig. 6). Generally, the 5′ limit of the ribose-protected region of mRNAs varies between residues −17 and −35, suggesting that weaker interactions may occur between ribosome and mRNA (Huttenhofer and Noller, 1994). Interestingly, the 5′ boundary of the ternary complex in thrS mRNA is located in domain 3. Reduced RNase T2 cleavages are observed in the region −53 to −50 concomitantly with protection at riboses −55 to −50 (Fig. 6). No base-specific protection occurred in domain 3, but weak enhanced reactivities are observed at position N3 of U-55 and at position N1 of A-56 and A-57.

It should be noted that most of the enzymatic cleavages and chemical reactivities remain unchanged in the hairpin structure of domain 2 encompassing nucleotides −17 to −46 (Fig. 6). Only the RNase V1 cleavages at positions −13 to −16, RNase T2 cleavages at −48, −47 and riboses −49 to −47 become protected in the ternary complex. The same pattern of reactivity was obtained using GC-d2 RNA, in which the bottom part of the stem has been stabilized by G-C basepairs (result not shown). These data indicate that the protections that occur in the edge of the helix of domain 2 most probably result from steric hindrance. In summary, the ribosome covers two non-contiguous regions, domain 1 (nucleotides U-12 to U +18) and the 3′ part of domain 3, leaving the hairpin loop of domain 2 accessible in the mRNA track.

Ribosomal protein S1 does not enhance translation initiation of thrS mRNA

Based on cross-linking experiments, it has been proposed that protein S1 interacts with U-rich sequences upstream of the SD region of natural mRNAs (Boni et al., 1991). Furthermore, RNAs selected for high-affinity S1 binding were found to be similar to sequences in mRNA regulatory regions, suggesting a potential role for S1 protein in translation initiation (Ringquist et al., 1995). The U-rich sequence of domain 3 is therefore a potential candidate for an S1 binding site. To answer this question, we used a specific allele of the rpsA gene that synthesizes a truncated S1 protein, which is less active as it causes slow growth (Shiba et al., 1986). The effect of the S1 protein on the U-rich single-stranded region in domain 3 was studied by comparing the effect of the rpsA mutation on β-galactosidase synthesized by the thrS–lacZ fusions described in Table 1. No significant effect of the rpsA allele on thrS expression was found using fusions either with (ILOΔ1, ILOΔ2) or without domain 3 (ILOΔ4), suggesting that no strong S1-specific effect could be associated with domain 3.

Table 1. . The effect of ribosomal protein S1 on thrS expression. β-Galactosidase levels (given in Miller units) are synthesized from either IBPC6361 (wt) or IBPC6371(rpsA) lysogenized with λMΔ20-10 its derivatives, λMΔ20-10ILOΔ1, λMΔ20-10ILOΔ2, λMΔ20-10ILOΔ4 and λMΔ20-10ILOΔ5.Thumbnail image of


A split thrS ribosomal binding site interrupted by a non-inhibitory hairpin structure

The purpose of this work was to investigate the effect of secondary structures in the thrS operator on translation initiation and to define the regions of the operator recognized directly by the ribosome. It has been shown by footprinting that the ribosome covers about 54 nucleotides encompassing the SD sequence and the AUG codon in several natural mRNAs (Huttenhofer and Noller, 1994). Therefore, it is expected that domain 2, located at the 5′ edge of the SD sequence, is part of the ribosome binding site. As the stem of domain 2 is not very stable (Moine et al., 1990), it was reasonable to assume that the ribosome may bind to the fraction of the mRNAs at which the stem of domain 2 is in an open conformation. However, this hypothesis is not supported by our present in vivo and in vitro experiments, which establish that the fully folded domain 2 is tolerated by the ribosome in thrS mRNA. The efficiency of translation is not even affected if domain 2 is stabilized by closing the stem with four consecutive G-C basepairs (mutants N1 and GC-d2, Fig. 1), and a footprinting experiment revealed that the hairpin loop of domain 2 is still folded and accessible in the initiation complex. However, translation is strongly decreased when transcription starts directly at domain 2 (mutant ILOΔ4, Fig. 3), indicating that a stretch of unpaired nucleotides of domain 3 are essential for efficient initiation. The inhibitory effect of domain 2, in the absence of upstream unpaired nucleotides, subsists as long as the structure of domain 2 is maintained (Figs 3 and 5). These data also indicate that the nature of the unpaired nucleotides in domain 3 is not important, as they can be replaced by other sequences (e.g. when domain 2 is opened) with an equivalent efficiency. This is also supported by the fact that point mutations or an inversion of the whole sequence in domain 3 did not affect translation (Brunel et al., 1992). However, the size of the loop is important, because deletion of 18–22 nucleotides strongly decreases translation, whereas deletion of 9 and 4 nucleotides does not significantly affect translation (Brunel et al., 1992). The ribosome footprint data indicate that the positive effect of domain 3 might simply be caused by direct interactions between the 30S subunit and the ribose–phosphate backbone of the mRNA, either to the 16S RNA or to some ribosomal protein. Ribosomal protein S1, which has been cross-linked to a U-rich sequence upstream of the SD of several mRNAs (Boni et al., 1991), does not seem to be specifically involved in the recognition of domain 3 by the ribosome (Table 1). However, more direct evidence is certainly required to demonstrate a possible effect of S1 protein on thrS mRNA recognition. Another essential element for ribosome recognition is the SD sequence. The strongest base-specific contacts are indeed located at the SD sequence and the AUG codon. Furthermore, mutation (XI-26), which increases the complementarity with the 3′ end of 16S RNA, enhances ribosome binding and, conversely, mutation (X-18), which decreases the interaction with 16S RNA, strongly inhibits translation (Brunel et al., 1992; Fig. 5). Therefore, an efficient initiation on thrS mRNA requires domain 1 (encompassing the SD sequence and the AUG codon) and several unpaired nucleotides located 40 nucleotides upstream of the SD sequence. These unpaired nucleotides are brought close to the SD sequence by the hairpin structure of domain 2.

Quite a few other mRNAs have non-inhibitory hairpin structures just upstream of the SD sequence. One may speculate that translation initiation in these cases relies on the same strategy. Indeed, the rIIB messenger RNA from phage T4 forms a 5-bp-long helix just upstream of the SD sequence, which has no effect on initiation, even if the hairpin is stabilized by four additional basepairs (Shinedling et al., 1987). It can be assumed that the unpaired nucleotides 5′ of the hairpin may contribute to the efficiency of translation. Conversely, in S. typhimurium pyrD mRNA, the absence of an upstream single-stranded sequence may explain why a stem–loop of 6 bp is inhibitory. Interestingly, at high CTP concentration, the transcription of pyrD starts at a CCCGGT sequence, which forms the 5′ strand of a stem–loop structure inhibiting mRNA translation. When CTP concentration is low, transcription is initiated at the first guanine leading to the formation of a 3 bp helix too weak to abolish downstream translation (Frick et al., 1990). Nevertheless, other examples indicate that longer stem–loop structures located upstream of the SD sequence are inhibitory even in the presence of 5′ unpaired nucleotides. In plasmid R1, it has been shown that a stable 48 nucleotide stem–loop located 2 or 6 nucleotides upstream of the SD is sufficient to prevent ribosome binding in vitro and to inhibit tap translation in vivo despite the presence of upstream unpaired nucleotides (Malmgren et al., 1996). A similar conclusion has emerged from a genetic analysis of the IncB plasmid pMU720 (Wilson et al., 1994). Therefore, the capacity of the ribosome to tolerate hairpin structures immediately upstream of the SD sequence depends on both the presence of an upstream single-stranded region and the size of the secondary structure. However, the three-dimensional folding of the hairpin may play a critical role, making tenuous any general statement about the exact threshold over which a structure is inhibitory, even in the presence of upstream unpaired nucleotides.

Consequences regarding the regulation of thrS expression

Although the present work primarily addresses questions about the role of secondary structures in the thrS RBS, it gives some insights into how thrS regulation works. We have shown previously that ThrRS binds, via symmetrical interactions, with domains 2 and 4, which both resemble the anticodon arm of tRNAThr (Romby et al., 1996). Although ThrRS binding to domains 2 and 4 is required for efficient translational control, domain 2 contains the main specific recognition elements, as its deletion causes a total loss of control (Brunel et al., 1993). The fact that the three consecutive A-U basepairs in domain 2 can be changed to G-C indicates that the bottom part of the helix does not contain any specific determinant for ThrRS binding. However, this helix is most probably required for an optimal presentation of the specific recognition elements located in the apical and the internal loops of domain 2. We have also shown previously that ThrRS does not interact either with the SD sequence or with the single-stranded sequence of domain 3 (Moine et al., 1990; Brunel et al., 1995). The protections against RNases in the 3′ part of domain 3 and in domain 1 most probably result from a steric hindrance effect (Moine et al., 1990).

The present data indicate that the ribosome recognizes both domain 1, which includes the SD sequence and the AUG codon, and the 3′ part of domain 3 which is essential for an efficient translation. ThrRS, on the other hand, essentially recognizes the two anticodon-like domains 2 and 4. In other words, regions involved in ribosome and ThrRS binding are not strictly overlapping, but interspersed. The geometry of the ThrRS–mRNA and 30S–mRNA interactions provides a simple molecular interpretation of the competition between the repressor and the 30S subunit for mRNA binding (Moine et al., 1990). Indeed, regulation seems to work mainly by a steric hindrance-guided process resulting from interspersed ribosome and repressor recognition sites. Nevertheless, changes in the spatial orientation of the different domains of thrS mRNA upon ThrRS binding may occur. In the unbound mRNA, the two structured domains 2 and 4 move freely within a radius defined by the length of domain 3. In the presence of ThrRS, domains 2 and 4 of the operator interact co-operatively with the two tRNAThr anticodon recognition sites on each subunit of the enzyme, freezing the mRNA in a defined conformation. The resulting complex may induce structural constraints in the connecting domain 3 that renders the unpaired nucleotides in domain 3 not available for ribosome binding. On the other hand, binding of the 30S subunit may constrain the orientation of the two anticodon-like domains in a conformation that does not allow the two domains to bind co-operatively to ThrRS, even if domain 2 is accessible in the initiation complex. In conclusion, thrS regulation, which involves competition between the repressor protein and the ribosome, probably relies on a combination of steric hindrance and changes in the spatial orientation of the four domains of thrS mRNA in both complexes. A more precise understanding of the regulatory mechanism will most probably require the determination of the high-resolution structure of the thrS mRNA–ThrRS complex.

Experimental procedures

General techniques and bacterial growth conditions

The E. coli strains, plasmids and bacteriophages are described in Table 2. Nucleotide sequences in the thrS region are found in GenBank under the name ECOTHRINF. General genetic and cloning techniques are as described by Miller (1972) and Sambrook et al. (1989). IBPC5723 and IBPC5731 were derived from IBPC5701 and IBPC5711, respectively, by selection for tetracycline sensitivity on CTF (chlortetracycline–fusaric acid) plates (Maloy and Nunn, 1981). Among the tetracycline-sensitive clones, some were Pro and most probably correspond to deletions that extend from aroL478::Tn10 to proC. The corresponding genotypes are called Δ(aroL478::Tn10-proC)1 and Δ(aroL478::Tn10-proC)2 in Table 2. Lysogens carrying the fusions were selected as described previously (Springer et al., 1985). For each phage, three independent lysogens were tested for β-galactosidase levels. Values varied as multiples of the lowest value, which was considered that of a monolysogen. Bacteria were grown in MOPS medium (Neidhardt et al., 1974) supplemented with glucose at 0.2% w/v or glycerol at 0.4% v/v. Amino acids and vitamins were supplemented as already published (Neidhardt et al., 1977). The tests for β-galactosidase were performed with toluene-permeabilized cells (Miller, 1972).

Oligonucleotide site-directed mutagenesis

Oligonucleotide site-directed mutagenesis was performed on M13mp8Δ20-10 using the U-containing DNA method (Kunkel et al., 1987). Sequencing was done from the thrS–lacZ boundary to about 50 nucleotides upstream of the thrS transcription initiation site using the dideoxynucleotide method (Sanger et al., 1977).

Phage and plasmid constructions

The derivatives of λMΔ20-10 carrying the thrS–lacZ fusions were constructed from M13mp8Δ20-10 or their mutant derivatives as reported previously (Brunel et al., 1992). The phages λMΔ20-10ILOΔ1, 2, 4 and 5 were constructed using the same protocol with M13mp8Δ20-10ILOΔ1, 2, 4, 5 and λSKS107T. λSKS107T is identical to λSKS107 but carries a mutated version of the gene for tRNA5Arg behind lacZ. The phage λSKS107T was constructed by ligating three different DNA pieces together: (i) the left arm of λcI857S7 up to its Ngo MI site; (ii) the BspEI–SacI fragment of pLccLacZ-Arg5; and (iii) the right arm of λSKS107 from its SacI site to cosL. The plasmid pLccLacZ-Arg5 is identical to pTLacZ-Arg5 (Lopez et al., 1994), except that the T7 late promoter has been replaced by a modified version (two C inserted between −2 and −3 of transcription) of the lac promoter (a generous gift from M. Dreyfus and P. Lopez). This plasmid carries the gene for tRNA5Arg with the A-73 to U mutation after lacZ. M13mp8Δ20-10ILOΔ1, 2, 4 and 5 were constructed in three steps. We first deleted the thrS promoter, which is located between two HincII sites on M13mp8Δ20-10 by digestion and religation. Sequencing of the recombinant (M13mp8Δ20-10ΔHII) showed that the HincII junction was abnormal (GTCAAC was replaced by GTTACCGGCAATCAGCAC). In a second step, the lac promoter was introduced in place of the thrS promoter. This was done by cloning an EcoRI–BamHI fragment carrying the lac promoter and operator from pILO into M13mp8Δ20-10ΔHII. The recombinant (M13mp8Δ20-10ILO) carries the lac promoter and operator twice: one copy from M13mp8 and the other from pILO. In a third step, M13mp8Δ20-10ILOΔ1–5 were made by oligonucleotide site-directed deletions between the pILO copy of the lac promoter and the thrS leader. Mutations other than the deletions were introduced using either the same or an independent second mutagenic oligonucleotide. All the recombinants were sequenced from upstream of the lac promoter to within thrS.

In vitro transcription was performed with plasmids in which the regulatory regions of thrS were transcribed from a T7 late promoter. These plasmids were constructed by cloning a HpaI (the site straddles the −10 box of the thrS promoter) to HindIII (located at the thrS–lacZ boundary) fragment from M13mp8Δ20-10 or their mutant derivatives between the HincII and HindIII sites of pTZ18R+.

Northern blot analysis

Extraction of total bacterial RNA was performed as described previously (Hajnsdorf et al., 1994).

RNA samples (20 or 25 μg) were electrophoresed on 10% polyacrylamide–7 M urea gels, blotted and UV cross-linked onto a nylon membrane (Hybond N+; Amersham) (Lopez et al., 1994). The blots were probed with two 32P-labelled oligonucleotides, one specific for 5S RNA and the other for tRNA5Arg carrying the A73→U mutation, as described previously (Lopez et al., 1994). The 5S probe was diluted isotopically in order to reduce the intensity of the 5S signal and allow probing with the two oligonucleotides together. The hybridization signals were quantified using a Phosphorimager (Molecular Dynamics).

Primer extension experiments

The lacZ 17-mer sequence primer (New England Biolabs) was labelled with 32P and used to prime reverse transcriptase. RNA samples (25 μg) were annealed to this primer, and the reactions were performed as described previously (Uzan et al., 1988). Sequencing was carried out with single-stranded DNA from M13mp8Δ20-10ILO and loaded on the same 6% polyacrylamide–urea sequencing gel.

RNA and protein preparations

Wild-type thrS mRNA and mutant RNAs (X-18 and XI-26) corresponding to nucleotides −192 to +61 (+1 being the A of the thrS translational initiation codon) were synthesized by in vitro transcription with T7 RNA polymerase from plasmids linearized with HindIII as described by Milligan et al. (1987). Mutant RNAs were transcribed from DNA templates containing a T7 promoter followed by the mRNA gene, which was obtained by polymerase chain reaction (PCR) amplification of the appropriate plasmid. GC-d2 DNA template was generated by two successive PCR amplifications. Primers used for the generation of the truncated mutants were: for GC-d2, CAATTTTTCTTTGTATGTGATCTTTGCTCTCCCTCACCACTCGTTTTAAGGATATA (first PCR, 5′ oligo), CTTGGGCTTACAGCGTGA (first PCR, 3′ oligo: P1), and TAATACGACTCACTATAGATTCGGCAACCAATTTAGCATTTTG (second PCR, 5′ oligo containing the T7 promoter); for the truncated mutants ILOΔ4 (wild type, X-18), TAATACGACTCACTATAGTTGTATGTGATCTTTCGTGTGGGTCACCACTGCAAC (5′ oligo containing the T7 promoter) and P1 as 3′ primer; for the truncated L7-3, TAATACGAC-

TCACTATAGTTGTATGTAATCTTTCGTGTGGGTCACCACTGCAAA and P1 as 3′ primer. Full-length RNA transcripts were purified by high-performance liquid chromatography (HPLC) using a Bio-sil TSK250 column as described previously (Moine et al., 1990). Before use, RNAs were renatured by incubation at 90°C for 1 min in water and slow cooling to 20°C in the appropriate buffer.

Threonyl-tRNA synthetase was purified from the E. coli overproducing strain IBPC6163λMΔ20-10-XII-25 at 4°C as described by Brunel et al. (1993). The enzyme is stored in a buffer containing 25 mM HEPES–NaOH, pH 7.5, 5 mM MgCl2, 5 mM β-mercaptoethanol, 20 mM KCl and 50% glycerol. The activity of the enzyme was measured by the rate of threonyl-tRNA formation as described by Théobald et al. (1988).

Toeprinting assays

The formation of a simplified translational initiation complex and the extension inhibition conditions were strictly identical to those described by Moine et al. (1990). 30S subunits were prepared from tight couples according to a procedure adapted from Makhno et al. (1988). Standard conditions contained 15 nM mRNA annealed to a 5′ end-labelled oligonucleotide complementary to nucleotides +47 to +61, 50 nM to 1 μM 30S subunits, 1 μM non-acylated initiator tRNAMet in 10 μl of buffer A (20 mM Tris-acetate, pH 7.5, 60 mM NH4Cl, 10 mM magnesium acetate, 3 mM β-mercaptoethanol). Incubation was for 15 min at 37°C. Reverse transcription was conducted with one unit of AMV reverse transcriptase for 15 min at 37°C. The reactions were stopped by the addition of 10 μl of loading buffer. Relative toeprinting (toeprint band over 5′ ends + toeprint) was calculated by scanning of the gel with the Bio-imager Analyser BAS 2000 (Fuji).

Enzymatic and chemical footprinting

Conditions for ternary complex formation were as described previously (Huttenhofer and Noller, 1994). The mRNA (15 nM) was incubated with renatured 30S subunits (250–500 nM) and initiator tRNAMet (1 μM) in 20 μl of buffer A at 37°C for 10 min and then for 20 min on ice. Enzymatic hydrolysis was performed on free and bound mRNAs in the presence of 0.1 unit of RNase V1 or 0.05 units of RNase T2 for 5 min at 25°C. Chemical probing was performed at 25°C by adding 2 μl of DMS (diluted 1:16 in ethanol), 5 μl of kethoxal (20 mg ml−1 in 20% ethanol) or 4 μl of CMCT (42 mg ml−1) for 5, 10 and 15 min respectively. Hydroxyl radical footprinting was done in 25 μl containing the free or bound mRNA at 4°C for 10 min in the presence of 1 μl of Fe(NH4)2(SO4)2, 1 μl of 0.1 M EDTA, 1 μl of 0.5% (v/v) H2O2 and 1 μl of 0.25 M dithiothreitol (DTT). The reactions were stopped by ethanol precipitation in the presence of 0.3 M sodium acetate (pH 5.5). The RNA was then redissolved in 0.3 M sodium acetate, extracted twice with phenol/chloroform and precipitated. Cleavages and modified bases were identified by primer extension as published previously (Moine et al., 1990). The 5′ end-labelled DNA primer is complementary to nucleotides +47 to +61. Incubation controls in the presence or in the absence of 30S subunits were run in parallel in order to detect nicks in the RNA and pauses of reverse transcriptase.

Aminoacylation assays

Initial rates of reaction (v) were measured at 37°C at various tRNA3Thr concentrations (90–750 nM) in the absence or in the presence of the different renatured mRNAs according to Romby et al. (1992): wild type (25 and 50 nM), ILOΔ4 (1 and 5 μM) and GC-d2 (25 and 50 nM).


We thank P. Lopez, I. Boni and M. Dreyfus for strains, plasmids and advice. We are grateful to M. De Smit, S. Lodmell and H. Moine for helpful discussions and critical reading of the manuscript. This work was supported by grants from the CNRS (UPR 9002, UPR 9073 and PCV 97-142) and the Association pour la Recherche sur le Cancer (number 9440) to M.S.