A Comparative Analysis of Perturbations Caused by a Gene Knock-out, a Dominant Negative Allele, and a Set of Peptide Aptamers*S

The study of protein function mostly relies on perturbing regulatory networks by acting upon protein expression levels or using transdominant negative agents. Here we used the Escherichia coli global transcription regulator Fur (ferric uptake regulator) as a case study to compare the perturbations exerted by a gene knock-out, the expression of a dominant negative allele of a gene, and the expression of peptide aptamers that bind a gene product. These three perturbations caused phenotypes that differed quantitatively and qualitatively from one another. The Fur peptide aptamers inhibited the activity of their target to various extents and reduced the virulence of a pathogenic E. coli strain in Drosophila. A genome-wide transcriptome analysis revealed that the “penetrance” of a peptide aptamer was comparable to that of a dominant negative allele but lower than the penetrance of the gene knock-out. Our work shows that comparative analysis of phenotypic and transcriptome responses to different types of perturbation can help decipher complex regulatory networks that control various biological processes.

In those organisms that do not lend themselves to classical genetics, the study of molecular regulatory networks owes a lot to a methodology often referred to as reverse genetics. Reverse genetics encompasses various approaches aimed at inducing three major classes of targeted perturbations within regulatory networks. The first consists of affecting protein expression levels by overexpressing gene products or by inhibiting their expression either partially (antisense technology and RNA interference) or totally (gene knock-out). The second consists of forming inactive protein complexes by use of dominant negative alleles, which usually titrate interacting proteins away from endogenous wild type proteins (1). The third consists of inhibiting protein function or protein interactions by use of specific ligands such as antibodies (2), nucleic acid aptamers (3), or small molecule inhibitors when available (4).
Peptide aptamers add to this arsenal of protein ligands (5). They are man-made combinatorial protein reagents that bind target proteins and can interfere with their function in living cells and organisms. They consist of conformationally constrained random sequence peptide loops displayed by a scaffold protein. Peptide aptamers can be selected by yeast twohybrid methods for their ability to interact with a given protein.
They can be selected also for their ability to confer a given phenotype to a cellular model. Typically they bind target proteins with a high specificity that allows them to discriminate between closely related members within a protein family or even between different allelic variants of a protein. Peptide aptamers have been selected against many proteins, and in every case, some aptamers have been shown to interfere with the function of their cognate target when expressed or introduced in cellular models or organisms (for reviews, see Refs. 6 -8).
Although all three above-mentioned classes of perturbation can provide important clues on protein function, they also present limitations that can produce misleading results. Gene knock-outs, which often produce either inconspicuous or dramatic phenotypes, do not always deliver direct information on protein function (9). Dominant negative alleles and inhibitory ligands can be prone to limited efficacy and/or lack of specificity.
Here we set out to compare the perturbations caused by peptide aptamers that bind a given protein, a dominant negative allele of the said protein, and the knock-out of its coding gene. We reasoned that an Escherichia coli global regulator would offer a convenient model to our endeavor. The ferric uptake regulator (Fur) 1 protein satisfied many criteria to serve as a good case study. Fur is an iron-dependent global regulator of gene expression in bacteria. In complex with Fe 2ϩ that activates the protein through a conformational change (10), Fur represses the transcription of target genes harboring a consensus operator sequence in their promoter (referred to as a Fur box) (11), although it can also positively regulate a smaller number of genes at a posttranscriptional level (12,13). About 100 genes from various E. coli strains have been shown to be regulated by Fur: most are involved in iron homeostasis; others act in metabolism and oxidative, nitrosative, and acidic stress responses (14 -17). Because fur deletion mutants of different pathogenic bacteria such as Vibrio cholerae (18) and Pseudomonas aeruginosa (19) show a decreased virulence, Fur is considered as a potential target for novel antibiotic therapies. Finally Fur has been extensively studied through the use of fur Ϫ strains, reporter genes, and fur mutants such as fur90 (H90Y) and fur51 (G51D), which were shown to negatively complement wild type fur (20). Although a total Fur inhibition achieved by gene knock-out is expected to affect the expression of all genes controlled by Fur, the variable levels of Fur inhibition expected from the use of a dominant negative allele or a set of peptide aptamers binding different surfaces of Fur with different affinities should produce finer perturbations.
We selected by a yeast two-hybrid method a set of four peptide aptamers that bind the Fur protein. We used different in vitro and in vivo phenotypic assays, and we performed a whole-genome transcriptome analysis to compare the perturbations caused by the Fur peptide aptamers, a fur dominant negative allele, and the fur gene knock-out.

E. coli Expression Plasmids
pBAD-Fur-We amplified the fur gene from E. coli genomic DNA using the oligonucleotides 5Ј-ATATGAATTCATGACTGATAACAATC-ACGC-3Ј and 5Ј-ATATCTCGAGTTATTTGCCTTCGTGCGCAT-3Ј that contained, respectively, an EcoRI and a XhoI site. We ligated the PCR product into EcoRI/XhoI-cut pBAD24, a plasmid that bears a pBR replication origin, the P BAD promoter of the arabinose operon, and its regulatory gene araC and that directs an arabinose-inducible expression of transgenes (21).
Two-hybrid Plasmids pHA3-We amplified the LexA cDNA of pEG202 using the oligonucleotides 5Ј-CCAAGCATACAATCAACTCCAAGCTTGAATTCCAT-GGGCTCGAGATGAAAGCGTTAACGGCCAGGC-3Ј that contained EcoRI, NcoI, and XhoI sites and 5Ј-CGGAATTAGCTTGGCTGCAGG-TCGACTTACAGCCAGTCGCCGTTGCGAATAACCCCAACCGCC-3Ј that contained a stop codon. We partially digested pEG202 with HindIII and XhoI and we introduced the amplified product into pEG202 by homologous recombination in yeast. The resulting plasmid, pHA3, bears a yeast 2 replication origin and a HIS3 marker and directs the expression of proteins whose carboxyl termini are fused to LexA.

Two-hybrid Selection of Fur-binding Peptide Aptamers
Library and strain construction is detailed in the Supplemental Experimental Procedures. We transformed 700 ml of MB210a yeast with 100 g of library to obtain 2.8 ϫ 10 7 transformants, and we transformed MB226␣ yeast with pEG202-Fur and pSH18-34. We performed a yeast-two hybrid selection essentially as described previously (25). We estimated the mating efficiency at 76% and the number of diploid exconjugants at 6 ϫ 10 8 . We plated 6 ϫ 10 7 diploids onto Ura Ϫ His Ϫ Trp Ϫ Leu Ϫ Ade Ϫ galactose/raffinose plates and incubated them for 7 days. We replica-plated onto Ura Ϫ His Ϫ Trp Ϫ X-gal galactose/raffinose plates. We picked five clones that grew in the absence of leucine and adenine and that displayed a ␤-galactosidase activity. Library plasmids were recovered and retransformed into EGY48␣. The interaction phenotypes were confirmed by a mating assay with EGY42a transformed with pEG202-Fur and pSH18-34 (23). We sequenced the aptamer genes and observed that we selected four different peptide aptamers.

In Vitro Binding Assay
We transformed BL21(DE3) E. coli with pGEX4T1-aptamer plasmids. We induced the expression of the GST fusions with 1 mM isopropyl 1-thio-␤-D-galactopyranoside for 3 h. We collected the bacteria and resuspended them into a lysis buffer (50 mM Tris, pH 8, 0.1 M NaCl, 1 mM EDTA, 5 mM MgCl 2 ) containing 1 mg/ml lysozyme. We froze and thawed the suspensions three times and sonicated them on ice. We centrifuged the lysates at 13,000 ϫ g for 30 min, collected the soluble fractions, and analyzed aliquots by SDS-PAGE to verify that the respective amounts of GST-aptamer fusions were similar. We thus immobilized equal amounts of GST-aptamer fusions onto 100 l of glutathione-Sepharose 4B (GE Healthcare) at room temperature for 1 h. We washed the beads with 10 ml of lysis buffer, and we eluted the bound proteins with lysis buffer containing 1.5 M NaCl. We loaded the eluates for SDS-PAGE, and we performed a Western blot analysis using an anti-Fur rabbit polyclonal antibody (raised against purified dimeric Fur protein, 1:1000 dilution). We revealed the blot using a horseradish peroxidase-linked rabbit antiserum and an ECL kit (Pierce).

Quantitative Yeast Two-hybrid Assay
We co-transformed EGY42a with pHB2-luc and pEG202-Fur. We transformed TB50␣ with pWP2C plasmids directing the expression of aptamer-B112 fusion proteins or with pJG4-5 plasmids directing the expression of B42-Fur wild type or B42-Fur90 fusion proteins. We mated the transformants and selected the diploid exconjugants. We performed the luciferase yeast two-hybrid assays as described 2 24 h after addition of galactose.

Drop-arrayed Yeast Two-hybrid Assay
We co-transformed TB50␣ with pSH18-34 and pJG4-5-Fur or pWP2C-aptamer plasmids. We transformed EGY42a with the different pEG202 bait plasmids. We mated the transformants on YPD (yeast peptone, bacto-peptone, and dextrose) solid medium, and we selected the diploid exconjugants by replica-plating onto Ura Ϫ His Ϫ Trp Ϫ glucose solid medium. We picked and resuspended the diploids into 50 l of sterile water. We deposited 3 l of each diploid onto a very dry Ura Ϫ His Ϫ Trp Ϫ X-gal galactose/raffinose plate.

Yeast Two-hybrid Competition Assay
We co-transformed TB50a with pBC104-aptamer or -Fur expression vectors and with pHB2-luc. We co-transformed TB50␣ with pEG202-Fur and pJG4-5-Fur. We mated the transformed strains on rich solid medium for 6 h, and we selected diploid exconjugants by replica-plating onto Ura Ϫ His Ϫ Trp Ϫ Leu Ϫ glucose medium. We performed the luciferase yeast two-hybrid assay as described 2 5 h after addition of galactose.

Growth Assay on High Concentration of Mn 2ϩ
We grew XL-1 blue transformants in TN medium containing 0.4% glycerol. We added 0.2% arabinose to half of the culture and grew it for another 1 h. We adjusted the A 600 to 0.6, and we performed serial dilutions of the cultures. We spotted 5 l of each dilution onto 0.4% glycerol, 100 g/ml ampicillin, 10 mM MnCl 2 , 2 mM MgCl 2 , and 50 M deferoxamine mesylate (Desferal) selection medium with or without 0.2% arabinose, and we observed the plates after 3 days.

Fur Transcriptional Repression Assay
We transformed QC6009 and QC6008 fur Ϫ bacteria with pBAD24 to confer ampicillin resistance. We transformed QC2949 and QC2146 with pBAD plasmids directing the expression of Fur90, aptamer cl20, and the Fur aptamers. We selected the transformants on TN medium (bacto tryptone, NaCl) supplemented with 0.4% glycerol, 2 mM MgSO 4 , and 100 g/ml ampicillin, and we streaked them onto the same medium. We replica-plated the streaks onto the same medium supplemented with 1 mM Fe(NH4) 2 (SO4) 2 and 1 mM X-gal with or without 0.2% arabinose. We observed the plates after 15 h.

Oxidative Stress Assay
We grew overnight cultures of E. coli QC2949 or 6009 transformants, and we adjusted the A 600 to 0.08 with TN medium supplemented with 0.4% glycerol. After 2 h, we added 0.2% arabinose, and after 30 min we added 2 mM H 2 O 2 to the cultures. We grew the cultures for 15 h, and we measured the A 600.

Drosophila Infection
We grew wild type Canton S or Tak1 [2] mutant flies at 25°C. We grew E. coli 1106 transformants to exponential phase (A 600 ϭ 0.8 -1) in arabinose-containing medium. We pelleted and resuspended the bacteria into PBS to reach a theoretical A 600 of about 200. We pricked 30 adult flies (5-10 days old) in the upper part of the thorax with a thin needle previously dipped into the above described bacterial suspensions as described previously (26).

RNA Extraction
We transformed QC6009 (fur Ϫ ) bacteria with pBAD24 and QC2949 (fur ϩ ) bacteria with pBAD24 containing Fur90, cl20, and F1. We grew overnight cultures in TN medium with ampicillin; we used these cultures to inoculate fresh cultures with a starting A 600 ϭ 0.02. We grew the bacteria to reach an A 600 ϭ 0.1, added 0.2% arabinose and 1 mM (NH4) 2 Fe(SO4) 2 ⅐6H 2 O, and grew the bacteria again to reach an A 600 ϭ 0.3-0.5. We extracted the RNA using a Qiagen RNeasy Midi kit (catalog number 75142) according to the instructions. We quantified the extracted RNA by UV spectrophotometry, and we checked the quality of the samples by agarose gel electrophoresis. We precipitated and resuspended the RNA to reach a concentration of 10 g/l.

RNA Labeling and Hybridization
Following the instructions of the MWG array application guide, we performed a direct Cy3-dCTP (Amersham Biosciences) labeling of single strand cDNA from 50 g of RNA, and we purified the labeled cDNAs using a Qiagen PCR purification kit. We labeled two biological replicates for each experimental condition. We quantified the amounts of cDNAs and the label incorporation rates using a Nano-Drop UV spectrophotometer. We then evaporated and resuspended the samples in 300 l of hybridization buffer. We mounted wholegenome E. coli K12 V2 OciChips (Ocimum Biosolutions) in Agilent hybridization chambers, and we incubated them in a rotating oven for 20 h at 42°C. We washed the chips as described in the MWG application guide, and we scanned them with a Genepix scanner (Axon Instruments, Molecular Devices Corp.). We measured Cy3 at 570 nm, setting laser power at 100% and photomultiplier tube power at 65-75%.

Microarray Analysis
We quantified the TIF images using the Genepix pro 6.0 software (Axon Instruments) and an "adaptative circular feature" quantification method. We normalized the data by the quantile method ("between array" normalization) using limmaGUI (27) in the R Bioconductor software package (open source). The normalized Cy3 channel intensities were log-transformed, and a gene expression analysis of variance (GeneANOVA) was performed as described previously (28). We retained those genes that showed a variation of expression of at least 2-fold in at least one of the perturbing conditions (fur Ϫ , Fur90, and/or F1 as compared with cl20) associated with a p value Ͻ0.01. We conducted additional statistical analysis (principal component analysis and unsupervised hierarchical classification) using the ANOVA and the MeV (version 3) softwares (29). We performed hierarchical clustering using R Bioconductor as described previously (30).

Fur Box Predictions
We downloaded the E. coli K-12 MG12655 complete genome sequence (U00096.fna) and gene annotation (U00096.ptt) from Gen-Bank TM . We extracted synonymous gene names from the additional U00096.gbk GenBank file. We identified all sequences matching the 19-bp E. coli consensus Fur box (GATAATGATAATCATTATC) (31), with a tolerance of five, six, or seven mismatches, using the pattern search algorithm implemented in the ICM software (Molsoft). We associated Fur box(es) to a gene when this sequence was located between 1 and 250 bp upstream of the transcription start of the gene itself or of an upstream gene within the same operon. We considered genes to belong to the same operon when all three following criteria were satisfied: (i) same strand location, (ii) transcription starts less than 4 kb apart, and (iii) gene names sharing the same first three letters. When applicable, we "manually" attributed Fur box(es) to those few genes (among the 196 retained genes) that did not conform to the above mentioned criteria but were known to belong to an operon.
To establish the statistical significance of the data shown in Fig. 7, we used the hypergeometric distribution to compare the frequency of a category in a group with that expected, knowing the amount of the category in the population. We computed the probability of finding the observed amount or more genes belonging to the considered category in a group. When the probability was less than 0.01, the category was considered highly enriched in the group (32)(33)(34).

RESULTS
Selection of Fur-binding Peptide Aptamers-We used an optimized yeast two-hybrid method and a new generation library (variable regions of 13 amino acids) (25) to select peptide aptamers for their ability to bind Fur. Fur homodimerization offered the opportunity to validate the use of this protein in a yeast two-hybrid assay. We detected a strong interaction phenotype when Fur was expressed both as bait (in fusion to LexA) and as prey (in fusion to the B42 activation domain) (Fig. 1A). We selected four different Fur-interacting aptamers, named F1 to F4. The sequences of their variable regions did not reveal any similarity with E. coli proteins (Table  I). All four aptamers showed a moderate to strong lacZ interaction phenotype against LexA-Fur (the bait used for the selection) contrary to cl20 (a control aptamer randomly picked from the library) and to Mn1 and Mn2 (two aptamers obtained from a phenotypic selection and that do not interact with Fur 3 ). In contrast to F1, -2, and -3, F4 bound a Fur-LexA fusion but failed to bind noticeably two dominant negative alleles of Fur (Fur90 and Fur51) (Fig. 1A). Interestingly we detected a strong interaction phenotype between Fur90, Fur51, and Fur, thus confirming the hypothesis that both mutants form inactive heterodimers with the wild type protein (20).
To confirm that these aptamers bound Fur in E. coli, we performed pulldown experiments. We successfully captured the endogenous Fur protein from BL21(DE3) E. coli expressing GST-Fur aptamer fusions (Fig. 1B).
Quantification and Mapping of Interactions-To compare more accurately the apparent binding affinities of the peptide aptamers, Fur and Fur90, for Fur, we performed yeast twohybrid assays using a new luciferase reporter gene (luc) that allows an easy and precise quantification of two-hybrid phenotypes (8). The luc interaction phenotypes measured between Fur and the peptide aptamers revealed that the apparent binding affinities of F1, -2, and -4 are similar to one another and higher than that of F3 (Fig. 1C). These results are globally consistent with the lacZ interaction phenotypes and with the amounts of Fur protein captured using the GSTaptamer affinity matrices (Fig. 1, A and B). The luc interaction phenotype corresponding to Fur homodimerization is comparable to that of aptamers F1, -2, and -4 but lower than the luc interaction phenotype measured between Fur and Fur90 (Fig.  1C). The apparent discrepancy with the yeast two-hybrid assay shown in Fig. 1A can be explained by the rapid saturation of the blue colorations obtained with a lacZ reporter system, whose dynamic range is much more limited than that of the luc reporter system. These luciferase two-hybrid assays thus indicate that peptide aptamers F1, -2, and -4 bind to Fur with an affinity that is comparable to that of Fur itself, higher than that of aptamer F3, and lower than that of the Fur90 dominant negative protein.
The results shown in Fig. 1A suggest that peptide aptamer F4 binds to Fur on a molecular surface that is distinct from the molecular surface(s) bound by aptamers F1, -2, and -3. To map the aptamer binding sites on Fur, we constructed a collection of bait plasmids expressing different truncations of Fur, and we performed yeast two-hybrid mating assays. To validate the different bait constructs, we used Fur itself and RG22, a peptide aptamer that interacts with LexA and that usually gives an interaction phenotype with most but not all LexA fusion proteins (25). Every truncated Fur bait construct gave an interaction phenotype with either RG22 or Fur or with both for one of them (Fig. 1D). The interaction phenotypes obtained between the different truncations and Fur are consistent with the structural knowledge on Fur homodimerization (10). Peptide aptamer F4 binds a molecular surface that is located on the carboxyl-terminal half of Fur, whereas peptide aptamers F1, -2, and -3 probably bind (a) molecular surface(s) formed by the folding of Fur and that involve(s) residues located in the amino-and carboxyl-terminal halves of the protein (Fig. 1D). This (or these) molecular surface(s) does not exist or is corrupted in the context of a Fur-LexA fusion protein (Fig. 1A).
Inhibition of Fur Homodimerization and Function-Fur exerts its transcriptional repressor activity by binding DNA regulatory sequences as a homodimer. Therefore, an obvious peptide aptamer-mediated Fur inhibitory mechanism would be either to prevent Fur homodimers from binding DNA or to prevent Fur homodimerization. We explored this latter hypothesis by testing the ability of peptide aptamers to inhibit the Fur homodimerization yeast two-hybrid phenotype. We co-expressed HA-NLS-aptamer fusions together with LexA-Fur and NLS-B42-HA-Fur. We validated our competition assay by showing that an HA-NLS-Fur fusion was able to inhibit FIG. 1. Fur peptide aptamer interaction assays. A, yeast two-hybrid mating assay between LexA fusions to the indicated Fur proteins and B112-aptamer or B42-protein fusions. cl20, Mn1, and Mn2 are negative control aptamers. B, Fur pulldown assays. We ran bacterial protein extracts on the indicated GSTaptamer solid phases. We analyzed the eluates by Western blot using an anti-Fur antibody. C, quantitative yeast two-hybrid assay. We quantified yeast two-hybrid phenotypes between LexA-Fur and peptide aptamers, Fur, and Fur90 prey proteins using a luciferase reporter gene. We performed two independent experiments. D, mapping of peptide aptamer binding sites on Fur. We performed drop-arrayed yeast two-hybrid mating assays between the peptide aptamers and different Fur truncations fused to LexA (depicted linearly). RG22 is a LexA peptide aptamer that interacts with most but not all LexA fusion constructs. A.U., arbitrary units; wt, wild type.

Aptamers
Variable regions 80% of the Fur homodimerization two-hybrid phenotype ( Fig.  2A). All aptamers tested failed to interfere with the Fur homodimerization two-hybrid phenotype except for aptamer F4, which produced a 55% inhibition of the two-hybrid signal ( Fig.  2A), while not inhibiting an unrelated protein interaction phenotype (not shown). These results suggest that Fur homodimerization is inhibited by peptide aptamer F4 but not by the other three aptamers, which bind to (a) distinct molecular surface(s). We then examined the capacity of the Fur-binding aptamers and the dominant negative allele to phenocopy fur Ϫ bacteria in allowing growth on a high concentration of Mn 2ϩ (35). We expressed peptide aptamers and Fur90 using pBAD24, a plasmid that bears an arabinose-inducible promoter. Aptamers F1 and -2 were as potent as Fur90 in their ability to allow bacterial growth in these conditions, whereas aptamers F3 and -4 did not allow growth (Fig. 2B).
Next we used a transcriptional repression assay to compare Fur inhibition caused by the gene knock-out, the dominant negative allele, and peptide aptamers. We used the bacterial strains QC6008 (fur Ϫ ) and QC2146 (fur ϩ ) (from Daniè le Touati). Both strains carry an integrated lacZ reporter gene placed under the control of the fiu promoter that contains four Fur boxes and is thus strongly repressed by Fur (36). We transformed QC6008 with pBAD24 to obtain a control devoid of any Fur activity. We transformed QC2146 with pBAD24 plasmids directing the expression of Fur90 and the aptamers to be tested. Although aptamer F3 showed no effect, Fur90 and aptamers F1 and -2 inhibited Fur repression activity to a similar extent. Aptamer F4 caused a slightly weaker inhibition (Fig. 2C). We obtained similar results using the bacterial strains QC6009 (fur Ϫ ) and QC2949 (fur ϩ ) containing a lacZ reporter gene controlled by the fhuF promoter, which contains two Fur boxes (not shown). These results show that aptamers F1, -2, and -4 act as Fur inhibitors in this transcriptional repression activity.
Oxidative Stress Sensitivity-Fur regulates the expression of numerous genes involved in the oxidative stress response among which are superoxide dismutases that play a key role in the protection against oxygen toxicity (37). As fur Ϫ strains are very sensitive to hydrogen peroxide (H 2 O 2 ) (14), we wished to determine whether Fur90 or the Fur aptamers would confer sensitivity to H 2 O 2 in a growth assay performed in liquid cultures. Although we confirmed the extreme sensitivity of the fur Ϫ strain, neither Fur90 nor the tested aptamers conferred any significant sensitivity to H 2 O 2 in this assay (Fig. 2D).
Virulence Assay in Animals-Fur has been shown to play an important role in the virulence of different pathogenic bacteria (18,38). We thus set out to determine whether Fur90 and the Fur-binding peptide aptamers affected the virulence of a pathogenic E. coli strain in an in vivo model system. It is now well established that Drosophila and mammals share conserved immune mechanisms, including the activation of NF-B-dependent signaling pathways (39). In addition, several FIG. 2. Fur homodimerization and functional assays. A, yeast two-hybrid competition assay. We co-expressed in yeast LexA-Fur, NLS-B42-HA-Fur, and different HA-NLS-aptamer or -Fur fusions. We quantified the Fur dimerization two-hybrid phenotypes using a luciferase reporter gene, and we normalized the results with the value obtained with aptamer cl20. We performed three independent experiments. B, bacterial growth on medium containing a high concentration of Mn 2ϩ . We spotted serial dilutions of E. coli XL-1 blue transformants expressing peptide aptamers or Fur90 onto TN medium containing 10 mM Mn 2ϩ and arabinose. C, Fur transcriptional repression assay. We transformed QC6008 (fur Ϫ ) strain with pBAD24 to confer ampicillin resistance and QC2146 (fur ϩ ) strain with pBAD24 expressing peptide aptamers, Fur90, or nothing. We streaked the transformants onto solid medium containing X-gal with or without arabinose. D, oxidative stress assay. We measured the growth of E. coli QC2949 (fur ϩ ) transformants expressing various aptamers or Fur90 and the growth of QC6009 (fur Ϫ ) transformed with pBAD24 in the presence or in the absence of H 2 O 2 . We performed four independent experiments. bacterial pathogens use similar virulence mechanisms against mammalian and non-mammalian hosts, including insects (26,40). We infected Drosophila flies with the E. coli 1106 pathogenic strain transformed with pBAD plasmids expressing Fur90 and the peptide aptamers to be tested. Because the 1106 strain did not kill wild type flies (not shown), we used TAK1 (transforming growth factor-␤-activated kinase 1) immunodeficient Drosophila mutants that are strongly impaired for the activation of NF-B-dependent antimicrobial peptide synthesis (41). In these flies, lethality occurred 3-4 days following septic thorax injury. Fur90-and Fur aptamer-expressing bacteria caused a slower mortality than control bacteria (Fig. 3). In the absence of sustained promoter induction by arabinose, the differential growth between the transformants eventually vanished, and all flies died (not shown). Thus, the transient expression of Fur90 and of the tested Fur aptamers decreased the virulence of pathogenic E. coli in flies 3-4 days after infection.
Genome-wide Transcriptome Analysis-To perform a comprehensive comparative analysis of the molecular perturbations exerted by the gene knock-out, the expression of a dominant negative allele, and the expression of a peptide aptamer, we carried out a global transcriptome study using E. coli pan-genomic oligonucleotide microarrays and RNA samples extracted from pBAD24-transformed QC6009 (fur Ϫ ) and from QC2949 (fur ϩ ) transformed with pBAD24 plasmids directing the expression of Fur90 and peptide aptamers cl20 and F1. We performed two independent experiments, and we retained those genes that showed a variation of expression of at least 2-fold in at least one of the perturbing conditions (fur Ϫ , Fur90, and/or F1) as compared with the cl20 condition and that passed a significance cutoff determined by an ANOVA (see "Experimental Procedures"). We thus retained 196 genes for further analysis ( Fig. 4 and Supplemental Table S1). Three genes provided built-in controls for these experiments. lacZ (Fur reporter gene integrated in the genome of both strains), trxA (peptide aptamer scaffold), and fur showed expected expression profiles (Fig. 5A), thus confirming the identity of the samples, the proper expression of the transgenes, and the strong Fur basal activity under our experimental conditions. 75 of the 196 selected genes have already been identified in previous transcriptome studies as Fur-and/or iron-regulated genes in E. coli (17,42), Helicobacter pylori (38,(43)(44)(45), Neisseria meningitidis (46,47), Pseudomonas aeruginosa (48,49), Campylobacter jejuni (50), Shewanella oneidensis (51,52), and Pasteurella multicoda (53). Conversely we unveiled 121 genes that have not been reported by previous studies. The expression of 106 and 78 genes was up-and down-regulated in at least one of the perturbing conditions, respectively (Fig.  4, A and B). Within the up-and down-regulated gene populations, we defined clusters according to the perturbing condition(s) that caused at least a 2-fold deregulation of expression (Fig. 4). The vast majority of the up-and down-regulated gene populations belonged either to the fur Ϫ -only (group 1) or to the fur Ϫ /Fur90/F1 (group 2) clusters, whose existence can be easily explained by straightforward biological hypotheses (summarized in Fig. 6A). A few typical examples of gene expression profiles from these two groups are shown in Fig. 5, B (up-regulations) and C (down-regulations).
We analyzed the general profiles of the up-and downregulation values in all three perturbing conditions. The downregulation values were generally lower than the up-regulation values, especially in the fur Ϫ condition. The down-regulation values were comparable in all three perturbing conditions. In contrast, the up-regulation values were comparable in the Fur90 and the F1 conditions and lower than that observed in the fur Ϫ condition (Fig. 4 and Supplemental Table S1).
We next investigated whether the predicted sensitivity of a gene to a regulation by Fur determined its belonging to the different groups. The predicted sensitivity of a Fur-controlled promoter to Fur can be approximated from the number of Fur boxes detected in the promoter sequence. We first applied different Fur box prediction settings to the entire E. coli genome to determine those settings that yielded the highest Fur box enrichment rates within the population of the 196 retained genes (see "Experimental Procedures" and Supplemental Fig.  S1). From these results, we retained the prediction setting that considered a 250-bp region upstream of the transcription start of each gene, tolerating up to five or six mismatches from the Fur box sequence consensus (31). As shown in Fig. 7, we observed a striking correlation between the number of predicted Fur boxes and the percentage of fur Ϫ -only genes. Among the 196 retained genes, the vast majority of those genes harboring two or more Fur predicted boxes (with a tolerance of five mismatches) were deregulated by the gene knock-out only. Examples of genes deregulated by the gene knock-out only include entA-F and fepA-G, which play a key role in iron homeostasis and whose expression is known to be tightly regulated by iron in a fur-dependent manner. Conversely genes such as exbB and exbD, which present only one predicted Fur box and which are involved in the uptake of various extracellular molecules (including iron), were found to be deregulated in all three perturbing conditions. DISCUSSION Using E. coli Fur as a case study, we compared the perturbations exerted by gene disruption, expression of a dominant   FIG. 3. Virulence assay. The percentage of surviving flies following septic injury was determined in 30 infected individuals 3 and 3.5 days post-infection in the first and second experiment, respectively. cl20 aptamer was tested in the first experiment only.
negative allele, and expression of a set of peptide aptamers, performing four functional assays and a whole-genome tran-scriptome study. The four selected aptamers show different apparent binding affinities for Fur and bind to different molecular surfaces. F4 binds to the carboxyl-terminal half of Fur, which includes the dimerization domain. This is thus consistent with the ability of F4 to inhibit Fur homodimerization. F1, -2, and -3 probably bind to (a) molecular surface(s) involving residues located in the amino-and carboxyl-terminal halves of the protein. Recent structural studies on Fur homodimers have revealed ionic interactions between carboxylates of the carboxyl-terminal part of one subunit and the NH 2 terminus of the other subunit (54). The metal-dependent activation mechanism of Fur has been shown to involve the disruption of this ionic interaction together with the folding of the amino-terminal helix (10,54). This suggests that aptamers F1, -2, and -3 bind Fur homodimers and may inhibit their metal-dependent activation mechanism.
F1 and F2 allowed growth on a high concentration of Mn 2ϩ and strongly inhibited Fur transcriptional repression activity, thereby phenocopying a fur Ϫ strain or a strain expressing Fur90. F4 did not allow growth on a high concentration of Mn 2ϩ and inhibited Fur repressor activity to a lesser extent than did F1 or F2. None of the tested aptamers were able to confer a detectable sensitivity to H 2 0 2 . F3, the weakest binder, failed to induce a detectable perturbation in all phenotypic assays but the virulence test in a Drosophila model of infection. The results of this assay suggest that a sustained expression of Fur peptide aptamers could stably reduce animal mortality caused by an infection with pathogenic bacteria. Hence they further validate Fur as an interesting target to pursue for antibacterial drug discovery, which could be guided by the use of the Fur peptide aptamers (8). FIG. 4. RNA expression data. Heat maps comparing the normalized log 2 of the ratios of Fur Ϫ , F90, and F1 signals on the cl20 control signal. Clustering was performed using a Manhattan distance and a Ward method of agglomeration. Colors vary from green for the lowest ratios to red for the highest ratios. A, up-regulated genes; B, down-regulated genes; C, unclassifiable and control genes. Gr1, "fur Ϫ -only" genes; Gr2, fur Ϫ / F90/F1 genes; Gr3, fur Ϫ /F90 genes; Gr4, fur Ϫ /F1 genes; Gr5, F90/F1 genes; Gr6, "F90-only" genes; Gr7, "F1-only" genes; Gr8, unclassifiable genes; Cont, control genes. Altogether these observations demonstrate that the use of a set of peptide aptamers that bind their target protein on different molecular surfaces with different affinities induces a broad range of perturbations of protein function. As shown recently in another study, peptide aptamers can also activate rather than inhibit the function of their target protein (55).
Our transcriptome study has revealed 121 genes that have not yet been reported as being regulated by Fur and/or iron in E. coli or other bacteria. Although it is possible that our selection cutoff was not stringent enough, we believe that a high confidence can be attributed to the genes that satisfy either of the following criteria: (i) function showing a conspicuous link to iron metabolism; (ii) co-involvement in a given regulatory pathway (for example, wzb and wzc, which code for a proteintyrosine phosphatase and kinase, respectively; nac and glnk, which are both involved in the regulation of nitrogen assimilation; pyrB and pyrI, which code for the aspartate carbamoyltransferase catalytic and regulatory subunits, respectively); (iii) presence of predicted or experimentally revealed Fur boxes in the promoter (36 of 121 newly described genes); or (iv) deregulation observed in two or three of the perturbing conditions used in our study. This latter criterion (satisfied by 71 of 121 newly described genes) provides a particularly high confidence as it rules out artifactual deregulations that may be caused by compensatory mechanisms (triggered by gene knock-out or overexpression) or off-target peptide aptamer effects. In total, 88 of the 121 newly described genes satisfy at least one of the above mentioned criteria and can thus be considered with a good confidence as genes regulated di- In this diagram, the nine unattributed genes, which showed opposite deregulation values according to the perturbing condition, were attributed to the various groups (marked by an asterisk). Hence some numbers differ from the numbers given in A.

FIG. 7.
Occurrence of fur ؊ -only genes and expected sensitivity to Fur. From the 196 retained genes, we took into consideration the 165 genes that were found to be deregulated at least in the fur Ϫ condition, and we split them into four groups according to the number of Fur boxes that were predicted in their promoters. For each group, we calculated the percentage of fur Ϫ -only genes. Enrichment values labeled with an asterisk are statistically significant (see "Experimental Procedures"). mis., mismatches. rectly or indirectly by Fur (Supplemental Table S1).
Not surprisingly, the perturbation that deregulated the highest number of genes was the gene knock-out, which generally induced more pronounced up-regulations than the expression of the dominant negative allele or of the peptide aptamer. In contrast, the down-regulation values, which were generally lower than the up-regulation values, were similar between all three perturbing conditions. The lower down-regulation values (as compared with up-regulation values) could be explained by the fact that Fur can exert a positive control on gene expression through an indirect, small RNA-based mechanism (13). However, Fur box(es) are predicted for 24 of the 70 genes that were down-regulated at least in the fur Ϫ condition (Supplemental Table S1). This suggests that in E. coli Fur might activate the transcription of a subset of genes also through a direct mechanism as already described in H. pylori (56).
Altogether these results show that the "penetrance" of the dominant negative allele and of the peptide aptamer is lower than the penetrance of the gene knock-out. This conclusion is further supported by the observation that the percentage of genes deregulated only by the gene knock-out increases with the predicted sensitivity of the gene promoters to Fur. A key determinant of the penetrance of an inhibitory ligand (dominant negative protein or peptide aptamer) is the stoichiometry of target molecules in the free state and in complex with the ligand. This stoichiometry is itself determined by the expression level and the binding affinity of the ligands. Here the lower penetrance of the dominant negative allele and of the peptide aptamer could be due to insufficient expression levels and binding affinities. However, the mRNA expression levels of the peptide aptamers and of Fur90 are very high and, for the latter at least, largely exceed the expression level of the target (Fig. 5A). Moreover the apparent binding affinity of Fur90 for Fur is very strong and higher than that of Fur itself (Fig. 1C). Therefore, although it might be possible to increase the penetrance of the peptide aptamer by conducting an in vitro evolution approach and obtaining a higher affinity mutant (24), this penetrance will probably not exceed that of Fur90 and will thus remain lower than that of the gene knock-out.
Another key determinant of the penetrance of an inhibitory ligand lies in its capacity of inhibiting every function of its target protein when the protein exerts multiple functions. In addition to its transcription regulation activity related to its capacity to bind DNA, Fur has been suspected to exert an iron sequestration function (57). If this hypothesis were correct, the lower penetrance of the peptide aptamers and of Fur90 could also be due to the fact that they can only inhibit the DNA binding-dependent activities of Fur but not its alternative function. More work will be needed to clarify this point.
Relating the phenotypes observed in the functional assays to the results of the transcriptome analysis should be particularly useful to identify Fur-controlled genes that govern various Fur-regulated processes. Those genes whose deregulation allows bacterial growth on a high concentration of manganese or inhibits bacterial virulence in flies should be mostly found among the genes that are up-or down-regulated by all three perturbing conditions. The genes that tend to confer protection to oxidative stress should be mostly found among the genes that are down-regulated by fur Ϫ only, whereas the genes that tend to confer sensitivity to oxidative stress should be mostly found among the genes that are down-regulated by Fur90 and F1 and up-regulated by fur Ϫ only (Fig. 6B). Many genes known or predicted to determine oxidative stress sensitivity conform to this prediction. Most genes involved in iron uptake (ent, fep, fec, and fhu genes) are up-regulated by fur Ϫ only. Increased intracellular iron levels are known to confer a high sensitivity to oxidative stress. The group of fur Ϫ -only, down-regulated genes includes katE (catalase that eliminates H 2 O 2 ), osmC (an osmotically induced protein that uses highly reactive cysteine thiol groups to elicit hydroperoxide reduction (58)), yggE (a protein that restores physiological defects caused by oxidative stress (59)), phoB (a positive regulator of polyphosphate accumulation, which confers resistance to oxidative damages (60)), ileS (isoleucyl-tRNA synthetase, down-regulation of which may cause a partial starvation for isoleucine, which has been shown to cause a repression of genes involved in oxidative stress protection (61)), and of course fur itself, whose gene product may exert a protective role by sequestering ferrous iron (57). The group of genes that are down-regulated by Fur90 and F1 (and slightly up-regulated in the fur Ϫ condition) includes codA and codB (cytosine deaminase and permease, respectively), which should increase oxidative mutagenesis and thus sensitivity to oxidative stress (62).
In conclusion, our work establishes that the use of peptide aptamers or of a dominant negative allele induces more limited phenotypic responses than the use of a gene knock-out. This can be accounted for by the incomplete penetrance of the transdominant agents compared with the gene knock-out as revealed by the transcriptome analysis. Our work also shows that a comparative analysis of the phenotypic and transcriptome responses to different types of perturbation can help identify regulatory network members that govern various biological processes.