C

3

4

5

6

7

8

9

10

11

A12

iIcThro

13

14

15

16

17

18

19

20

©21

K22

23

124

25

f26

s27

s28

s29

r30

p31

a32

s33

r34

R35

p36

T37

a38

1 12 d

PR

OO

F

ARTICLE IN PRESS+ModelBAC 5929 1–9

Computational Biology and Chemistry xxx (2007) xxx–xxx

Ligand-binding prediction in the resistance-nodulation-celldivision (RND) proteins

Armando Hernandez-Mendoza a, Carmen Quinto a,Lorenzo Segovia b, Ernesto Perez-Rueda a,∗

a Departmento de Biologıa Molecular de Plantas, Instituto de Biotecnologia, Universidad Nacional Autonoma de Mexico,A.P. 565-A Cuernavaca, Morelos 62210, Mexico

b Departamento de Ingenierıa Celular y Biocatalisis, Instituto de Biotecnologia, Universidad Nacional Autonoma de Mexico,A.P. 565-A Cuernavaca, Morelos 62210, Mexico

Received 18 May 2006; received in revised form 7 February 2007; accepted 8 February 2007

bstract

The resistance-nodulation-cell division (RND) protein family is a ubiquitous group of proteins primarily present in bacteria. These proteins,nvolved in the transport of multiple drugs across the cell envelope in bacteria, exhibit broad substrate specificity and act like efflux pumps.n this work, a protein belonging to the RND protein family, AcrB of Escherichia coli was used as a working model to predict in silico theompounds transported by 47 RND proteins. From AcrB we extracted and clustered 14 amino acids directly involved in substrate interactions.his clustering provides enough information to identify 16 groups that correlates with the ligand they extrude, such as proteins expelling aromatic

Dydrocarbons (SrpB cluster) or proteins expelling heavy metals (CnrA cluster). The relationship between conserved, cluster-specific and variable

esidues indicates that although the ligand-binding domain is conserved in structure, it has enough flexibility to recognize specifically a diversityf molecules.

2007 Elsevier Ltd. All rights reserved.

o 39

2 40

41

i 42

a 43

d 44

a 45

f 46

c 47

t 48

t 49

i 50

d 51

OR

RE

CTEeywords: RND-proteins; AcrB; Ligand binding prediction

. Introduction

In general, homologous proteins share common biologicalunctions but exhibit different specificity towards their sub-trates. The identification of residues that participate in thepecificity of these interactions is useful for functional analy-is, and for predictive studies. Recent analyses have shown thatesidues involved in specific recognition between interactingartners are frequently found at conserved positions (Johnsonnd Church, 2000). These positions are evident in multipleequence alignments, where the distribution of amino acidseflects functional and structural constraints in the proteins.ecently, a number of algorithms addressing ligand-binding

UN

C

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligand-proteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

redictions have been developed, such as the Evolutionaryrace method, that exploits information on protein sequencend structure (Lichtarge et al., 1996), whereas alternative meth-

∗ Corresponding author. Tel.: +52 56 22 76 10; fax: +52 777 3 17 23 88.E-mail address: erueda@ibt.unam.mx (E. Perez-Rueda).

a 52

i 53

b 54

atta

476-9271/$ – see front matter © 2007 Elsevier Ltd. All rights reserved.oi:10.1016/j.compbiolchem.2007.02.003

ds only use protein sequence information (Berezin et al.,004).

The resistance-nodulation-cell division (RND) protein familys a ubiquitous group of proteins described in bacteria, archaeand eukarya, which is involved in the transport of multiplerugs across the cell envelope (Paulsen et al., 1996; Paulsen etl., 1997). The RND proteins form complexes with membraneusion proteins (MFP) in the periplasm and with outer membranehannels of TolC superfamily to accomplish the transport fromhe cell to the extracellular medium. Diverse RND efflux sys-ems have been functionally characterized, which are involvedn the transport of and resistance to antibiotics, hydrophobicyes, and detergents, among others (Poole, 2004). In addition,large number of point mutations causing altered substrate or

nhibitor specificities in different multidrug transporters haveeen reported (Hearn et al., 2006; Murakami et al., 2004; Yu et

binding prediction in the resistance-nodulation-cell division (RND)7.02.003

l., 2005) as well as crystallographic data of purified AcrB in 55

he presence of several ligands (Yu et al., 2003). These results 56

ogether do not only suggest that these proteins form direct 57

tomic interactions with the molecules of substrates but a dif- 58

ED

IN+ModelC

2 nal B

f59

s60

61

f62

t63

c64

o65

c66

m67

c68

s69

s70

p71

i72

c73

i74

w75

e76

n77

h78

w79

280

281

82

s83

s84

b85

s86

o87

k88

(89

w90

d91

S92

293

l94

95

d96

197

K98

a99

o100

V101

s102

r103

a104

r105

i106

n107

F108

109

t110

n 111

p 112

a 113

1 114

s 115

l 116

d 117

a 118

t 119

s 120

o 121

o 122

2 123

124

p 125

w 126

c 127

t 128

a 129

l 130

f 131

e 132

p 133

d 134

a 135

l 136

S 137

138

S 139

t 140

p 141

I 142

o 143

a 144

3 145

3 146

147

A 148

F 149

t 150

i 151

t 152

o 153

m 154

b 155

o 156

d 157

R 158

NC

OR

RE

CT

ARTICLEBAC 5929 1–9

A. Hernandez-Mendoza et al. / Computatio

erential interaction can be achieved as a consequence of theubstrate nature.

In order to predict ligand molecules in the RND proteinamily in bacteria, we clustered 14 columns that correspondo amino acids selected from the AcrB protein of Escherichiaoli, whose 3D structure has been determined in the presencef four structurally diverse ligands (Murakami et al., 2002). E.oli AcrB is an excellent archetype to predict ligand bindingolecules, since it recognizes many structurally unrelated toxic

ompounds, such as antibiotics, disinfectants, dyes, and simpleolvents (Nikaido and Zgurskaya, 2001). In this work, a multipleequence alignment (MSA) was constructed with 47 RND-likerotein sequences and 14 columns were extracted correspond-ng to residues directly involved in substrate interactions into theentral cavity (Yu et al., 2003) of AcrB and to residues locatedn the central pore (Murakami et al., 2002). These 14 columnsere used to generate a tree that shows diverse RND multidrug

fflux family clusters that correlate with the ligand they recog-ize and extrude across the membrane cell. Similar approachesave shown the advantage of anchoring functional annotationsithin a protein family context (Johnson and Church, 2000).

. Methods

.1. Sequence analysis

The E. coli AcrB protein sequence, whose crystallographictructure was solved at 3.5 A, was used as seed in a Blastearch with an E-value ≤ 10−5 to identify RND protein mem-ers in the NR-database. Five hundred and sixty-five proteinequences were retrieved from this sequence comparison, butnly 43 RND-proteins were selected whose ligands are wellnown. In addition, RmiB and ORF2pC (Rhizobium etli), Orf2R. leguminosarum) and AmeB (Agrobacterium tumefaciens)ere included in the analysis because we have experimental evi-ences supporting a different function than previously reported.ee below.

.2. Multiple sequence alignments (MSA) andigand-binding-site analysis

Forty-seven RND proteins were aligned using ClustalX withefault conditions and manually edited (Thompson et al., 1997,994). In this alignment the AcrB protein sequence from E. coli12 was used as reference to identify and extract 10 amino

cid residues involved in ligand binding. Those residues are 6 Ar less from their ligands (forming preferentially hydrophobic,an der Waals, or electrostatic interactions) in the AcrB proteintructure (1IWG) (Yu et al., 2003). Four additional amino acidesidues identified experimentally by mutational analysis werelso included (Murakami et al., 2004). In total, 14 amino acidesidues were selected to perform the analysis: five residues aren the central pore (D101, V105, N109, Q112 and P116), and

U

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligandproteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

ine in the central cavity (L25, K29, D99, V382, A385, F386,388, F458 and F459) (see Table A.1, Fig. A.1 in Appendix A).

In a second step, columns from the MSA corresponding tohe 14 ligand-binding residues were selected and used to build a

Atin

PR

OO

F

PRESSiology and Chemistry xxx (2007) xxx–xxx

ew alignment. This new alignment was analyzed by maximumarsimony (MP), fitch (F) and neighbor-joining (NJ) methods,nd their corresponding trees were generated. From this analysis,6 clusters subsequently emerged that correlated with the sub-trates they extrude. The MP tree clusters were used to predictigands in those proteins whose ligands are unknown or contra-ictories. Alternatively, 100 random MSAs with the same aminocid composition and length than the original were constructedo evaluate the previously described clustering. We must empha-ize that phylogenetic tools were only used to cluster similar setsf residues, not to make inferences on the evolutionary historyf these proteins.

.3. Performance evaluation

Finally, to evaluate the performance of the ligand-bindingredictions, the clusters identified by MP (“observed clusters”)ere compared to their corresponding annotated ligands. This

omparison was useful to calculate the following values: (1)rue positives (TP): proteins with (at least one) common lig-nd clustered together; (2) false positives (FP): proteins whoseigand was completely different to the rest of the cluster; (3)alse negatives (FN): proteins included in a cluster with differ-nt ligands; (4) sensitivity, Sn = TP/(TP + FN), is the fraction ofroteins recovered in the inferred clustering; (5) positive pre-ictive value, PPV = TP/(TP + FP), is the fraction of the proteinsnd ligands in the inferred clusters that belongs to the annotatedigand binding; (6) Accuracy, Ac = (Sn + PPV)/2, is the PPV andn average.

In this analysis the classical definition of specificityp = TN/(TN + FP) was not used, because our evaluation cri-

eria was based on the rate of true negatives (TN), defined asroteins whose ligand has not been experimentally described.ndeed, the number of ligands is typically smaller than the sizef the cluster (16 clusters), and the percentage of TN should belways closer to 1, which would favorably bias in the evaluation.

. Results

.1. Identification and selection of RND proteins

In order to identify members of the RND family, the E. colicrB protein sequence was used as seed to scan the NR database.ive hundred and sixty-five proteins were identified as poten-

ial RND-proteins and 43 proteins functionally characterizedn diverse bacteria were selected, ranging from proteobacteriao bacteroidetes (Fig. 1). These proteins share in average 40%f identity among them, and the information on their ligandolecules is known. Around 90% of the 565 proteins identified

y BLAST search have been annotated as hypothetical, unknownr uncharacterized proteins, emphasizing the importance of pre-ict their probable ligand compound. In addition, four proteinsmiB and ORF2pC (R. etli), Orf2 (R. leguminosarum) and

-binding prediction in the resistance-nodulation-cell division (RND)7.02.003

meB (A. tumefaciens) whose ligands are unknown or con- 159

radictories were considered in this analysis. AmeB might be 160

nvolved in ß-lactams and detergents extrusion, however its phe- 161

otype is not completely clear when the protein is deleted (Peng 162

CTE

D P

RO

OF

ARTICLE IN PRESS+ModelCBAC 5929 1–9

A. Hernandez-Mendoza et al. / Computational Biology and Chemistry xxx (2007) xxx–xxx 3

Fig. 1. Ligand prediction for the 47 RND proteins. (A) Tree generated with MP using the Phylip package program Protpars, with the 14 residue columns involved in theligand interaction reported for AcrB. Bars in red show the clusters identified. (B) Description of the clusters identified. Columns are as follow: protein name includedin the cluster, residues aligned and used to build the tree; substrate preferentially recognized by the proteins, and organisms where the protein has been characterized.In red are the residues from the central pore and in black residues from the central cavity. Substrates included—AH: aromatic hydrocarbons; AC: acriflavine; AG:aminoglycosides; BKC: benzalkonium chloride; BL: ß-lactams; BS: bile salts; CM: chloramphenicol; CP: cephalosporin; CV: crystal violet; COU: coumestrol;CHA: cholic acid; CHE: chenodeoxycholic acid; DC: deoxylchloate; EB: ethidium bromide; EM: erythromycin; FA: fatty acids; FQ: fluoroquinolones; FU: fusidica ML:h ium dt bromi

a163

w164

d165

R166

m167

s168

i169

m170

p171

s172

173

n174

f175

a176

f177

m178

o 179

t 180

R 181

i 182

b 183

e 184

o 185

c 186

s 187

c 188

e 189

b 190

NC

OR

REcid; H33342: hoechst 33342; HM: heavy metals (e.g., Co2+, Zn2+and Ag+);

ydrocarbon; PU: puromycin; RH6G: rhodamine 6G; RF: rifampicin; SDS: sodrimethoprim; TR: triclosan; TX: Triton X-100; TPP: tetraphenylphosphonium

nd Nester, 2001). RmiB and ORF2pC were included becausee have experimental evidences that support the notion theyo not play a role in the rhizobia-legume nodulation process.ecently, we identified in R. etli an operon resembling the RNDultidrug efflux system, rmiRABC, however, we did not found a

ubstrate extruded by this putative pump (Hernandez-Mendoza,n press). The clustering analysis proposed here do not onlyight give us clues about the most probable ligands of these

roteins, but also might suggest us the existence of additionalubstrates.

As it has been previously reported, RND proteins are orga-ized mostly in operons, together with genes from the membrane

U

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligand-proteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

usion protein (MFP) family (Dinh et al., 1994), and occasion-lly transcribed together with genes from the outer membraneactor (OMF) family (Paulsen et al., 1997). In order to deter-ine how many efflux systems can be functionally traced per

thod

macrolides; MB: methylene blue; NO: novobiocin; PAH: polycyclic aromaticodecyl sulfate; TAU: taurocholic acid; TBT: tributyltin; TC: tetracycline; TM:de; UK: unknown; V: vanadium. In question marks are probable substrates.

rganism we made an exhaustive genomic context analysis usinghe GeConT program (Ciria et al., 2004). Interestingly, diverseND, MFP and OMF protein genes organized in operons were

dentified. Indeed we found an average of six efflux systems peracterial genome. In Pseudomonas aeruginosa and P. putida, 13fflux systems were identified (E-value ≤ 2.3E−49), of whichnly six have been experimentally characterized to date. In E.oli K12, seven RND transporters were identified based onequence comparisons (E-value ≤ 3.2–43), six of them fullyharacterized. Finally, in A. tumefaciens, eight probable RND-fflux pumps were predicted (E-value ≤ 3.04E−44) though theiological function has been experimentally determined for only

binding prediction in the resistance-nodulation-cell division (RND)7.02.003

wo of them (Palumbo et al., 1998; Peng and Nester, 2001). The 191

igh number of RND-proteins in diverse organisms does not 192

nly suggest a high duplication rate but also a probable high 193

iversity of compounds that can be expelled by those systems. 194

ED

OF

ARTICLE IN PRESS+ModelCBAC 5929 1–9

4 A. Hernandez-Mendoza et al. / Computational Biology and Chemistry xxx (2007) xxx–xxx

Table 1Conserved and variable residues in the RND-protein clusters

e proclust

uded.

I195

g196

c197

M198

b199

b200

3201

202

m203

1204

c205

l206

f207

a208

s209

a210

r211

n212

r213

e214

a215

e216

fi217

o218

h219

E220

221

1222

o223

a224

p225

o226

v227

t 228

p 229

a 230

c 231

c 232

e 233

p 234

1 235

t 236

r 237

m 238

h 239

240

p 241

R 242

m 243

c 244

( 245

s 246

a 247

i 248

s 249

t 250

a 251

a 252

t 253

t 254

c 255

256

a 257

p 258

NC

OR

RE

CT

“X” denotes 100% identical residues. The first column has a representativper position: for instance, F458 is conserved in all proteins included in the(black letters); CP: central pore (red letters). Orphan groups were not incl

n addition, the organization in operons of the MFP and RNDenes suggests a probable co-evolution process to protect theell from toxic compounds. This finding also suggest that RND,FP and OMF act in conjunction to mediate transport across

oth membranes of the cell envelope, as it has been evidencedy genetic experiments (Dinh et al., 1994).

.2. Ligand prediction in members of the RND family

In order to predict the most probable ligands in 47 selectedembers of the RND family, 14 columns out of approximately

300 in a MSA were picked out, as described in Section 2. Theseolumns correspond to the amino acids involved in E. coli AcrBigand-binding, five residues from the central pore, and ninerom the central cavity (Fig. 1 and Table 1). These residuesre distributed throughout the primary amino acid structure inuch a way that there is no local sequence motif that may bessociated to a particular binding function. The columns cor-esponding to the 14 selected residues were extracted into aew MSA to complete the clustering analysis. Still, additionalesidues might be involved in ligand binding, we only consid-red 14 amino acid residues since they have been well analyzeds involved in physical interactions with diverse compoundsxtruded by AcrB of E. coli. Indeed, Hearn et al. (2006) identi-ed a diversity of potential residues (most of them included inur analysis) that affect the selectivity for polycyclic aromaticydrocarbons, antibiotics and solvents in P. fluorescens cLP6amhB, supporting the importance of the selected residues.

Based on the MP analysis of the new MSA we identify6 groups of proteins that correlate with the ligand preferencebserved for several RND proteins (Fig. 1). The calculated aver-

U

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligandproteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

ge sensitivity for all clusters analyzed reaches 94.0% and aositive predictive value (PPV) of 88.0%, leading to an accuracyf 91.0%. This indicates the significance of our analysis. Theariability observed among these residues should explain why

dcao

PR

Otein for the cluster. Second to fifteenth columns indicate the conservationer “AcrD”. Cluster-specific residues are in parenthesis. CC: central cavity

hese proteins recognize such diversity of ligands. In Fig. 1, weresent 14 clusters identified by MP, and in Table 1 we presentsummary of conserved positions per cluster. These clusters

ontained variability in the 14 residues of the central cavity andentral pore, and some positions (such as F458) can be consid-red as universal and some other as cluster specific. From thiserspective, diverse clusters conserve between 0 and 13 out of4 residues involved in ligand binding. We do not only suggesthat this variability inside clusters is enough to differentiate theirespective ligands but also that conserved positions per clusteright define the major recognition of the ligands, however this

ypothesis should be further explored.Alternatively, to contrast the 14-columns tree clusters we

erformed a similar analysis by using MP with the completeND-protein sequences. The results obtained show that theajority of the RND-proteins are grouped within a single large

losely related cluster, while proteins extruding heavy metalsHM) are clustered together (data not shown), as previouslyhown (Paulsen et al., 1996). This result suggests that thenalysis of the complete RND protein sequences alignment isnfluenced by the organism speciation rather than by substratepecificity. Conversely, random MSA clustering did not correlateo the ligand-binding substrates observed with the 14 functionalmino acid residues, except for the HM cluster (which, as shownbove, is divergent from the other sequences) suggesting thathe conservation of the selected residues in their specific posi-ions is influenced by the interaction with the molecules of theompounds extruded by the protein.

In summary, we found that 14-columns with selected aminocids involved in ligand interaction are good enough to grouproteins with similar ligands, and they might be useful to pre-

-binding prediction in the resistance-nodulation-cell division (RND)7.02.003

ict ligands in proteins whose substrate is unknown. A tree 259

onstructed with the whole sequences and random columns 260

re not enough informative, hence showing the significance 261

f our approach. The results obtained for certain groups are 262

D

IN+ModelC

nal B

d263

(264

3265

266

n267

i268

p269

L270

s271

T272

a273

r274

a275

a276

d277

v278

a279

p280

l281

n282

f283

p284

3285

c286

287

r288

e289

(290

p291

2292

t293

t294

t295

w296

s297

v298

l299

s300

t301

R302

i303

A304

3305

306

C307

t308

e309

1310

i311

t312

(313

e 314

t 315

H 316

g 317

d 318

b 319

b 320

h 321

w 322

3 323

324

t 325

g 326

1 327

– 328

m 329

o 330

a 331

( 332

m 333

i 334

w 335

g 336

337

L 338

N 339

i 340

r 341

s 342

3 343

344

( 345

i 346

a 347

2 348

a 349

s 350

t 351

l 352

p 353

4 354

355

a 356

a 357

T 358

a 359

t 360

NC

OR

RE

CTE

ARTICLEBAC 5929 1–9

A. Hernandez-Mendoza et al. / Computatio

epicted below. Additional groups are described in Appendix ATable A.1).

.3. Proteins with broad substrate specificity (AcrB cluster)

A group of three proteins including E. coli AcrB, P. aerugi-osa MexB and Salmonella enterica serovar Typhimurium AcrBs shown in Fig. 1B. These proteins have been described as RND-roteins with broad substrate specificity (Baucheron et al., 2002;i et al., 1995; Sulavik et al., 2001). The MP tree using exclu-ively the 14 residues clustered together these proteins (Fig. 1A).his result indicates that these residues are enough to differenti-te proteins with preference for similar ligands. The 14 analyzedesidues are highly conserved among members of this groups only K29 varies (Table 1). The three experimentally char-cterized proteins bind similar ligands: antibiotics, dyes andetergents (Nikaido and Zgurskaya, 2001), and all might haveery broad substrate specificity (Fig. 1B). The substrate diversityssociated to the group suggests that the observed residues mightarticipate in the major recognition to their ligands. Whereasess conserved residues (such as K29 and alternative residuesot considered here) might have specific ligand-contacts at dif-erent conditions, i.e. they might be exposed in response to theresence of a particular ligand.

.4. Proteins expelling aromatic hydrocarbons (SrpBluster)

The analysis of the cluster (SrpB) demonstrates the accu-acy of the approach described in this work, since five proteinsxtruding aromatic hydrocarbons (AH), such as P. putida TtgHtoluene tolerance) (Rojas et al., 2001), SrpB (solvent-resistantump) (Kieboom and de Bont, 2001), SepB (Phoenix et al.,003), and TtgE proteins (Rojas et al., 2001) (Fig. 1B) were clus-ered together. In addition, P. stutzeri TbtB extrudes both AH andributyltin, an organotin compound (Jude et al., 2004), was clus-ered in this group. Twelve amino acid residues are conservedithin this group. From these residues two positions are cluster

pecific V382 and A385, while D99 and F388 are completelyariable (Table 1), suggesting that these changes are essential forigand selectivity, i.e. they might participate discriminating theirpecific ligands, while conserved residues might be performinghe major contacts with AH. We suggest that new hypotheticalND-proteins whose amino acid sequence patterns will be sim-

lar to the proteins included in this cluster might be extrudingH or related compounds.

.5. Proteins expelling heavy metals (CnrA cluster)

The CnrA cluster (Fig. 1A) includes five proteins (CnrA,zcA, CusA, SilA and MexK) sharing 31.5% amino acid iden-

ity. These proteins bind cations or heavy metals (HM) (Guptat al., 2001; Liesegang et al., 1993; Munson et al., 2000; Nies,

U

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligand-proteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

995), except for the P. aeruginosa MexK, which is involvedn the resistance to tetracycline and to the amphiphilic drugriclosan, a broad-spectrum biocide (Chuanchuen et al., 2002)Fig. 1B). We found that proteins included in this group do not

bgao

PR

OO

F

PRESSiology and Chemistry xxx (2007) xxx–xxx 5

xhibit any conserved residue from the 14 selected, representinghe most divergent cluster identified (Table 1). It is intriguing thatM and MexK proteins are included in the same cluster. We sug-est that MexK might recognize and extrude HM compounds,ue its inclusion in the cluster; however this hypothesis shoulde further explored. In summary, CnrA-proteins might exhibit ainding pocket able to bind molecules with similar size or mightave similar physicochemical properties not discernible in thisork.

.6. Proteins expelling antibiotics (MexD cluster)

Three proteins expelling fluoroquinolones, chloramphenicol,etracycline and aminoglycoside antibiotics were included in thisroup: AdeB and MexD that discharge antibiotics (Gotoh et al.,998; Magnet et al., 2001) and our hypothetical protein Orf2pCidentified in our laboratory – that does not have an experi-entally evident substrate so far (Fig. 1B). The results obtained

ut of this cluster analysis show the predictive potential of thispproach, i.e. tetracycline, aminoglycoside, fluoroquinolonesnorfloxacin or ciprofloxacin) or chemically similar compounds,ight be specifically recognized by Orf2plC. This proposal

s based exclusively in the clustering analysis described here,here AdeB, MexD and Orf2pC are included in a particularroup (Fig. 1A).

In this cluster, only five amino acid positions are conserved:25, D99, N109, F388 and F458 (Table 1). All of them, except109, are located in the central cavity where most of the chem-

cal interactions with the ligands occur. Thus, these conservedesidues could be participating in the specific recognition of theirubstrates.

.7. Proteins expelling unknown compounds (AmeB cluster)

Finally, RagC (B. japonicum), AmeB (A. tumefaciens), RmiBR. etli), and orf2 (R. leguminosarum bv. viciae) were includedn the same cluster. RagC might be involved in heavy met-ls or antibiotics extrusion (Krummenacher and Narberhaus,000), AmeB in extrusion of ß-lactams and detergents (Pengnd Nester, 2001), and RmiB and Orf2 do not exhibit an evidentubstrate. Though these proteins were included in a same clus-er we do not have enough evidences supporting their probableigands, however they might have HM or Ac as substrates orrobably another not considered here.

. Discussion

In the present work, we evidenced that 14 ligand-bindingmino acids, out of approximately 1300 columns from a MSA,re informative enough for ligand prediction in 47 RND proteins.hese residues are distributed throughout the primary aminocid structure in such a way that there is no local sequence motifhat may be associated to a particular binding function. On this

binding prediction in the resistance-nodulation-cell division (RND)7.02.003

asis, several defined groups we made showed specificity for a 361

iven ligand. The PPV in average is equal to 88.0%, suggesting 362

high level of confidence in these predictions. The data here 363

btained not only suggest that the substrate specificity is deter- 364

IN+ModelC

6 nal B

m365

p366

l367

u368

a369

t370

a371

t372

t373

a374

c375

c376

fi377

b378

379

R380

t381

t382

r383

r 384

m 385

p 386

m 387

A 388

389

w 390

f 391

A 392

r 393

394

TC

G

X

M

A

M

A

S

S

E

A

ARTICLEBAC 5929 1–9

A. Hernandez-Mendoza et al. / Computatio

ined by the properties of these 14 residues, but also that therotein fold adapts to accommodate a considerable diversity ofigands, i.e. that each ligand, in association to the protein clusterses, a slightly different subset of binding residues. Thus, anncestral generic ligand-binding domain would have divergedo expel diverse ligands and, as a result, only a particular set ofctive site residues have been conserved throughout evolution. Inhis regard, the AcrB group contains diverse conserved residueshat can be classified in variable or not conserved; conservedre those residues with identical physicochemical properties asompared to AcrB of E. coli and residues conserved in a specificluster. All these residues together correlate to its substrate speci-city, probably by using a discriminatory mechanism mediatedy conserved and cluster-specific residues.

The approach presented here, will help to understand the

UN

CO

RR

EC

TED

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligandproteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

ND protein specificity in bacteria and other organisms, ando decipher the evolution of these proteins in the context ofheir cognate ligands. Finally, this procedure not only describeesidues functionally important which sequence pattern does not

ctHa

able A.1lusters identified by a 14-residues cluster in RND proteins

roup Proteins Main ligand

epB XepB Ethidium bromide, puromyrifampicin

trD MtrD Ethidium bromide, �-lactafatty acids, Triton X-100

crD Escherichia coli AcrD Antibiotics with amphiphilproperties, such as glycosid

Burkholderia cepacia CeoBexW Pseudomonas aeruginosa MexW, MexI Novobiocin

E. coli MdtC Ethidium bromideHaemophilus influenzae AcrB ß-Lactams

meB Bradyrhizobium japonicum RagC Unknown

Rhizobium etli RmeBAgrobacterium tumefaciens AmeBR. leguminosarum bv. viciae Orf2

meE Stenotrophomonas maltophilia SmeE Macrolides

E. coli YhiV Ethidium bromidedeY Serratia marcescens SdeY Macrolides

B. pseudomallei BpeB Isoflavonoid coumestrolA. tumefaciens IfeB

mhB P. putida TtgB, ArpB Chloramphenicol

P. fluorescens EmhB HAß-Lactams

crF E. coli AcrF Detergents (SDS, bile salts

Erwinia amylovora AcrB Dyes (crystal violet, acriflaEnterobacter aerogenes AcrB

F

PRESSiology and Chemistry xxx (2007) xxx–xxx

eflect an evident sequence motif, but also provides diverse ele-ents to be expanded in new hypothetical RND proteins and

otentially to alternative protein families where ligand bindingodifies allosterically the protein.

cknowledgements

AHM was supported by a grant (90288) from CONACyT. Weould like to thank Enrique Merino Perez and Edmundo Calva

or their discussion and comments.

ppendix A. Ligand-binding prediction in theesistance-nodulation-cell division (RND) proteins

Based on the selection and their posterior clustering of 14

PR

OO

-binding prediction in the resistance-nodulation-cell division (RND)7.02.003

olumns involved in ligand interaction, 16 groups were iden- 395

ified in which almost all members share a common substrate. 396

enceforth, we describe the rest of the groups reported in this 397

rticle (Table A.1 and Fig. A.1). 398

Conservedpositions

Notes

cin, Orphan group (Ikeda and Yoshimura, 2002)

ms, Orphan group (Rouquette-Loughlin et al., 2002)

ices

4 CP Elkins and Nikaido (2002) and Nair et al. (2004)

4 CC2 CP MexI has been also associated to quorum

sensing (Aendekerk et al., 2002). MexW, MdtCand AcrB (Li et al., 2003; Nagakubo et al., 2002;Sanchez et al., 1997)

1 CP Proteins of this cluster have been exclusivelyidentified in Rhizobiaceae (Krummenacher andNarberhaus, 2000; Peng and Nester, 2001)

2 CC

5 PC Chang et al. (2004) and Nishino and Yamaguchi(2001)

4 CC3 CP IfeB could have a role in macrolide resistance,

although this possibility has not been explored(Palumbo et al., 1998). SdeY and BpeB (Chan etal., 2004; Chen et al., 2003)

4 CC

5 CP Changes in residues located in the EmhB centralcavity change substrate specificity (Hearn et al.,2006).TtgB, ArpB (Kieboom and de Bont, 2001;Mosqueda and Ramos, 2000)

8 CC

) 5 CP At least two members use similar compounds(Burse et al., 2004; Nishino and Yamaguchi,2001; Pradel and Pages, 2002)

vine) 6 CC

RE

CTE

D P

RO

OF

ARTICLE IN PRESS+ModelCBAC 5929 1–9

A. Hernandez-Mendoza et al. / Computational Biology and Chemistry xxx (2007) xxx–xxx 7

Table A.1 (Continued )

Group Proteins Main ligand Conservedpositions

Notes

SmeB S. maltophilia SmeB Unknown 3 CP Li et al. (2002), Masuda et al. (2000) and Mooreet al. (1999)

P. aeruginosa MexY Aminoglycoside 2 CCB. pseudomallei AmrB Macrolide

SdeB S. marcescens SdeB Fluoroquinolones 2 CP Kohler et al. (1997), Kumar and Worobec(2005), Lin et al. (2002) and Maseda et al. (2000)

Campylobacter jejuni CmeB ß-Lactams 5 CCP. aeruginosa MexF

NolG Sinorhizobium meliloti NolG Novobiocin and deoxycholate(MdtB)

1 CP There is no experimental evidence supportingthe proposed NolG (Baev et al., 1991) functionas a putative lipo-chito-oligosaccharide (Nodfactor) exporter (Saier et al., 1994). Based onthis analysis, we propose that NolG might beinvolved in extrusion of antibiotics such asfluoroquinolones and novobiocin or the bile saltdeoxicholate rather than in theRhizobium-legume symbiosis process

E. coli MdtB 4 CC MdtB (Nagakubo et al., 2002)

Columns are as follow: the first column has a representative protein for the cluster, second to fifth indicate the bacteria and proteins included in the cluster, mainligands associated to the protein cluster, number and location of conserved residues in the cluster (CP, central pore; CC, central cavity), and observations.

Fig. A.1. Mapping of the selected residues on the AcrB structure. (A) PyMol ribbon diagram of the E. coli AcrB multidrug efflux pump (PDB 1IWG), showing ther 09), QR resid

R399

B400

401

402

403

B404

405

406

C407

408

C409

410

411

D 412

413

414

G 415

416

417

418

G 419

420

NC

OResidues selected for this study: L (25), K (29), D (99), D (101), V (105), N (1

ibbon diagram of two AcrB subunits indicating the localization of the selected

eferences

aucheron, S., Imberechts, H., Chaslus-Dancla, E., Cloeckaert, A., 2002. TheAcrB multidrug transporter plays a major role in high-level fluoroquinoloneresistance in Salmonella enterica serovar typhimurium phage type DT204.Microb. Drug Resist. 8, 281–289.

erezin, C., Glaser, F., Rosenberg, J., Paz, I., Pupko, T., Fariselli, P., Casadio, R.,Ben-Tal, N., 2004. ConSeq: the identification of functionally and structurallyimportant residues in protein sequences. Bioinformatics 20, 1322–1324.

U

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligand-proteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

iria, R., Abreu-Goodger, C., Morett, E., Merino, E., 2004. GeConT: genecontext analysis. Bioinformatics 20, 2307–2308.

huanchuen, R., Narasaki, C.T., Schweizer, H.P., 2002. The MexJK efflux pumpof Pseudomonas aeruginosa requires OprM for antibiotic efflux but not forefflux of triclosan. J. Bacteriol. 184, 5036–5044.

H

(112), P (116), V (382), A (385), F (386), F (388), F (458) and F (459). (B)ues into the cavity region. Colors represent selected residues per protomer.

inh, T., Paulsen, I.T., Saier Jr., M.H., 1994. A family of extracytoplasmicproteins that allow transport of large molecules across the outer membranesof gram-negative bacteria. J. Bacteriol. 176, 3825–3831.

otoh, N., Tsujimoto, H., Tsuda, M., Okamoto, K., Nomura, A., Wada,T., Nakahashi, M., Nishino, T., 1998. Characterization of the MexC-MexD-OprJ multidrug efflux system in DeltamexA-mexB-oprM mutants ofPseudomonas aeruginosa. Antimicrob. Agents Chemother. 42, 1938–1943.

upta, A., Phung, L.T., Taylor, D.E., Silver, S., 2001. Diversity of silver resis-tance genes in IncH incompatibility group plasmids. Microbiology 147,

binding prediction in the resistance-nodulation-cell division (RND)7.02.003

3393–3402. 421

earn, E.M., Gray, M.R., Foght, J.M., 2006. Mutations in the central cavity 422

and periplasmic domain affect efflux activity of the resistance-nodulation- 423

division pump EmhB from Pseudomonas fluorescens cLP6a. J. Bacteriol. 424

188, 115–123. 425

ED

IN+ModelC

8 nal B

J426

427

428

J429

430

431

K432

433

434

K435

436

437

L438

439

440

L441

442

443

L444

445

446

M447

448

449

450

M451

452

453

M454

455

456

M457

458

459

N460

461

462

N463

464

P465

466

467

P468

469

P470

471

472

473

P474

475

P476

477

478

479

P480

481

R482

483

484

485

S486

487

488

489

T490

491

492

493

T 494

495

496

497

Y 498

499

500

Y 501

502

503

F 504

A 505

506

507

B 508

509

510

511

B 512

513

514

C 515

516

517

C 518

519

520

521

C 522

523

524

E 525

526

527

528

I 529

530

531

K 532

533

534

535

K 536

537

538

L 539

540

541

L 542

543

544

545

L 546

547

548

M 549

550

551

M 552

553

NC

OR

RE

CT

ARTICLEBAC 5929 1–9

A. Hernandez-Mendoza et al. / Computatio

ohnson, J.M., Church, G.M., 2000. Predicting ligand-binding function infamilies of bacterial receptors. Proc. Natl. Acad. Sci. U.S.A. 97, 3965–3970.

ude, F., Arpin, C., Brachet-Castang, C., Capdepuy, M., Caumette, P., Quentin,C., 2004. TbtABM, a multidrug efflux pump associated with tributyltinresistance in Pseudomonas stutzeri. FEMS Microbiol. Lett. 232, 7–14.

ieboom, J., de Bont, J., 2001. Identification and molecular characterization ofan efflux system involved in Pseudomonas putida S12 multidrug resistance.Microbiology 147, 43–51.

rummenacher, P., Narberhaus, F., 2000. Two genes encoding a putative mul-tidrug efflux pump of the RND/MFP family are cotranscribed with an rpoHgene in Bradyrhizobium japonicum. Gene 241, 247–254.

i, X.Z., Nikaido, H., Poole, K., 1995. Role of mexA-mexB-oprM in antibi-otic efflux in Pseudomonas aeruginosa. Antimicrob. Agents Chemother. 39,1948–1953.

ichtarge, O., Bourne, H.R., Cohen, F.E., 1996. An evolutionary trace methoddefines binding surfaces common to protein families. J. Mol. Biol. 257,342–358.

iesegang, H., Lemke, K., Siddiqui, R.A., Schlegel, H.G., 1993. Characteri-zation of the inducible nickel and cobalt resistance determinant cnr frompMOL28 of Alcaligenes eutrophus CH34. J. Bacteriol. 175, 767–778.

agnet, S., Courvalin, P., Lambert, T., 2001. Resistance-nodulation-celldivision-type efflux pump involved in aminoglycoside resistance in Acine-tobacter baumannii strain BM4454. Antimicrob. Agents Chemother. 45,3375–3380.

unson, G.P., Lam, D.L., Outten, F.W., O’Halloran, T.V., 2000. Identifica-tion of a copper-responsive two-component system on the chromosome ofEscherichia coli K-12. J. Bacteriol. 182, 5864–5871.

urakami, S., Nakashima, R., Yamashita, E., Yamaguchi, A., 2002. Crys-tal structure of bacterial multidrug efflux transporter AcrB. Nature 419,587–593.

urakami, S., Tamura, N., Saito, A., Hirata, T., Yamaguchi, A., 2004.Extramembrane central pore of multidrug exporter AcrB in Escherichia coliplays an important role in drug transport. J. Biol. Chem. 279, 3743–3748.

ies, D.H., 1995. The cobalt, zinc, and cadmium efflux system CzcABC fromAlcaligenes eutrophus functions as a cation-proton antiporter in Escherichiacoli. J. Bacteriol. 177, 2707–2712.

ikaido, H., Zgurskaya, H.I., 2001. AcrAB and related multidrug efflux pumpsof Escherichia coli. J. Mol. Microbiol. Biotechnol. 3, 215–218.

alumbo, J.D., Kado, C.I., Phillips, D.A., 1998. An isoflavonoid-inducible effluxpump in Agrobacterium tumefaciens is involved in competitive colonizationof roots. J. Bacteriol. 180, 3107–3113.

aulsen, I.T., Brown, M.H., Skurray, R.A., 1996. Proton-dependent multidrugefflux systems. Microbiol. Rev. 60, 575–608.

aulsen, I.T., Park, J.H., Choi, P.S., Saier Jr., M.H., 1997. A family of gram-negative bacterial outer membrane factors that function in the export ofproteins, carbohydrates, drugs and heavy metals from gram-negative bacte-ria. FEMS Microbiol. Lett. 156, 1–8.

eng, W.T., Nester, E.W., 2001. Characterization of a putative RND-type effluxsystem in Agrobacterium tumefaciens. Gene 270, 245–252.

hoenix, P., Keane, A., Patel, A., Bergeron, H., Ghoshal, S., Lau, P.C., 2003.Characterization of a new solvent-responsive gene locus in Pseudomonasputida F1 and its functionalization as a versatile biosensor. Environ. Micro-biol. 5, 1309–1327.

oole, K., 2004. Efflux-mediated multiresistance in gram-negative bacteria.Clin. Microbiol. Infect. 10, 12–26.

ojas, A., Duque, E., Mosqueda, G., Golden, G., Hurtado, A., Ramos, J.L.,Segura, A., 2001. Three efflux pumps are required to provide efficienttolerance to toluene in Pseudomonas putida DOT-T1E. J. Bacteriol. 183,3967–3973.

ulavik, M.C., Houseweart, C., Cramer, C., Jiwani, N., Murgolo, N., Greene,J., DiDomenico, B., Shaw, K.J., Miller, G.H., Hare, R., Shimer, G., 2001.Antibiotic susceptibility profiles of Escherichia coli strains lacking mul-

U

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligandproteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

tidrug efflux pump genes. Antimicrob. Agents Chemother. 45, 1126–1136.hompson, J.D., Gibson, T.J., Plewniak, F., Jeanmougin, F., Higgins, D.G.,

1997. The CLUSTAL X windows interface: flexible strategies for multi-ple sequence alignment aided by quality analysis tools. Nucl. Acids Res. 25,4876–4882.

554

M

PR

OO

F

PRESSiology and Chemistry xxx (2007) xxx–xxx

hompson, J.D., Higgins, D.G., Gibson, T.J., 1994. CLUSTAL W: improvingthe sensitivity of progressive multiple sequence alignment through sequenceweighting, position-specific gap penalties and weight matrix choice. Nucl.Acids Res. 22, 4673–4680.

u, E.W., Aires, J.R., McDermott, G., Nikaido, H., 2005. A periplasmic drug-binding site of the AcrB multidrug efflux pump: a crystallographic and site-directed mutagenesis study. J. Bacteriol. 187, 6804–6815.

u, E.W., McDermott, G., Zgurskaya, H.I., Nikaido, H., Koshland Jr., D.E.,2003. Structural basis of multiple drug-binding capacity of the AcrB mul-tidrug efflux pump. Science 300, 976–980.

urther reading

endekerk, S., Ghysels, B., Cornelis, P., Baysse, C., 2002. Characterization ofa new efflux pump, MexGHI-OpmD, from Pseudomonas aeruginosa thatconfers resistance to vanadium. Microbiology 148, 2371–2381.

aev, N., Endre, G., Petrovics, G., Banfalvi, Z., Kondorosi, A., 1991. Six nodula-tion genes of nod box locus 4 in Rhizobium meliloti are involved in nodulationsignal production: nodM codes for d-glucosamine synthetase. Mol. Gen.Genet. 228, 113–124.

urse, A., Weingart, H., Ullrich, M.S., 2004. The phytoalexin-inducible mul-tidrug efflux pump AcrAB contributes to virulence in the fire blight pathogenErwinia amylovora. Mol. Plant Microb. Interact. 17, 43–54.

han, Y.Y., Tan, T.M., Ong, Y.M., Chua, K.L., 2004. BpeAB-OprB, a multidrugefflux pump in Burkholderia pseudomallei. Antimicrob. Agents Chemother.48, 1128–1135.

hang, L.L., Chen, H.F., Chang, C.Y., Lee, T.M., Wu, W.J., 2004. Contributionof integrons, and SmeABC and SmeDEF efflux pumps to multidrug resis-tance in clinical isolates of Stenotrophomonas maltophilia. J. Antimicrob.Chemother. 53, 518–521.

hen, J., Kuroda, T., Huda, M.N., Mizushima, T., Tsuchiya, T., 2003. AnRND-type multidrug efflux pump SdeXY from Serratia marcescens. J.Antimicrob. Chemother. 52, 176–179.

lkins, C.A., Nikaido, H., 2002. Substrate specificity of the RND-type mul-tidrug efflux pumps AcrB and AcrD of Escherichia coli is determinedpredominantly by two large periplasmic loops. J. Bacteriol. 184, 6490–6498.

keda, T., Yoshimura, F., 2002. A resistance-nodulation-cell division familyxenobiotic efflux pump in an obligate anaerobe Porphyromonas gingivalis.Antimicrob. Agents Chemother. 46, 3257–3260.

ohler, T., Michea-Hamzehpour, M., Henze, U., Gotoh, N., Curty, L.K., Pechere,J.C., 1997. Characterization of MexE-MexF-OprN, a positively regulatedmultidrug efflux system of Pseudomonas aeruginosa. Mol. Microbiol. 23,345–354.

umar, A., Worobec, E.A., 2005. Cloning, sequencing, and characterizationof the SdeAB multidrug efflux pump of Serratia marcescens. Antimicrob.Agents Chemother. 49, 1495–1501.

i, X.Z., Zhang, L., Poole, K., 2002. SmeC, an outer membrane multidrug effluxprotein of Stenotrophomonas maltophilia. Antimicrob. Agents Chemother.46, 333–343.

i, Y., Mima, T., Komori, Y., Morita, Y., Kuroda, T., Mizushima, T., Tsuchiya,T., 2003. A new member of the tripartite multidrug efflux pumps, MexVW-OprM, in Pseudomonas aeruginosa. J. Antimicrob. Chemother. 52, 572–575.

in, J., Michel, L.O., Zhang, Q., 2002. CmeABC functions as a multidrugefflux system in Campylobacter jejuni. Antimicrob. Agents Chemother. 46,2124–2131.

aseda, H., Yoneyama, H., Nakae, T., 2000. Assignment of the substrate-selective subunits of the MexEF-OprN multidrug efflux pump ofPseudomonas aeruginosa. Antimicrob. Agents Chemother. 44, 658–664.

asuda, N., Sakagawa, E., Ohya, S., Gotoh, N., Tsujimoto, H., Nishino, T.,2000. Contribution of the MexX-MexY-oprM efflux system to intrinsicresistance in Pseudomonas aeruginosa. Antimicrob. Agents Chemother. 44,

-binding prediction in the resistance-nodulation-cell division (RND)7.02.003

2242–2246. 555

oore, R.A., DeShazer, D., Reckseidler, S., Weissman, A., Woods, D.E., 1999. 556

Efflux-mediated aminoglycoside and macrolide resistance in Burkholderia 557

pseudomallei. Antimicrob. Agents Chemother. 43, 465–470. 558

IN+ModelC

nal B

M559

560

561

N562

563

564

565

N566

567

568

N569

570

P 571

572

573

R 574

575

576

577

S 578

ARTICLEBAC 5929 1–9

A. Hernandez-Mendoza et al. / Computatio

osqueda, G., Ramos, J.L., 2000. A set of genes encoding a second tolueneefflux system in Pseudomonas putida DOT-T1E is linked to the tod genesfor toluene metabolism. J. Bacteriol. 182, 937–943.

agakubo, S., Nishino, K., Hirata, T., Yamaguchi, A., 2002. The putativeresponse regulator BaeR stimulates multidrug resistance of Escherichiacoli via a novel multidrug exporter system, MdtABC. J. Bacteriol. 184,4161–4167.

air, B.M., Cheung Jr., K.J., Griffith, A., Burns, J.L., 2004. Salicylate induces

UN

CO

RR

EC

TED

Please cite this article in press as: Hernandez-Mendoza, A. et al., Ligand-proteins, Computat. Biol. Chem. (2007), doi:10.1016/j.compbiolchem.200

an antibiotic efflux pump in Burkholderia cepacia complex genomovar III(B. cenocepacia). J. Clin. Invest. 113, 464–473.

ishino, K., Yamaguchi, A., 2001. Analysis of a complete library of puta-tive drug transporter genes in Escherichia coli. J. Bacteriol. 183, 5803–5812.

S

PRESSiology and Chemistry xxx (2007) xxx–xxx 9

radel, E., Pages, J.M., 2002. The AcrAB-TolC efflux pump contributes tomultidrug resistance in the nosocomial pathogen Enterobacter aerogenes.Antimicrob. Agents Chemother. 46, 2640–2643.

ouquette-Loughlin, C., Stojiljkovic, I., Hrobowski, T., Balthazar, J.T., Shafer,W.M., 2002. Inducible, but not constitutive, resistance of gonococci tohydrophobic agents due to the MtrC-MtrD-MtrE efflux pump requires TonB-ExbB-ExbD proteins. Antimicrob. Agents Chemother. 46, 561–565.

aier Jr., M.H., Tam, R., Reizer, A., Reizer, J., 1994. Two novel families of

PR

OO

F

binding prediction in the resistance-nodulation-cell division (RND)7.02.003

bacterial membrane proteins concerned with nodulation, cell division and 579

transport. Mol. Microbiol. 11, 841–847. 580

anchez, L., Pan, W., Vinas, M., Nikaido, H., 1997. The acrAB homolog of 581

Haemophilus influenzae codes for a functional multidrug efflux pump. J. 582

Bacteriol. 179, 6855–6857. 583