Use of an inducible regulatory protein to identify members of a regulon: application to the regulon controlled by the leucine-responsive regulatory protein (Lrp) in Escherichia coli

Procedures were developed to facilitate the identification of genes that belong to a given regulon and characterization of their responses to the regulator. The regulon controlled by the Escherichia coli leucine-responsive regulatory protein (Lrp) was studied by isolating random transcriptional fusions to lacZ, using lambda placMu53 and a strain in which lrp is under isopropylthio-beta-D-galactopyranoside (IPTG)-inducible control. Fusions exhibiting IPTG-responsive beta-galactosidase activity were cloned by integrating the suicide vector pIVET1 via homologous recombination at lacZ, followed by self-ligating digested chromosomal DNA. We verified the patterns of lacZ expression after using the plasmid clones to generate merodiploid strains with interrupted and uninterrupted copies of the same sequence. If the merodiploid expression pattern was unchanged from that shown by the original fusion strain, then the cloned fusion was responsible for the regulatory pattern of interest; a difference in the expression pattern could indicate that the original strain carried multiple fusions or that there were autogenous effects of having interrupted the fused gene. Using these procedures, we generated a fusion library of approximately 5 x 10(6) strains; approximately 3,000 of these strains were screened, yielding 84 Lrp-responsive fusions, and 10 of the 84 were phenotypically stable and were characterized. The responses of different fusions in a given operon to in vivo Lrp titrations revealed variations in expression with the position of insertion. Among the newly identified members of the regulon is an open reading frame (orf3) between rpiA and serA. Also, expression of a fusion just downstream of dinF was found to be Lrp dependent only in stationary phase.

Our understanding of cell physiology is changing from the extremely reductionist viewpoint of cells as collections of individual macromolecules specified by strings of self-sufficient genes toward the more complex view of interacting proteins and RNAs specified by cross-signaling global regulons. Regulons are sets of genes controlled by a common regulator, and since many genes are controlled by multiple regulators, there can be considerable overlap between regulons (11,47). We are interested in the roles played by global regulons in integrating a cell's responses to its environment.
Escherichia coli is perhaps the best-studied organism in the biosphere, yet it was not until 1991 that the leucine-responsive regulatory protein (Lrp) was recognized as being a global regulator. It had been suggested previously that L-leucine might be an important regulatory signal, since it was seen to affect the expression of several unrelated operons (24,53). A single gene, lrp, was found to be required for most of these leucine effects; lrp is located at 20 min on the E. coli chromosome and specifies an 18.8-kDa polypeptide that forms a homodimer (64). Lrp is a sequence-specific DNA-binding protein that contains a predicted helix-turn-helix motif (51,64). Lrp binds and bends regulatory sequences upstream of several operons in order to exert its effect on expression (3,13,16,17,20,25,26,39,55,56,(61)(62)(63). In this regard, it is a classical gene regulator. The abundance of Lrp in the cell (ϳ3,000 copies/haploid genome) (34,64) and its degree of DNA sequence specificity (17) are intermediate between specific DNA regulators like the lac repressor and general chromosomal organizers like H-NS and HU (33,49).
A most intriguing aspect of Lrp regulation is the variety of ways in which leucine, the effector, can modulate the regulatory effect of Lrp. For operons that are activated by Lrp, leucine may antagonize the activation (e.g., ilvIH [52]), it may be required for the activation (e.g., fimB-and fimE-promoted switching [25]), or it may have relatively little effect on activation (e.g., gltBDF [19,20]). These same three patterns of leucine responsiveness are seen for operons that are repressed by Lrp, as exemplified by tdh (40,54), livJ (27), and ompC (21). Lrp thus stands in sharp contrast to gene regulators such as the lac repressor, where the coregulator modulates the effect of the regulator in only one way. In a comparison of two Lrp-regulated operons, leucine antagonized the activation of ilvIH more strongly than that of gltBDF (20). Gel mobility shift assays indicate that the affinity of Lrp for the upstream region of gltBDF is greater than for ilvIH, with leucine reducing the amount of DNA binding in both cases; however, at an Lrp concentration matching the effective in vivo level, the higheraffinity gltBDF promoter region is nearly saturated and leucine shows little effect, while the ilvIH promoter region is not saturated and leucine has a profound effect (20). Based on these results, Ernsting et al. (20) proposed that the leucine sensitivity of positively regulated operons is a function of the intrinsic affinity of Lrp for the target operator, with higher-affinity operators showing lower sensitivity to leucine. This model was supported by in vivo studies that monitored the effects of leucine and varied Lrp levels on a gltB::lacZ operon fusion (12). To further test this model, as well as to understand the links between the Lrp regulon and other global regulons, we sought to obtain a battery of Lrp-controlled operons varying in leucine sensitivity and Lrp affinity.
Initially members of the Lrp regulon were identified by comparing two-dimensional gels of cell extracts of wild-type and lrp mutants (19) and by screening a placMu9 translational fusion library in the presence or absence of leucine (38). Our interest in leucine-insensitive operons, which were originally discovered by using two-dimensional electrophoresis of cell proteins (19), appeared to rule out our using fusion responses to leucine as a primary screening method. Furthermore, we are interested in identifying members of the Lrp regulon that range in affinity for Lrp. Thus, we made our fusion library by using a strain in which the lrp gene itself was inducible. We also developed approaches to cloning, confirming, and sequencing placMu53 fusions. Using these approaches, fusions could be subjected to an in vivo titration with Lrp, providing extensive information on the response of the targeted gene. Interestingly, the majority of fusions that we isolated did in fact respond to changes in the leucine concentration when subsequently tested, as predicted by the hypothesis (20) that all Lrp-controlled genes are leucine sensitive to some extent, with the degree of sensitivity a function of the affinity of Lrp for the relevant control region.
Construction and screening of a lacZ transcriptional fusion library. A promoterless lacZ gene was randomly fused into various locations of the chromosome of strain AAEC546 (p lacUV5 ::lrp) (10) by using placMu (14). The phage used to construct the transcriptional fusions, placMu53, carries the c end of phage Mu (including the genes MucI ts62 , ner [cII], and A) and the terminal sequences from the Mu S end (␤ end) but lacks a functional B gene (Fig. 1). In the absence of a helper phage supplying B protein, placMu53 transposes at a very low frequency and thus generates extremely stable and temperature-resistant lysogens (14). The presence of kan in placMu53 allows positive selection for integration into the chromosome independent of the Lac phenotype, allowing isolation of lacZ fusions under conditions where the expression of the fused gene is too low to give a detectably Lac ϩ phenotype. There are translation termination codons in all three reading frames between the S end and the translation initiator for lacZ.
FIG. 1. Chromosomally integrated bacteriophage placMu53. The fusion library was generated by using placMu53 and placMu507 (helper phage) (14) as described in Materials and Methods. Random integration of placMu53 is achieved by using its bacteriophage Mu integration sequences (S and c), and Mu B function is provided by the helper phage. In the correct orientation, lacZ expression is driven from the promoter of the unknown gene ‫;)ء(‬ XЈ is part of the unknown gene X that has been disrupted by integration of placMu53. The lacZ open reading frame is fused to a short portion of trpA and preceded by trpB. Not all placMu53 genes are shown, and genes derived from bacteriophage (in parentheses) are not to scale. At the S-end fusion junction, a 48-bp inverted repeat is indicated by the hairpin structure. Helper phage supplying Mu B G. Weinstock (14) a The allele number assigned to each of the various lacZ fusions corresponds to the nucleotide position of the fusion, using the numbering from an appropriate GenBank or EMBL entry. The dinfϩ103 allele has a fusion 103 bp past the 3Ј end of dinf.
Strain AAEC546 was grown overnight in LB medium supplemented with tetracycline and 0.2% maltose, and the cells in a 1-ml portion were infected with 10 8 PFU of placMu53 and 10 9 PFU of placMu507 (helper phage). After incubation for 30 min at room temperature, the cells were washed three times with LB and brought up in 1 ml of dil (8 mM MgSO 4 , 10 mM Tris-HCl [pH 7.5]). Portions were spread, undiluted, onto LB agar plates containing tetracycline and kanamycin to yield dense but not confluent growth; at the same time, dilutions were plated to allow an estimate of CFU/milliliter in the infected culture. After overnight incubation of the plates, 1 to 2 ml of LB was added, the plates were scraped, and the suspended cells were pooled and frozen without further growth to generate a library of ϳ5 ϫ 10 6 fusion strains.
This library was screened for IPTG-dependent lacZ expression by replica plating onto six different agar media, each containing X-Gal: LB, LB-IPTG, minimal glucose, minimal glucose-IPTG, minimal glucose-isoleucine-valineleucine, and minimal glucose-isoleucine-valine-leucine-IPTG. No antibiotics were included in the screening plates, as antibiotics can interfere with ␤-galactosidase activity (31). Lrp-responsive fusions were screened for stability by streaking out a frozen culture and retesting 10 independent colonies on each of the six agar media.
In vivo titrations with Lrp. Cultures were grown in glucose MOPS minimal medium (46) supplemented with isoleucine, valine, leucine (where indicated), and varied concentrations of IPTG. Samples were removed at various times from cultures in logarithmic growth. ␤-Galactosidase activity was determined by measuring o-nitrophenyl-␤-D-galactopyranoside (ONPG) hydrolysis in permeabilized cells (44,52). While the cultures were growing logarithmically (optical density at 600 nm [OD 600 ] of between 0.15 and 0.5), a series of 0.25-ml samples were removed and permeabilized in an equal volume of aqueous cetyl trimethylammonium bromide (200 g/ml) and sodium deoxycholate (100 g/ml) by incubation at 4°C overnight. Permeabilized cells (0.5 ml) were mixed with 0.5 ml of assay buffer (0.1 M sodium phosphate [pH 7.0], 1 mM MgSO 4 , 2 mM MnSO 4 , 50 mM ␤-mercaptoethanol) and 0.15 ml of ONPG stock solution (4 mg/ml). The assay mixtures were then incubated at 25°C until visibly yellow, at which point the reactions were terminated by adding 0.325 ml of 1 M Na 2 CO 3 . Cells were removed by centrifugation, and the A 420 of the supernatant was measured. ␤-Galactosidase activity was calculated by dividing the A 420 by the time allowed for the reaction and by the volume of the permeabilized cells used in the reaction. To determine fusion activity, the slope from linear regression of a plot of activity versus A 600 of the culture was determined; this slope is expressed in arbitrary units: (1,000 ϫ A 420 min Ϫ1 ml Ϫ1 )/OD 600 of the culture.
Cloning and sequencing of fusions. The cloning strategy is illustrated in Fig. 2. The fusion strains were transformed with pIVET1 by electroporation. Since the fusion strains do not produce the pi protein (22), which is required for replication of pIVET1, stable expression of ␤-lactamase requires pIVET1 integration into the chromosome by homologous recombination. To clone each fusion, overnight cultures were grown from five different integrant colonies (to increase the probability that an integrant via lacZ was present), and chromosomal DNA was isolated by using a Purgene DNA isolation kit (Gentra Systems, Inc.). This DNA was digested with BglII and self-ligated at a concentration of 0.025 g/l for 16 h at 16°C in the presence of 1 U of T4 DNA ligase (Life Technologies). The ligation mix was used to transform strain DH5␣ ⅐ pir. Plasmid DNA from the carbenicillin-resistant (Cb r ) transformants was screened for putative clones by restriction mapping. Desired clones should lack an ϳ1-kb BglII-EcoRI fragment and show increased overall size as assessed by EcoRI digestion (Fig. 2). DNA from selected clones was used to transform AAEC546; integration would result from homologous recombination between the plasmid and chromosomal copies of the promoter regions upstream of the reporter lacZ gene. Such transformants, now merodiploid for the fused gene, were retested for effects of IPTG on expression by streaking onto medium containing or lacking IPTG as described above.
Clones were sequenced by using a RapidWell DNA sequencing kit (U.S. Biochemicals) and primers L19 and U173 (Table 2; Fig. 2C). The S end of Mud1, the progenitor of placMu53, contains a 48-bp hairpin presumably generated by illegitimate transposition (43). When primer L19 was used, progression of DNA polymerases through this hairpin structure was found to be inefficient and yielded strong bands in all lanes through the hairpin and faint (but readable) bands beyond the fusion junction. This was true even when we used thermostable polymerases and elevated temperatures (not shown). The sequence obtained from both primers was sufficient to allow unambiguous identification of the fusion junction, and in some cases this identification was confirmed using insertspecific primers.

Construction and screening of a transcriptional fusion library.
To screen fusions for responsiveness to Lrp without depending on use of the coregulator leucine, we prepared a lacZ transcriptional (operon) fusion library in E. coli AAEC546. This strain, constructed by Blomfield et al. (10), carries the lrp gene under the control of a p lacUV5 promoter. Strain AAEC546 also has its native lrp gene disrupted by a Tn10 insertion, and the lac operon has been deleted. In this strain, lrp expression is repressed by LacI and is induced by IPTG. It has been found that the Lrp levels in this strain are undetectable when IPTG is omitted from the medium and can be titrated up to a maximum of two-thirds of the level in wild-type strains when increasing amounts of IPTG are added to the medium (12).
To increase the sensitivity of our screen, the fusion library was screened by replica plating onto three media that each contained or lacked 0.2 mM IPTG: rich medium and minimal glucose medium containing or lacking leucine. Fusion strains that showed IPTG-responsive expression on one or more of these three media were subjected to further study. To test for phenotypic stability, frozen stocks of individual fusion strains were streaked onto a rich medium containing kanamycin (to select for placMu53), and isolated colonies were tested on all six plates for the original pattern of expression. Only fusions for which 10 of 10 tested colonies showed the original regulatory pattern on all six plates were studied further. Of more than 3,000 fusions initially screened by replica plating, 84 (ϳ3%) showed Lrp-responsive regulatory patterns in primary screening; some of these could have been siblings. Of the 84 fusions tested, 10 were found to be phenotypically stable by our criteria.
In  (41) replicates autonomously only in a host that produces pi protein (22). S, bacteriophage Mu-derived S integration sequence; ‫,ء‬ promoter for the fused gene. (A) Integration of pIVET1 into the chromosome of a placMu fusion strain. When pIVET1 is introduced into a lacZ fusion strain, stable carbenicillin resistance results from integration via homologous recombination. (B) Genetic map of the integrant strain generated by the recombination shown in panel A. Chromosomal DNA was isolated from a pool of several Cb r integrants, cut with BglII, circularized by self-ligation, and introduced into strain DH5␣ ⅐ pir to allow autonomous replication. After carbenicillin selection, putative plasmid clones were isolated. (C) Genetic map of a plasmid containing the cloned fusion. Clones that yielded appropriate restriction maps and phenotypes were sequenced by using primers U173 and L19. (D) The plasmid shown in panel C was introduced into strain AAEC546, and the genetic map of the expected tandem merodiploid strain is shown. This strain was used to confirm the expression pattern of the cloned fusion.
changing Lrp concentrations on fusion expression. In strain DB3, which differs from AAEC546 only by the presence of a gltB::lacZ fusion, the concentration of Lrp has been quantitated on Western blots as a function of the concentration of IPTG in glucose minimal medium; near-maximal expression of lrp is achieved in the presence of 50 M IPTG (12). For each concentration of IPTG used, we measured fusion activity versus OD 600 of the culture, as shown in Fig. 3. The linearity of plots such as this one confirm that the cultures were in pseudosteady state (stable exponential) growth (1,23), and the slopes provide an accurate measure of fusion expression. In Fig. 4, these slopes were plotted against the IPTG concentration. Figure 4 summarizes the expression patterns of seven of the stable Lrp-responsive transcriptional fusions. Of the three fusions not shown, one yielded a pattern similar to that shown in Fig. 4G, and two gave responses similar to those shown in Fig.  4C to E. Transcription of the fusions shown in Fig. 4A and B is stimulated by Lrp, and leucine antagonizes the stimulation. These two fusions were found to be in the same gene (gltB [see below]), and while they exhibit similar regulatory patterns, they differ significantly in the level of transcription at a given IPTG concentration. Transcription of the fusions shown in Fig. 4C to E is stimulated by Lrp in the absence of leucine and depressed by Lrp in the presence of leucine; the leucine-dependent reduction in expression is shown more clearly in Fig. 4F, where the "ϩLeu" data from Fig. 4C to E have been replotted against an expanded y axis. Surprisingly, given the results of the original plate screening, Lrp and leucine had no discernible effects on transcription of the fusion shown in Fig. 4G (but see below). Transcription of the fusion shown in Fig. 4H is depressed by Lrp, and leucine antagonizes the depression.
It is possible that the internal cellular concentration of IPTG in these experiments was affected by varying expression of lacY (the lactose permease gene that is downstream of lacZ in placMu53). While IPTG diffuses across the bacterial membrane without the need for a permease, LacY can concentrate IPTG (44a). Any LacY-dependent effects would be most pronounced in those fusions having the highest levels of expression, and then only under conditions giving rise to high levels of expression. LacY concentrations will not affect the measured levels of fusion expression in the presence of a saturating concentration of IPTG or in the absence of IPTG. In sum, the presence of LacY could affect the apparent concentrations of IPTG required for half-maximal effects on fusion expression but will not affect the general patterns shown in Fig. 4.
Cloning and identification of the Lrp-responsive fusions. While methods have been developed to sequence placMu fusions without intermediate cloning steps (57,59), cloning allows phenotypic verification that the fusion being sequenced is actually the desired Lrp-responsive fusion. It provides more information than, and obviates the need for, P1 transduction of the fusion into a clean background. The cloning method that we devised makes use of the suicide vector pIVET1 (Fig. 2), a plasmid that replicates only in the presence of the pi protein, provided in host strain DH5␣ ⅐ pir by a pir ϩ prophage. In a recA ϩ pir background, such as strain AAEC546, Cb r transformants can be obtained by pIVET1 integration into the chromosome via homologous recombination ( Fig. 2A). In the case of placMu53 fusion strains, such homologous recombination is most likely to take place via lacZ (as shown in Fig. 2A and  B). For brevity, we refer to the resulting strains as integrants and to the original placMu53 strains as fusion strains. Plasmid pIVET1 contains unique restriction sites, such that digestion of the integrant chromosomal DNA followed by circularization with T4 DNA ligase can yield a functional Cb r plasmid carrying lacZ and some portion of the chromosome upstream of the fusion point (Fig. 2C). Eight of ten stable fusions were successfully cloned in this manner, and the resulting plasmids were used as templates for DNA sequencing to identify the site of the fusion.
It was important to verify that the cloned fusion exhibits the expected pattern of regulation in response to Lrp and leucine. Plasmid DNA from positive clones was used to transform AAEC546, which is Lac Ϫ and lacks pir. Cb r integrants resulting from homologous recombination at the location of the original fusion are tandem fused/wild-type merodiploids (Fig.  2D). These merodiploid strains were tested for the expression pattern shown by the original fusion. If the merodiploid pattern of expression was unchanged from that shown by the original fusion strain, then the cloned fusion is directly responsible for the regulatory patterns of interest; a difference in the expression pattern could indicate that the original strain carried multiple fusions or that there were autogenous effects of having interrupted the fused gene. Figure 4D shows a detailed comparison of the expression patterns of one fusion strain and its corresponding merodiploid; they yield virtually identical patterns and levels of expression. The results for five of the eight fusion/merodiploid pairs are shown in Table 3, and the merodiploids and corresponding fusion strains have closely matching activities. The three strain pairs not shown in Table   FIG  3 have activities that differ substantially between merodiploid and fusion strains, and these are discussed below. We next determined the sequence of the DNA flanking the insertions. We used a primer (L19 [Fig. 2C]) complementary to trpB (in placMu53, lacZ is actually fused to a short segment of trpA corresponding to the amino terminus, preceded by trpB [ Fig. 1]). This strategy worked, though it provided only a relatively short sequence across the fusion point for reasons described in Materials and Methods. We also sequenced the opposite end of the cloned fragment by using a primer (U173 [ Fig. 2C]) that corresponds to the region upstream of the bla gene. Sequences obtained by using each primer were used to search the GenBank and EMBL databases, using the program BLAST (2). As shown in Fig. 5A, sequences from six of the eight clones matched database entries. For a given clone, comparing the corresponding sequences from primers L19 and U173 confirmed these matches in three ways (not shown): they match the same regions, the U173 sequence matches are opposite in polarity to the L19 matches, and the U173 and L19 matches are separated by the distance predicted by restriction analysis of the plasmid clones. In one case, the quality of the match led us to confirm the identification with a specific sequencing primer (Fig. 5B). In another case (not shown), the sequence obtained from a clone matched the E. coli trp operon. However, the merodiploid strain gave a regulatory pattern completely different from that of the original fusion, and others have reported recombination between chromosomal and placMu-carried trp DNA (9); we thus have to reclone the original fusion in this case.
The fusions that have been identified fall into three categories: fusions to genes known to be in the Lrp regulon and showing expected patterns of expression, fusions to genes known to be in the regulon but showing unexpected patterns of expression, and fusions to new loci including open reading frames of unknown function. Two of our fusions, gltB4527:: lacZ and gltB4702::lacZ, were found to be in the gltBDF operon, and their expression is stimulated by Lrp (Table 3). gltB specifies the large subunit of glutamate synthase and is known to be activated by Lrp in a relatively leucine-insensitive manner (19,20). Thus, our strategy for construction and cloning of fusions yields known members of the Lrp regulon that are regulated in the expected manner. Two other fusions, livG6919::lacZ and livK3663::lacZ, were found to be in the livKHMGF operon, which is also known to be part of the Lrp regulon (27,38,59). livG codes for a membrane-associated subunit of the leucine transport system, while livK specifies a periplasmic binding protein associated with this transport system. As previously noted for certain livJ and livH translational fusions (59), fusions in livK ( Fig. 4D and F) Table 1, is from another study and has no corresponding merodiploid strain.
b In each case, the first number indicates the units of LacZ activity from the original placMu fusion strain, and the second number represents units of LacZ activity from the corresponding merodiploid strain. Assays and activity units are described in Materials and Methods. Each activity value is the slope from linear regression of activity versus culture density; at least five points define each slope, and the correlation coefficients were in all cases greater than 0.9. Cultures were grown in MOPS minimal glucose medium containing isoleucine, valine, and, where indicated, 0.05 mM IPTG and/or 10 mM leucine.
c In this case the concentration of IPTG, when added, was 0.2 mM. the presence of leucine. Two newly identified fusions occur in orf3 (EMBL accession no. X66836) and downstream of dinF (GenBank accession no. U00006). orf3 lies between rpiA, the gene for the constitutive form of ribose phosphate isomerase, and serA, the gene for 3-phosphoglycerate dehydrogenase. Two groups have shown that serA transcription is activated by Lrp (40,54); we find that transcription of orf3 is depressed by Lrp (Fig. 4H). dinF is not known to be controlled by Lrp but is a member of the SOS regulon; the role of dinF has not yet been determined (60).

Regulation of a fusion downstream of dinF by Lrp.
We further explored regulation of the fusion that yielded the data in Fig. 4G, because the effects of Lrp on ␤-galactosidase expression were so minimal (at least during logarithmic growth) that we wondered how this fusion had been identified by our screen. Since screening colonies on plates is essentially an analysis of stationary-phase cells, we decided to determine the expression of this fusion in liquid culture during the transition to stationary phase. We confirmed that while the culture is actively growing, expression of fusion dinFϩ103::lacZ is essentially the same whether or not IPTG or leucine is present in the medium (Fig. 6). However, as the cultures enter stationary phase, expression of this fusion is substantially lower when IPTG (Lrp) is present; this depression of expression is antagonized by leucine (Fig. 6).
A surprising and yet unexplained observation is that the patterns of expression of the original fusion strain and the merodiploid strain are essentially identical, while the amounts of expression in each case differ nearly 10-fold (compare Fig.  6C and D). As mentioned above, one possible explanation for this type of discrepancy is that the fusion interrupts an autogenous regulatory circuit.
Because this fusion is downstream of a member of the SOS regulon, we also tested its response to the DNA-damaging agent mitomycin. The results (not shown) reveal that fusion expression increases somewhat following mitomycin addition, whether or not IPTG or leucine is present, but that expression of a control fusion (to gltB) does not show this increase. This modest induction is similar to that for the Lrp-regulated L-  serine deaminase (sdaA) gene when cells were treated with mitomycin, nalidixic acid, or UV irradiation (48).
Positional effects of insertions into the gltBDF operon. Even in cases of genes already known to be regulated by Lrp, important features have become apparent from having multiple fusions at various locations within the gene or operon. gltBDF expression is stimulated by Lrp, and the stimulation is antagonized by leucine (19,20). The gltB4527::lacZ and gltB4702:: lacZ fusions both showed the expected regulatory patterns, though the highest observed level of expression of gltB4702:: lacZ was ϳ4-fold lower than that of gltB4527::lacZ (Table 3). This difference could result from the extents to which the responses are sigmoidal: in 20 M IPTG, expression of gltB4527::lacZ is already maximal (Fig. 4A) whereas that of gltB4702::lacZ is only slightly above its baseline level (Fig. 4B). The results shown in Fig. 4B also suggest that gltB4702::lacZ expression might increase further with further increases in Lrp concentration (not achievable in strain AAEC546). To further explore the differences between these two fusions, we studied an additional gltB::lacZ fusion located near the beginning of gltB. Strain DB3, an AAEC546 derivative, contains a Mud1 (kan, lac) fusion in gltB at position 1594 (12,43). The results for gltB1594::lacZ (DB3) were similar to those for gltB4527:: lacZ and ϳ4-fold higher than those for gltB4702::lacZ (Table  3), implying a drop in expression associated with the 175 bp of DNA between positions 4527 and 4702. The sequence between the two fusion points contains no obvious transcription terminators. It appears that something within this 175-bp region either attenuates transcription under the influence of Lrp or affects Lrp-mediated transcript initiation at the gltBDF promoter. This observation illustrates the value of examining multiple fusions even within a single gene.
Positional effects of adjacent insertions into operon clusters. Operons with related functions sometimes occur in clusters; an example is the adjacent livJ and livKHMGF operons that specify the high-affinity branched-chain amino acid transport system (Fig. 7) (35). As described above, we observed marked effects on expression with the position of an insertion within a single gene. In addition, we and others (27,59) have observed that multiple insertions into the same operon cluster can reveal differences in both the patterns of regulation and the extent of expression.
(i) Positional effects of insertions into the liv operons. Two of our fusions were found to involve the livK and livG genes (Fig. 7). Expression of both livK3663::lacZ and livG6919::lacZ was strongly stimulated by Lrp in the absence of leucine but depressed by Lrp in the presence of leucine (Table 3; Fig. 4D to F). The liv genes appear to be organized into two operons. The livJ gene was previously shown to be repressed by Lrp, with leucine being required for repression (27), which differs in part from the regulatory pattern shown by our fusions and by a livH::lacZ fusion (coordinates not reported) cited in another study (59): all fusions from all three studies show depression of expression mediated by Lrp in the presence of leucine, but results differ in the absence of leucine (no effect versus stimulation). It should be noted that only Haney et al. confirmed their results by assays of leucine transport activity (27). The three characterized fusions in the livKHMGF operon (Fig. 7) are thus far the only reported examples of Lrp both stimulating and depressing expression dependent on whether leucine is present. A possible physiological rationale for livJ having a distinct regulatory pattern from livKHMGF is suggested by the facts that livK specifies a leucine-binding protein, while livJ codes for a protein that binds all three branched-chain amino acids (the livHMGF genes specify the transport complex with which both LivJ and LivK interact) (4,35,45).
(ii) Regulation of orf3 by Lrp. One of our fusions, which shows Lrp-dependent repression (Fig. 4H), lies within orf3 about 100 bp upstream of the translational start of serA (Fig.  8A). serA is an Lrp-controlled gene (40,50,54) transcribed from two promoters, P1 (activated by Lrp) and P2 (repressed by Lrp) (50,65); our fusion lies between the two promoters. However, orf3 begins upstream of both identified serA promoters and appears to be transcribed from a third Lrp-repressed promoter (labeled P3 in Fig. 8) (9a). Just upstream of orf3 and in the same orientation is the gene for the constitutive form of ribose phosphate isomerase, rpiA (30). The possible effects of transcriptional readthrough from rpiA on orf3 expression have not been determined.

DISCUSSION
A set of procedures for characterizing regulons. Several approaches have been used to identify members of the Lrp regulon, including two-dimensional gel electrophoresis of extracts from Lrp ϩ and Lrp Ϫ strains (19) and screening lacZ fusion libraries in the presence or absence of leucine (38). Both approaches identified important members of the regulon but suffered some limitations. In the two-dimensional gel approach, most Lrp-responsive polypeptides could not be identified and polypeptides with pI values above ϳ8 could not easily be studied. The leucine screening approach was also limited in its ability to identify genes whose regulation by Lrp is relatively insensitive to leucine.  (35) is shown with base pair locations (where known) of gene fusions from this study (shown in bold) and from earlier studies (27,59). For two fusions, the affected gene was reported but insertion coordinates were not published (59). The chart indicates effects of Lrp alone or of Lrp in the presence of exogenous leucine, relative to expression in an Lrp Ϫ (lrp::Tn10) background; ϩ indicates stimulation of expression, 0 indicates no effect, and Ϫ and ϪϪ indicate partial and profound reduction of expression. The simplest model consistent with the available information would involve three promoters (circles): one promoter for livJ, one for the livKHMGF cluster, and one for the oppositely oriented yhhK. Deletion of yhhK (originally called orf-19) results in pantothenate auxotrophy (35). Two lacZ fusions (CP5 and SH152) show expression patterns that are dissonant with this three-promoter model (27).
FIG. 8. Genetic map of the region including the rpiA (ribose phosphate isomerase) and serA (3-phosphoglycerate dehydrogenase) genes. The amplified region of the map shows the transcription terminator following rpiA (30) (shown as a hairpin), reported Lrp-binding sites (50,65), and the serA promoters P1 and P2 (50,65). The apparent promoter identified in this study, P3, is indicated by a dashed arrow.
We are particularly interested in identifying genes whose regulation by Lrp is relatively insensitive to leucine and in obtaining genes that range in the degree of their responsiveness to Lrp. Hence, we constructed a lacZ transcriptional (operon) fusion library in a strain in which the lrp gene itself is inducible by IPTG. We screened this library in media either containing or lacking IPTG and looked for members which showed differential expression of lacZ under these conditions. Selected fusions were tested for stability of the Lrp effect, and the most stable ones were selected for further study.
We next sought to clone and identify the genes that yielded Lrp-responsive fusions. We developed an approach to clone and sequence chromosomal placMu53 fusions that offers some advantages over recently published methods. One published method involves directly sequencing phage DNA prepared from UV-induced placMu lysogens, using a primer specific to the Mu c end (57). Another method involves inverse PCR amplification of the upstream region of a placMu insertion followed by sequencing of the PCR product by using a primer specific to the Mu S end (59). Both of these methods are straightforward and useful; however, neither yields plasmid clones of the fusions and thus neither can rule out misidentifications due to DNA rearrangements or to multiple fusions having occurred in the same cell. Further, neither method provides for a way to distinguish between normal regulatormediated effects and possible additional effects of having disrupted the target gene. Our method has the advantages of confirming that the DNA that was cloned and sequenced actually contains a gene regulated by Lrp and revealing effects of the gene disruption on expression of that gene.
Differences between the expression patterns shown by corresponding merodiploid and nonmerodiploid fusion strains would be expected for autogenously regulated genes (for example), since in the latter case there may be no intact copy of the fused autoregulatory gene. For most of the fusions that we characterized, the regulatory patterns of the merodiploids and the corresponding fusion strains were similar or identical, and the effects of Lrp and leucine were also very close in most cases (Table 3; Fig. 4H). One pronounced difference that we did see between a fusion strain and its corresponding merodiploid involved the fusion downstream of dinF; the effects of Lrp, leucine, and growth phase are virtually identical in the two strains, but there is a ϳ10-fold difference in the total amount of LacZ activity (Fig. 6). A second case of fusion/merodiploid difference in expression revealed the occurrence of a DNA rearrangement (recombination between the placMu and host trp sequences, as has been seen by others [9]).
Sampling the membership of the Lrp regulon. While we have found some interesting new regulon members, many of our fusions were in genes or operons previously known to be regulated by Lrp. The gltBDF and liv operons were shown by electrophoretic and genetic means, respectively, to be Lrp regulated (19,27) and were also detected in an earlier study using a variant of placMu that yields translational fusions (38). There are several possible explanations for the fact that so many of the same genes are found in various surveys of the Lrp regulon. The simplest explanation is that perhaps this regulon is already well defined. The alternative and, in our view, more likely explanation is that the survey methods are biased in some way. One possible source of bias is that placMu may have some site preference for integration (57), but we have not observed any obvious sequence preference at our fusion junctions. Second, the genes that are most readily detected in such screenings are presumably those that show the strongest effects of a particular regulator, and both gltBDF and the liv operon fall into this category. Third, many genes regulated by Lrp may be essential or physiologically important, such that their disruption is lethal or greatly slows growth. In contrast, while mutation of gltBDF results in an Ntr Ϫ phenotype (an inability to grow with very low concentrations of inorganic nitrogen as the sole source of nitrogen), our libraries were constructed in a rich medium; and while mutation of the liv operon eliminates a high-affinity transport system for branched-chain amino acids, there is still another low-affinity system as well as endogenous biosynthesis.
A fourth possible source of bias in the various surveys of the Lrp regulon involves the preference for genetically stable fusions. Of 3,000 clones initially tested, 84 showed Lrp-dependent differential expression in the initial round of screening, which is about 3% of the total. This number of fusions was about what we expected given the frequency of Lrp responsiveness among E. coli polypeptides (19), considering that over half of the fusions would be in the wrong orientation or in untranscribed regions. Of these 84, only 10 were found to give consistent Lrp-dependent regulatory patterns. In some cases, such genetic instability may be instructive: mutations in metK, which codes for an adenosylmethionine synthetase, lead to much lower growth rates, and metK strains accumulate secondary mutations in lrp that restore the growth rates (39). metK is not known to be regulated by Lrp, and there is no satisfactory explanation for the secondary mutations in lrp, but now that we have a good method to clone and characterize these fusions and generate merodiploids, characterization of the unstable fusions could well identify genes that previous screens have missed.
One of our original goals was to find additional leucineindependent operons (defined as showing a leucine effect on expression of less than twofold). We found none among the set of fusions described here. However, strain AAEC546, even when maximally induced, produces only about 70% of the wild-type level of Lrp (12). Reduced levels of Lrp are expected to increase the leucine sensitivity of expression of Lrp-regulated operons (12,20). Thus, our having used a strain with reduced Lrp levels decreased the likelihood that what are normally leucine-independent fusions would appear so in our screens. Our results are consistent with the model that all Lrp-regulated genes are leucine responsive at sufficiently low Lrp concentrations (20).
Growth phase-dependent regulation by Lrp. A fusion downstream of dinF showed an unusual pattern of expression, being depressed by Lrp only on the transition to stationary phase (Fig. 6). In other words, this fusion is not simply responding to a transition to stationary phase, but rather the transition to stationary phase is required to reveal Lrp dependence of expression. In strain AAEC546, lrp is under the control of p lac , but in a wild-type strain, Lrp levels are reported to rise in stationary phase (34); thus, the stationary-phase effects that we observed with the fusion in strain AAEC546 might be more dramatic in a background wild type for lrp. Lrp has previously been shown to affect the expression of some other genes preferentially expressed in stationary phase: otsBA, bolA, osmY, and csiD (29) and aidB (36). Lrp is reported to stimulate stationary-phase expression of csiD, to depress expression of osmY in all growth phases in a minimal medium, and to prevent induction of osmY in rich medium prior to stationary phase (29,37); also, Lrp represses RpoS-dependent transcription of aidB (36). Regardless of exactly how Lrp is acting, the pattern shown by the dinFϩ103::lacZ fusion is consistent with the role that Lrp plays in controlling osmY in rich medium, a role that Hengge-Aronis et al. refer to as transition state regulator, controlling the time of expression rather than the extent of expression (29,37). This result underscores the importance of screening fusions under a variety of growth conditions. Summary. There are two main advantages provided by our approach to characterizing members of a regulon. First, this approach provides phenotypic confirmation of the relevance of the DNA being sequenced. Second, the identified fusions can be immediately characterized by an in vivo titration with the regulatory protein (Lrp, in this case), not only revealing whether positive or negative effects are being seen but also providing an indication of the relative affinity of various control regions for the regulator. Such information is important for any regulatory protein that has different concentrations under different growth conditions, for example, Lrp (34), Fis (8), integration host factor (42), H-NS (18), and cyclic AMP receptor protein (32). Combining many of these in vivo titration results, to give a more extensive version of Fig. 4, would provide a useful and interesting snapshot of cell physiology as it is modulated by a regulator. We have identified some previously unknown members of the regulon, but it remains to be determined whether their modulation by Lrp is direct. Even in operons previously shown to belong to the Lrp regulon, the new fusions have revealed apparent complexities that remain to be investigated.