Expression Levels Influence Ribosomal Frameshifting at the Tandem Rare Arginine Codons AGG_AGG and AGA_AGA in Escherichia coli

Abstract

The rare codons AGG and AGA comprise 2% and 4%, respectively, of the arginine codons of Escherichia coli K-12, and their cognate tRNAs are sparse. At tandem occurrences of either rare codon, the paucity of cognate aminoacyl tRNAs for the second codon of the pair facilitates peptidyl-tRNA shifting to the +1 frame. However, AGG_AGG and AGA_AGA are not underrepresented and occur 4 and 42 times, respectively, in E. coli genes. Searches for corresponding occurrences in other bacteria provide no strong support for the functional utilization of frameshifting at these sequences. All sequences tested in their native context showed 1.5 to 11% frameshifting when expressed from multicopy plasmids. A cassette with one of these sequences singly integrated into the chromosome in stringent cells gave 0.9% frameshifting in contrast to two- to four-times-higher values obtained from multicopy plasmids in stringent cells and eight-times-higher values in relaxed cells. Thus, +1 frameshifting efficiency at AGG_AGG and AGA_AGA is influenced by the mRNA expression level. These tandem rare codons do not occur in highly expressed mRNAs.

The expression of a minority of genes in probably all organisms utilizes a change in the translational reading frame at a specific site(s). The ribosomal frameshifting involved occurs at shift-prone sequences, and the proportion of ribosomes that participate in frameshifting is greatly elevated in many cases by recoding signals embedded in the mRNA. The “programmed” ribosomal frameshifting in these instances is used for regulatory purposes or to synthesize an additional useful product in a fixed ratio to that of standard decoding (reviewed in references 7, 9, 43, and 56). In some cases, however, only low levels (1 to 2%) of specific frameshifting are important, and the presence, if any, of stimulatory signals aside from the frameshift site and flanking 3′ codon is not evident (10, 23, 67).

Under standard conditions, shift-prone sites exhibit levels of frameshifting orders of magnitude higher than the level of background error frameshifting (1, 35). In addition, certain sequences which exhibit low levels of frameshifting under standard conditions have this level greatly elevated under amino acid starvation (20, 21, 38). Though no examples are known to date, some instances of the latter may be advantageous to the cell in more elaborate ways than simply leading to termination and ribosome recycling. However, under standard conditions, frameshifting at most shift-prone sites, which is not positively utilized for gene expression, likely yields erroneous frameshift products whose synthesis is a waste of energy. Consequently, one might expect shift-prone sequences to be selected against and therefore underrepresented in coding sequences, especially those of highly expressed genes. Conversely, when shift-prone sequences occur, analysis for their possible utilization in gene expression is merited (25, 52).

Here we investigate occurrences of two “shifty” sequences in Escherichia coli K-12 genes, determine the level of frameshifting at the sequences, and assess whether the frameshifting at any of these occurrences is likely utilized for gene expression. The two shift-prone sequences are the tandem rare arginine codons AGG_AGG and AGA_AGA. In an earlier investigation which started with overexpression of a mammalian gene in E. coli, Spanjaard and van Duin (55) found that translation of AGG_AGG (or AGA_AGA), when present in highly expressed mRNA, yields up to 50% frameshifting. AGG and AGA are decoded by separate (54) and sparse (anticodons ^3′UCC^5′ and ^3′UCU*^5′, respectively, where U* is 5-methylaminomethyl-2-thiouridine) tRNAs. Sequestration of the minor tRNA^Arg from the limited local pool by the first codon reduces its availability for the second codon of the pair, which then can be considered a “hungry” codon (a term coined by J. Gallant). When the second codon of the pair is at the ribosomal A site, the longer-than-usual time for arrival of its cognate tRNA increases the chance for the dissociation of peptidyl-tRNA, which may re-pair to mRNA at the overlapping +1 frame codon (GGA or GAA). Even though base pairing with the new P site codon is not optimal, the availability of the tRNA cognate to the new A site codon makes frameshifting favorable (6, 46). When this happens, ribosomes continue protein synthesis in the new frame. The introduction of an extra copy of tRNA_3′UCC5′^Arg or tRNA_3′UCU*5′^Arg diminishes frameshifting on AGG_AGG and AGA_AGA, respectively (54). This indicates that limitation of the cognate tRNA is responsible for the “shiftiness” of the AGG_AGG and AGA_AGA sequences. Since separate tRNAs decode AGG and AGA in E. coli, it is not surprising that the mixed codon pairs AGG_AGA and AGA_AGG exhibit undetectable or greatly reduced frameshifting (54).

Previous studies have shown the importance of tRNA balance for frameshifting (2, 21, 45, 58). A corollary is that the abundance of mRNAs containing sequences prone to frameshifting due to tRNA limitation is expected to be equally important. Such mRNAs, when expressed at high levels, can sequester a significant fraction of the rare tRNAs at the first of the tandem codons, thereby significantly depleting the pool of free tRNAs and increasing the frequency of frameshifting. Spanjaard and van Duin (55) noted an unspecified decrease in frameshifting when the mRNA concentration was reduced, although no data were presented. Here, the efficiency of frameshifting in the native-gene context was also examined under different mRNA expression levels.

MATERIALS AND METHODS

Bacterial strains and plasmids.

The E. coli strains and plasmids used in this study are listed in Table 1. Gene sequences (yhaC, smf, fecC, lhr, recF, t150, and emrK) were amplified by PCR from E. coli genomic DNA by use of primers with appropriate overhanging restriction sites and cloned into the BamHI/EcoRI sites in between glutathione S-transferase (GST) and maltose binding protein (MBP) of pGHM57, so that MBP was in the +1 frame relative to GST. Additionally, the emrK sequence (coding for multidrug efflux protein) was cloned into the HindIII/ApaI sites in between GST and lacZ of pSKAGS, so that lacZ was either in the +1 frame relative to GST (frameshift reporter) or in the 0 frame (in-frame control). Further, the GST-emrK fusions were amplified by PCR from the constructions in the pSKAGS vector and cloned into the NdeI site of pLA2 in between P_araB and lacZ. All plasmid constructions were confirmed by DNA sequencing on automated sequencing machines (model ABI3730).

TABLE 1.

E. coli strains and plasmids used in this study

Strain or plasmid	Genotype and/or characteristic(s)	Reference or source	Experiment used
Parental strains
DH5α	endA1 recA1 relA1 gyrA96 hsdR17 (rK− mK+) phoA supE44 thi-1 Δ(lacZYA-argF)U169 Φ80 Δ(lacZ)M15 F−	Lab stock	Pulse-chase
CP78 MC1000	thr leu his arg thi Δlac	17 (obtained from J. Gallant)	Nonintegrated β-Gal assay
CP79 MC1000	thr leu his arg thi Δlac relA2	17 (obtained from J. Gallant)	Nonintegrated β-Gal assay
BW27786	lacIqrrnB3 ΔlacZ4787 hsdR514 DE(araBAD)567 DE(rhaBAD)568 DE(araFGH) Φ(ΔaraEp Pcp13-araE)	34 (obtained from E. coli Genetic Stock Center)
New strains
JFG001	BW27786 ParaB-GST-emrK-lacZ(+1), lacZ in +1 frame relative to GST	This study	Integrated β-Gal assay
JFG002	BW27786 ParaB-GST-emrK-lacZ(0), lacZ in 0 frame relative to GST	This study	Integrated β-Gal assay
Parental vectors
pGHM57	Plac GST-MBP pBR322 ori bla lacIq	29
pSKAGS	Plac GST-lacZ pBR322 ori bla	66
pLacIq	lacIqkan	Gift from R. Maldonado
pLA2	ParaB lacZ kan, λ integration site, oriR	26 (obtained from E. coli Genetic Stock Center)
pINT-ts	λ integrase	26 (obtained from E. coli Genetic Stock Center)
New vectors
pGHM-yhaC	pGHM57 GST-yhaC-MBP	This study	Pulse-chase
pGHM-smf	pGHM57 GST-smf-MBP	This study	Pulse-chase
pGHM-fecC	pGHM57 GST-fecC-MBP	This study	Pulse-chase
pGHM-lhr	pGHM57 GST-thr-MBP	This study	Pulse-chase
pGHM-recF	pGHM57 GST-recF-MBP	This study	Pulse-chase
pGHM-t150	pGHM57 GST-t150-MBP	This study	Pulse-chase
pGHM-emrK	pGHM57GST-emrK-MBP	This study	Pulse-chase
pSKAGS-emrK+1	pSKAGS GST-emrK-lacZ, lacZ in +1 frame relative to GST	This study	Nonintegrated β-Gal assay
pSKAGS-emrK-0	pSKAGS GST-emrK-lacZ, lacZ in 0 frame relative to GST	This study	Nonintegrated β-Gal assay
pLA2-GST-emrK+1	pLA2 GST-emrK-lacZ, lacZ in +1 frame relative to GST	This study	Integrated β-Gal assay
pLA2-GST-emrK-0	pLA2 GST-emrK-lacZ, lacZ in 0 frame relative to GST	This study	Integrated β-Gal assay

Plasmid integration.

The JFG001 and JFG002 integrant strains were constructed as described by Haldimann and Wanner (26). Briefly, electrocompetent BW27786 cells harboring pINT-ts helper plasmid were transformed with pLA2(GST-emrK) vectors, incubated for 1 h at 37°C and for 30 min at 42°C, spread onto agar containing 10 μg/ml kanamycin, and incubated at 37°C. Colonies were tested by PCR for copy number according to the method of Haldiman and Wanner (26). Single-copy integrants were then purified once nonselectively, tested for antibiotic resistance for stable integration and loss of the helper plasmid, and retested for copy number.

Estimate of frameshift efficiency by pulse-chase analysis.

Overnight cultures of strains expressing the appropriate construct were grown in MOPS (morpholinepropanesulfonic acid)-glucose (44) containing 100 μg/ml ampicillin and all amino acids (150 μg/ml each) except methionine and tyrosine and diluted 1:50 in 300 μl of the same medium. After 2 h of incubation at 37°C, cultures were induced with 1 mM IPTG (isopropyl-β-d-thiogalactopyranoside) for 10 min. The cells were pulse-labeled for 2 min by the addition of 7.5 μCi [³⁵S]methionine in 30 μl of medium, chased for 2 min by the addition of 30 μl of 50-mg/ml cold methionine, chilled on ice, and harvested by centrifugation. The pellets were resuspended in 50 μl cracking buffer (6 M urea, 1% sodium dodecyl sulfate, 50 mM Tris-HCl [pH 7.2]) and heated at 95°C for 5 min. Five-microliter aliquots were loaded on 4 to 12% NuPAGE gels (Invitrogen, Inc.) and electrophoresed, under conditions recommended by the manufacturer, in MOPS-sodium dodecyl sulfate buffer (Invitrogen, Inc.). Gels were exposed overnight and visualized with a Molecular Dynamics PhosphorImager. The amounts of termination and frameshift products were quantified by ImageQuant. The frameshifting efficiency was estimated as the ratio of the amount of frameshift product to the sum of the termination and frameshift products.

Estimate of frameshift efficiency by β-galactosidase assays.

Overnight cultures of CP78/CP79 harboring the pSKAGS-emrK constructs and pLacI^q or JFG001 and JFG002 integrant strains were grown in MOPS-glucose (44) containing appropriate antibiotics and all amino acids (150 μg/ml each) except tyrosine and diluted 1:50 in 800 μl of the same medium. After 2 h of incubation at 37°C, half of the cultures were induced (with either 1 mM IPTG or 0.02% or 0.2% l-arabinose), and incubation continued for another hour. Cultures were chilled on ice, and β-galactosidase assays were performed as described by Miller (42). Assays were performed in triplicate on three separate days with at least two independent clones/integrants. The frameshifting efficiency was calculated as the ratio of β-galactosidase activity of the cells harboring the frameshift reporter to the β-galactosidase activity of the cells harboring the in-frame control construct.

Random-genome analysis.

The CodonShuffle and the DicodonShuffle programs were obtained from Christopher Burge (33). The NC_000913.ffn file (the 24 June 2004 updated version is gi 49175990), which contains the nucleotide sequences of all E. coli K-12 protein-coding genes, was downloaded from the National Center for Biotechnology Information's website (http://www.ncbi.nlm.nih.gov/). The MotifCounter program was written with Java2 (http://java.sun.com/) and BioJava (http://www.biojava.org/) (25).

The DicodonShuffle or the CodonShuffle program was used to generate 1,000 randomized nucleotide files from the original NC_000913.ffn file. The MotifCounter was then used to search through the 1,000 new files and to count how many times a motif occurred in each genome. The results were further analyzed in Microsoft Excel for the distribution of a motif in 1,000 random genomes.

Calculation of the total number of genes in which error frameshifting occurs.

AGA_AGA occurs 42 times in the genome; however, in three genes, it occurs twice. Therefore, there are 39 AGA_AGA-containing genes. AGG_AGG occurs four times in the genome in four different genes. CCC_TGA occurs 19 times in the E. coli K-12 genome; however, frameshifting was experimentally verified to occur in only nine genes. A_AAA_AAG occurs 70 times in E. coli K-12: 9 times in programmed frameshifting cases, 2 times in non-protein-encoding genes (misannotated; see reference 25), and 1 time in a gene that also contains CCC_TGA and therefore has been taken into account already. We presume that all genes that contain AGG_AGG, AGA_AGA, and A_AAA_AAG exhibit some level of frameshifting, since frameshifting was experimentally detected in all of the genes tested that contain these sequences. Therefore, the total number of genes in which error frameshifting occurs is 108 (39 AGA_AGA plus 4 AGG_AGG plus 9 CCC_TGA plus 56 A_AAA_AAG). There are 4,242 protein-encoding genes annotated in E. coli K-12 (NC_000913.ffn; gi 49175990), of which 108 genes constitute about 2.5%.

RESULTS

Occurrence of tandem AGG and AGA codons in E. coli K-12 genes.

A search for AGG_AGG and AGA_AGA sequences in annotated E. coli K-12 open reading frames (ORFs) revealed that there are 4 occurrences of AGG_AGG and 44 occurrences of AGA_AGA (genes containing these sequences are listed in Table 2). However, two of the genes containing AGA_AGA, yjbF and pnp, were excluded from the final count of the AGA_AGA occurrences and from the analysis which follows. For both genes, the predicted start site is not conserved among different species and lacks an appropriate Shine-Dalgarno sequence upstream. The probable initiation site is located downstream of AGA_AGA. In fact, a downstream start site was previously reported for pnp (47). Both genes were annotated so that they contained AGA_AGA in the E. coli K-12 genome version available in August 2003 when the work was initiated and in the final version from 24 June 2004 (gi 49175990) used in the text. However, several intermediate releases (i.e., gi 16127994) had start sites annotated downstream of AGA_AGA for both genes. Therefore, we presume that these genes do not contain AGA_AGA and that the total count of AGA_AGA is 42. (Two more genes, b2655 and livJ, were annotated to contain AGA_AGA in their coding sequences in the August 2003 genome version. Alignment of the b2655 sequences from different bacteria showed that there is a nucleotide insertion upstream of the AGAAGA sequence in the E. coli K-12 gene. In other bacteria, AGAAGA in b2655 is read in the frame NNA_GAA_GAN. When this portion of the gene was amplified by PCR from the DH5α strain [a derivative of K-12] and sequenced, we found that the frameshift mutation was not present and that AGAAGA was read in the same frame as in other bacteria. The June 2004 release of the E. coli K-12 genome has the corrected sequence. Our analysis of livJ suggested that the start codon is located downstream of AGA_AGA; a subsequent genome version also had the start codon annotated downstream of AGA_AGA. Consequently, it is not surprising that Boycheva et al. [12] reported 45 occurrences of AGA_AGA in the E. coli K-12 genome, since different releases do vary slightly.)

TABLE 2.

E. coli K-12 genes containing AGG_AGG or AGA_AGA sequences

Gene name	Protein function	Gene length (nt)	Positions of tandem codons (nt)	No. of sense codons after +1 shift	Conservation of tandem codona	Protein conservationa
AGG_AGG-containing genes
ninE	Unknown	171	154, 160	113	K-12, O157, phage 82	K-12, O157, bacteriophages
yhaC	Unknown	1,188	409, 415	1	E.c., S.f.	E.c., S.f.
smf	Putative Rossmann-fold nucleotide-binding protein	1,125	1084, 1090	471	E.c., S.f., S.t., Y.p.	Conserved
fecC	Citrate-dependent iron(III) transport	999	985, 991	90	K-12	Conserved
AGA_AGA-containing genes
intE	Prophage e14 integrase	1,128	556, 561	7	K-12	Conserved
rfbX	Putative O-antigen transporter	1,248	25, 30	24	K-12	Conserved
yagM	CP4-6 prophage	855	685, 690	45	K-12	Conserved
yacH	Putative membrane protein	1,854	1837, 1842	8	K-12	Conserved
ydaU	Rac prophage	858	370, 375	15	K-12	K-12
ymfK	e14 prophage, putative phage repressor	675	652, 657	10	K-12	K-12
rfaS	Lipopolysaccharide core biosynthesis	936	640, 645	2	K-12	K-12
ymfH	e14 prophage	312	196, 201	7	K-12	K-12
ydfO	Qin prophage	426	289, 294	46	K-12, CFT073	E.c.
gspA	Putative export protein A	1,470	10, 15	40	K-12, CFT073	Conserved
ycdF	Unknown	231	199, 204	44	K-12, CFT073, S.f.	K-12, CFT073, S.f.
ygeP	Unknown	300	280, 285	5	K-12, O157	K-12, O157
yjcF	Unknown	1,293	325, 330	17	K-12, O157, S.f.	K-12, O157, S.f.
yhiJ	Unknown	1,623	799, 804	4	K-12, O157, S.f.	K-12, O157, S.f.
t150	IS150 putative transposase	852	586, 591	33	K-12, S.f.	Conserved
t150	IS150 putative transposase	852	778, 783	3	K-12, S.f.	Conserved
ylbH	Unknown	711	307, 312	5	E.c.	E.c.
b1459	Unknown	201	181, 186	5	E.c.	Conserved
ydeN	Putative sulfatase	1,716	19, 24	4	E.c.	Conserved
emrK	Multidrug resistance protein K	1,164	37, 42	8	E.c.	Conserved
intC	Putative prophage Sf6-like integrase	1,158	484, 489	26	E.c.	Conserved
sfmF	Putative fimbria-like protein	516	4, 9	15	E.c.	Conserved
yfcC	Putative S-transferase	1,542	52, 57	8	E.c.	Conserved
ygeH	Putative invasion protein	1,377	1210, 1215	5	E.c.	Conserved
yhiU	Putative membrane protein	1,158	7, 13 + 11, 16	3 or 4	E.c.	Conserved
ybcK	DLP12 prophage, putative recombinase	1,527	73, 78	5	E.c.	Conserved
ybcK	DLP12 prophage, putative recombinase	1,527	1000, 1005	8	E.c.	Conserved
ybfL	Putative receptor	858	694, 699	17	E.c.	Conserved
ydcC	H repeat-associated protein	1,137	973, 978	17	E.c.	Conserved
yhhI	H repeat-associated protein	1,137	973, 978	17	E.c.	Conserved
yjgR	Putative nucleotide triphosphate hydrolase	1,503	1492, 1497	60	E.c.	Conserved
yqeI	Putative sensory transducer	810	292, 297	14	E.c.	Conserved
intR	Rac prophage, putative transposase	1,236	757, 7632	2	E.c., S.f.	Conserved
ynbB	Putative phosphatidate cytidiltransferase	897	577, 582	15	E.c., S.f.	Conserved
lhr	Enzyme; DNA replication and repair	4,617	2797, 2802	15	E.c., S.f.	Conserved
yddW	Unknown	1,320	37, 42	4	E.c., S.f.	Conserved
yifQ	Unknown	726	703, 708	12	E.c., S.f.	Conserved
mdoC	Membrane protein for succinylation of osmoregulated periplasmic glucans	1,158	580, 585	10	E.c., S.f.	Conserved
ybjR	N-Acetylmuramoyl-l-alanine amidase	831	4, 9	45	E.c., S.f.	Conserved
recF	Gap repair protein	1,074	394, 399	35	E.c., S.f., S.t.	Conserved
trmD	tRNA (guanine-1-)-methyltransferase	768	655, 660	9	E.c., S.f., S.t., Y.p.	Conserved
yjbFb	Putative membrane-associated protein	669	22, 27	2	E.c., S.f.	Conserved
pnpb	Polynucleotide phosphorylase	2,205	7, 12	7	E.c., S.f.	Conserved

^aBacterial names are abbreviated as follows: K-12, E. coli K-12; O157, E. coli O157:H7; CFT073, E. coli CFT073; E.c. includes all three E. coli species (K-12, O157:H7, and CFT073); S.f., Shigella flexneri 2a strain 301; S.t., Salmonella enterica serovar Typhimurium LT2; Y.p., Yersinia pestis KIM. A protein is considered conserved if its homologues are present in at least four different genera.

^bThe genes yjbF and pnp were excluded from the total count of the AGA_AGA occurrences and all other considerations in the text for the reasons discussed in the text.

A statistical analysis was performed to find out whether AGG_AGG and AGA_AGA are over- or underrepresented in coding regions. For this purpose, the DicodonShuffle program was employed (33). The DicodonShuffle program generates random genomes from the real genome in question (i.e., E. coli K-12) by randomizing mRNA sequences while preserving the sequences of the encoded protein, the codon usage, and the dinucleotide composition of the original message. Consequently, the random genome maintains the same amino acid constraints as the real genome but is relieved from the nucleotide constraints. One thousand random genomes were generated, and the distribution of sequences in question was analyzed (Fig. 1). The mean values of AGG_AGG and AGA_AGA occurrences in the 1,000 generated genomes are 5.1 and 38.3, respectively. The observed numbers in real genomes of 4 for AGG_AGG and 42 for AGA_AGA are within the distribution range and are anticipated in the absence of selective pressure (Fig. 1A). It should be noted that the value 38.3 predicted here by the DicodonShuffle program for the AGA_AGA occurrences is strikingly different from the expected value of 6.2 reported by Boycheva et al. (12). However, in their estimate, they took into account only the codon usage bias of the native genome. When we generated the random genomes using the CodonShuffle program (33), which does not take into account the dinucleotide composition and is similar to an approach used before (25, 51), the AGA_AGA occurrence was predicted to be 7.2. Thus, apparently the dinucleotide composition of the E. coli K-12 genome contributes greatly to the occurrence of AGA_AGA.

FIG. 1.

Distribution of occurrences of AGGAGG and AGAAGA sequences in 1,000 computer-generated random genomes.

Distribution analysis of AGGAGG in other frames.

AGGAGG is a strong Shine-Dalgarno sequence, in contrast to AGAAGA, and can functionally interact with translating ribosomes in a frame-independent manner (see below). Therefore, its distribution was examined in other translational frames in E. coli K-12 genes to determine whether its representation is biased. The sequence NAG_GAG_GNN occurs 168 times in E. coli genes. In 1,000 random genomes generated by the DicodonShuffle program, the sequence NAG_GAG_GNN occurs 396 times per genome on average, and the lowest count is 342 times in one genome (Fig. 1C). The sequence NNA_GGA_GGN occurs 47 times in coding regions of E. coli K-12. In 1,000 random genomes, the same sequence occurs 88 times per genome on average, and the lowest count is 61 times in a genome (Fig. 1D). Thus, the AGGAGG sequence is significantly underrepresented in the other two frames.

Comparative analysis of AGG_AGG- and AGA_AGA-containing genes.

In genes that exploit programmed ribosomal frameshifting for expression purposes, the shift site is often conserved (the +1 frameshifting utilized in decoding many eubacterial RF2 genes is an example [8]). Where the ORF in the new frame after the frameshift site is long, the sequence of the product from that frame is also conserved (for example, in RF2 and antizyme [8, 30, 41]). (Exceptions are candidates for hypothetical nonproduct roles for frameshifting [3]). Such conservation makes it easier to list viable candidates for this type of programmed frameshifting than when the ribosome quickly terminates and the purpose is to generate a product lacking significant domains present in the full-length product of standard decoding (e.g., the gamma subunit of DNA polymerase III encoded by E. coli dnaX [11, 18, 61]). Of course, as also illustrated by dnaX decoding, frameshifting may be important for gene expression but only somewhat conserved. In some bacteria, the same product mix and ratio are achieved by a different mechanism, e.g., transcription slippage (37).

All E. coli K-12 genes that contain tandem AGG or AGA codons were compared with homologous genes in other bacteria for conservation of the frameshift site and the amino acid sequence encoded in the +1 frame after the shift site. However, in none of the genes containing AGG_AGG or AGA_AGA can the utilization of frameshifting for gene expression purposes be predicted with high certainty.

Efficiency of frameshifting on AGG_AGG and AGA_AGA in their native contexts.

As described in the introduction, the efficiency of frameshifting on AGG_AGG and AGA_AGA tandem codons at different expression levels merits study. We first attempted to achieve the lowest practical expression levels. For that, the lacZ reporters with the AGA_AGA-containing sequence of the multidrug efflux gene emrK were integrated into the E. coli chromosome. The frameshifting efficiency was assayed by β-galactosidase assays. In this setting, frameshifting is detectable at about 0.9 to 1% (Table 3). Thus, the translation of an AGA_AGA-containing mRNA synthesized from a single-copy gene from the P_araB promoter still results in a detectable level of frameshifting, albeit about 50 times lower than that reported previously (55). Although it is possible to titrate the level of inducer (l-arabinose) and achieve lower levels of mRNA expression, the data would be difficult to reliably quantitate since, in the performed experiment, β-galactosidase activity is rather low even for the in-frame control (165 and 287 Miller units with 0.02% and 0.2% arabinose induction, respectively).

TABLE 3.

Frameshifting efficiencies on AGA_AGA in an emrK context under different expression conditions

Plasmid/strain	Reporter copy no.	Promoter	Inducer	β-Galactosidase activity (Miller units)a		Average frameshifting level (%) ± SD
				Frameshift reporter	In-frame control
pGHM-emrK/DH5α	15-20	Ptac	1 mM IPTG	NA	NA	7.4 ± 3.4
pSKAGS-emrK/CP79	15-20	Ptac	None	41 ± 2	1,847 ± 105	2.2 ± 0.2
			1 mM IPTG	376 ± 33	4,698 ± 545	8.0 ± 1.2
pSKAGS-emrK/CP78	15-20	Ptac	None	10 ± 1	690 ± 67	1.5 ± 0.1
			1 mM IPTG	63 ± 6	4,559 ± 348	1.4 ± 0.2
pSKAGS-emrK/BW27786	15-20	Ptac	None	20 ± 2	1,449 ± 160	1.4 ± 0.2
			1 mM IPTG	345 ± 55	8,850 ± 465	3.9 ± 0.7
JFG001 (derived from BW27786)	1	ParaB	0.02% arabinose	1.6 ± 0.1	165 ± 15	1.0 ± 0.1
			0.2% arabinose	2.6 ± 0.3	287 ± 13	0.9 ± 0.1

^aNA, not applicable.

Next, the levels of frameshifting were assayed when the same sequences were expressed from pSKAGS-based plasmid-borne constructs (with pBR322 ori). The expression of a GST-emrK-lacZ fusion in this vector is driven by a P_tac promoter and, since the promoter is leaky, can be measured with and without IPTG induction (approximately a fivefold difference in the expression levels judging from expression levels from the in-frame controls). The β-galactosidase assays were then performed with the same strain as above. In the absence of IPTG induction, the frameshifting level is 1.4%, while in the presence of inducer, the frameshift level is 3.9%. The respective values for the in-frame controls are 1,449 and 8,850 Miller units (Table 3).

With the same pSKAGS-emrK constructs, frameshifting efficiency was also measured in relA⁺ and relA mutant isogenic strains (CP78 and CP79), since starvation-promoted frameshifting is influenced by relA gene inactivation (19, 39). In the absence of IPTG induction, frameshifting is 1.5% and 2.2% in these relA⁺ and relA mutant strains, respectively (Table 3). In the presence of the IPTG inducer, frameshifting levels on the AGA_AGA-containing sequence in the relA⁺ strain did not show any significant change. On induction in the relA mutant strain, however, frameshifting increased to 8%. A similar value of frameshifting, 7.4%, was obtained for this sequence when frameshifting was measured by pulse-chase assay in the DH5α strain (also a relA mutant) in the presence of the inducer.

The pulse-chase method is simpler and more reliable, since both products, of termination and frameshifting, can be visualized. Therefore, we employed pulse-chase analysis to examine frameshifting in cassettes with the sequence context from several genes in which the tandem rare codons naturally occur. Frameshifting was measured from cassettes with the sequence contexts from three AGG_AGG-containing (yhaC, smf, and fecC) and three other AGA_AGA-containing (lhr, recF, and t150) genes. Although tentatively proposed only in one case of prokaryotic +1 frameshifting (69), 3′ mRNA structures might stimulate +1 frameshifting (and are known for eukaryotic antizyme +1 frameshifting [30, 41]). Therefore, all sequences were examined for the presence of potential stem-loop structures downstream of the shift site. When no strong candidate was identified, sequences from the genes encompassing at least 30 nucleotides upstream and downstream of the shift site (or up to, but excluding, the stop codon in the +1 frame) were included in the constructs. Pulse-chase experiments were performed at 37°C. Frameshifting was detected in all genes examined. The efficiency of frameshifting ranges from 3 to 11% (Fig. 2), which is much lower than that reported before, though the prior analysis was performed at 42°C (55). Since frameshifting efficiencies can be up to fivefold higher at 42°C than at 37°C on some sequences (O. Gurvich, unpublished data), the pulse-chase analysis was redone at 42°C. With the AGG_AGG and AGA_AGA sequences in their native flanking sequence contexts, the frameshifting values obtained at 42°C are up to two times higher than those at 37°C under the conditions tested (Fig. 2). However, they are still much lower than the 50% level reported previously. We were unable to discern from the original Spanjaard and van Duin paper (55) the flanking sequence, mainly from reporter genes, used in that work and so were unable to include it for comparative purposes in the present experiments. There may well be avoidance in the natural gene context of features that would stimulate high levels of erroneous frameshifting at tandem AGG and tandem AGA codons, which are fortuitously present in other sequences.

FIG. 2.

Frameshifting efficiencies on AGG_AGG and AGA_AGA tandem codons in E. coli K-12 genes at 37°C (black bars) and at 42°C (grey bars).

DISCUSSION

As foreshadowed in the study of Spanjaard and van Duin (55), and investigated here, the level of frameshifting at the tandem rare codon AGA_AGA (and presumably at AGG_AGG) is influenced by the codon's expression level. When expression was tested in the same strain (BW27786), there was a fourfold decrease in frameshifting when AGA_AGA was integrated in the chromosome and expressed from a P_araB promoter than when it was borne on a pBR322-based plasmid and expressed from an induced tac promoter. The relative expression, as quantified by comparing β-galactosidase activities from the in-frame controls, is 53-fold lower from the integrated reporter induced with 0.02% arabinose than from the plasmid-borne cassette induced with 1 mM IPTG. The higher expression of the AGA_AGA-containing mRNA likely results in the sequestration of the sparse cognate tRNA into separate translating complexes. Thus, more ribosomes translating AGA_AGA are paused at the first AGA codon, and more of them can escape stalling by shifting into the +1 frame. These results are in good agreement with a previous report (68) that overexpression of mRNA enriched with AGA and AGG codons results in a translation inhibition which is reversible by increasing the levels of the rare tRNA^Arg.

As shown above, the frameshifting is exacerbated upon induction in the relA mutant, rather than in the otherwise identical relA⁺, strain. Masucci et al. (39) proposed that starvation-promoted frameshifting is higher in relA mutant cells than in relA⁺ cells due to accumulation of undermodified tRNA during amino acid starvation. Several tRNA modifications prevent +1 frameshifting (62), and the accumulation of the hypomodified tRNAs in relA mutant cells is the likely cause of the higher frameshifting under relaxed conditions. The above-described experiments were not performed under amino acid starvation per se. However, the overexpression of mRNA containing AGA_AGA increases the demand for the cognate tRNA dramatically and is therefore likely to trigger a response similar to that of starvation. The relA⁺ strain is better able to adjust the levels of cognate tRNA and so to keep frameshifting levels lower than those of the relA mutant cells.

All gene sequences tested supported +1 frameshifting with efficiencies of 3 to 11%. Nevertheless, this plasmid-borne frameshifting is still at least 4.5-fold less than the 50% value reported by Spanjaard and van Duin (55). Their study also used a pBR322-based plasmid. However, there are several discrepancies which together likely account for the different values obtained. Spanjaard and van Duin used a P_L promoter (whose induction required 42°C thermal inactivation of a thermolabile repressor; the elevated temperature itself may have modestly elevated the frameshifting level compared to 37°C). They also had an unknown sequence context which may have contained additional rare arginine codons which would affect the cognate tRNA pool levels. Moreover, although Spanjaard and van Duin carried out their experiments with two completely different strains, both strains carry mutations conferring streptomycin resistance. Certain streptomycin-resistant ribosomes translate more accurately and more slowly (50). Such hyperaccurate ribosomes are expected to spend more time with their A sites vacant. As shown by Sipley and Goldman (53), such ribosomes are more prone to shift into the +1 frame (though this does not imply that the leakiness of all frameshift mutants will be elevated in the streptomycin-resistant strains).

None of the E. coli K-12 chromosomal genes that contain AGG_AGG or AGA_AGA belong to the subset of highly expressed genes. Low expression could be due to any of several causes, e.g., low promoter strength or low translation rate due to suboptimal codon usage. The latter situation would be similar to that of the AGA_AGA-containing trmD gene, which is transcribed at a high level but is translated poorly and expressed only at about 80 protein molecules per cell (13). Some of the AGG_AGG- or AGA_AGA-containing genes are most likely not expressed under standard conditions but become activated depending on growth medium and physiological state. This is certainly true for emrK, the drug efflux gene that we used for comparison studies, the transcription of which is dependent on the drug presence and growth phase (16, 40, 60). Thus, at a given time, most likely only a subset of these genes is expressed, and the total amount of mRNAs containing AGA_AGA (or AGG_AGG) is likely to be somewhat lower than that from the P_araB promoter used in the current study (at 0.02% arabinose induction). Consequently, the frameshifting efficiencies exhibited during expression of these genes in E. coli are presumably lower. However, the natural frameshifting levels are also likely to vary depending on growth condition and physiological state. When E. coli enters the stationary phase, frameshifting levels increase (5, 65). In contrast, Saccharomyces cerevisiae Ty1-programmed frameshifting decreases on entering the stationary phase, probably due to a disproportionate decrease in overall demand for the sparse tRNA that mediates the shift (57).

The most likely efficiencies of frameshifting at AGA_AGA and AGG_AGG in the mRNAs in which they naturally occur are at least 1 to 2 orders of magnitude higher than the level estimated for general coding sequences (35). Even so, any results of selection against occurrences of these sequences in genes that are not highly expressed are not evident. AGG_AGG is predicted to occur five times, but in reality, it occurs four times in coding regions of E. coli K-12. In three of these occurrences, in fecC, smf, and ninE, the AGG_AGG sequence likely serves as a Shine-Dalgarno sequence for the initiation of translation of the downstream gene. The benefit of this sequence for translation initiation likely outweighs any negative consequences of its frameshifting propensity, probably because the products of frameshifting are not toxic and/or rapidly degraded. In the fourth gene, yhaC, there is no obvious Shine-Dalgarno role for the AGG_AGG sequence. It could be that this sequence is involved in the formation of some RNA secondary structure and therefore is selected for. Indeed, when the nucleotide sequence of this part of the gene was analyzed by mfold (70), a potential strong stem-loop structure with an AGGAGG sequence involved in base pairing was predicted. However, it is hard to assess by covariation analysis whether the stem-loop is functional, since this gene exists only in E. coli and Shigella flexneri and the nucleotide sequences of these genes are nearly identical.

Pairing of AGGAGG with rRNA likely influences its occurrence within coding sequences.

AGGAGG is one of the strongest Shine-Dalgarno sequences and, by virtue of its binding to 16S rRNA in 30S ribosomal subunits, it directs translation initiation. However, when it occurs in coding sequences, it is also sensed by translating 70S ribosomes, due to the scanning of mRNA by the anti-Shine-Dalgarno sequence in these ribosomes (63, 64). This can result in mRNA-rRNA pairing, which can potentially stall ribosomes. The pushing or pulling effect of the internal Shine-Dalgarno sequence in these instances, with the effect and directionality depending on the spacing length from the shift site (4, 36), might be expected to result in selective pressure against occurrences of AGGAGG when frameshifting is not functionally favored, though evidence for this has not been forthcoming.

As the present work shows, AGGAGG is indeed significantly underrepresented in the two non-zero frames (Fig. 1C and D). The corresponding data from zero frame occurrences of AGG_AGG is insignificant with respect to this point, since the numbers are so small.

Twenty-five percent of the tandem AGA codons occur early in genes and likely affect ribosome loading.

The presence of rare codons close to the translational start site can have a strong negative effect on gene expression. Ribosomal stalling at early rare codons can lead to mRNA release from ribosomes (22) and termination of translation. Interestingly, this phenomenon is utilized for regulation of expression by a substantial number of genes in E. coli. Prior work showed that the rare codon AGA or AGG preferentially occurs early in a group of essential genes with a variety of functions (14, 15). Tandem rare codons near the 5′ end of a coding sequence confer an even stronger inhibitory effect on translation (15, 24, 49). It is only a small extension to now report that 11 of the 42 (26%) AGA_AGA tandem codons occur within the first 25 codons (Fig. 3 shows the distribution of the AGA_AGA positions within the genes). Intriguingly, half of these genes have a membrane-associated function. In one of the genes, yhiU, which encodes a putative membrane protein, AGA occurs as the triplet AGA_AGA_AGA.

FIG. 3.

Positions of AGA_AGA tandem codons normalized to the gene length.

Other features relevant to AGA_AGA occurrences.

There are several features which may be relevant to occurrences of AGA_AGA. Suboptimal aminoacylation levels of sparse tRNAs (which decode rare codons) are a sensitive indicator of amino acid starvation. Starved cells need to preferentially synthesize essential proteins, and it is likely advantageous to abort the synthesis of less important proteins. Transfer mRNA (tmRNA)-mediated tagging that occurs at ribosomes stalled at rare codons may permit a redirection of ribosomes to more important tasks. tmRNA-mediated tagging occurs at tandem AGA codons, including in the yjgR gene as shown in Table 2, and may involve mRNA cleavage (27, 28, 48, 59). Consistent with the redirection possibility, none of the genes with AGA_AGA are highly expressed (32), and none are known to be important under starvation conditions (Table 2). The occurrence of AGA_AGA early in genes (discussed above) is also preferable since cells will not waste energy synthesizing long peptides destined for degradation. A certain level of +1 frameshifting at AGA_AGA codons, whose level is dependent on sequence context, will partially circumvent tmRNA-mediated aborted synthesis. In most locations, such circumvention may be disadvantageous (exceptions may occur when AGA_AGA is positioned near the end of the gene or the 3′ ends of sequences encoding domains that are functional on their own).

Occurrences of AGA_AGA may also partially reflect the evolutionary history, by lateral transfer or direct descent, from an ancestor in which AGA was not shift prone due to the presence of multiple genes encoding its cognate tRNA. AGA codons occur at high frequency in toxin-encoding genes of E. coli O157:H7 (31), and the cognate tRNA^Arg (anticodon UCU) occurs in eight copies in this bacterium (as annotated in the genome NC_002695). Therefore, AGA_AGA is not shift prone in O157:H7 and is not avoided. There are 62 genes containing AGA_AGA in O157:H7, and 28 of them have AGA_AGA-containing homologues in K-12. It could be that some of these 28 genes came to K-12 by horizontal gene transfer from O157:H7. Since none of these genes belong to the subset of highly expressed genes, evolutionary pressure is probably not strong enough to eliminate these shift-prone sequences from the K-12 genome. Furthermore, nearly 30% of the AGA_AGA occurrences in the E. coli K-12 genome are in prophages and mobile elements (Table 2).

Perspective.

Any generalizations from the lack of evidence for utilization of frameshifting at tandem rare arginine codons need to be treated cautiously. Though no examples have yet been found, future work may reveal utilization of frameshifting at codons that become shift prone due to the limitation of the corresponding aminoacyl-tRNA following amino acid starvation. The penetrating work of Gallant and colleagues (19) clearly points out the possibilities. In conclusion, the present and a previous work (25) show that an error frameshift event(s), at levels considerably over the background rate, occurs during the translation of at least 2.5% of E. coli genes (see Materials and Methods). A number of other shift-prone sequences are known to occur in E. coli, and more are likely to be identified. It may well be that future studies will show that there is no strong selection against shift-prone sequences in general in ORFs that are not highly expressed.

ADDENDUM IN PROOF

A recent publication (R. Leipuviene and G. R. Björk, RNA 11:796-807, 2005) demonstrates that the tRNA cognate for at least one particular codon 5′ of a single AGA shifts to the +1 frame when there is either diminished quantity or functionality of the AGA-decoding tRNA. The degree of reduction of in-frame decoding of a single AGA needed before defects of cell division or other attributes become apparent (Leipuviene and Björk; K. Sakamoto, S. Ishimaru, T. Kobayashi, J. R. Walker, and S. Yokoyama, J. Bacteriol. 186:5899-5905, 2004) is pertinent to the decoding of tandem AGA_AGA (or AGG_AGG), as discussed elsewhere (Leipuviene and Björk).