Critical Features of a Conserved RNA Stem-loop Important for Feedback Regulation of RNase E Synthesis*

RNase E is an important regulatory enzyme that governs the principal pathway for mRNA degradation inEscherichia coli. This endonuclease controls its own synthesis via a feedback mechanism in which the longevity ofrne (RNase E) mRNA is modulated by a cis-acting sensory element that responds to changes in cellular RNase E activity. Previous research has shown that this element is an RNA stem-loop (hp2) within the 5′-untranslated region of the rne transcript. Here we report studies involving mutational analysis and phylogenetic comparison that have identified the features of rne hp2 important for its function. These comprise an internal loop flanked on one side by a 2-bp stem and a hairpin loop and on the other side by a longer stem whose sequence is inconsequential. A search of bacterial genome sequences suggests that regulation by an hp2-like element may be a unique evolutionary adaptation of the rne transcript that is not shared by other mRNAs.

Messenger RNA degradation is an important mechanism for controlling gene expression in all organisms. The rate of mRNA decay directly affects the steady-state concentration of mRNA, thereby influencing rates of protein synthesis. mRNA lifetimes can differ markedly within a single cell. In Escherichia coli, for example, mRNA half-lives can be as short as a fraction of a minute or as long as an hour, with a typical half-life being 2-4 min (1). The stability of a given message need not be fixed and may vary in response to growth conditions (2)(3)(4)(5).
In E. coli, mRNA decay generally involves the sequential action of endonucleases and 3Ј-exonucleases (1). For most E. coli mRNAs, degradation begins with internal cleavage by RNase E (6 -11). This endonuclease, 1061 amino acids in length, possesses broad cleavage-site specificity, cutting RNA in a variety of single-stranded regions that are AU-rich (12,13). The rate at which RNase E cleaves mRNA is strongly influenced by RNA features that are distinct from its sites of cleavage. These features include structural elements within the 5Ј-untranslated region (UTR), 1 5Ј-terminal phosphorylation, and bound ribosomes (14 -16).
In E. coli, RNase E is an essential protein whose underproduction or overproduction can impair cell growth (17,18). Syn-thesis of this important endonuclease is tightly controlled by an autoregulatory mechanism that modulates the longevity of the transcript of the RNase E gene (rne) in response to changes in cellular RNase E activity (19). Feedback regulation of RNase E synthesis is mediated in cis by the 361-nucleotide (nt) rne 5Ј-UTR, which can confer a comparably high degree of sensitivity to cellular RNase E levels onto heterologous mRNAs to which it is fused (19,20). Thus, expression of a reporter gene comprising the rne 5Ј-UTR and the initial portion of the rne coding region joined in-frame to lacZ can differ by more than 30-fold in cells containing low versus high RNase E activity, and this sensitivity to RNase E is virtually abolished when all of the 5Ј-UTR upstream of the ribosome binding site is deleted (19,20). The rne 5Ј-UTR contains six structural domains ( Fig.  1), two of which, the stem-loop structures hp2 and hp3, are important for feedback regulation (20). Of these two stemloops, hp2 is the more potent sensor of cellular RNase E activity. When fused directly upstream of the rne ribosome binding site, hp2 alone is sufficient to direct efficient feedback regulation by RNase E (20).
To better understand the mechanism of RNase E autoregulation, we have used mutational analysis and phylogenetic comparison to examine the features of rne hp2 that are important for its ability to mediate feedback control. These studies indicate that an 8-nt internal loop near the top of hp2 plays a central role in this regulatory process.
Plasmid pEZ⌬114 -337 is a pSC101 derivative bearing an rne-lacZ fusion that comprises the rne promoter, all of the 361-nt rne 5Ј-UTR except nucleotides 114 -337, and the first 181 rne codons fused in-frame to the tenth codon of lacZ (20). To construct plasmid pEZ8, a unique ApaI restriction site was created in pEZ⌬114 -337 by inserting two guanidylate residues between nucleotides 53 and 54 and two cytidylate residues between nucleotides 106 and 107 of the rne 5Ј-UTR, resulting in two additional G-C base pairs at the base of rne hp2. Plasmid pEZ8hp2Hinf was constructed by inserting a DNA fragment encoding hp2 of Haemophilus influenzae rne mRNA (GGGCCCTGTTGGTGAA-AACTCAATGCAGCAATTGGCATAAGACATTGATAATCAACATTAG-CTGTACA) between the ApaI and BsrGI restriction sites of pEZ8. In plasmid pEZ8⌬hp2, the segment between the ApaI and BsrGI sites of pEZ8, including hp2 and ss2, was replaced with a short linker (GGG-CCCGCGGCCCAGTTAGCTGTACA), whereas in pEZ8NcoI, the same pEZ8 segment was replaced with a different linker (GGGCCCGTGCC-ATGGTGCTGTACA). Plasmids pEZ8hp2mut4, pEZ8hp2mut2T, pEZ8hp2mut3T, and pEZ8hp2mut2B are derivatives of pEZ8 that each encode mutations in the hairpin loop at the top of rne hp2 (AAUG 3 GGAA, AGAG, AUUCG, and GAUA, respectively). Plasmids pEZ8hp2IL-5U, pEZ8hp2IL-5G, pEZ8hp2IL-2C, and pEZ8hp2IL-2C,3A are derivatives of pEZ8 that each encode mutations in the 8-nt internal loop near the top of rne hp2 (CA-GUAAGA 3 CA-GUUAGA, CA-GUGAGA, CC-GUAAGA, and CC-AUAAGA, respectively). Plasmid pEZ8hp2ϩ2bp is a pEZ8 derivative encoding a mutant reporter mRNA in which two base pairs have been inserted into the 2-bp stem near the top of rne hp2 (GCAATGGC 3 GCGCAATGGCGC). Plasmid pEZ8hp2bot-syn was constructed from pEZ8 by replacing the segment between the ApaI and BsrGI sites with the sequence GGGCCCGCA-TGCAGCAATGGCGTAAGACATGCGGGCCCAGTTAGCTGTA C A . Plasmid pEZ8hp2CG is a pEZ8 derivative that encodes a G-C 3 C-G base pair substitution just beneath the 8-nt internal loop of hp2.
Construction and Screening of rne hp2 Internal Loop Libraries-To identify the important nucleotides in the 8-nt internal loop of rne hp2, two combinatorial libraries were created in which the sequence of the top four (hp2IL-top) or bottom four (hp2IL-bot) nucleotides of this internal loop were randomized. Oligonucleotide primer Hp2comp (5Ј-CG-TGGGGGTGTACA-3Ј, 32 pmol) was annealed to an equimolar amount of either of two complementary primers that encoded the entire hp2-ss2 region: Hp2IL-toplib (5Ј-CATTTTGGGCCCGACCGATCATCCACGC-NGCAATGGCNNNAGACGTATTGATCTTTCAGGCCCAGTTAGCTG-TACACCCCCACG-3Ј) or Hp2IL-botlib (5Ј-CATTTTGGGCCCGACCGA-TCATCCACGNAGCAATGGCGTANNNCGTATTGATCTTTCAGGCC-CAGTTAGCTGTACACCCCCACG-3Ј). The annealed oligonucleotides were extended for 1 h at 37°C with T4 DNA polymerase (10 units) and all four dNTPs (250 M each) in T4 DNA polymerase buffer (New England BioLabs). The extended products (86 bp) were purified using a High Pure purification kit (Roche Molecular Biochemicals), digested with ApaI and BsrGI, repurified, and ligated between the ApaI and BsrGI sites of pEZ8NcoI. Each library was transformed into competent E. coli WM1/FЈ cells containing plasmid pRNE101, and the cells were plated on LB-agar supplemented with X-gal (60 g/ml), ampicillin (100 g/ml), and kanamycin (100 g/ml). The high level of RNase E activity present in these cells allowed reporter mRNAs sensitive to RNase E overproduction to be identified by their pale color when compared with otherwise identical cells containing pEZ8 (white colonies) or pEZ8⌬hp2 (blue colonies). Reporter plasmids sensitive to RNase E overproduction were purified and transformed into E. coli strains CH1828 and CH1827 ϩ pRNE101 for quantitative measurements of ␤-galactosidase activity.
Measurement of Repression Ratios-␤-Galactosidase assays were performed as previously described (21). For each rne-lacZ reporter, a repression ratio was calculated by dividing the ␤-galactosidase activity in CH1828 host cells (low RNase E activity) by the ␤-galactosidase activity in CH1827 ϩ pRNE101 host cells (high RNase E activity). The reported repression ratios are each the average of three or more measurements. Error estimates correspond to the standard deviation of these measurements.
Computer Search for rne hp2-like Elements-The computer program RNAbob (available at www.genetics.wustl.edu/eddy/software/#rnabob) was used to search for elements resembling rne hp2 in the genomes of E. coli, Yersinia pestis, and Haemophilus influenzae (genomic sequence files downloaded from ncbi.nlm.nih.gov/genomes/Bacteria). The search parameters specified an 8-nt internal loop consisting of the sequence CD-RUHAGA (where D ϭ A, G, or U; R ϭ A or G; H ϭ A, C, or U), flanked below by a Ն3-bp stem and above by a 2-bp stem and a 4-to 6-nt hairpin loop whose sequence was not constrained. Strict search criteria allowed only canonical base pairs in the 3-and 2-bp stems, as observed in nature. Only elements present on the coding strand of each gene were considered.

RESULTS
Evolutionary Conservation of rne hp2 Function-Autoregulation of RNase E synthesis is mediated by rne hp2, an RNA stem-loop that functions post-transcriptionally as a sensor of cellular RNase E activity (20). Elucidating the essential features of this 57-nt 5Ј-UTR element would be useful not only for understanding the mechanism of RNase E feedback regulation but also for efforts to identify other genes whose expression might be similarly regulated by this important ribonuclease.
Despite extensive sequence divergence, the 5Ј-UTR of the Haemophilus influenzae and E. coli rne transcripts share significant secondary-structure homology (20). We first wished to address whether hp2 of Haemophilus rne mRNA could substitute functionally for the corresponding E. coli stem-loop in mediating RNase E feedback regulation in E. coli. These studies were performed in the context of an rne-lacZ reporter (ez⌬114 -337; Fig. 2) in which all of the E. coli rne 5Ј-UTR, except hp3, and the first 181 codons of the rne coding region were fused in-frame to lacZ. Expression of ez⌬114 -337 in E. coli is very sensitive to the cellular level of RNase E activity, which modulates the longevity of the reporter transcript (19,20). Previous experiments have shown that, in the absence of hp3, feedback regulation of rne-lacZ expression is almost entirely dependent on the presence of hp2, thereby allowing the effects of hp2 mutations to be fully manifested (20).
To facilitate the construction of hp2 substitution mutants, we first introduced two additional G-C base pairs at the base of hp2 to create a unique ApaI restriction site (GGGCCC) in the reporter gene (Fig. 2). These additional base pairs did not significantly affect the sensitivity of the reporter mRNA to feedback regulation by RNase E, as determined by comparing ␤-galactosidase production from this new reporter (ez8) in isogenic lacZ Ϫ E. coli host strains with low (CH1828) or high (CH1827 ϩ pRNE101) levels of cellular RNase E activity. The observed repression ratio (R, the ratio of ␤-galactosidase activity in the CH1828 host versus the CH1827 ϩ pRNE101 host), which is a direct measure of the degree to which expression of the rne-lacZ fusion can be inhibited by RNase E, was approximately the same for ez8 (R ϭ 10.4 Ϯ 2.0) and the original ez⌬114 -337 reporter (R ϭ 11.1 Ϯ 2.1). As expected, when hp2 was instead replaced with an unrelated stem-loop (UUGGGC-CCGCGGCCCAG), sensitivity to cellular RNase E activity was nearly abolished (R ϭ 2.4 Ϯ 0.2 for ez8⌬hp2).
We then replaced hp2 of ez8 with the corresponding stemloop of the H. influenzae rne transcript (ez8hp2Hinf; Fig. 2). Despite numerous differences in sequence and secondary structure between the Haemophilus and E. coli stem-loops, this hp2 replacement had almost no effect on feedback regulation (R ϭ 9.8 Ϯ 2.3). This finding suggests that the features of hp2 that are critical for feedback regulation in E. coli are conserved in the Haemophilus stem-loop.
To determine which of the conserved features of rne hp2 are most important for feedback regulation, this stem-loop was divided conceptually into three subdomains. One was an 8-nt internal loop near the top of hp2, a region of the stem-loop whose sequence is especially well conserved between Hae- mophilus and E. coli (Fig. 2). The other two subdomains were the hairpin loop and 2-bp stem above this internal loop and the long imperfect stem below it. Each of these subdomains was examined by mutational analysis.
Sequence Requirements Within the Internal Loop of rne hp2-To identify which nucleotides in the 8-nt internal loop of E. coli rne hp2 (CA-GUAAGA) are most important for feedback regulation, we created two combinatorial libraries in which the sequence of each half of the internal loop was randomized in the context of the ez8 reporter (hp2IL-top: CN 2 -N 3 N 4 N 5 AGA and hp2IL-bot: N 1 A-GUAN 6 N 7 N 8 ). We chose to divide the internal loop into halves, as the maximum complexity of each library would equal 256 (4 4 ), a relatively manageable number of sequence combinations. In contrast, simultaneously randomizing all eight positions in a single library would greatly increase the potential complexity of the library to 65,536 (4 8 ), making it much less likely that all sequence combinations would be represented.
The libraries were each transformed into E. coli cells containing a high concentration of RNase E due to the presence of a multicopy rne plasmid (pRNE101). The high level of RNase E activity in these cells allowed reporter mRNAs susceptible to feedback repression to be identified on the basis of the pale color of bacterial colonies grown on LB-agar plates containing the chromogenic ␤-galactosidase substrate X-gal. The color of these transformants was compared with that of colonies with an rne-lacZ reporter either containing (ez8) or lacking (ez8⌬hp2) a functional hp2 sequence. Reporter plasmids were purified from colonies that were as pale as those containing the parent plasmid (pEZ8) and retransformed into isogenic E. coli strains containing low or high RNase E activity. For each plasmid, a repression ratio was calculated by comparing levels of ␤-galactosidase activity in the two host strains.
A total of 1202 colonies were screened from the hp2IL-bot library (CA-GUAAGA 3 N 1 A-GUAN 6 N 7 N 8 ). Nearly all of these colonies were as blue as colonies expressing a reporter deficient for feedback regulation (ez8⌬hp2). This finding seemed to imply that the sequence requirements for the bottom half of the internal loop were strict. Indeed, only two colonies in this combinatorial library contained reporters with repression ratios similar to that of a reporter (ez8) that has a wild-type internal loop sequence (R ϭ 12.7 Ϯ 1.9 for hp2IL-bot#7 and 15.2 Ϯ 3.4 for hp2IL-bot#8 versus 10.4 Ϯ 2.0 for ez8; Fig. 3). DNA sequencing revealed that the internal loops of these two FIG. 2. Feedback regulation of rne-lacZ chimeras. A, sequence and secondary structure of the 5Ј-UTR of ez8 mRNA. This 5Ј-UTR differs from that of E. coli rne mRNA in that hp3 and the first nine nucleotides of ss3 have been deleted, a single nucleotide (C) has been inserted near the 5Ј boundary of the remaining ss3 segment to create a BsrGI site, and two G-C base pairs have been added near the bottom of hp2 to create an ApaI site. The Shine-Dalgarno element and translation initiation codon are underlined. B, effect of hp2 substitutions on feedback regulation. Plasmids encoding ez⌬114 -337, ez8, ez8⌬hp2, or ez8hp2Hinf mRNA were introduced into an isogenic pair of lacZ Ϫ E. coli strains: CH1827 (rne ϩ ) containing the multicopy rne ϩ plasmid pRNE101 (high RNase E activity) and CH1828 (ams-1; low RNase E activity). Cellular ␤-galactosidase activity was measured in extracts prepared from log-phase bacterial cultures, and repression ratios were calculated by dividing the ␤-galactosidase activity in CH1828 cells by the ␤-galactosidase activity in CH1827 cells containing pRNE101. The reported values are each the average of three or more measurements. Errors correspond to the standard deviation of these measurements. Each of the four hp2 variants compared in these experiments is shown, with gray rectangles enclosing the regions that differ from wild-type rne hp2 of E. coli.
reporters were identical in sequence to that of wild-type E. coli hp2. These results indicate that only the wild-type sequence of nucleotides is tolerated at positions N 1 (C), N 6 (A), N 7 (G), and N 8 (A) of the internal loop.
In contrast to the hp2IL-bot library, the hp2IL-top library (CA-GUAAGA 3 CN 2 -N 3 N 4 N 5 AGA) gave rise to colonies with a wide range of color intensities on X-gal plates. Among a total of 3580 colonies, several contained reporter genes that were functional for feedback regulation (defined as having a repression ratio greater than 6.0), with repression ratios ranging from 7 to 15. This variation in the repression ratios suggested more relaxed sequence requirements in the upper half of the internal loop. Sequencing of eleven library members that were functional for feedback regulation confirmed this prediction (Fig. 4). Although the nucleotide at position N 4 (U) was invariant, sequence variation was observed at positions N 2 (A, G, or U), N 3 (A or G), and N 5 (A, C, or U). Although the wild-type E. coli sequence was not isolated in this screen, the repeated isolation of four variants (hp2IL-2G, hp2IL-2U, hp2IL-3A, and hp2IL-5C) suggested that we had identified most internal loop sequences in the library that were competent for feedback regulation.
Additional information about the key features of the internal loop was gleaned from close inspection of the repression ratios for each functional sequence variant within the hp2IL-top library and from the creation of additional site-directed mutants. For example, the potential for Watson-Crick base pairing between positions N 2 and N 3 appears to partially impair the ability of hp2 to mediate feedback regulation. In the double mutant hp2IL-2U,3A (CA-GUAAGA 3 CU-AUAAGA), where Watson-Crick base pairing is possible between uracil at N 2 and adenine at N 3 , there is a 35% reduction in the repression ratio relative to the wild-type sequence (R ϭ 6.9 Ϯ 0.4 for hp2IL-2U,3A versus 10.4 Ϯ 2.0 for ez8; Fig. 4). In contrast, the same substitutions made individually at N 2 or N 3 lack the potential for Watson-Crick base pairing and have no detrimental effect on feedback regulation (R ϭ 10.6 Ϯ 1.3 for hp2IL-2U and 12.6 Ϯ 0.8 for hp2IL-3A; Fig. 4).
In none of the functional sequence variants was cytosine observed at N 2 . Although cytosine at this position would be capable of pairing with a wild-type guanine base at N 3 , it would not be able to form a Watson-Crick base pair with adenine at N 3 , a nucleotide substitution that is itself well-tolerated (hp2IL-3A in Fig. 4). To determine the effect of cytosine at N 2 in either of these sequence contexts, we constructed two additional variants: hp2IL-2C (CA-GUAAGA 3 CC-GUAAGA) and hp2IL-2C,3A (CA-GUAAGA 3 CC-AUAAGA). In the context of guanine at N 3 , the presence of cytosine at N 2 severely impaired feedback regulation, causing the reporter to be scarcely more sensitive to cellular RNase E activity than a reporter lacking hp2 altogether (R ϭ 3.6 Ϯ 0.2 for hp2IL-2C versus 2.4 Ϯ 0.2 for ez8⌬hp2; Fig. 5). Eliminating the potential for Watson-Crick base pairing between N 2 and N 3 by substituting both cytosine at N 2 and adenine at N 3 resulted in only a partial restoration of feedback regulation (R ϭ 4.7 Ϯ 0.4 for hp2IL-2C,3A versus 3.6 Ϯ 0.2 for hp2IL-2C and 12.6 Ϯ 0.8 for hp2IL-3A; Figs. 4 and 5). These findings indicate that cytosine is intrinsically not well tolerated at position N 2 , quite apart

FIG. 4. Analysis of a combinatorial library of rne-lacZ reporters bearing a randomized sequence in the upper half of the hp2 internal loop.
Members of the hp2IL-top library that appeared to be sensitive to feedback regulation on the basis of colony color were introduced into the isogenic E. coli strains CH1828 (low RNase E activity) and CH1827 ϩ pRNE101 (high RNase E activity). Cellular ␤-galactosidase activity and repression ratios were measured as in Fig. 2. Nucleotides that deviated from the sequence of wild-type E. coli rne hp2 are underlined. To the right is a diagram of the upper portion of hp2; the nucleotides in the internal loop that were randomized in the hp2IL-top library (N 2 , N 3 , N 4 , and N 5 ) are enclosed in boxes. Four hp2 variants (IL-2G, IL-2U, IL-3A, and IL-5C) were isolated from the library on two independent occasions. FIG. 3. Analysis of a combinatorial library of rne-lacZ reporters bearing a randomized sequence in the lower half of the hp2 internal loop. Members of the hp2IL-bot library that appeared to be sensitive to feedback regulation on the basis of colony color were introduced into the isogenic E. coli strains CH1828 (low RNase E activity) and CH1827 ϩ pRNE101 (high RNase E activity). Cellular ␤-galactosidase activity and repression ratios were measured as in Fig. 2. Among the 1202 library members that were screened, only two (#7 and #8) were found to be sensitive to feedback regulation, and both of these had an internal loop that was identical in sequence to that of wild-type E. coli rne hp2. To the right is a diagram of the upper portion of hp2; the nucleotides in the internal loop that were randomized in the hp2IL-bot library (N 1 , N 6 , N 7 , and N 8 ) are enclosed in boxes. from its base pairing potential.
Position N 5 shows considerable sequence flexibility, with adenine (wild-type), cytosine, or uracil present at this position in functional transcripts. The reporter transcript with cytosine at position N 5 was fully sensitive to feedback repression by RNase E (R ϭ 10.7 Ϯ 0.6 for hp2IL-5C; Fig. 4). The presence of uracil at this position only partially impaired feedback regulation when this substitution was accompanied by guanine at position N 2 , which by itself does not affect feedback regulation (R ϭ 6.8 Ϯ 0.7 for hp2IL-2G,5U versus 14.7 Ϯ 1.2 for hp2IL-2G; Fig. 4). The same uracil substitution had a more deleterious effect in the context of the naturally occurring adenine at position N 2 (R ϭ 4.1 Ϯ 0.2 for hp2IL-5U; Fig. 5). Introducing guanine at position N 5 severely impaired feedback regulation (R ϭ 1.9 Ϯ 0.2 for hp2IL-5G; Fig. 5), as expected from the absence of this substitution among the functional reporters in the hp2IL-top library (Fig. 4). Other nucleotide substitutions that did not appear among the functional hp2 variants in the hp2IL-top and hp2IL-bot libraries (IL-4G, IL-6U, IL-7U, IL-7A) were likewise found to be deleterious for feedback regulation (Fig. 5).
The Hairpin Loop at the Top of rne hp2-Other features of rne hp2 that may contribute to feedback regulation include the hairpin loop (AAUG) and 2-bp stem at the top. The importance of the hairpin loop was evaluated by mutational analysis (Fig.  6). Mutating all four nucleotides of this loop (hp2mut4: AAUG 3 GGAA) significantly impaired but did not abolish feedback regulation (R ϭ 4.7 Ϯ 0.4 for hp2mut4 versus 2.4 Ϯ 0.2 for ez8⌬hp2), as did replacing only the top two nucleotides of the loop with two dissimilar nucleotides (hp2mut2T: AAUG 3 AGAG; R ϭ 4.8 Ϯ 0.5). Nonetheless, there is no sequence feature of the hairpin loop that is absolutely required. Thus, substituting three dissimilar nucleotides for the top two nucleotides of the loop (hp2mut3T: AAUG 3 AUUCG) did not impair feedback regulation (R ϭ 9.9 Ϯ 1.0), and mutating the bottom two nucleotides of the loop (hp2mut2B: AAUG 3 GAUA) caused only a mild reduction in the repression ratio (R ϭ 8.4 Ϯ 0.8). These findings indicate that the hairpin loop at the top of rne hp2 makes a significant contribution to feedback regulation by RNase E but in a manner that cannot readily be predicted by merely examining the loop sequence. A significant regulatory defect was also observed when the 2-bp stem that separates the hairpin loop from the 8-nt internal loop was doubled in length (R ϭ 4.0 Ϯ 0.5 for hp2ϩ2bp), suggesting that the distance and orientation of the hairpin loop with respect to the internal loop are also important.
The Lower Stem of rne hp2-The internal loop and hairpin loop of E. coli rne hp2 sit atop a tall, imperfectly base-paired stem. To determine whether the sequence of this lower portion of hp2 contributes to feedback regulation, we replaced all of hp2 below the internal loop (a total of 41 nucleotides, including 14 Watson-Crick and wobble (G-U) base pairs and three small internal loops) with 13 base pairs differing in sequence from wild-type at all but the three bottom-most positions and the position immediately beneath the internal loop (hp2bot-syn; Fig. 7). The resulting reporter transcript remained fully sensitive to feedback regulation (R ϭ 12.1 Ϯ 0.3). The ez8 reporter also retained its sensitivity to feedback regulation when the G-C base pair directly below the internal loop was mutated to a C-G pair (hp2CG; R ϭ 9.1 Ϯ 1.0). These findings indicate that neither the sequence of the stem below the internal loop of rne hp2 nor the base-pairing imperfections within this lower stem are important for efficient feedback regulation in E. coli.
Phylogenetic Analysis of rne hp2-The mutational studies described above identify features of rne hp2 that are important for RNase E autoregulation in E. coli. Natural sequence variation among bacterial species provides an independent source of information as to the key features of this stem-loop. We Derivatives of ez8 bearing site-directed mutations within the hairpin loop and uppermost stem of rne hp2 were introduced into the isogenic E. coli strains CH1828 (low RNase E activity) and CH1827 ϩ pRNE101 (high RNase E activity). Cellular ␤-galactosidase activity and repression ratios were measured as in Fig. 2. Only the upper portion of hp2 is shown. Substituted or inserted nucleotides that deviated from the sequence of wild-type E. coli rne hp2 are enclosed in gray rectangles.

FIG. 5. Effect of site-directed mutations in the internal loop of rne hp2.
Derivatives of ez8 bearing site-directed mutations within the internal loop of rne hp2 were introduced into the isogenic E. coli strains CH1828 (low RNase E activity) and CH1827 ϩ pRNE101 (high RNase E activity). Cellular ␤-galactosidase activity and repression ratios were measured as in Fig. 2. Nucleotides that deviated from the sequence of wild-type E. coli rne hp2 are underlined. To the right is a diagram of the upper portion of wild-type hp2; the nucleotides in the internal loop that were targeted for sitedirected mutagenesis are enclosed in boxes.
compared the sequences of rne hp2 from nine species (Fig. 8). These phylogenetic data indicate that the sequence and secondary structure of the upper portion of hp2, comprising the internal loop, the hairpin loop, and the 2-bp stem between them, is conserved in evolution and therefore likely to have an important function.
Especially well conserved in sequence is the 8-nt internal loop, which has the consensus sequence CD-RUAAGA (where D ϭ A, G, or U and R ϭ A or G, with no potential for Watson-Crick base pairing between these two nucleotides). Except for the absence of phylogenetic variation at N 5 , this consensus sequence agrees perfectly with the results of mutational analysis. Also well conserved in length and sequence is the 2-bp stem above this internal loop, consistent with our finding that feedback regulation is impaired when this short stem is extended. The hairpin loop at the top of rne hp2 shows somewhat greater evolutionary variation, ranging in size from four to five nucleotides with the consensus sequence AN(N)NG (where N is a nucleotide of unspecified identity). The lack of sequence variation at the first and last positions of the hairpin loop is unexpected in view of our finding that these two nucleotides can be mutated without adversely affecting feedback regulation in E. coli.
Except for the highly conserved G-C base pair immediately beneath the internal loop, the lower portion of hp2 shows the greatest variability in sequence and secondary structure, ranging in size from 32 nt (including 12 base pairs) to 41 nt (including 17 base pairs) and bearing diverse internal loops at disparate locations. This high degree of variability is in agreement with our finding that the sequence and secondary structure of the lower stem of hp2 can be changed significantly without impairing feedback regulation. DISCUSSION An important step in elucidating post-transcriptional gene regulation is to identify the features of mRNAs that determine their lifetimes in vivo. Previous studies have shown that RNase E tightly autoregulates its synthesis in E. coli via a mechanism in which rne hp2, a sensory element within the 5Ј-UTR, modulates rne mRNA longevity in response to changes in cellular RNase E activity (19,20). We have now identified evolutionarily conserved features in the upper portion of rne hp2 that are crucial for its ability to mediate efficient feedback regulation of RNase E synthesis. These include an 8-nt internal loop and the 2-bp stem and hairpin loop above it.
The core recognition element within rne hp2 is an internal loop comprising two nucleotides on one strand and six nucleotides on the other. Point mutations within this internal loop can abolish hp2 function in E. coli. At five of the eight positions (N 1 ϭ C; N 4 ϭ U; N 6 ϭ A; N 7 ϭ G; N 8 ϭ A), no sequence variation is tolerated, whereas at three other positions (N 2 ϭ A, G, or U; N 3 ϭ A or G; N 5 ϭ A, C, or U) some sequence variation is permitted (Fig. 9A). Consistent with these findings is the limited sequence variation that occurs naturally for this rne internal loop in several related bacterial species, where the sequence is strictly conserved at six of the eight positions and observed to vary only at positions N 2 (A, G, or U) and N 3 (A or G) (Fig. 9B).
Above the internal loop of rne hp2, between nucleotides N 2 and N 3 , are a 2-bp stem and a 4 -5 nt hairpin loop that also contribute to hp2 recognition. In the rne transcripts of related bacterial species, the sequence of this hairpin loop varies at all but the first and last positions (Fig. 9B). Nonetheless, nucleotide substitutions at these two conserved positions can have little effect on hp2 function in E. coli, whereas substitutions at the nonconserved positions can significantly reduce the efficiency of feedback regulation when they deviate from variations observed naturally in other bacterial species (Figs. 6 and 8; note that the hairpin loop of hp2mut3T is identical to that of rne hp2 in Actinobacillus actinomycetemcomitans, whereas the hairpin loop of hp2mut2T is unlike any known hp2 variant in nature). The length of the short stem between the hairpin loop and internal loop is critical, hp2 function being markedly impaired by increasing the size of this stem from 2 to 4 bp (Fig. 6). On the other hand, the sequence of this 2-bp stem is inconsequential (20) despite its evolutionary conservation. Together, FIG. 8. Comparison of rne hp2 from nine bacterial species. A hp2-like element was identified upstream of the rne coding region in nine bacterial species. A sequence and likely secondary structure is shown for each. Sequence data for Vibrio cholera was previously published (24). Preliminary sequence data for Actinobacillus actinomycetemcomitans was obtained from The Institute for Genomic Research website at www.tigr.org. Preliminary sequence data for Klebsiella pneumoniae was obtained from the Genome Sequencing Center, Washington University website at genome.wustl.edu. FIG. 7. Effect of mutations in the lower stem of rne hp2. Derivatives of ez8 bearing site-directed mutations within the lower stem of rne hp2 were introduced into the isogenic E. coli strains CH1828 (low RNase E activity) and CH1827 ϩ pRNE101 (high RNase E activity). Cellular ␤-galactosidase activity and repression ratios were measured as in Fig. 2. Substituted nucleotides that differed from wild-type E. coli rne hp2 are enclosed in gray rectangles.
these findings suggest that the shape of the short stem and hairpin loop at the top of hp2 may be more important for recognition than their sequence per se and that the crucial features of this shape have been conserved during evolution despite sequence divergence.
The 8-nt internal loop of hp2 is situated atop a tall, imperfectly paired stem whose sequence, mispairings, and length are poorly conserved in evolution. Our data show that the sequence of this lower stem, including a highly conserved G-C base pair directly beneath the 8-nt internal loop, is unimportant (Fig. 7). Consequently, this portion of hp2 can be replaced with a perfectly base-paired stem having a different sequence without affecting feedback regulation. The lower stem of hp2 may serve simply as a pedestal for the upper portion of hp2, helping to ensure its proper folding and to improve its accessibility to the cellular factor that mediates its function.
Together, our findings suggest a model for rne hp2 recognition in which RNase E or a hypothetical protein cofactor makes intimate, sequence-specific contact with the 8-nt internal loop and backbone contact with the upper stem and hairpin loop (Fig. 9C). In vitro experiments suggest that hp2 itself is not a target for RNase E cleavage. 2 Nonetheless, by increasing the local concentration of RNase E, binding to hp2 could help the ribonuclease gain access to one or more cleavage sites elsewhere in the rne transcript, thereby accelerating rne mRNA degradation. The fact that base pairs can be removed from (hp2bot-syn) or added to (ez8) the bottom of hp2 without impairing its function suggests that the precise spatial relationship between the upper portion of this stem-loop and other regions of the rne 5Ј-UTR is relatively inconsequential.
Besides hp2, the only other domain of the rne 5Ј-UTR that is known to contribute appreciably to feedback regulation is rne hp3 (Fig. 1); however, this branched stem-loop is not required, and its role is not known (20). Previous experiments have shown that a putative RNase E cleavage site in the singlestranded segment (ss1) that precedes hp2 can be removed without impairing regulation (19,20). Likewise, the rne coding region appears not to contribute in cis to feedback regulation, because an rne-lacZ reporter in which the fifth rne codon has been fused to lacZ is tightly regulated by RNase E (23).
In comparison to a number of other E. coli genes, expression of the rne gene is unusually sensitive to the cellular level of RNase E activity (19). The identification of the key sequence features of rne hp2 provides a basis for searching the genomes of E. coli and related microorganisms for other genes encoding similar elements that may render those genes particularly sensitive to feedback regulation by RNase E. Using strict search criteria based on the essential features of hp2 to examine the sequences of the E. coli, Yersinia pestis, and H. influenzae genomes (see "Experimental Procedures"), rne was the only gene found to encode a hp2-like stem-loop in all three species. Two other mRNAs in E. coli and one in Y. pestis have the potential to form a similar stem-loop, but in each case this potential is of questionable significance due to its lack of conservation in either of the other two bacterial species that were examined. Using more relaxed criteria, additional candidates for possible hp2 homologs were identified, but again none were conserved in all three species. We conclude that hp2 most likely is a unique evolutionary adaptation of the rne gene that enables its expression to be tightly regulated by cellular RNase E activity.
Examination of the RNase E genes of bacterial organisms that are only distantly related to E. coli suggests that their rne transcripts may lack a recognizable homolog of hp2 upstream of the coding region. Either RNase E synthesis is not autoregulated in those species, or its autoregulation is mediated in cis by an element distinct from hp2.
The precision with which E. coli cells regulate RNase E synthesis is crucial for properly controlling rates of mRNA decay and for sustaining cell growth (18). Knowing the critical features of rne hp2 should be of considerable value in elucidating the molecular mechanism by which this regulatory stemloop senses the cellular concentration of RNase E and modulates rne gene expression through changes in rne mRNA longevity.