Targeting virulence: salmochelin modification tunes the antibacterial activity spectrum of β-lactams for pathogen-selective killing of Escherichia coli

New antibiotics are required to treat bacterial infections and counteract the emergence of antibiotic resistance.


Introduction
Bacterial infections, the rise in antibacterial resistance in hospital and community settings, and the paucity of new antibiotics in the current drug pipeline create a worldwide public health crisis. 1,2 New strategies to diagnose and treat bacterial infections as well as counteract the emergence and spread of antibiotic resistance in bacterial pathogens are urgently needed to reduce morbidity and mortality, as well as the economic burden, caused by these infections. 3,4 The discovery of narrowspectrum antibiotics that target select pathogens is one important and necessary facet of this large and complex problem. 2,5,6 Pathogen-specic antibiotics that minimally perturb the normal microbial ora are expected to reduce the likelihood of colonisation by pathogenic and drug-resistant microbes during or aer antibiotic treatment, and prevent lifethreatening secondary infections such as those caused by Clostridium difficile. 5,7 Moreover, the availability of narrowspectrum antibiotics, coupled with rapid diagnostics, is expected to reduce the use of broad-spectrum therapeutics and thereby slow down the evolution of drug resistance. 2,5,7 Among current and emerging microbial threats, Gram-negative bacteria, including pathogenic Escherichia coli, Klebsiella pneumoniae, Acinetobacter baumannii, and Salmonella spp., pose a challenge for antibiotic drug discovery. 1,2,8 These strains have an outer membrane that serves as a permeability barrier and prevents the cellular entry of many antibiotics. 9 In this work, we report a stealth antibiotic delivery strategy that overcomes the outer membrane permeability barrier of Gram-negative E. coli and targets pathogenicity by hijacking the iron import machinery utilised by virulent strains during colonisation in the mammalian host.
Iron is an essential nutrient for almost all bacterial pathogens. 10,11 Because iron exhibits low solubility in aqueous solutions at physiological pH and enables Fenton chemistry, the levels of "free" iron in mammals (ca. 10 À24 M in serum) 12 are tightly regulated by homeostatic mechanisms, which include the expression of the iron transport and storage proteins transferrin and ferritin. 13 Most bacterial pathogens require micromolar concentrations of iron to colonise and cause disease, and bacterial iron acquisition machineries contribute to virulence. 10,14 One way that bacteria scavenge iron in the host environment is to biosynthesize and export siderophores, secondary metabolites that chelate Fe(III) with high affinity. 15 The ferric siderophores are recognised and transported into the cell by dedicated uptake machinery. In this work, we consider the catecholate siderophore enterobactin 1 (Ent, Fig. 1a), its glucosylated congeners 2-4 (GlcEnt, Fig. 1a), and the outer membrane receptors for these iron chelators. Ent is biosynthesized by all E. coli and the ferric complex is transported across the outer membrane by the TonB-ExbB-ExbD-dependent outer membrane receptor FepA (Fig. 1b). 12 In addition to Ent, many pathogenic E. coli as well as Salmonella spp. biosynthesize salmochelins, C-glucosylated derivatives of Ent (Fig. 1a). 16 The iroA gene cluster (iroBCDEN) 14,17,18 encodes enzymes that tailor the Ent scaffold to provide the salmochelins (IroBDE), and proteins for salmochelin transport (IroCN). Expression of genes encoded by the iroA locus contributes to virulence by providing Gram-negative pathogens with additional iron acquisition machinery and enabling the pathogens to overcome the host innate immune response. 19,20 In the battle against such invading pathogens, the mammalian host mounts a metal-withholding response and secretes lipocalin-2 (Lcn2), a 22 kDa antimicrobial protein that captures ferric Ent. 19,21,22 Gram-negative pathogens that utilise salmochelins for iron acquisition readily evade this innate immune mechanism because the salmochelins cannot be sequestered by Lcn2. 19 Because bacteria utilise siderophores to acquire nutrient iron during infection, these molecules, as well as the corresponding biosynthetic and transport machineries, provide opportunities for antibiotic development. 10,11,[23][24][25][26][27] The notion of using siderophores or siderophore mimics to deliver antibacterial cargo into bacterial cells has garnered attention over several decades. [28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43][44] Our approach to siderophore-based targeting focuses on harnessing native siderophore platforms used by pathogens in the human host for cargo delivery, and we seek to modify these scaffolds in ways that minimally perturb iron binding and receptor recognition. We have designed and utilised a monofunctionalized Ent platform to assemble a variety of Ent-cargo conjugates, and we reported that the Ent uptake machinery (FepABCDG) provides a means to transport smallmolecule cargo, including antibiotics in clinical use, into E. coli. 44,45 For instance, the Ent-b-lactam conjugates 5 and 6 ( Fig. 1c) target and kill E. coli expressing FepA. 44 Because all E. coli use Ent for iron acquisition, the Ent-b-lactam conjugates target and kill both non-pathogenic and pathogenic E. coli strains. Some E. coli are commensal microbes, comprising <1% of the total microbial community in the human gut, that biosynthesize vitamin K that is needed by the host. 46,47 Thus, the ability to target pathogenic E. coli has utility for minimally perturbing the normal ora. Inspired by prior investigations of native siderophore transport, 16,48 we hypothesised that salmochelin-antibiotic conjugates will be specically recognised by IroN, the outer membrane receptor for the salmochelins, and afford a strategy for overcoming the outer membrane permeability barrier, achieving narrow-spectrum antibacterial activity against pathogenic E. coli, and evading capture by Lcn2.
In this work, we report the design and chemoenzymatic preparation of siderophore-b-lactam conjugates based on salmochelin platforms, and demonstrate targeting of b-lactam antibiotics to pathogenic E. coli that harbour the iroA gene cluster and express IroN. Salmochelin-inspired GlcEnt-b-lactam conjugates based on the diglucosylated Ent (DGE, Fig. 1a) platform provide selective antibacterial activity against pathogenic E. coli and up to 1000-fold enhanced potency relative to the parent b-lactam antibiotics. Moreover, the salmochelininspired conjugates remain antibacterial in the presence of Lcn2. These investigations establish a chemoenzymatic route to functionalized salmochelins and provide a new approach for transforming a broad-spectrum antibiotic in clinical use into a narrow-spectrum therapeutic that targets microbial pathogens on the basis of siderophore receptor expression.

Results and discussion
Design and syntheses of GlcEnt-b-lactam conjugates We present a family of salmochelin-inspired GlcEnt-b-lactam conjugates 7-10 that exhibit ampicillin (Amp) or amoxicillin (Amx) attached to either monoglucosylated Ent (MGE, 2) or diglucosylated Ent (DGE, 3) by a stable polyethylene glycol (PEG 3 ) linker. The design of the GlcEnt-b-lactam conjugates 7-10 builds upon Ent-Amp/Amx 5 and 6 ( Fig. 1C). 44 These conjugates are based on a monosubstituted Ent platform where one catechol moiety is modied at the C5 position for cargo attachment. We sought to install glucose moieties at the C5 position of one or both of the unfunctionalized catechol rings to afford MGE-Amp/Amx 7 and 8 and DGE-Amp/Amx 9 and 10, respectively. Although the total chemical syntheses of salmochelins have been reported, nine steps are required to achieve the requisite glucosylated 2,3-dihydroxybenzoic acid building block. 49 We therefore established a chemoenzymatic approach that employs enzymes involved in salmochelin biosynthesis, which affords the desired glucosylated conjugates and requires only one additional step compared to the reported preparation of Ent-Amp/Amx.
IroB and MceC are C-glucosyltransferases that catalyse Cglucosylation of Ent at the C5 positions of the catechol rings.
MceC is encoded by the MccE492 gene cluster of K. pneumoniae RYC492, and has 75% amino acid sequence identity with IroB. 50 IroB catalyses up to three C-glucosylation events, affording MGE, DGE and TGE as products (Fig. 1a). 51 MceC, in contrast, produces only MGE and DGE. 52 On the basis of these observations, we hypothesised that both IroB and MceC would accept monofunctionalized Ent as a substrate, providing a preparative route to 7-10. Initial activity assays where either IroB or MceC was incubated with Ent-PEG 3 -N 3 11, UDP-Glc, and Mg(II) revealed that both enzymes accept Ent-PEG 3 -N 3 11 as a substrate and afford MGE-PEG 3 -N 3 12 and DGE-PEG 3 -N 3 13 as products ( Fig. S1 and S2 †). Accumulation of 12 was observed in the MceC-catalysed reactions, whereas 13 accumulated in reactions catalysed by IroB. When Ent-Amp/Amx 5 and 6 were employed as substrates, complex product mixtures were obtained. LC/MS analysis of the mixtures revealed the desired products as well as multiple byproducts, including products of b-lactam decomposition. We therefore performed large-scale Cglucosylation reactions employing Ent-PEG 3 -N 3 11 as a substrate to afford milligram quantities of MGE-PEG 3 -N 3 12, and DGE-PEG 3 -N 3 13 (Scheme 1). We subsequently employed copper-catalysed azide/alkyne cycloaddition to install the blactam moieties (Scheme 1). 44 This route achieved the monoand diglucosylated conjugates 7-10 in high purity and in yields of 26-59% from 7 following HPLC purication. As expected, the GlcEnt-b-lactam conjugates 7-10 bind iron. 53 Each Fe(III) complex exhibits a broad absorption band (ca. 400-700 nm, MeOH) characteristic of ferric Ent and its derivatives (Fig. S3 †). 44,45,54 DGE-b-lactam conjugates target pathogenic E. coli expressing IroN To evaluate whether GlcEnt-Amp/Amx 7-10 target pathogenic E. coli expressing IroN, we compared the antibacterial activities of the parent antibiotics Amp/Amx, Ent-Amp/Amx 5 and 6, and GlcEnt-Amp/Amx 7-10. We selected ve E. coli strains on the basis of siderophore receptor expression (Table S1 †). E. coli CFT073 55 and UTI89 56 harbour the iroA gene cluster, biosynthesize and utilise salmochelins for iron acquisition in the host, and cause urinary tract infections. 57,58 In contrast, E. coli H9049 is a clinical isolate that does not have the iroA cluster. 22 E. coli K-12 59 and E. coli B 60 are non-pathogenic laboratory strains that also lack the iroA cluster. To ascertain the effect of iron limitation on antibacterial activity, we performed antibacterial activity assays in the absence or presence of the metal-ion chelator 2,2 0 -dipyridyl (DP, 200 mM). This concentration of DP inhibits E. coli growth (Fig. S4 †). These assays revealed that DGE-Amp/Amx 9 and 10 target pathogenic E. coli that express IroN.
Amp/Amx exhibit minimum inhibitory concentration (MIC) values of 10 mM against the ve E. coli strains (AEDP, Fig. 2 and S5-S9 †). Under conditions of iron limitation, Ent-Amp/Amx provide 100-to 1000-fold enhanced activity against all ve strains (50% MHB, +DP). These results are in agreement with our prior studies of Ent-Amp/Amx killing of E. coli. 44 Glucosylation affords strain-dependent antimicrobial activity that correlates with IroN expression ( Fig. 2a and b, S5 and S6 †). Like Ent-Amp/Amx 5 and 6, GlcEnt-Amp/Amx 7-10 provide 100-and 1000-fold enhanced antimicrobial activity against E. coli UTI89 and E. coli CFT073, respectively (+DP). The susceptibility of E. coli CFT073 to GlcEnt-Amp/Amx remains enhanced in the absence of DP, as observed previously for Ent-Amp/Amx. 44 The antibacterial activity of GlcEnt-Amp/Amx 7-10 against E. coli H9049, K-12, and B is attenuated relative to that of Ent-Amp/ Amx 5 and 6 (+DP, Fig. 2c-e and S7-S9 †). Moreover, for these non-pathogenic strains, the MIC values of (Glc)Ent-Amp/Amx follow the trend Ent-Amp/Amx < MGE-Amp/Amx < DGE-Amp/ Amx. The MGE modication provides enhanced potency relative to Amp/Amx because growth reduction (K-12) or complete growth inhibition (H9049 and B) occurs at 1 mM MGE-Amp/Amx (+DP). In contrast, the DGE-b-lactam conjugates exhibit negligible antibacterial activity against the three strains that lack IroN (MIC > 10 mM). The growth medium contains z4 mM iron (Table S2 †) and we attribute the growth inhibition observed at 10 mM DGE-Amp/Amx to iron deprivation that results from DGE-Amp/Amx sequestering the iron in the growth medium (Table S2 †).
In the antibacterial activity assays described above, we treated the bacterial cultures with the apo conjugates and expected that the siderophore moieties chelate iron from the growth medium, allowing for recognition of the ferric-siderophore complexes by FepA and IroN. We previously reported that preloading of Ent-Amp/Amx with Fe(III) prior to antibacterial activity assays against E. coli K-12 had negligible effect on the MIC value. 44 Here we report that preloading of MGE-Amp/ Amx and DGE-Amp/Amx also has a negligible effect on the growth inhibitory properties (Fig. S10 †). This result is expected given that the concentration of iron in the growth medium far exceeds the MIC values obtained for the conjugates under conditions where FepA and IroN are expressed. Lastly, mixtures of unmodied Amp/Amx and (Glc)Ent 1-3 against E. coli CFT073 and UTI89 provide the same MIC values as Amp/Amx alone and conrm that the enhanced antibacterial activity of (Glc)Ent-Amp/Amx 5-10 requires the covalent attachment of blactams to the siderophore scaffolds ( Fig. S11 and S12 †).
Siderophore modication accelerates killing of pathogenic E. coli CFT073 E. coli CFT073 is rapidly killed by conjugates 5-10 ( Fig. 3a and S13 †). The OD 600 value of E. coli CFT073 culture (10 8 CFU mL À1 ) treated with 5 mM (Glc)Ent-b-lactam is reduced to almost the baseline value (z0.04) aer 1 h, which corresponds to a z2-fold log reduction in CFU mL À1 , whereas the change in OD 600 and CFU mL À1 for E. coli CFT073 treated with 50 mM Amp/Amx is negligible over this time period. In contrast, siderophore modication has negligible effect on the time-kill kinetics observed for E. coli UTI89 ( Fig. 3b and S14 †); the (Glc) Ent-b-lactam conjugates provide similar proles as observed for Amp/Amx. This result is reminiscent of our prior observations for E. coli K-12 where attachment of Ent to Amp/Amx provided only a modest increase in the time-kill kinetics compared to the parent antibiotics. 44 The origin of this straindependence is unclear and warrants further investigation. Nevertheless, these data show that glucosylation of Ent-Amp/ Amx does not alter the time-kill kinetics of Ent-Amp/Amx for either E. coli CFT073 or UTI89, and (Glc)Ent-Amp/Amx 5-10 kill CFT073 more rapidly than UTI89.

Siderophore competition supports recognition of (Glc)Ent-blactam conjugates by IroN
To investigate the interactions between (Glc)Ent-b-lactam conjugates 5-10 and the siderophore receptors FepA and IroN of E. coli CFT073 and UTI89, we performed modied antimicrobial activity assays where varying concentrations (0-10 mM) of (Glc)Ent 1-3 were combined with 100 nM (Glc)Ent-Amp/Amx 5-10 ( Fig. 4 and S15 †). These mixtures provide a means to probe competition between exogenous native siderophores and the conjugates for receptor recognition because siderophore uptake of the former molecules results in growth promotion whereas the latter afford growth inhibition. The competition assays establish that Ent and MGE attenuate the antibacterial activity of all (Glc)Ent-b-lactam conjugates 5-10 to varying degrees, whereas DGE only inhibits the activity of the glucosylated congeners 7-10. Moreover, DGE fully attenuates DGE-Amp/Amx 9-10 but not MGE-Amp/Amx 7 and 8. These conclusions are drawn from the following observations: (i) a 100-fold molar excess of Ent recovers the growth of E. coli CFT073 treated with Ent/MGE-Amp/Amx 5-8 to levels comparable to that of the untreated control ( Fig. 4a and S15a †). In contrast, a 100-fold excess of Ent provides only partial growth recovery of E. coli CFT073 treated with DGE-Amp/Amx 9 and 10. (ii) A 100-fold excess of MGE fully recovers the growth of E. coli CFT073 treated with (Glc)Ent-Amp/Amx 5-10 ( Fig. 4b and S15b †). (iii) A 100-fold molar excess of DGE does not recover the growth of E. coli CFT073 treated with Ent-Amp/Amx 5 and 6, whereas it provides partial and complete growth recovery of E. coli CFT073 treated with MGE-Amp/Amx 7 and 8 and DGE-Amp/Amx 9 and 10, respectively ( Fig. 4c and S15c †). In total, this work indicates that FepA recognises and delivers Ent/MGE-Amp/Amx 5-8 but not DGE-Amp/Amx 9 and 10, whereas IroN binds and transports all conjugates based on the three siderophore scaffolds. Competition assays employing E. coli UTI89 afford overall trends that are similar to those observed for E. coli CFT073 except that lower concentrations of exogenous siderophores effectively block the antibacterial action of (Glc)Ent-Amp (Fig. 4d-f and S15d-f †).
GlcEnt-Amp/Amx kill pathogenic E. coli in the presence of other microbes that include non-pathogenic E. coli and commensal Lactobacilli To further probe the activity spectrum and investigate strain selectivity of GlcEnt-Amp/Amx, we performed mixed-species Time-kill kinetics of (Glc)Ent-Amp against E. coli CFT073 and UTI89. (a and b) Time-kill kinetics of (Glc)Ent-Amp 5/7/9 against (a) uropathogenic E. coli CFT073 (z10 8 CFU mL À1 ) treated with 50 mM Amp or 5 mM (Glc)Ent-Amp 5/7/9 and (b) uropathogenic E. coli UIT89 (z10 8 CFU mL À1 ) treated with 50 mM Amp or 50 mM (Glc)Ent-Amp 5/ 7/9. All assays were performed in 50% MHB medium supplemented with 200 mM DP (T ¼ 37 C) (mean AE standard deviation, n $ 3). The data for (Glc)Ent-Amx 6/8/10 and for the assays performed in the absence of DP are presented in Fig. S13-S14. † assays to determine whether these conjugates will selectively kill pathogenic E. coli that express IroN cultured in the presence of other organisms. These assays conrmed that GlcEnt-Amp/Amx 7-10 selectively kill pathogenic E. coli that express IroN in the presence of E. coli strains that do not express this receptor. Treatment of co-cultures of pathogenic E. coli (CFT073 or UTI89, transformed with the chloramphenicol resistance plasmid pSG398) and non-pathogenic E. coli K-12 (transformed with the kanamycin resistance plasmid pET29a) with 100 nM Ent-Amp/ Amx 5 and 6 results in complete killing of both strains (Fig. 5a-d and S16a-d †). In contrast, treatment of the co-cultures with 100 nM GlcEnt-Amp/Amx 7-10 affords killing of the uropathogenic E. coli concomitant with E. coli K-12 survival (Fig. 5a-d and S16ad †). Taken together, these results demonstrate that GlcEnt-blactam conjugates 7-10 provide strain-specic targeting of the antibacterial cargo to virulent E. coli that express IroN.
(Glc)Ent-Amp/Amx 5-10 also target pathogenic E. coli in the presence of commensal microbes. Lactobacilli are Gram-positive commensal bacteria of the human gastrointestinal tract, and are also found in the urinary and genital tracts. 61 Some Lactobacilli reduce recurrent urinary tract infections in women. 62 Lactobacilli have little-to-no minimal metabolic iron requirement, and do not employ enterobactin or salmochelins for iron acquisition. 63,64 Lactobacillus rhamnosus GG (ATCC 53103), a human commensal that is considered to be a probiotic, is susceptible to b-lactam antibiotics, and we obtained a MIC value of 10 mM for Amp/Amx against this strain (1 : 1 MRS/ MHB medium, AEDP) (Fig. S17 †). In contrast, 10 mM (Glc)Ent-Amp/Amx 5-10 have negligible effect on L. rhamnosus GG growth (Fig. S17 †). Treatment of E. coli CFT073 and L. rhamnosus GG co-cultures with (Glc)Ent-Amp/Amx 5-10 affords selective killing of E. coli CFT073 ( Fig. 5e and f, S16e and f †).
We previously reported that modication of Amp/Amx with Ent attenuated the activity of the b-lactam against Staphylococcus aureus ATCC 25923. 44 In the current work we obtained a similar result with GlcEnt-Amp/Amx, and found that the salmochelin modication lowers the antibacterial activity of Amp/ Amx against S. aureus by 10-fold (Fig. S18 †). Moreover, treatment of E. coli CFT073 and S. aureus co-cultures with DGE-Amp/ Amx 9,10 affords selective killing of E. coli CFT073 ( Fig. S19a and b, S20a and b †). Selective killing of E. coli CFT073 cocultured with Acinetobacter baumannii ATCC 17961 also coli CFT073 with UTI89 in these assays afforded similar selectivity trends (Fig. S22 and S23 †). In total, the mixed-species assays provide support for DGE-based targeting of the antibacterial cargo to IroN-expressing strains.

GlcEnt-Amp/Amx kill E. coli in the presence of lipocalin-2
To ascertain whether GlcEnt-Amp/Amx 7-10 overcome Lcn2 sequestration, in analogy to Lcn2 evasion by the salmochelins, 14,19 we conducted antibacterial assays with E. coli CFT073 in the absence or presence of Lcn2 or bovine serum albumin (BSA, control). These assays were conducted in modied M9 medium, 65 and 100 nM (Glc)Ent-Amp/Amx 5-10 provide complete growth inhibition of E. coli CFT073 in this medium (Fig. 6). A 10-fold excess of Lcn2 attenuates the antibacterial activity of Ent-Amp/Amx 5 and 6, in agreement with prior work. 44 In contrast, Lcn2 has negligible effect on the antimicrobial activity of GlcEnt-Amp/Amx 7-10 against E. coli CFT073 ( Fig. 6 and S24 †).

GlcEnt-Amp/Amx exhibit low cytotoxicity to human T84 cells
The cytotoxicity of apo and iron-bound GlcEnt-Amp/Amx 7-10 (#10 mM) against human T84 colon epithelial cells was evaluated by the MTT assay. In all cases, the cell viability was $80% of that of the untreated control, indicating that the conjugates exhibit negligible cytotoxicity to T84 cells over a 24 h period (Fig. S25 †).

Conclusions
In this work, inspired by the siderophore recognition strategies utilised by E. coli for iron acquisition in the host, we report a siderophore-based approach for antibiotic delivery that targets strains that express IroN, a siderophore receptor that contributes to virulence. First, we establish that the tailoring enzymes IroB and MceC can C-glucosylate monofunctionalized Ent and therefore be employed in chemoenzymatic synthesis to afford functionalized salmochelins. Next, we demonstrate that GlcEnt-b-lactam conjugates are recognised by siderophore transport machinery, target IroN, provide $100-fold enhanced antibacterial activity against uropathogenic E. coli relative to the parent b-lactams, afford killing of virulent E. coli in the presence of non-pathogenic E. coli and other commensal strains, and overcome the enterobactin-sequestering host-defense protein Lcn2. Our results establish that conjugation of a broad-spectrum antibiotic to a siderophore tunes the activity prole of the parent antibiotic. With the appropriate choice of siderophore, the antibacterial activity spectrum can be modulated to afford species-and strain-specic targeting. In broad terms, targeting pathogens is important for pharmaceutical development, which will ultimately afford treatment options that minimally perturb the commensal microbiota. 66,67 IroN was rst discovered in Salmonella 18 and subsequently identied in other Enterobacteriaceae. Our current work focuses on antibiotic delivery to uropathogenic E. coli that harbour the iroA gene cluster, and we expect that this strategy will be applicable to other pathogens that employ salmochelins for iron acquisition. At present, 121 completely sequenced E. coli genomes are available, which include 46 human pathogens. A BLAST search using iroN from E. coli CFT073 afforded hits with $99% sequence identity for three uropathogenic E. coli (UTI89, 536, and 83792), adherent invasive E. coli UM146, the meningitis isolate E. coli IHE3034, and a carbapenemase-producing isolate E. coli ECONIH1 (Table S4 †). The probiotic E. coli Nissle 1917 and the laboratory reference strain for antimicrobial testing E. coli ATCC 25922 were the only other E. coli revealed as hits. Studies of the distribution of siderophore biosynthetic machinery in E. coli isolated from feces of healthy mammals indicate that z20% of the commensal isolates produce salmochelins. 68 This observation suggests that one potential limitation of GlcEntbased antibiotic delivery is that a fraction of commensal E. coli harbour the iroA cluster are susceptible and, conversely, that some pathogenic E. coli do not. Regarding the former possibility, the healthy gut is considered to be a reservoir for E. coli that cause infections of the urinary tract, 58,69-71 and the ability to target such pathobionts using siderophores may be advantageous in certain cases. In addition to Salmonella and E. coli, BLAST revealed that the genomes of the human pathogens Shigella dysenteriae 1617 and Sd197, Enterobacter cloacae, Klebsiella pneumoniae, and Enterobacter aerogenes encode iroN (Table S4 †). Thus, it will be informative to determine whether DGE also provides targeted antibiotic delivery to these problematic strains.
Our current investigations also provide fundamental insights into siderophore recognition and transport. Prior studies of siderophore uptake in Salmonella revealed that both FepA and IroN recognise and transport Ent. 72 Our competition assays employing uropathogenic E. coli are in agreement with this observation, and indicate that both receptors deliver Ent-Amp/Amx 5 and 6 into E. coli. Moreover, our competition data suggest that MGE 2 and MGE-Amp/Amx 7 and 8 are recognised and transported by FepA as well as IroN of E. coli. In contrast, DGE 3 only competes with GlcEnt-Amp/ Amx 7-10 and most effectively blocks the activity of DGE-Amp/Amx 9 and 10. These observations support exclusive transport of DGE-Amp/Amx 9 and 10 through IroN. Indeed, prior studies demonstrated that IroN is required for transporting salmochelin extracts isolated from several S. enterica strains, 16 and in vitro activity assays reveal that IroB accumulates DGE 3. 51 We previously reported that E. coli CFT073 exhibits greater sensitivity to Ent-Amp/Amx 5 and 6 than E. coli UTI89, 44 and we observe the same trend with GlcEnt-Amp/Amx 7-10. The physiological origins of this observation remain unclear. One possible explanation may be differences in the siderophore biosynthetic and uptake machineries employed by these two uropathogens. E. coli CFT073 expresses a third catecholate siderophore receptor, IhA, 73 whereas E. coli UTI89 biosynthesizes yersiniabactin, a siderophore mainly used by Yersinia spp. 74 Alternatively, as-yet unidentied factors may account for these trends, and further studies are warranted to understand these observations.
In closing, this investigation establishes that siderophores and the siderophore uptake machinery employed by virulent bacteria provide a powerful approach for targeting pathogenesis in the context of antibacterial drug discovery. Narrow-spectrum and species-specic antibiotics are needed for treating infections where the causative agent is known and, when coupled with rapid diagnostics, will ultimately reduce the onset of secondary infections and evolution of antibiotic resistance. 2,5,7 The current study focuses on targeting broad-spectrum b-lactam antibiotics to pathogenic E. coli on the basis of iron acquisition machinery that is employed by these pathogens during colonisation in the host. We establish that native salmochelins can be used as scaffolds for "Trojan horse" antibiotic delivery to hijack the iron acquisition machinery that contributes to pathogenicity. It will be important to ascertain whether this salmochelin-inspired strategy is applicable to other Gramnegatives, such as Salmonella and K. pneumoniae, which cause human disease and utilise salmochelins for iron acquisition. Leveraging this strategy to target other antibacterial cargos and thereby modulate activity and mitigate off-target effects is another important avenue of future chemical and biological investigation.

Instrumentation
Analytical and semi-preparative high-performance liquid chromatography (HPLC) were performed using an Agilent 1200 series HPLC system outtted with a Clipeus reverse-phase C18 column (5 mm pore size, 4.6 Â 250 mm; Higgins Analytical, Inc.) at a ow rate of 1 mL min À1 and an Agilent Zorbax reverse-phase C18 column (5 mm pore size, 9.4 Â 250 mm) at a ow rate of 4 mL min À1 , respectively. The multi-wavelength detector was set to read the absorbance at 220, 280, and 316 (catecholate absorption) nm. HPLC-grade acetonitrile (MeCN) and tri-uoroacetic acid (TFA) were purchased from EMD and Alfa Aesar, respectively. For HPLC analyses, solvent A was 0.1% TFA/ H 2 O and solvent B was 0.1% TFA/MeCN, unless stated otherwise. The HPLC solvents were prepared with HPLC-grade MeCN and TFA, and Milli-Q water (18.2 mU cm), and ltered through a 0.2 mm lter before use. For analytical HPLC to evaluate conjugate purity, the entire portion of each HPLC-puried compound was dissolved in a mixture of 1 : 1 MeCN/H 2 O and an aliquot was taken for HPLC analysis. The remaining solution was subsequently lyophilized.
High-resolution mass spectrometry was performed by using an Agilent LC-MS system comprised of an Agilent 1260 series LC system outtted with an Agilent Poroshell 120 EC-C18 column (2.7 mm pore size) and an Agilent 6230 TOF system housing an Agilent Jetstream ESI source. For all LC-MS analyses, solvent A was 0.1% formic acid/H 2 O and solvent B was 0.1% formic acid/ MeCN (LC-MS grade, Sigma-Aldrich). The samples were analysed using a solvent gradient of 5-95% B over 10 min with a ow rate of 0.4 mL min À1 . The MS proles were analysed and deconvoluted by using Agilent Technologies Quantitative Analysis 2009 soware version B.03.02.
Optical absorption spectra were recorded on a Beckman Coulter DU800 spectrophotometer (1 cm quartz cuvettes, Starna). A BioTek Synergy HT plate reader was used to record absorbance at 600 nm (OD 600 ) for antimicrobial activity assays and absorbance at 550 nm for cytotoxicity assays.

Synthesis of DGE-Amx (10)
As described for MGE-Amp with the exception that Amx-alkyne 15 and DGE-PEG 3 -N 3 13 were used instead of Amp-alkyne 14 and MGE-PEG 3 -N 3 12. All (Glc)Ent 1-3 and siderophore-antibiotic conjugates 5-10 were stored as DMSO stock solutions at À20 C. The stock solution concentrations for (Glc)Ent-Amp/Amx 5-10 ranged from 2 to 5 mM. These values were determined by diluting the DMSO stock solution in MeOH and using the reported extinction coefficient for enterobactin in MeOH (316 nm, 9500 M À1 cm À1 ). 76 To minimize multiple freeze-thaw cycles, the resulting solutions were divided into 50 mL aliquots and stored at À20 C. The b-lactam moieties and enterobactin trilactone are susceptible to hydrolysis, and aliquots were routinely analysed by HPLC to conrm the integrity of the samples.

General microbiology materials and methods
Information pertaining to all bacterial strains used in this study is listed in Table S1. † Freezer stocks of all Escherichia coli strains (E. coli K-12, B, H9049, CFT073, and UTI89), Staphylcoccus aureus ATCC 25923, and Acinetobacter baumannii ATCC 17961 were prepared from single colonies in 25% glycerol/Luria Broth (LB) medium. Freezer stocks of Lactobacillus rhamnosus GG ATCC 53103 were prepared from single colonies in 25% glycerol/de Man, Rogosa, and Sharpe (MRS) medium.
LB, MRS, 5Â M9 minimal medium and agar were purchased from BD. Mueller Hinton Broth (MHB) was purchased from Fluka. Recombinant human Lcn2 was purchased from R&D System (Minneapolis, MN). The iron chelator 2,2 0 -dipyridyl (DP) was purchased from Sigma-Aldrich. All growth medium and Milli-Q water used for bacterial cultures or for preparing solutions of the enterobactin-antibiotic conjugates were sterilised by using an autoclave. A DP stock solution (200 mM) was prepared in DMSO and used in the bacteria growth assays requiring iron-depleted conditions. Working dilutions of the siderophore and siderophore-antibiotic conjugate stock solutions were prepared in 10% DMSO/H 2 O. For all assays, the nal cultures contained 1% v/v DMSO. Sterile polypropylene culture tubes and sterile polystyrene 96-well plates used for culturing were purchased from VWR and Corning Incorporated, respectively. The optical density at 600 nm (OD 600 ) was recorded on a Beckman Coulter DU800 spectrophotometer or by using a Bio-Tek Synergy HT plate reader.
Growth studies of E. coli in the presence of DP Overnight cultures of E. coli were prepared by inoculating 5 mL of Luria Broth (LB) medium with bacterial freezer stocks. The cultures were incubated at 37 C in a tabletop incubator with shaking at 150 rpm for 16-18 h. The overnight culture was diluted 1 : 100 into 5 mL of fresh LB medium containing DP (200 mM) and incubated at 37 C with shaking at 150 rpm until OD 600 reached 0.6. The cultures were subsequently diluted to an OD 600 value of 0.001 using 50% MHB medium (11.5 g L À1 ). A 90 mL aliquot of the diluted culture was combined with a 10 mL aliquot of a 10Â solution of DP (0, 0.25, 0.5, 1, 2, 4, and 8 mM) in a 96-well plate, which was wrapped in Paralm and incubated at 30 C with shaking at 150 rpm. Bacterial growth was determined at t ¼ 0, 2, 4, 6, 8, 10, and 20 h by measuring the OD 600 using a BioTek Synergy HT plate reader. Each well condition was prepared in duplicate and at least three independent replicates were conducted on different days and using two different DP stock solutions. The resulting mean OD 600 values are reported and the error bars represent the standard deviation.

General procedure for antimicrobial activity assays
Overnight cultures of E. coli, S. aureus, and A. baumannii were prepared by inoculating 5 mL of Luria Broth (LB) medium with bacterial freezer stocks. The cultures were incubated at 37 C in a tabletop incubator with shaking at 150 rpm for 16-18 h. The overnight culture was diluted 1 : 100 into 5 mL of fresh LB medium containing DP (200 mM) and incubated at 37 C with shaking at 150 rpm until OD 600 reached 0.6. The cultures were subsequently diluted to an OD 600 value of 0.001 using 50% MHB medium (11.5 g L À1 ) with or without DP (200 mM). A 90 mL aliquot of the diluted culture was combined with a 10 mL aliquot of a 10Â solution of Amp/Amx or (Glc)Ent-Amp/Amx 5-10 in a 96-well plate, which was wrapped in Paralm and incubated at 30 C with shaking at 150 rpm for 19 h. Bacterial growth was determined by measuring the OD 600 using a BioTek Synergy HT plate reader. Each well condition was prepared in duplicate and at least three independent replicates were conducted on different days and using different synthetic batches of each conjugate. The resulting mean OD 600 values are reported and the error bars represent the standard deviation.
For L. rhamnosus GG ATCC 53103, the bacterial culture was grown in MRS medium overnight. The resulting culture was diluted 1 : 50 into 5 mL of fresh MRS medium containing DP (200 mM) and incubated at 37 C with shaking at 150 rpm until OD 600 reached 1.0. The culture was subsequently diluted to an OD 600 value of 0.004 in 1 : 1 MRS/MHB medium with or without DP (200 mM). The antibacterial activity assays were performed as described above for E. coli.

Time-kill kinetic assays
A 5 mL overnight culture of E. coli CFT073 or UTI89 grown in LB medium was diluted 1 : 100 into 5 mL of fresh LB medium with 200 mM DP and incubated at 37 C with shaking at 150 rpm until OD 600 reached z0.3. The culture was centrifuged (3000 rpm Â 10 min, rt) and the resulting pellet was resuspended in 50% MHB and centrifuged (3000 rpm Â 10 min, rt). The resulting pellet was resuspended in 50% MHB with or without DP (200 mM) and the OD 600 was adjusted to 0.3. A 90 mL aliquot of the resulting culture was combined with a 10 mL aliquot of a 10Â solution of Amp/Amx or (Glc)Ent-Amp/Amx 5-10 in a 96-well plate, which was wrapped in Paralm and incubated at 37 C with shaking at 150 rpm. The OD 600 values were recorded at t ¼ 0, 1, 2, and 3 h. In a parallel experiment, a 10 mL aliquot of the culture was taken at t ¼ 0, 1, 2, and 3 h and serially diluted by using sterile phosphate-buffered saline (PBS) and plated on LB agar to obtain colony forming units (CFU mL À1 ). Each well condition was repeated at least three times independently on different days. The resulting mean OD 600 or CFU mL À1 is reported and the error bars are the standard deviation.

Antimicrobial activity assays in the presence of unmodied (Glc)Ent
These assays were performed following the general procedure described above except that varying concentrations of Ent, MGE, or DGE were mixed with Ent-Amp/Amx 5 and 6, MGE-Amp/Amx 7-8, or DGE-Amp/Amx 9 and 10. Ent was synthesized following a literature procedure, 75 MGE 2 and DGE 3 were prepared from Ent using MceC and IroB as described for MGE-PEG 3 -N 3 12 and DGE-PEG 3 -N 3 13. Stock solutions of (Glc)Ent 1-3 were prepared in DMSO and stored at À20 C.

Mixed-E. coli assays
The pET29a plasmid (kanamycin resistance) was transformed into E. coli K-12, and the pHSG398 plasmid (chloramphenicol resistance) was transformed into E. coli CFT073 and UTI89, by electroporation. Overnight cultures of the bacterial strains were prepared by inoculating 5 mL of LB medium containing the appropriate antibiotic (kanamycin, 50 mg mL À1 ; chloramphenicol, 34 mg mL À1 ) with bacterial freezer stocks, and the cultures were incubated at 37 C in a tabletop incubator shaker set at 150 rpm for 16-18 h. Each overnight culture was diluted 1 : 100 into 5 mL of fresh LB medium containing 200 mM DP, but no antibiotics, and incubated at 37 C with shaking at 150 rpm until OD 600 reached 0.6. The cultures were diluted to an OD 600 value of 0.001 in 50% MHB separately or in a 1 : 1 mixture (10 6 CFU mL À1 for each strain), with or without 200 mM DP. No antibiotic marker was included in these cultures. Aliquots of these cultures were serially diluted by using sterile PBS and plated on LB agar plates with or without corresponding antibiotic to conrm the CFU of the starter cultures. A 90 mL aliquot of each culture was combined with a 10 mL aliquot of a 1 mM solution of Amp/Amx or (Glc)Ent-Amp/Amx 5-10 in a 96-well plate. The plate was then wrapped in Paralm and incubated at 30 C with shaking at 150 rpm for 19 h. Bacterial growth was evaluated by measuring OD 600 as well as counting colonies formed on LB agar with or without kanamycin/chloramphenicol aer serial dilution with sterile PBS. Each well condition was repeated at least three times independently on different days. The resulting mean OD 600 and CFU mL À1 values are reported and the error bars are the standard deviation.

Mixed-species assays
These assays were performed following the mixed-E. coli assay procedure except that E. coli CFT073 and L. rhamnosus GG ATCC 53103 were used. A 5 mL culture of E. coli CFT073 or L. rhamnosus GG was grown for 16-18 h in LB or MRS medium, respectively. The overnight culture was diluted 1 : 100 (E. coli) or 1 : 50 (L. rhamnosus GG) into 5 mL of fresh LB or MRS medium with 200 mM DP and incubated at 37 C with shaking at 150 rpm until OD 600 reached 0.6 (E. coli) or 1.0 (L. rhamnosus GG). The cultures were diluted to an OD 600 value of 0.001 (E. coli) or 0.004 (L. rhamnosus GG) in 1 : 1 MRS/MHB containing 200 mM DP separately or in a 1 : 1 mixture (10 6 CFU mL À1 for each strain). Aliquots of these cultures were serially diluted by using sterile PBS and plated on LB and MRS agar plates to conrm the CFU of the starter culture. A 90 mL aliquot of each culture was combined with a 10 mL aliquot of a 10 mM solution of Amp/Amx or (Glc)Ent-Amp/Amx 5-10 in a 96-well plate, which was wrapped in Paralm and incubated at 30 C with shaking at 150 rpm for 19 h. Bacterial growth was assayed by both measuring OD 600 and counting colonies formed on LB and MRS agar plates aer serial dilution with sterile PBS. Each well condition was repeated at least three times independently on different days. The resulting mean OD 600 and CFU mL À1 values are reported and the error bars are the standard deviation. Comment: E. coli CFT073 forms colonies more quickly than L. rhamnosus GG on LB agar plates, whereas L. rhamnosus GG colonies appear more quickly than those of E. coli CFT073 on MRS agar plates, and these behaviours allow for each strain to be monitored independently over a 24 h period.
The assays were also performed by co-culturing E. coli CFT073 or UTI89 with S. aureus ATCC 25923 or A. baumannii ATCC 17961. A 5 mL culture of each individual bacterial strain was grown for 16-18 h in LB. The overnight culture was diluted 1 : 100 into 5 mL of fresh LB with 200 mM DP and incubated at 37 C with shaking at 150 rpm until OD 600 reached 0.6. The cultures were diluted to an OD 600 value of 0.001 in 50% MHB containing 200 mM DP separately or in a 1 : 1 mixture (10 6 CFU mL À1 for each strain). A 90 mL aliquot of each culture was combined with a 10 mL aliquot of a 10 mM solution of Amp/Amx or (Glc)Ent-Amp/Amx 5-10 in a 96-well plate, which was wrapped in Paralm and incubated at 30 C with shaking at 150 rpm for 19 h. Bacterial growth was assayed by both measuring OD 600 and counting colonies formed on HardyCHROM UTI plates aer serial dilution with sterile PBS. Plating E. coli strains on these plates results in pink colonies, whereas S. aureus and A. baumannii provide white colonies. Each well condition was repeated at least three times independently on different days.
The resulting mean OD 600 and CFU mL À1 values are reported and the error bars are the standard deviation.

Antimicrobial activity assays in the presence of lipocalin 2
Cultures of E. coli CFT073 were grown in modied M9 minimal medium 65 (Na 2 HPO 4 6.8 g L À1 , KH 2 PO 4 3 g L À1 , NaCl 0.5 g L À1 , NH 4 Cl 1 g L À1 , 0.4% glucose, 2 mM MgSO 4 , 0.1 mM CaCl 2 , 0.2% casein amino acids, and 16.5 mg mL À1 of thiamine) for 16-18 h. The overnight culture grew to saturation and was diluted 1 : 100 into 5 mL of fresh modied M9 minimal medium and incubated at 37 C with shaking at 150 rpm until OD 600 reached 0.6. The OD 600 of the culture was adjusted to 0.001, and the culture was further diluted 1 : 100 with the M9 medium in two steps (1 : 10 Â 1 : 10). The corresponding CFU was determined to be z10 4 CFU mL À1 by plating on LB agar plates. Lipocalin 2 (Lcn2, R&D Systems) was diluted into PBS, pH 7.4 to a concentration of 20 mM and frozen at À20 C until use. Bovine serum albumin (BSA, Sigma-Aldrich) was prepared in PBS, pH 7.4 to achieve a concentration of 20 mM. A 90 mL aliquot of the diluted culture was combined with a 5 mL aliquot of a 20Â solution of (Glc)Ent-Amp/Amx 5-10 and a 5 mL aliquot of Lcn2 or BSA in a 96-well plate, which was wrapped in Paralm and incubated at 37 C with shaking at 150 rpm for 24 h. Bacterial growth was determined by OD 600 . Each well condition was repeated at least three times independently on different days. The resulting mean OD 600 is reported and the error bars are the standard deviation.

Cytotoxicity assays
The human colon epithelial T84 cell line was purchased from ATCC and cultured in 1 : 1 DMEM/F12 medium with 10% fetal bovine serum, and 1% penicillin and streptomycin (v/v, ATCC). The cells were grown to approximately 95% conuency and treated with 3 mL of trypsin-EDTA (Corning). A 12 mL portion of fresh medium was added to the detached cells, and the T84 cell suspension was centrifuged (600 rpm Â 5 min, 37 C). The supernatant was discarded and the cell pellet was resuspended in 6 mL of the fresh culture medium. The concentration of cells was quantied by using a manual hemocytometer (VWR International) and adjusted to 1 Â 10 5 cells per mL. A 90 mL aliquot of T84 cells were then added to 96-well plates and incubated at 37 C and 5% CO 2 for 24 h. Stock solutions (10Â) of Amp/Amx or (Glc)Ent-Amp/Amx 5-10 were prepared in sterile-ltered 10% DMSO/H 2 O and 10 mL of each solution was added to the appropriate well. The plate was incubated at 37 C and 5% CO 2 for another 24 h. 3-[4,5-Dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide (MTT, Alfa Aesar) was dissolved in sterile PBS and the concentration was adjusted to 5 mg mL À1 . The resulting yellow solution was ltered through a 0.2 mm lter and a 20 mL aliquot of the resulting MTT solution was added to each well. The plate was incubated at 37 C and 5% CO 2 for 4 h and the supernatant was removed from each well. DMSO (100 mL) was added to each well and the absorbance at 550 nm was recorded by using a plate reader. Blank readings were recorded on wells that contained only the medium. The assay was repeated in triplicate on different days, and the mean and standard deviation are reported.

Competing financial interests
A patent application covering GlcEnt-Amp/Amx has been led.