virG, an Agrobacterium tumefaciens Transcriptional Activator, Initiates Translation at a UUG Codon and Is a Sequence-Specific DNA-Binding Protein

The Agrobacterium tumefaciens Ti plasmid virG locus, in conjunction with virA and acetosyringone, activates transcription of the virulence (vir) genes. Insertional and deoxyoligonucleotide-directed mutagenesis studies showed that both octopine and nopaline Ti plasmid virG genes initiate translation at a UUG codon. VirG protein initiated at this UUG codon was found to be 241 amino acid residues in length and had an apparent molecular mass of 27.1 kilodaltons. A SalmoneUa typhimurium trp-virG transcriptional fusion was constructed to overproduce VirG. Agrobacterium cells containing this gene fusion showed a large increase in virG activity in the presence of virA and acetosyringone. Since the trp promoter is not under virA-virG control, this result indicates that modification of VirG is necessary for its full activity. VirG overproduced in Escherichia coli was purified from inclusion bodies. It was found to be a DNA-binding protein that preferentially bound DNA fragments containing the 5' nontranscribed regions of the virA, -B, -C, -D, and -G operons. Significant specific binding to the 5' nontranscribed region sequences of virE was not detected. DNase I footprinting of the upstream regions of virC-virD and virG showed that VirG binds to sequences around the vir box region. The virulence (vir) genes of the Agrobacterium

The virulence (vir) genes of the Agrobacterium tumefaciens Ti plasmid encode proteins required for the pathogenic action of the bacterium (reviewed in references 10, 18, and 44). Of the seven vir operons identified thus far, only virA and virG are expressed in free-living bacteria. Interaction of bacteria with plant cells is required for the transcriptional activation of the other virulence genes (29,30). These vir genes probably mediate the processing and transfer of a segment of the Ti plasmid (T-DNA) from the bacteria to the plant. The T-DNA is found integrated in the plant nuclear genome and encodes, among others, genes responsible for the biosynthesis of the phytohormones auxin and cytokinin. The hormonal imbalance, a result of expression of T-DNAencoded genes, causes the tumorous growth associated with crown gall tumor disease.
Expression of the vir genes is controlled at at least two levels. These genes are regulated positively by virA and virG in conjunction with plant phenolic compounds such as acetosyringone (AS) and hydroxyacetosyringone (29). These compounds, normally produced in metabolically active plant cells, are present at higher levels in wounded plant cells. In addition, virC and virD are regulated by a chromosomal locus (ros) in a negative manner (1,2,33).
Regulation of the regulatory genes, virA and virG, is more complex. virA is constitutively expressed in free-living bacteria and is induced by AS in a virA-virG-dependent manner (23,40). virG is transcribed from two promoters producing transcripts that differ in length at the 5' end by 50 nucleotides. The shorter transcript is produced by free-living bacteria, whereas the longer one is induced in a virA-virG-dependent manner (31). Transcription of virG is also induced by pH shift or by phosphate starvation (35,40). The promoter(s) that is activated by these stresses is not known; however, this activation is not virA-virG dependent (40).
Gene fusion studies and DNA sequence analysis indicate * Corresponding author.
that one of the regulatory proteins, VirA, spans the bacterial inner membrane (14,16,41), whereas the other, VirG, is probably a cytosolic protein (42). VirA is hypothesized to interact with the plant phenolics in its periplasmic domain, leading to the activation of its cytoplasmic domain. Activated VirA is then expected to activate VirG by modification. Modified VirG probably functions as a transcriptional activator (42). Recently, nopaline Ti plasmid VirG was shown to bind DNA nonspecifically when the nontranscribed region of the virE operon was used as a probe (21). Our recent deletion mutagenesis studies indicate that a conserved tetradecameric vir box sequence, 5'-dPuT/ATD CAATTGHAAPy-3' (D = A, G, or T; H = A, C, or T), is required for induction of the virB promoter (3). This sequence is presumed to be the VirG-binding site. All inducible vir genes of both octopine (pTiA6) and nopaline (pTiC58) Ti plasmids contain this sequence, whereas the noninducible nopaline Ti plasmid virG lacks it (3,32,39). DNA sequence analysis of pTiA6 virG indicates that it can potentially encode a 267-residue polypeptide (42). This study also indicates that there are several potential translation initiation sites for virG mRNA. In this study, we demonstrate that a UUG codon, not an AUG codon as presumed earlier (42), is the translation initiation codon for pTiA6 virG mRNA. The same codon is also used for translation initiation of the pTiC58 virG mRNA. We also report the overproduction of VirG in Escherichia coli and its purification and demonstrate that VirG binds specifically to the upstream regions of the vir promoters and protects sequences in the vicinity of the vir boxes in DNase I protection assays.

MATERIALS AND METHODS
virG assay plasmids. The pTiA6 virG was isolated as a 1.25-kilobase-pair (kb) SmaI-HindIII fragment from plasmid pSW104 (42) and was cloned into plasmid vector pUC119 (36) to const-uct pGP109. Plasmid pGP119 was constructed by introduction of the pTiA6 virA and a virB-lacZ gene 1242 PAZOUR AND DAS fusion into the wide-host-range vector pTJS75 (3,26). Both plasmids pGP109 and pGP119 were linearized at respective unique EcoRI sites and ligated together to construct pGP159. Plasmid pGP159 contains pTiA6 virA, virG, and a virB-lacZ gene fusion. This plasmid, when introduced into avirulent, Ti plasmid-less A. tumefaciens A136, supports synthesis of 3-galactosidase in the presence of AS (3). All virG mutants were constructed in plasmid pGP109 or a derivative, fused to pGP119, and introduced into A. tumefaciens A136 for ,Bgalactosidase assays.
pTiA6 virG mutant plasmids. Plasmids pGP206 and pGP207 were constructed by linearization of pGP109 with Eco47III, followed by ligation with 8-base-pair (bp) XbaI linker DNA (dCTCTAGAG) and by linearization of pGP109 with TthIII1, followed by mung bean nuclease treatment and ligation in the presence of XbaI linker DNA, respectively.
pTiC58 virG mutant plasmids. The pTiC58 virG was isolated as a 2.9-kb EcoRI restriction fragment from cosmid clone pTHE7 (9) and was cloned into the EcoRI site of pUC119 to construct pGP204. sspGP204 DNA was mutagenized with oligo C to construct plasmid pGP214. sspGP214 DNA was mutagenized with oligo D to construct pGP216. The virG containing the 2.9-kb EcoRI fragment of plasmid pGP204, pGP214, or pGP216 was cloned into the EcoRI site of pGP119 to construct plasmid pGP217, pGP218, or pGP219, respectively. pTiA6 virG-overproducing plasmids. The pTiA6 virG gene product was overproduced in E. coli from a Salmonella typhimurium trp-virG fusion plasmid. The virG structural gene was isolated as a 0.95-kb DraI-HindIII restriction fragment from plasmid pGP172. pGP172 has a G --T mutation at position +113 (transcription initiation site, +1) that creates a unique Dral site. The DraI-HindIII fragment, after filling in with Klenow enzyme in the presence of deoxynucleoside triphosphates, was cloned into the SmaI site of pAD9 to construct pGP183. Plasmid pAD9, a derivative of pAD7 (5), contains the S. typhimurium trp promoter as a 151-bp Hindlll fragment and the E. coli rpoC transcription terminator sequences (28) as a 225-bp EcoRI fragment in plasmid pUC8. The trp promoter fragment lacks the attenuator sequences (43) and therefore is constitutively expressed in a trpR host. Plasmid pGP198 was constructed in an analogous manner except that the virG sequences were isolated from pGP195. This plasmid contains an ATG start codon for virG instead of the wild-type TTG codon.
Plasmid pGP202 was constructed by cloning a 1.6-kb PvuII fragment containing the trp-virG fusion of pGP198 into the SmaI site of pUC18 Chl (a chloramphenicol-resistant pUC plasmid). Plasmid pGP208 is a pGP202 derivative in which a 93-bp XhoI fragment within the virG structural gene was deleted. This deletion removes 31 amino acids from VirG and therefore is expected to yield a 210-residue polypeptide.
Other plasmids. The wide-host-range derivatives of pGP183 and pGP198 were constructed by fusing these plasmids with EcoRI-cleaved pGP119. The two plasmids were designated pGP188 and pGP199, respectively. Plasmid pAD1137 was constructed by cloning the 188-bp Sau3A fragment (14; A. Das, unpublished data) that spans from -183 to +5 (transcription initiation site, + 1 [4]) of the virA gene into the BamHI site of pAD1090. Plasmid pAD1138 contains a 330-bp virB fragment that spans the region from -108 to +221. An AvaII-Hinfl restriction fragment (37) containing this region, after filling in, was cloned into the HinclI site of pUC7 to construct pAD1025. The insert was isolated as a BamHI fragment and was cloned into the BamHI site of plasmid pAD1090 to construct plasmid pAD1138. Plasmids pAD1139 and pAD1140 were constructed by cloning the 315-bp Sau3A fragment (11) that contains the region from -241 to +74 of the virC operon or -203 to + 112 of the divergent virD operon into the BamHI site of pAD1090. The orientation of this fragment, which places the virC promoter oriented toward the rpoC terminator, is designated pAD1139; the orientation that places the virD promoter oriented toward the rpoC terminator is designated pAD1140. Plasmid pAD1090 is a pUC13 Chl derivative in which a PvuII-HindIII fragment containing the E. coli lac promoter sequences were deleted and the E. coli rpoC terminator sequences were cloned as an EcoRI restriction fragment. This plasmid contains a short duplication of the pUC13 polylinker region.
Plasmid pAD1073 contains a 201-bp fragment that spans from -146 to +55 of the virG gene. It was constructed by cloning an EcoRI-RsaI restriction fragment from pSW104 (42) that had been filled in with Klenow enzyme into the SmaI site of pAD1072. Plasmid pAD1074 was constructed by cloning the filled-in 618-bp SalI-StuI fragment (38) containing from -193 to +425 of the virE operon into the SmaI site of pAD1072. Plasmid pAD1072 was constructed by cloning the 225-bp EcoRI fragment of pAD9 containing the E. coli rpoC terminator sequences into the EcoRI site of pGEM1 (Promega Corp.).
Purification of VirG. E. coli GP202 was grown overnight at 30°C in TB medium (34) containing 25 ,ug of chloramphenicol THE virG GENE ENCODES A DNA-BINDING PROTEIN per ml. The cells were diluted to an optical density at 600 nm of 0.1 in the same medium and grown for 12 h at 37°C. The cells were collected by centrifugation, washed twice with cold 0.8% NaCl, and frozen at -80°C. The cell paste was thawed on ice in protein buffer (25 mM Tris [pH 8.0],0.1 mM EDTA, 0.02 mM phenylmethylsulfonyl fluoride, 1 mM dithiothreitol, 10% glycerol) and lysed in a French pressure cell at 10,000 lb/in2. Inclusion bodies and the membranes were pelleted by centrifugation at 10,000 x g for 15 min at 4°C. The pellet was suspended in protein buffer and centrifuged twice to remove contaminating soluble proteins. The pellet was resuspended in protein buffer and was made 1% in Sequanal-grade sodium dodecyl sulfate (SDS) (Pierce Chemical Co.), 5% 3-mercaptoethanol, and 10% glycerol. The proteins were solubilized by incubation at 37°C for 30 min, loaded onto a preparative 10% SDS-polyacrylamide gel (13), and electrophoresed overnight at 4°C. After staining of the gel with cold 0.25 M KCl-1 mM DTT the VirG band was excised and electroeluted (8). The electroeluted protein was dialyzed against protein buffer, made 50% in glycerol, and stored at -20°C.
DNA-protein binding assay. DNA-binding ability was measured by a gel retardation assay. The reaction mixture contained, in a total volume of 20 ,ul, 2.5 ng of radiolabeled DNA (see below), 50 mM KCl, 10 mM Tris hydrochloride (pH 7.4), 1 mM EDTA, 5% glycerol, 2 ,ug of bovine serum albumin, and VirG, as indicated. Incubation was for 15 min at 30°C. After incubation, the reaction mixture was loaded onto a 6% polyacrylamide gel. The gel was electrophoresed for 2 h at 400 V in 0.5x TBE, dried, and subjected to autoradiography.
Probe for examining binding to the virA promoter was prepared by digesting pAD1137 with ApaLI, EcoRI, and PvuII, producing four fragments, of which the 0.7-, 0.23-, and 0.09-kb fragments were labeled. The 0.7-kb fragment contained the virA sequences. Plasmid pAD1138 was digested with EcoRI, HindIII, and HaeII to prepare probe for virB. This digest produced six fragments, of which 0.38-, 0.23-, 0.16-, 0.05-, and 0.03-kb fragments were labeled, and the 0.38-kb fragment contained the virB sequences. To examine binding to DNA from the divergent operons, virC and virD pAD1139 was digested with EcoRI, Hindlll and HaeII, producing six fragments, of which the 0.37-, 0.23-, 0.16-, 0.05-, and 0.03-kb fragments were labeled. The 0.37kb fragment contained the virC and virD sequences. This fragment migrated slower than expected in a native polyacrylamide gel. Probe for virE was prepared by digesting pAD1074 with EcoRI, HindIII, and HaeII, producing nine fragments, of which the 0.63-, 0.28-, 0.23-, and 0.1-kb fragments were labeled. The 0.63-kb fragment contained the virE sequences. Plasmid pAD1073 was digested with EcoRI and HaeII to prepare probe for virG. This produced eight fragments, of which 0.5-, 0.23-, and 0.1-kb fragments were labeled. Of these, the 0.5-kb fragment contained the virG sequences. All probes for gel retardation were end labeled by filling in with Klenow enzyme in the presence of [a-32P]dATP. Because of different labeling efficiencies of the various ends, the specific activities of individual restriction fragments varied considerably.
DNase I footprinting. VirG was incubated with approximately 1 fmol of end-labeled pAD1073 or pAD1140 DNA as described above. The reaction mixtures were then cooled in ice, MgCl2 was added to 5 mM, and 0.4 U of RQ1 DNase (Promega Corp.) was added. After 2 min at 30°C, the reactions were terminated by phenol-chloroform extraction and made 0.4 M in LiCl, and the DNA was precipitated with 2 volumes of ethyl alcohol, using 20 Kg of glycogen as a carrier. The precipitate was dried and suspended in 95% formamide-20 mM EDTA with bromophenol blue and xylene cyanol. Samples were denatured by boiling and electrophoresed on a 6% urea-polyacrylamide DNA sequencing gel. After electrophoresis, the gels were dried and subjected to autoradiography. The probe DNA was prepared by digestion of pAD1073 and pAD1140 DNA with XbaI and PstI, followed by filling in with Klenow enzyme in the presence of [a-32P]dCTP. Restriction endonuclease XbaI cuts pAD1073 DNA 166 bp upstream of the virG transcription initiation site and pAD1140 DNA 76 bp downstream of the virC transcription initiation site (which is also 205 bp upstream of the virD start site). PstI cuts each of these plasmids just upstream of the XbaI site and cannot be radiolabeled by Klenow enzyme. A DNA sequencing reaction of M13mpl8 DNA was electrophoresed simultaneously for sizing of DNA fragments.
Other procedures. Plasmids were mobilized from E. coli to A. tumefaciens A136 by triparental mating procedure (6). A. tumefaciens A136 is an avirulent strain that lacks a Ti plasmid. Agrobacteria were grown in ABmes medium (AB salts, 0.2% glucose, 25 mM phosphate, 25 mM morpholineethanesulfonic acid [MES] [pH 5.5]) for induction of the vir genes. P-Galactosidase enzyme activity was measured according to published procedures (40). Protein concentration was determined by the method of Bradford (Bio-Rad Laboratories).
Restriction and modification enzymes and deoxyoligonucleotide linkers were from New England Bio-Labs, Inc., and Pharmacia, Inc. Radioisotopes were from Amersham Corp.

RESULTS
Identification of the octopine Ti plasmid pTiA6 virG translation initiation codon. The DNA sequence of the virG locus of the A. tumefaciens Ti plasmid pTiA6 and the transcription initiation site for this virG mRNA have been determined (4,31,42). These analyses indicate that within the first 164 nucleotides of virG mRNA there are three commonly used translation initiation codons in the same open reading frame: two AUGs and a GUG at positions 48 to 50, 162 to 164, and 87 to 89, respectively (transcription start site, + 1 [4]). None of these potential start sites are preceded by sequences characteristic of a ribosome-binding site (27) normally found to precede bacterial translation initiation sites. There are two UUG codons in this open reading frame as well. UUGs are used rarely as translation initiation codons in procaryotes (7,15). The two UUGs in virG are located at positions 111 to 113 and 126 to 128. The second one is preceded by a putative ribosome-binding site sequence, AAGG, at positions 115 to 118. To overproduce the virG-encoded gene product (VirG) in E. coli, we considered it essential to identify the virG translation initiation codon. If the wrong translation initiation codon is assumed, an in vitro gene fusion could result in an N-terminal extension or deletion that could yield a functionally inactive product. The approach taken to identify the virG translation initiation codon was to modify all potential codons by insertional and site-directed mutagenesis and then assay for virG activity by measuring its ability to induce a chimeric virB-lacZ gene fusion in vivo in the presence of virA and AS.
To determine whether either of the two ATGs (for simplicity, the DNA sequence is referred to for discussion of the mutants) is used for virG translation initiation, insertion mutants were constructed. An  inserted after the first ATG (pGP209) and just before the second ATG (pGP210). The linker insertion would cause a frameshift mutation if translation initiates upstream of the insertion site. The results (Table 1) were as follows. Whereas plasmid pGP209 encoded a functional VirG, that encoded in plasmid pGP210 was functionally inactive. As expected, the wild type VirG produced in plasmid pGP159 was functionally active. This result indicated that neither of the two ATG codons is used for virG translation initiation, leaving the GTG codon as a prime candidate. To test this possibility, the GTG was changed to GTT (pGP157.1). This change maintained the valine codon in the polypeptide sequence and did not affect virG activity either (Table 1). Therefore, the GTG codon is not the sole translation initiation site for virG. We considered the possibility that either the first ATG or the GTG codon could function as the translation initiation codon. To test this, we constructed a double mutant, plasmid pGP165, that contains the GTG -* GTT mutation and an XbaI linker insertion blocking the first ATG. The VirG encoded by this plasmid was functionally active, indicating that neither the ATG nor the GTG serves as virG translation initiation site (Table 1).
Having ruled out the ATGs and GTG as virG translation initiation codons, we examined the rare initiation codons (TTGs). When the first TTG was changed to TTT (pGP179), no effect on virG activity was observed ( Table 1). The same mutation at the second TTG codon (pGP176), however, completely abolished virG activity. This indicated that the second TTG at positions 126 to 128 is the virG translation initiation codon. The TTG -* TTT mutation causes a Leu --Phe codon change. Therefore, the possibility remained that loss of virG activity in pGP176 may be due to this point mutation in a protein initiating translation upstream. To test this, we changed the second TTG to a TTA, maintaining this as a leucine codon (pGP190). This change greatly attenuated virG activity, confirming our earlier deduction that this codon is the virG translation initiation codon (Table 1). We then introduced a TTG-to-ATG mutation to study its effect on virG activity and for overproduction of VirG. The TTG -> ATG change (pGP196.1) had little effect on virG activity (Table 1); however it was not induced as highly as was the wild type.
Analysis of the effect of multiple mutations on virG activity provided additional support for our conclusion that the second TTG is the virG translation initiation codon. Plasmid pGP177 has all potential translation initiation codons blocked except for the second TTG codon. Agrobacteria containing pGP177 exhibited normal induced levels of virG activity (Table 1). When this TTG was modified to TTT, as in pGP187, virG activity was almost completely lost. Changing the TTG to an ATG (pGP197) had little effect on virG activity.
Nopaline Ti plasmid pTiC58 virG initiates translation at a UUG codon. virG genes from pTiA6 and pTiC58 are highly homologous both in DNA sequence and in polypeptide sequence; each can complement a mutation in the other gene (21). There are some differences, however. The first potential translation initiation site of pTiC58 virG, an ATG codon, is located one codon downstream of the GTG of the pTiA6 virG. The GTG codon is not conserved, but both of the TTG codons (see above) are conserved. In a study analogous to that which identified the pTiA6 virG translation initiation site, we sought to determine whether the second TTG of pTiC58 virG functions as its translation initiation codon. To check this, the same oligonucleotides that were used to mutagenize the pTiA6 virG were used to introduce mutations at the second TTG codon of pTiC58 virG. These oligonucleotides also introduced an A-to-G change two codons upstream of the second TTG (Table 2). This additional mutation did not change the primary sequence of the encoded polypeptide. Mutations that changed the second TTG either to a TTT (Leu -> Phe) (pGP218) or to a TTA (Leu -* Leu) (pGP219) led to a complete loss of virG activity (Table 2).
Since the TTT --TTA change would not change the primary sequence of the encoded polypeptide, these data indicate that the virG gene of pTiC58 also initiates translation at the second TTG codon.
Activation of VirG is necessary for its transcriptional activator function. The current model of vir gene induction is that VirA in response to plant phenolic compounds modifies VirG to activate it. Activated VirG then becomes a transcriptional activator that turns on the expression of all vir genes, including itself (39,42). We wished to examine the role of AS in the modification of VirG without the complication presented by activated VirG inducing its own expression. To maintain a constant level of VirG in the cell in both the presence and absence of AS, we constructed a transcriptional fusion (plasmid pGP188) in which the pTiA6 virG structural gene was fused to the S. typhimurium tryptophan (trp) operon promoter. This promoter was expressed in agrobacteria at a low level in ABmes minimal medium ( Table  3) and was not induced by AS (data not shown). Thus, any AS-induced expression of virB in this strain will be due to the modification of VirG. A 13-fold increase in virB-lacZ expression was observed in the presence of AS in agrobacteria containing pGP188; cells containing plasmid pGP199, a plasmid containing ATG as the virG translation initiation codon instead of the wild-type TTG codon, showed a similar high level of induction (Table 3). However, the TTG --ATG change led to a 3-to 5-fold increase in both basal and induced levels of virG activity even though the net induction essentially remained the same. Thus, it appears that VirG can partially induce expression of the vir genes in the absence of modification, but for maximal expression, VirG modification is essential.
These constructs also provide additional support for our conclusion that the second TTG is the virG translation initiation site. Only 16 bp upstream of the second TTG codon of the pTiA6 virG gene are present in these fusion genes and as such lack the first ATG, GTG, and TTG codons. They are fused to the trp promoter in a manner such that there is no upstream ATG or GTG codons in frame. There is one TTG trinucleotide in frame in the trp sequences, but it is not preceded by a ribosome-binding site sequence and therefore would not be expected to serve as a translation initiation site.
Overexpression and purification of VirG. E. coli GP202 containing a trp-virG transcriptional fusion was used for VirG overproduction. VirG produced in this strain was sequestered into inclusion bodies and upon cell lysis was pelleted with the membrane fraction (Fig. 1, lane 1). To confirm that this protein was VirG, we constructed an in-frame deletion of 93 bp within the virG structural gene. The new plasmid, pGP208, did not synthesize the 27-kilodalton VirG protein but produced a new protein of 23 kilodaltons, as expected (Fig. 1, lane 2). For purification, the cell membrane fraction containing VirG was solubilized with SDS and electrophoresed on a preparative SDS-polyacrylamide gel. The VirG band was excised from the gel, and the protein was electroeluted. The protein obtained was estimated to be greater than 90% pure, with a small amount of contamination by OmpA (Fig. 1, lane 3). A yield of approximately 50% was obtained in these studies.
VirG is a sequence-specific DNA-binding protein. Genetic evidence indicates that VirG is a positive regulator of vir gene transcription. This property predicts that VirG may bind to the regulatory region sequences of the genes it controls. To study this activity in vitro, VirG binding to DNA restriction fragments containing promoter regulatory sequences of the inducible genes virA, virB, virC, virD, virE, and virG was examined by a gel retardation assay. Purified VirG was incubated with end-labeled DNA fragments, electrophoresed, and subjected to autoradiography. By comparing the levels of VirG required to retard a particular DNA fragment, the specificity of VirG binding could be assessed in this assay. VirG bound specifically to DNA restriction fragments containing the 5' nontranscribed region sequences of virA, virB, virC-virD, and virG (Fig. 2, lanes 3, 6, 10, and 19). It should be noted that the virC and virD operons are divergent from each other and have common 5' nontranscribed regions. The 315-bp fragment used in this study contains both virC and virD regulatory sequences. Consequently, whether VirG bound to virC, virD, or both sequences could not be determined. VirG appeared to bind virE sequences with much reduced specificity (lanes 13 to 16). In several other experiments using other restriction fragments, we did not observe specific binding to virE nontranscribed region sequences (data not shown). The low-level binding to virE sequences (lane 13 to 16) may be due to specific binding or nonspecific binding. Since the virE-containing fragment (lane 13, band e) was approximately three times longer than the vector sequences (band v), nonspecific binding would retard band e significantly more than it would retard band v. In comparison, the restriction fragment containing virA sequences (lane 1, band a) was slightly larger than the virE fragment (lane 13, band e). The virA fragment was completely retarded by 70 ng of VirG (lane 3, band a), whereas under similar conditions the virE fragment was only partially retarded (lane 15, band e). In all cases, we observed partial or complete retardation of other radiolabeled fragments at higher protein concentrations (lanes 4, 8, 12, 16, and 20). These results are in agreement with results previously obtained with other DNAbinding proteins. At a low VirG concentration (20 ng), almost complete retardation of the virC-virD-containing fragment was observed (lane 10, band c-d). Under similar conditions, only partial or no retardation of virA-, virB-, virE-, and virG-containing fragments was noted (lanes 2, 6, 14, and 18). This higher affinity of VirG for the virC-virD sequences may be due to the presence of the regulatory sequences of the two operons in the same restriction fragment or due to stronger binding to either or both sequences. The VirG used in these studies had a low level of OmpA contamination (Fig. 1, lane 3). To determine whether OmpA was contributing to DNA binding, we isolated VirG from an E. coli ompA strain and performed DNA-binding studies. No difference in DNA binding was observed in these studies (data not shown).
Gel retardation assays clearly showed that VirG bound specifically to DNA fragments containing the regulatory regions of virA, virB, virC, virD, and virG. To localize the VirG-binding sites on these fragments, a DNase I protection assay was used. End-labeled DNA fragments containing the regulatory regions of the virC-D and virG genes were incu- bated with VirG, treated with DNase I, denatured, and electrophoresed in a urea-polyacrylamide sequencing gel (Fig. 3A).
When the virC-D regulatory region sequences were used as a substrate, regions from -35 to -40 and from -50 to -59 were protected against DNase I digestion (Fig. 3A, lanes 3 to 5). Two additional bands at positions -22 and -25 were also protected. Similar studies using a fragment containing the virG regulatory sequences showed that regions from -16 to -27, from -48 to -60, and from -68 to -72 were protected (lanes 8 and 9). The DNA sequences of the 5' nontranscribed regions, the sequences protected by VirG against DNase I degradation, and the locations of the central core regions of the vir box sequences are shown in Fig. 3B.

DISCUSSION
DNA sequence analysis by Winans et al. (42) predicted that virG of pTiA6 could encode a polypeptide of 267 amino acids. This prediction was based on the assumption that the first AUG of the largest open reading frame is used for virG translation initiation. We have shown that a UUG codon, located 26 codons downstream of the first AUG codon, is used to initiate virG translation (Table 1). A UUG codon located in an analogous position is also used to initiate translation of the virG mRNA of pTiC58 ( other members of two-component regulatory systems such as OmpR, SfrA, and PhoB over their entire lengths (24,42). The UUG codon is known to function as a translation initiation codon in E. coli and other procaryotes (7,15). In several cases, it has been demonstrated that translation initiation at UUG is inefficient. Changing the native translational initiation codon UUG of E. coli adenylate cyclase to GUG or AUG led to a twoor sixfold increase in enzyme activity (22). Similarly, alteration of the UUG translation initiator of the E. coli ribosomal protein S20 to an AUG appears to increase translatability of the cognate mRNA (19). It is likely that agrobacteria use a similar mechanism to down-regulate virG expression by the use of an inefficient UUG translation initiation codon.
The studies of Rogowsky et al. (23) demonstrate that virG can function as a transcriptional activator in the absence of induction by AS. Our studies support these results (Table 3). These results also indicate that modification of VirG is an important part of the induction process. Induction with AS caused a 12-to 15-fold increase in virB expression when the level of virG transcription was kept constant by the use of a trp-virG fusion (Table 3). Since transcription from the trp promoter is not modulated by VirA and VirG, the level of virG mRNA, and therefore VirG protein, must remain the same in both the absence and presence of AS. The increased activity of VirG must therefore be due to a modification(s) of the protein.
Genetic evidence suggests that VirG is a transcriptional activator that requires VirA and AS for function. It is proposed that VirA in response to AS modifies VirG to change it from an inactive to an active state. Activated VirG would be expected to bind to DNA in the regulatory region and to interact with RNA polymerase. Using protein produced in E. coli, we examined the binding of VirG to inducible vir promoters. Because of the absence of VirA and AS in the overproducing E. coli strain, this protein would not be expected to be activated by modification. Unmodified VirG bound specifically to vir regulatory sequences ( Fig. 3  and 4). However, we did not detect specific binding to the virE regulatory region. This result is in agreement with the data of Powell et al. (21), who observed that VirG is a DNA-binding protein that binds nonspecifically to the virE regulatory region. The reason for the lack (or reduced level) of VirG binding to virE sequences is not apparent. vir gene induction studies indicate that virE is induced to a high level (30), yet VirG did not show significant binding to a restriction fragment containing the virE regulatory region sequences. Several plausible explanations can account for this. First, the VirG used in this study is probably in the unmodified form, although some modification by an E. coli homolog of VirA, such as EnvZ, cannot be ruled out. Modified VirG may have a higher affinity, greater specificity, or both. Second, the position of the vir box relative to the promoter (-10, -35 region) sequence is important for induction of at least one vir gene, virB (3). A combination of the location of vir box sequence and binding affinity for a vir box sequence(s) together probably control the induction of individual vir genes.
DNase I protection studies showed that the binding sites of VirG lie around the vir box sequences of virC-D and virG. Both virC and virG contain multiple vir box sequences (Fig.  3B). DNase I footprinting analysis showed that VirG does not bind to all of these sequences. One binding site of VirG mapped to the second of the five vir box sequences of virC. The second site (-36 to -41) mapped to sequences between the highly conserved hexameric region (dCAATTG) of two vir box sequences. VirG bound to two of the three vir box sequences (boxes 1 and 3) of its own upstream region. In a recent study, S. Winans (Cornell University, Ithaca, N.Y.) has reported that vir boxes 1 and 3 are required for virG induction and that vir box 2 is not (Fallen Leaf Lake Conference, Lake Tahoe, Calif., 1989).