Genome-Scale Analyses of Escherichia coli and Salmonella enterica AraC Reveal Noncanonical Targets and an Expanded Core Regulon
Escherichia coli AraC is a well-described transcription activator of genes involved in arabinose metabolism . 
Using complementary genomic approaches , chromatin immunoprecipitation ( ChIP ) - chip , and transcription profiling , we identify direct regulatory targets of AraC , including five novel target genes : ytfQ , ydeN , ydeM , ygeA , and polB . 
Strikingly , only ytfQ has an established connection to arabinose metabolism , suggesting that AraC has a broader function than previously described . 
We demonstrate arabinose-dependent repression of ydeNM by AraC , in contrast to the well-described arabinose-dependent activation of other target genes . 
We also demonstrate unexpected read-through of transcription at the Rho-independent terminators downstream of araD and araE , leading to significant increases in the expression of polB and ygeA , respectively . 
AraC is highly conserved in the related species Salmonella enterica . 
We use ChIP sequencing ( ChIP-seq ) and RNA sequencing ( RNA-seq ) to map the AraC regulon in S. enterica . 
A comparison of the E. coli and S. enterica AraC regulons , coupled with a bioinformatic analysis of other related species , reveals a conserved regulatory network across the family Enterobacteriaceae comprised of 10 genes associated with arabinose transport and metabolism . 
Escherichia coli AraC is the founding member of a large family of transcription factors ( TFs ) found across a wide range of bacterial species ( 1 ) . 
AraC was first identified in 1959 by virtue of the requirement of araC for the metabolism of L-arabinose ( 2 ) and is the first-described positive regulator of transcription ( 3 , 4 ) . 
E. coli AraC activates transcription of the araBAD , araFGH , araE , and araJ transcripts in the presence of its inducer , L-arabinose ( 5 ) . 
AraC binds DNA as a dimer . 
Dimerization occurs between adjacent DNA sites when AraC binds arabinose . 
In the absence of arabinose , AraC represses transcription of araBAD and araC by forming a repression loop mediated by dimerization of distally bound AraC monomers ( 5 , 6 ) . 
Chromatin immunoprecipitation ( ChIP ) - chip and ChIP sequencing ( ChIP-seq ) are widely used techniques for genome-wide mapping of protein-DNA interactions in vivo . 
Surprisingly , these methods have been used only sparingly to study bacterial systems ( 7 ) . 
ChIP-chip and ChIP-seq studies of bacterial TFs have identified novel regulatory interactions , even for well-studied proteins ( 7 , 8 ) . 
Furthermore , TF binding sites have been identified in unexpected locations , such as inside genes ( 9 ) , upstream of genes that are not detectably regulated by the TF , and in genomic regions that lack a canonical DNA sequence motif for the TF ( 7 , 10 ) . 
Transcription profiling uses microarrays or RNA-seq to determine differences in genome-wide RNA levels between two growth conditions and/or strains ( 11 ) . 
This approach is often used to identify regulatory targets of TFs by comparing RNA levels in wild-type cells and cells lacking the TF . 
In contrast to ChIP methods , transcription profiling identifies all genes regulated by a TF and the level and direction of regulation . 
However , transcription profiling identifies both direct and indirect regulatory targets . 
By combining ChIP methods and transcription profiling , it is possible to identify all direct regulatory targets of a TF for a given growth condition ( 11 ) . 
We refer to the set of direct regulatory targets as a regulon . 
Many TFs , including AraC , are highly conserved between E. coli and other species in the family Enterobacteriaceae . 
This suggests that DNA-binding specificity is the same for TF homologues across the family , and that TF regulon gene function is likely to be conserved . 
Most studies of regulon evolution have focused simply on whether regulon members ( i.e. , target genes ) have homologues in related species . 
In contrast , very few studies have determined whether conserved genes are regulated by the TF ( 12 ) . 
The beststudied TF in this regard is PhoP , a two-component regulator that is conserved across the family Enterobacteriaceae . 
Regulation of only three PhoP target genes is conserved across the family , although in any given species there are many more than three PhoP-regulated genes ( 13 ) . 
Most PhoP target genes in any given species lack homologues in other species or the genes are conserved but are only regulated by PhoP in one or two species . 
The latter phenomenon is known as network rewiring ( 12 ) . 
Most of the known AraC regulon members in E. coli are conserved across other Enterobacteriaceae members , but the extent of rewiring is unknown . 
Given that much of our understanding of regulon evolution is based on studies of a single TF , PhoP , it is important to experimentally compare regulons for additional TFs between related species ( 12 ) . 
Genome-scale approaches have not been previously used to identify AraC-regulated genes . 
We hypothesized that despite the extensive prior work on the AraC regulon , there are likely to be previously undescribed AraC-regulated genes and novel modes of regulation by AraC . 
In this work , we use a combination of ChIP-chip and transcription profiling with microarrays to identify all binding sites and all direct regulatory targets of E. coli AraC . 
In addition to identifying a novel mechanism of repression by AraC , our genomic approach reveals unexpected read-through of transcription terminators in AraC-activated transcripts and AraC-regulated genes with no connection to arabinose metabolism . 
We also identify all binding sites and all direct regulatory targets of AraC in the related species Salmonella enterica using a combination of ChIP-seq and RNA-seq . 
These targets include two novel , cotranscribed , AraC-activated genes ( STM14_0178 and STM14_0177 ) that encode a putative arabinoside transporter and an - L-arabinofuranosidase II precursor . 
We rename these genes araT and araU . 
Together with a bioinformatic analysis of other Enterobacteriaceae species , these data identify a conserved AraC regulon that includes 7 previously described AraC-regulated genes ( araB , araA , araD , araE , araF , araG , and araH ) as well as three novel targets identified in this work ( ytfQ , araT , and araU ) . 
Moreover , our data indicate only limited rewiring of the AraC regulatory network in the Enterobacteriaceae . 
MATERIALS AND METHODS
Strains and plasmids . 
Bacterial strains and plasmids used in this work are listed in Table 1 . 
Cells were grown in LB ( 1 % NaCl , 1 % tryptone , 0.5 % yeast extract ) . 
All oligonucleotides used in this work are listed in Table S1 in the supplemental material . 
AMD054 was constructed using Red recombineering as described previously ( 14 ) . 
The PCR product used for recombineering was generated with oligonucleotides JW464 and JW465 , using pKD13 ( 14 ) as the template . 
SAC003 ( MG1655 araC-TAP ) was constructed by P1 transduction of the kanamycin resistance ( Kanr ) gene-linked araC-TAP from DY330 araC-TAP ( 15 ) . 
The Kanr gene was removed using pCP20 as described previously ( 16 ) . 
SAC001 ( MG1655 araC ) and AMD115 were constructed by P1 transduction of the Kanrlinked araC from BW25113 araC ( 17 ) into MG1655 and AMD054 , respectively . 
The Kanr gene was removed using pCP20 as described previously ( 16 ) . 
Note that SAC001 and AMD115 also contain the ( araD-araB ) 567 mutation that lacks the araBAD operon . 
AMD187 ( E. coli MG1655 araC-3 FLAG ) , JTW010 ( E. coli MG1655 with ytfQ AraC site mutation , araC-3 FLAG ) , and CB005 ( S. enterica serovar Typhimurium 14028s araC-3 FLAG ) were constructed using the FRUIT recombineering system ( 18 ) . 
The PCR product used to generate the initial tagged strains was made using oligonucleotides JW1141 and JW1142 for E. coli and JW2895 and JW2901 for S. enterica , with pAMD135 as the template . 
For construction of JTW010 , the thyA-containing PCR product for insertion upstream of ytfQ was amplified with oligonucleotides JW601 and JW602 using pAMD001 as the template . 
The PCR product for replacing thyA with mutated sequence was constructed using SOEing PCR ( 19 ) with oligonucleotides JW599 , JW600 , JW603 , and JW604 , using a colony of MG1655 as a template . 
All lacZ reporter gene fusions were constructed in plasmid pAMD-BA-lacZ using the oligonucleotides listed in Table S1 in the supplemental material . 
PCR products were cloned as SphI-HindIII-digested fragments . 
pAMD-BA-lacZ has been described previously ( 20 ) , but its construction has not been described in detail . 
pAMD-BA-lacZ is a derivative of pBAC-BA-lacZ ( Addgene plasmid 13423 ) in which the NotI-HindIII fragment has been replaced with a PCR product ( cut with NotI and HindIII ) containing an intrinsic terminator from E. coli rrfB and additional restriction sites ( BamHI , XhoI , and SphI ) . 
This PCR product was generated using oligonucleotides JW659 and JW660 , with E. coli genomic DNA as the template ( colony PCR ) . 
lacZ in this plasmid does not have a start codon or Shine-Dalgarno sequence , so fusions must be made translationally , as is the case for pAMD086 and pAMD007 , or cloned fragments must include a Shine-Dalgarno sequence and start codon , as is the case ( AGAAGGAG ATATACATATG ) for pAMD124 and pAMD132 . 
Oligonucleotides used to generate PCR products for cloning of lacZ fusions for regions upstream of araE , ytfQ , and ydeN were JW679 and JW680 ( araE ) , JW675 and JW678 ( ytfQ ) , JW1438 and JW2391 ( ydeN , 371 to 1 ) , and JW1438 and JW1635 ( ydeN , 371 to 14 ) . 
The sequences of ytfQ and ydeN upstream sequences , indicating the pieces cloned into the lacZ fusion plasmid , are shown in Fig . 
S1 and S2 , respectively , in the supplemental material . 
lacZ fusion plasmids to address transcription termination ( pJTW064 , pJTW055 , pJTW060 , pJTW062 , and pJTW061 ) were cloned using SOE-ing PCR ( 19 ) . 
A constitutive promoter was amplified from pAMD001 ( 18 ) using oligonucleotides JW3415 and JW3379 . 
These were joined using SOEing PCR with PCR products amplified with oligonucleotides JW3381 and JW3416 ( araE terminator ) , JW3476 and JW3478 ( tppB terminator ) , or JW3424 and JW3425 ( ahpF terminator ) . 
Final PCR products were cloned into pAMD-BA-lacZ using the In-Fusion method ( Clontech ) . 
The mutant tppB terminator construct was isolated serendipitously as a result of a mutation introduced during the cloning of the wild-type construct . 
Analysis of binding site conservation . 
Sequences surrounding AraC binding sites upstream of E. coli araB , araF , araE , araJ , and ytfQ and within dcp ( 30 bp upstream sequence and 30 bp downstream sequence in addition to the 19-mer AraC site ) were individually aligned with equivalent regions ( i.e. , the sequence 500 bp upstream of the homologous gene , or for the site within E. coli dcp , the entire homologous gene ; for S. enterica araT , sequence was taken from 500 to 100 with respect to the gene start , since these genes may be misannotated ) from S. enterica , Citrobacter rodentium ICC168 , Enterobacter sp . 
strain 638 , Klebsiella pneumoniae 342 , and Cronobacter sakazakii ES15 using ClustalW ( 21 ) . 
Similarly , the AraC site upstream of S. enterica araT was aligned with homologues from the same list of species . 
The number of matches to each position of each AraC site was determined , and the fraction of all species with a match to the reference sequence at each position was calculated . 
For each AraC binding site , the multispecies collection of aligned sites was used to compute the information content of each position ( 22 ) to generate conservation pro-files . 
- Galactosidase assays . 
Two to 3 ml cells was grown in LB or LB plus 0.2 % arabinose at 37 °C to an optical density at 600 nm ( OD ) of 0.8 to 600 1.0 , and the OD600 was recorded . 
Eight hundred l cells was pelleted at full speed in a microcentrifuge for 1 min ( 80 l was used for strongly active fusions , and this was corrected for at the final calculation step ) . 
Cell pellets were resuspended in 800 l Z buffer ( 0.06 M Na2HPO4 , 0.04 M NaH2PO4 , 0.01 M KCl , 0.001 M MgSO4 ) plus 50 mM - mercaptoethanol ( added fresh ) . 
Twenty l chloroform and 10 l 0.1 % SDS was added to the cells , followed by vortexing for 5 s. Assays were started by addition of 160 l o-nitrophenyl - - D-galactopyranoside ( ONPG ; 4 mg/ml in distilled H2O ) and stopped by addition of 400 l 1 M Na2CO3 upon development of an appropriate yellow color . 
The reaction time was noted . 
Samples were centrifuged at full speed in a microcentrifuge to pellet the chloroform . 
The OD420 of the supernatant was recorded . 
Arbitrary assay units were calculated as 1,000 [ A420 / ( A600 ) ( total time ) ] . 
RNA purification . 
RNA was purified from cells using a modified version of the hot phenol method that has been described previously ( 11 ) . 
Cells were grown in LB or LB plus 0.2 % arabinose at 37 °C to an OD600 of 0.6 to 0.8 . 
One ml cells was mixed with 400 l ice-cold 95 % ethanol and 5 % phenol-chloroform-isoamyl alcohol ( 25:24:21 mix ) . 
Cells were pelleted in a microcentrifuge for 1 min at full speed and washed once with Tris-buffered saline . 
Cell pellets were resuspended in 400 l RNA lysis buffer ( 2 % SDS , 4 mM EDTA ) and boiled for 3 min . 
Four hundred l acid phenol-chloroform-isoamyl alcohol mix ( pH 4.3 ) was added and incubated at 65 °C for 6 min and on ice for 5 min . 
Samples were centrifuged , and the aqueous layer was extracted once more with phenol-chloroform-isoamyl alcohol mix ( pH 4.3 ) . 
RNA was precipitated with 1 ml 100 % ethanol and 40 l 3 M sodium acetate . 
RNA was pelleted in a microcentrifuge for 10 min at full speed and washed once with room temperature 75 % ethanol . 
RNA pellets were air dried and resuspended in water and treated with 10 U of DNase I ( NEB ) in 500 l for 1 h at 37 °C . 
RNA was then phenol extracted and ethanol precipitated as described above . 
Transcription profiling using microarrays . 
RNA was purified from MG1655 ( wild-type ) or SAC001 ( araC ) cells grown in LB with or without 0.2 % arabinose at 37 °C . 
cDNA synthesis , labeling , hybridization to Affymetrix GeneChip E. coli Genome 2.0 microarrays , washing , and scanning were performed according to the manufacturer 's ( Affymetrix ) instructions . 
Triplicate data sets for each strain/condition pair were analyzed using GeneSpring software ( Agilent ) to calculate fold changes and P values . 
Only genes with 4-fold changes and P values of 0.1 are shown in Tables 1 and 2 . 
5 = RACE . 
RNA was purified from MG1655 cells grown in LB . 
5 = Rapid amplification of cDNA ends ( RACE ) was performed using the FirstChoice RLM-RACE kit ( Ambion ) according to the manufacturer 's instructions . 
Oligonucleotides JW1485 and JW1486 , specific to ydeN , were used in conjunction with oligonucleotides provided by the manufacturer ( GCTG ATGGCGATGAATGAACACTG and CGCGGATCCGAACACTGCGTT TGCTGGCTTTGATG , respectively ) . 
Northern blotting . 
Ten g RNA was run per lane on a 1 % agarose , 1 3 - ( N-morpholino ) propanesulfonic acid ( MOPS ) , 2 % formaldehyde gel at 70 V for 4 h. RNA was blotted by capillary action onto Magna nylon transfer membrane ( GE Water & Process Technologies ) and fixed by UV 5 irradiation . 
Membranes were incubated with 10 cpm PCR-generated double-stranded DNA ( dsDNA ) probe overnight in hybridization buffer ( 0.525 M Na2HPO4 , 7 % SDS , 1 mM EDTA , 10 mg/ml bovine serum albumin [ BSA ] ) and washed twice with wash buffer 1 ( 40 mM Na2HPO4 , 5 % SDS , 1 mM EDTA ) , wash buffer 2 ( 40 mM Na2HPO4 , 1 % SDS , 1 mM EDTA ) , and wash buffer 3 ( 0.2 % SDS , 0.2 SSC [ 1 SSC is 0.15 M NaCl plus 0.015 M sodium citrate ] ) at 55 °C ( 23 ) . 
Blots were visualized by phosphorimaging . 
Oligonucleotides used to generate PCR products for probe labeling were JW243 and JW1399 for araE and JW2387 and JW2388 for ygeA . 
RNA-seq . 
RNA was purified from 1 ml cells grown in LB with or without 0.2 % arabinose at 37 °C to an OD600 of 0.6 to 0.8 . 
Duplicate samples were prepared from independent biological replicates for each condition/strain . 
rRNA was removed using the RiboZero kit ( Epicentre ) . 
Strand-specific DNA libraries for Illumina sequencing were prepared using the ScriptSeq 2.0 kit ( Epicentre ) . 
Sequencing was performed using an Illumina HiSeq instrument ( University at Buffalo Next Generation Sequencing Core Facility ) . 
Sequences were aligned to the 14028s genome using the CLC Genomics Workbench , and differences in expression between conditions/strains were determined using the Pathogen Portal RNA-seq Analysis Pipeline ( 24 ) that includes Bowtie ( version 2.02 ; for aligning reads to reference genomes ) ( 25 ) , Cufflinks ( version 2.02 ; for transcript mapping ) , and CuffDiff ( for comparing expression of transcripts between samples ) ( 26 ) with default settings . 
2 Genea Arabinosec araCd araA 9.6 7.6 araB 9.4 7.3 araD 7.8 7.0 araE 5.8 6.1 araG 4.8 4.9 araJ 4.1 4.7 araH 4.6 4.6 araF 4.8 4.3 araHb 4.8 4.0 ygeA 3.5 3.4 isrB 2.9 2.9 cstA 2.3 2.0 melA 2.2 2.0 aldB 2.5 2.1 fucI 2.0 2.2 tdcF 2.0 2.2 tdcA 2.1 2.2 xylF 2.6 2.5 gudX 2.6 2.5 tdcE 2.6 2.6 garR 2.7 2.8 tdcC 2.7 2.9 tdcB 3.3 3.1 ydeN 3.1 3.1 tdcD 2.4 3.1 yjhA 3.1 3.1 tnaL 3.5 3.2 garD 3.3 3.3 garL 3.9 3.4 tnaA 3.0 3.7 garP 3.2 3.8 malG 3.0 3.9 malF 3.8 4.1 tnaB 3.9 4.2 gudP 4.2 4.3 malE 3.6 4.5 malM 4.1 4.6 malK 4.5 5.1 lamB 4.6 5.2 a Arabinose-responsive genes in E. coli were defined by a 4-fold change ( significant difference ) in growth with or without arabinose in wild-type ( MG1655 ) cells and a 4-fold significant difference between wild-type ( MG1655 ) and araC ( SAC001 ) cells in the presence of arabinose . 
Direct regulatory targets of AraC are indicated by boldface . 
Previously described regulatory targets of AraC are shaded in gray . 
b araH is represented twice on the microarray . 
c Fold change in mRNA level for wild-type cells grown with or without arabinose . 
d Fold difference in mRNA level between wild-type and araC cells grown in the presence of arabinose . 
Reverse transcription-PCR ( RT-PCR ) . 
To assess terminator read-through downstream of araE and araD , RNA was purified from MG1655 cells grown in LB plus 0.2 % arabinose . 
RNA was reverse transcribed using SuperScript III reverse transcriptase ( Invitrogen ) with 100 ng random hexamer according to the manufacturer 's instructions . 
A control reaction omitted the reverse transcriptase . 
One-twentieth of the cDNA ( or negative control ) was used as a template in a PCR with appropriate primers ( see Table S1 in the supplemental material ) . 
Oligonucleotides used for PCR were JW435 and JW436 for araE-ygeA and JW1366 and JW1367 for araD-polB . 
ChIP , ChIP-chip , and ChIP-seq . 
ChIP methods are presented in the supplemental material . 
Accession numbers . 
Microarray and sequencing data sets are available in the supplemental material ( E. coli ChIP-chip ) or through the EBI / EMBL ArrayExpress repository under the following accession numbers : E. coli transcription profiling , E-MTAB-1916 ; S. Typhimurium ChIP-seq , E-MTAB-1915 ; S. Typhimurium RNA-seq , E-MTAB-1901 . 
The Agilent microarray design used for E. coli ChIP-chip is available through ArrayExpress under accession number A-MEXP-2346 . 
RESULTS
Genome-wide mapping of AraC binding sites in E. coli . 
E. coli AraC-regulated genes have been identified previously through a variety of genetic approaches ( 3 , 27 -- 29 ) . 
Here , we used two complementary genomic approaches to comprehensively identify members of the AraC regulon . 
First , we mapped the genome-wide binding of TAP ( tandem affinity purification ) - tagged AraC ( tagged at its native locus in an unmarked strain ) using chromatin immunoprecipitation ( ChIP ) coupled with custom-designed oligonucleotide microarrays ( ChIP-chip ; see Table S2 in the supplemental material ) . 
We identified seven putative target loci for AraC : upstream of araB-araC , araE , araF , araJ , ytfQ , ydeN , and within dcp . 
These included all previously described AraC target loci , with the exception of xylA , which we believe is not a direct target of AraC under these growth conditions ( see below ) . 
AraC association has not been previously described for ytfQ , ydeN , or dcp . 
We validated the ChIP-chip data using ChIP coupled with quantitative real-time PCR ( ChIP/qPCR ) . 
To demonstrate that ChIP signal was not an artifact of the TAP tag , we constructed an unmarked derivative of MG1655 that expresses a C-terminally 3 FLAG-tagged AraC from its native locus . 
ChIP/qPCR verified significant association of AraC with all regions tested in the presence of arabinose ( Fig. 1A ; araJ was not tested ) . 
Association of AraC with all regions was reduced in the absence of arabinose , with no association detected for ydeN ( Fig. 1A ) . 
Thus , our data suggest that the overall affinity of AraC for its DNA sites is increased by association with arabinose . 
This is particularly important for AraC binding upstream of ydeN , since this interaction appears to be completely dependent upon arabinose . 
The known consensus sequence for AraC ( Fig. 1B ) is based on extensive footprinting and mutagenesis studies of the araBAD , araC , araE , araFGH , and araJ promoters ( 30 -- 34 ) . 
From our validated AraC ChIP targets , we inferred a de novo position-specific weight matrix ( PSWM ) for AraC using MEME , a bioinformatic tool that identifies overrepresented motifs in multiple unaligned sequences ( 35 ) . 
The top-scoring motif predicted by MEME is a good match to the known AraC motif ( Fig. 1B ) . 
MEME identified many , but not all , of the known AraC binding sites . 
This is unsurprising , since cooperative interactions of AraC dimers stabilize binding to some nonconsensus DNA sites at previously described target loci ( 32 ) . 
Effects of AraC and arabinose on global gene expression in E. coli . 
We used transcription profiling with Affymetrix high-density microarrays to determine the effects of AraC and arabinose on RNA levels genome wide . 
Wild-type or araC mutant cells were grown in the absence or presence of 0.2 % L-arabinose . 
Table 2 lists the genes whose expression changed significantly by 4-fold in wild-type cells upon addition of arabinose and whose expression differed significantly by 4-fold between wild-type and araC cells in the presence of arabinose . 
As expected , expression of known AraC-regulated genes , i.e. , araB , araA , araD , araE , araF , araG , araH , and araJ , increased substantially upon addition of arabinose in wild-type cells and was substantially higher in wildtype cells than araC cells in the presence of arabinose ( Table 2 ) . 
Novel AraC-regulated genes identified using this approach are discussed below . 
We did not detect significant AraC-dependent or arabinose-dependent regulation of xylA , a previously described AraC-regulated gene ( 36 ) , nor did we detect binding of AraC upstream of xylA . 
Hence , we believe that xylA is not a direct regulatory target of AraC under the conditions tested here ( cells were grown in tryptone broth in the other study ) . 
Genes regulated indirectly by arabinose and AraC . 
Many of the genes regulated by AraC/arabinose ( Table 2 ) are not associated with binding of AraC , as determined by the ChIP-chip experiment . 
We conclude that these genes are indirectly regulated by arabinose and/or AraC . 
Almost all of these indirectly regulated genes are repressed by AraC/arabinose , and they include genes associated with maltose metabolism ( malE , malF , malG , malK , malM , and lamB ) , threonine metabolism ( tdcA , tdcB , tdcC , tdcD , and tdcE ) , D-glucarate/D-galactarate metabolism ( garD , garL , garP , and garR ) , and tryptophan metabolism ( tnaA , tnaB , and tnaL ) . 
Only one indirect target gene , isrB , is upregulated 4-fold by both AraC and arabinose . 
isrB was originally annotated as a small RNA but has more recently been shown to encode a small membrane protein ( 37 ) . 
The mechanisms by which these genes are indirectly regulated by AraC and/or arabinose are unclear . 
Arabinose-independent repression of ytfQ transcription by AraC . 
The ChIP-chip analysis identified binding of AraC upstream of ytfQ and ppa ( divergently transcribed genes ) . 
The MEME analysis identified a putative AraC binding site centered at positions 133.5 and 94.5 relative to the previously mapped transcription start sites of ytfQ and ppa , respectively ( Fig . 
S1 in the supplemental material ) ( 38 ) . 
To determine experimentally whether this is the true AraC site upstream of ytfQ , we performed a ChIP experiment in a wild-type strain and in a strain in which the putative AraC binding site was mutated . 
Association of AraC , as determined by ChIP/qPCR , was significantly reduced by mutation of the putative DNA site ( Fig. 1C ) . 
We conclude that this is a genuine DNA site for AraC . 
We did not detect significant regulation of ytfQ or ppa by AraC or arabinose in the transcription profiling experiment ; however , ytfQ encodes a transporter that binds arabinose and galactose ( 39 ) , consistent with ytfQ being a regulatory target of AraC . 
We constructed a translational fusion of ytfQ to a lacZ reporter gene and performed - galactosidase assays with or without arabinose in a wild-type and a araC strain . 
We detected a small ( 1.5-fold ) but significant increase in expression in the araC strain ( see Fig . 
S3 in the supplemental material ) , suggesting that AraC directly represses transcription of ytfQ , albeit weakly . 
This apparent repression did not depend upon the addition of arabinose ( see Fig . 
S3 ) . 
Arabinose-dependent repression of ydeNM transcription by AraC . 
The ChIP-chip analysis identified binding of AraC upstream of ydeN ( Fig. 1A ) . 
The relatively low resolution of ChIP-chip precluded precise identification of the binding site ( s ) . 
We also showed in the transcription profiling experiment that expression of ydeN is reduced in the presence of arabinose and reduced in the presence of araC ( Table 2 ) . 
Similarly , expression of ydeM , the downstream gene , decreased 3.2-fold in the presence of arabinose and was reduced 7.3-fold by the presence of araC . 
This suggests that ydeN and ydeM are transcribed as a two-gene operon that is repressed by AraC . 
In the absence of arabinose , we did not detect AraC association upstream of ydeN ( Fig. 1A ) , nor did we detect any significant difference in expression of ydeN or ydeM between wild-type and araC mutant cells in the absence of arabinose . 
ChIP/qPCR analysis of RNA polymerase ( RNAP ) at ydeN confirmed that transcription decreases in the presence of arabinose and that this decrease is dependent upon araC ( Fig. 2 ) . 
Thus , ydeNM is a novel AraC-regulated operon that is directly repressed by AraC in an arabinose-dependent manner . 
We mapped the 5 = end of the ydeNM transcript using 5 = RACE and constructed transcriptional fusions to a lacZ reporter gene with fragments starting at position 371 and ending at position 1 or 14 with respect to the transcription start site . 
The longer fragment , from 371 to 14 , showed 3-fold arabinose-depen-dent repression by AraC ( Fig. 3 ) . 
In contrast , the shorter fragment , from 371 to 1 , showed no repression by AraC , suggesting association of AraC with the sequence around the transcription start site ( Fig . 
S2 in the supplemental material ) , although no site matching the AraC motif could be identified in this region . 
ydeN and ydeM encode a predicted sulfatase and a predicted sulfatase maturase , respectively ; thus , they have no apparent connection to arabinose metabolism . 
To determine whether either ydeN or ydeM is required for normal regulation of AraC-activated genes , we constructed a translational fusion of the araE upstream region to lacZ and measured - galactosidase activity in a wildtype strain and in isogenic strains containing deletions of either ydeN or ydeM . 
We did not detect any substantial difference in - galactosidase activity relative to the wild-type strain in either mutant ( see Fig . 
S4 in the supplemental material ) . 
AraC binding within dcp is not associated with detectable regulation of transcription . 
We detected binding of AraC within dcp ( Fig. 1A ) . 
The predicted binding site is located far from the 5 = end of any gene , including dcp itself ( see Fig . 
S5 in the supplemental material ) , suggesting that it is not associated with regulation of an annotated gene . 
Intriguingly , association of AraC with the site in dcp , as measured by ChIP/qPCR , is the highest of all AraC-bound regions in the E. coli genome ( Fig. 1A ) . 
To determine whether the AraC site within E. coli dcp is associated with transcription regulation , we used ChIP/qPCR to measure association of RNAP in the presence and absence of arabinose in a wild-type and a araC strain ( Fig. 2 ) . 
We did not detect any significant differences in RNAP association , suggesting that under these growth conditions , AraC does not regulate expression of a transcript that initiates within dcp . 
tivated transcripts . 
In the transcription profiling experiment , we found that expression of ygeA is significantly induced by arabinose and is dependent on araC ( Table 2 ) . 
ygeA is located immediately downstream of araE , in the same orientation , suggesting that some RNAP reads through the terminator downstream of araE and transcribes ygeA . 
We tested this hypothesis using RT-PCR to detect RNA that spans the araE and ygeA genes . 
Despite the presence of a strong predicted terminator , we were able to detect RNA species that included both araE and ygeA , consistent with terminator read-through ( Fig. 4A ) . 
ChIP/qPCR analysis of RNAP dem-onstrated high levels of RNAP association within ygeA at both the 5 = and 3 = ends , in the presence but not the absence of arabinose and dependent upon araC ( Fig. 2 ) . 
Northern blotting using probes specific to araE and ygeA also demonstrated read-through of the terminator downstream of araE ( Fig. 4B ) , although the level of read-through transcript was lower than that of araE transcript . 
We also detected an araC-independent transcript by Northern blotting that is likely due to initiation of transcription immediately upstream of ygeA ( Fig. 4B ) . 
Using densitometry analysis , we determined that the araE-ygeA read-through product is 11 % as abundant as the araE transcript . 
In contrast , the ChIP/qPCR data ( Fig. 2 ) indicate that 50 % of RNAP complexes read through the terminator downstream of araE . 
Together , these data suggest that the read-through transcript is less stable than that for araE alone . 
Using the transcription profiling data , we analyzed the differences in expression with or without arabinose and in the presence or absence of araC for the genes immediately downstream of araD , araH , and araJ . 
Only polB , the gene immediately downstream of araD , showed a 2-fold change in expression . 
Specifically , expression of polB increased 2.6-fold in the presence of arabinose and was 2.5-fold higher in wild-type cells than in araC cells . 
This suggests that RNAP also reads through the terminator downstream of araD . 
We confirmed this using RT-PCR ( Fig. 4A ) and ChIP/qPCR of RNAP ( Fig. 2 ) , as described above for araE-ygeA . 
From the ChIP/qPCR data , we estimate that 30 % of RNAP complexes read through the terminator downstream of araD . 
A recent study predicted sites of Rho-independent termination based on RNA sequence and structure ( 40 ) . 
The sequence between araE and ygeA ranked 286th on the list of 1,058 predicted terminators , suggesting that it should function effectively to terminate transcription . 
To experimentally test the ability of this sequence to terminate transcription , we constructed a lacZ reporter fusion that includes the predicted terminator with limited flanking sequence downstream of a strong , constitutive promoter ( Fig. 5A ) ( 41 ) . 
As controls , we constructed fusions with either no terminator sequence or predicted terminators and limited flanking sequence for the ahpF and tppB genes , ranked 293rd and 638th on the list of 1,058 predicted terminators , respectively ( Fig. 5A ) ( 40 ) . 
While the ahpF and tppB terminators reduced expression by 98 % and 99 % , respectively , the araE terminator reduced - galactosi-dase activity by only 56 % ( Fig. 5B ) . 
We also tested a mutant version of the tppB terminator that contains a point mutation in the upstream stem of the terminator stem-loop . 
This mutant terminator reduced - galactosidase activity by 89 % ( Fig. 5B ) . 
Thus , the araE terminator is only weakly effective and does not even function as well as a mutant version of a terminator that has lower predicted strength . 
Genome-wide mapping of AraC binding sites in S. enterica . 
We mapped the genome-wide binding of C-terminally FLAG-tagged AraC in S. enterica subsp . 
enterica serovar Typhimurium strain 14028s using ChIP coupled with deep sequencing ( ChIP-seq ) . 
We identified five putative target loci for AraC : upstream of araB-araC , araE , araJ , STM14_0178 ( araT ) , and within sseD . 
We validated the ChIP-seq data using ChIP/qPCR . 
Thus , we con-firmed significant association of AraC with all regions identified by ChIP-seq ( Fig. 6 ) . 
Effects of AraC and arabinose on global gene expression in S. enterica . 
We used RNA-seq to determine the effects of AraC and arabinose on genome-wide RNA levels in S. enterica . 
Wild-type or araC mutant cells were grown in the presence or absence of 0.2 % L-arabinose . 
Table 3 lists the 16 genes whose expression changed significantly ( false discovery rate [ FDR ] , 0.05 ) by 4-fold in wild-type cells upon addition of arabinose and whose expression differed significantly ( FDR , 0.05 ) by 4-fold between wild-type and araC cells in the presence of arabinose . 
Of the 16 regulated genes , 9 are direct regulatory targets based on the association of AraC with regions upstream of these genes , as determined by ChIP-seq . 
All of the direct regulatory targets are positively regulated by AraC and arabinose . 
No direct targets were identified that are regulated by AraC in the absence of arabinose . 
We did not detect any significant change in expression of sseD or the surrounding genes , suggesting that , like E. coli dcp , this gene contains an AraC binding site that is not associated with regulation of transcription under the conditions tested . 
It is important to note , however , that sseD falls within Salmonella pathogenicity island 2 ( SPI2 ) , a region that is transcriptionally silenced by H-NS under the conditions used in our work ( 42 ) . 
Thus , it is possible that AraC regulates transcription from the site within sseD under conditions that derepress SPI2 . 
The direct regulatory targets of AraC include all classical ara genes that are conserved in S. enterica , with the exception of araH . 
Note that araH is part of the araFGH operon in E. coli but araF and araG are not conserved in S. enterica . 
As we have shown for E. coli , ygeA is a direct regulatory target of AraC in S. enterica ( cotrans-cribed with araE ) . 
Lastly , STM14_0178 and STM14_0177 are direct regulatory targets of AraC . 
STM14_0178 and STM14_0177 do not have close homologues in E. coli and are predicted to encode an arabinoside transporter and an - L-arabinofuranosidase II precursor , respectively . 
Thus , it is likely that S. enterica metabolizes arabinosides as a source of arabinose . 
Based on their predicted functions , we rename STM14_0178 and STM14_0177 araT ( arabinoside transporter ) and araU ( arabinofuranosidase II precursor ) , respectively . 
The AraC site location upstream of araT can be estimated with 20-bp accuracy from the ChIP-seq data ( predicted AraC sites upstream of araE and araJ are within 20 bp of the corresponding ChIP-seq peaks ) ( see Fig . 
S6 in the supplemental material ) . 
Two regions upstream of araT have sequences similar to the AraC consensus motif . 
The location of one of these regions is precisely aligned with the ChIP-seq peak , suggesting that this sequence is bound by AraC under the conditions tested . 
The more upstream conserved sequence that resembles an AraC binding site falls outside the region predicted by the ChIP-seq data ; hence , it may bind AraC under other growth conditions , e.g. , in the absence of arabinose . 
The end of the downstream putative AraC site is only 21 bp from the annotated gene start for araT , a distance inconsistent with activation of araTU transcription by AraC . 
However , the RNA-seq data strongly suggest that the transcription start site is downstream of the annotated gene start for araT . 
Hence , the 
Genea Arabinoseb araCc ygeA 4.1 4.3 araJ 3.8 4.1 yjcB 3.8 3.0 ycfR 2.7 2.9 dctA 2.4 3.7 mglC 2.4 3.9 mglA 3.2 4.7 ygbM 2.5 5.3 a Arabinose-responsive genes in S. enterica were defined by 4-fold change in expression ( significant difference ) under growth with or without arabinose in wild-type ( 14028s ) cells and 4-fold change ( significant difference ) between wild-type ( 14028s ) and araC ( AMD485 ) cells in the presence of arabinose . 
Direct regulatory targets of AraC are indicated by boldfaced text . 
b Fold change in mRNA level for wild-type cells grown with or without arabinose . 
c Fold difference in mRNA level between wild-type and araC cells grown in the presence of arabinose . 
translation start site of araT is likely to be incorrectly annotated , and the downstream putative AraC site is likely to be located upstream of position 40 with respect to the araTU transcription start site . 
This site position is consistent with transcription activation by AraC using a mechanism similar to that described for E. coli AraC-activated genes . 
Conservation of the AraC regulon across the family Entero-bacteriaceae . 
AraC is highly conserved across the family Entero-bacteriaceae , which includes E. coli and S. enterica . 
The two helix-turn-helix DNA-binding domains are particularly well conserved , e.g. , they are 100 % identical between E. coli and S. enterica . 
Hence , AraC likely binds with similar DNA sequence specificity across all Enterobacteriaceae species . 
To determine whether regulation of AraC target genes is conserved across the family Enterobacteria-ceae , we aligned sequence surrounding E. coli and/or S. enterica AraC sites identified in this work with equivalent regions from four other Enterobacteriaceae species ( Citrobacter rodentium , Enterobacter sp . 
strain 638 , Klebsiella pneumoniae , and Cronobacter sakazakii ; all alignments are shown in Fig . 
S7 in the supplemental material ) . 
S. enterica sseD is not conserved in any of the other species , and E. coli ydeN is only conserved in S. enterica ; hence , these regions were not analyzed . 
Conservation of AraC sites was observed for araBAD , araFGH , araE , ytfQ , and araTU ( Fig. 7 ; also see Fig . 
S6 ) . 
No conservation of AraC sites was observed for araJ or dcp . 
Conservation was highest for two regions of the AraC binding site : positions 4 to 7 and 13 to 19 . 
This is consistent with the information content of the motif derived from the E. coli AraC ChIP-chip data and with the known consensus sequence ( Fig. 1B ) . 
DISCUSSION
E. coli AraC is one of the best-studied TFs in any bacterial species and was the first described transcriptional activator ( 3 , 4 ) . 
With the exception of xylA , the last AraC-regulated gene to be identified was araJ , more than 30 years ago ( 27 ) . 
We combined two complementary genomic approaches to expand the known E. coli AraC regulon . 
Specifically , we identified three novel binding targets of AraC ( upstream of ytfQ and ydeN and within dcp ) and five novel AraC-regulated genes ( ytfQ , ydeN , ydeM , ygeA , and polB ) . 
Strikingly , regulation of four of the five novel target genes is mechanistically distinct from that observed previously for other AraC-reg-ulated genes . 
Thus , our data demonstrate the power of integrating ChIP-chip/ChIP-seq and transcription profiling as an unbiased and comprehensive approach to identify regulatory networks . 
ChIP-chip identifies noncanonical AraC binding sites . 
Despite the extensive history of research on E. coli AraC , we identified several novel AraC-bound regions and several novel AraC-regu-lated genes . 
It is perhaps unsurprising that our unbiased , genomic approach identified AraC sites and AraC-regulated genes that differ functionally from those identified previously , as this would explain why they were missed in previous studies . 
Specifically , we identified AraC binding sites that ( i ) repress rather than activate transcription in an arabinose-dependent manner ( ydeN ) , ( ii ) result in little or no observed regulation under standard laboratory growth conditions ( ytfQ and dcp ) , and ( iii ) are located within a gene ( dcp ) . 
We also identified AraC-regulated genes that are transcribed due to read-through of inefficient Rho-independent terminators ( ygeA and polB ) . 
Previous ChIP-chip studies in bacteria have identified many TF binding sites within genes ( 7 ) . 
The most striking example in E. coli is RutR , for which 80 % of binding sites are intragenic ( 9 ) . 
With the exception of binding sites close to the 5 = end of genes ( 43 ) , very few intragenic TF binding sites have a described function . 
We identified a binding site for AraC inside dcp , a gene that encodes dipeptidyl carboxypeptidase . 
Given the lack of conservation of this putative AraC binding site in other Enterobacteriaceae species and the lack of detectable regulation by AraC at this site , we conclude that the site is unlikely to have regulatory function under the tested growth conditions . 
We identified an analogous AraC site in S. enterica inside sseD . 
We propose that these binding sites have ( i ) regulatory function under a different growth condition , ( ii ) a function unrelated to transcription , or ( iii ) no function . 
Novel E. coli AraC binding sites that repress transcription . 
We identified two E. coli transcripts that are directly repressed by AraC : ytfQ and ydeNM . 
ytfQ encodes a galactose/arabinose transporter ; thus , it has a clear connection to the established function of AraC in regulating arabinose metabolism . 
Repression of ytfQ by 
AraC is weak ( 1.5-fold ; see Fig . 
S3 in the supplemental material ) , indicating that either AraC has only a minor effect on ytfQ expression or that more substantial regulation by AraC is associated with other growth conditions . 
AraC has previously been shown to repress its own transcription by binding to a region overlapping the araC promoter elements ( 32 ) . 
This repression occurs independently of the addition of arabinose . 
The location of the AraC binding site upstream of ytfQ is too far upstream of the transcription start site to repress transcription by directly occluding RNAP . 
We propose that AraC bound at this site interacts with additional regulatory proteins , perhaps another monomer of AraC , bound closer to the transcription start site . 
GalR has been shown to regulate ytfQ ( 44 ) ( Fig . 
S1 in the supplemental material ) . 
However , we detected no effect of GalR on regulation of ytfQ by AraC ( data not shown ) . 
Unlike AraC-dependent repression of araC and ytfQ , repression of ydeN occurs only in the presence of arabinose ( Table 2 and Fig. 3 ) . 
This is consistent with our ChIP data showing binding of AraC upstream of ydeN only in the presence of arabinose ( Fig. 1A ) . 
Although arabinose-dependent repression by AraC has not been observed before , there are clear parallels with arabinose-de-pendent activation of araBAD transcription . 
Arabinose binding to AraC alters its DNA binding properties ( 5 ) . 
At the araC-araBAD intergenic region , AraC forms a repression loop in the absence of arabinose due to the dimerization of distally bound AraC mono-mers . 
In the presence of arabinose , dimerization occurs at adjacent sites , breaking the repression loop and activating transcription of araBAD ( 6 ) . 
This change in DNA binding is due to a rearrangement of the N-terminal arabinose-binding/dimeriza-tion domain and the C-terminal DNA-binding domain relative to one another ( 5 ) . 
We propose that the DNA binding properties of AraC allow it to bind at ydeN only in the presence of arabinose . 
Our reporter fusions indicate that maximal repression by AraC requires sequence between 1 and 15 relative to the transcription start site ( Fig. 3 ) . 
This strongly suggests the presence of an AraC binding site overlapping the transcription start site , consistent with a role in transcriptional repression . 
We propose that AraC binds as a dimer to adjacent sites overlapping the transcription start site . 
Thus , arabinose-dependent repression of ydeNM by AraC would use the same mechanism as arabinose-dependent activation of araBAD . 
Read-through of inefficient transcription terminators contributes to the E. coli AraC regulon . 
ygeA and polB are positively regulated by AraC and arabinose due to partial read-through of Rho-independent terminators ( Fig. 2 , 4 , and 5 ) . 
We analyzed published microarray data from another group that used arabinose to induce overexpression of various proteins unrelated to AraC . 
Consistent with our own work , both ygeA and polB were in the top 5 % of all genes when ranked by the level of arabinose induction ( 45 ) . 
An equivalent analysis for ydeN showed that it is in the bottom 0.5 % of all genes ( 45 ) . 
From the Northern blot ( Fig. 4B ) it is clear that , in the presence of arabinose , the majority of ygeA mRNA is in the form of the read-through transcript , suggesting that read-through is physiologically relevant . 
Many predictions have been made for intrinsic terminators in E. coli and other species ( 40 , 46 -- 50 ) . 
Sequences downstream of araE and araD have been predicted to form terminators . 
This is especially true for the terminator downstream of araE , which has a long , G/C-rich stem-loop followed by a 10-mer sequence with 8 U 's ( Fig. 4 ) . 
However , both the araE and araD terminators are only weakly effective . 
For the araE terminator this is unlikely to be due to alternative structures influenced by upstream sequence , since a minimal region is insufficient to terminate in the reporter assay we used ( Fig. 5 ) . 
Thus , our data suggest that terminator predictions are often inaccurate . 
Regulatory functions for AraC beyond arabinose metabo-lism . 
We have identified 7 novel AraC-regulated genes in E. coli and S. enterica . 
S. enterica araT and araU encode a likely transport / metabolism system for arabinosides . 
This suggests that S. enterica can use arabinosides as a carbon source by metabolizing them to arabinose . 
Only one other novel AraC-regulated gene identified in this work , E. coli ytfQ , has a known connection to arabinose me-tabolism ( 39 ) . 
Furthermore , araJ is a long-established member of the AraC regulon but has no known connection to arabinose me-tabolism ( 51 ) . 
It is possible that some or all of the novel AraC-regulated genes have as-yet-unidentified connections to arabinose metabolism , although this seems especially unlikely for polB , which encodes a well-characterized DNA polymerase . 
In addition , deletion of ydeN or ydeM did not substantially affect araE expression ( see Fig . 
S4 in the supplemental material ) , suggesting that AraC and intracellular arabinose levels are unaffected by the absence of these genes . 
Regulation of polB by AraC is particularly intriguing given the well-established function of polB in DNA replication and repair ( 52 ) . 
A 6-fold increase in polB expression is sufficient to give a detectable increase in the spontaneous mutation rate independent of the SOS response ( 53 ) . 
We were not able to detect a significant increase in the spontaneous mutation rate by growth in the presence of arabinose ( data not shown ) , but polB expression increases only 2.6-fold . 
While it is likely that an increase in the spontaneous mutation rate would be below our detection threshold , the effect of arabinose on polB expression could contribute to genome variability during long-term growth . 
Conservation of the AraC regulon . 
The PhoP regulon is by far the best studied with respect to conservation . 
Only three genes are consistently regulated by PhoP across the family Enterobacteria-ceae ( 13 ) . 
In contrast , our data indicate that most members of the AraC regulon are conserved in this family . 
This `` core '' regulon is comprised of araBAD , araFGH , araE , ytfQ , and araTU . 
Three of these genes , ytfQ , araT , and araU , have not previously been described as AraC targets . 
The conservation of regulation of ygeA and polB by transcriptional read-through is more difficult to assess . 
araE-ygeA synteny is not well conserved , suggesting that ygeA is not a conserved AraC regulon member . 
We did not detect regulation of polB by AraC in S. enterica . 
However , there is a two-gene insertion between araD and polB in S. enterica . 
In contrast , most other Enterobacteriaceae species maintain the araD-polB synteny . 
Hence , polB regulation by AraC may be widely conserved . 
Strikingly , one of the conserved regulatory targets of AraC , araTU , is absent from E. coli . 
This highlights the risk associated with making inferences on TF regulons if experimental data are only available for one species . 
An analysis of AraC regulon conservation based only on E. coli target genes would have missed araTU . 
Similarly , an analysis of AraC regulon conservation based only on S. enterica target genes would have missed araFGH . 
The importance of using experimental data from multiple species is especially high for TFs that have degenerate binding motifs , such as AraC , since binding sites can not easily be predicted from DNA sequence alone . 
Conclusions . 
Our unbiased mapping of the AraC regulons of 
E. coli and S. enterica has revealed new functions and new mechanisms of action for this storied regulator . 
Our data suggest that AraC regulates functions beyond arabinose metabolism . 
Furthermore , unlike the PhoP regulon , most AraC regulatory targets are conserved across related species , although conservation is limited to genes required for the transport and metabolism of arabinose . 
Our work highlights the importance of genome-scale approaches in the study of bacterial gene expression . 
ACKNOWLEDGMENTS
We thank David Grainger , members of the Wade laboratory , Robert Schleif , and members of Keith Derbyshire and Todd Gray 's group for helpful discussions . 
We thank David Grainger , Todd Gray , Keith Derbyshire , and Rick Wolf for comments on the manuscript . 
We thank Chunhong Mao for assistance with RNA-seq analysis . 
We thank the Wad-sworth Center Bioinformatics Core , the Wadsworth Center Applied Genomic Technologies Core , and the University at Buffalo Next Generation Sequencing Core Facility for technical assistance . 
This work was supported by National Institutes of Health ( NIH ) grant 1DP2OD007188 and Wadsworth Center start-up funds ( J.W. ) , U.S. National Science Foundation grant MCB-1158056 ( I.E. ) , and appointments ( C.B. and B.P. ) to the Emerging Infectious Diseases ( EID ) Fellowship Program administered by the Association of Public Health Laboratories ( APHL ) and funded by the Centers for Disease Control and Prevention ( CDC ) .