Effect of loop deletions on the binding and transport of ferric enterobactin by FepA

Authors


Phillip E. Klebba. E-mail peklebba@ou.edu; Tel. (+1) 405 325 4969; Fax (+1) 405 325 6111.

Abstract

The siderophore ferric enterobactin enters Escherichia coli through the outer membrane (OM) porin FepA, which contains an aqueous transmembrane channel that is normally occluded by other parts of the protein. After binding the siderophore at a site within the surface loops, FepA undergoes conformational changes that promote ligand internalization. We assessed the participation of different loops in ligand recognition and uptake by creating and analysing a series of deletions. We genetically engineered 26 mutations that removed 9–75 amino acids from nine loops and two buried regions of the OM protein. The mutations had various effects on the uptake reaction, which we discerned by comparing the substrate concentrations of half-maximal binding (Kd) and uptake (Km): every loop deletion affected siderophore transport kinetics, decreasing or eliminating binding affinity and transport efficiency. We classified the mutations in three groups on the basis of their slight, strong or complete inhibition of the rate of ferric enterobactin transport across the OM. Finally, characterization of the FepA mutants revealed that prior experiments underestimated the affinity of FepA for ferric enterobactin: the interaction between the protein and the ferric siderophore is so avid (Kd < 0.2 nM) that FepA tolerated the large reductions in affinity that some loop deletions caused without loss of uptake functionality. That is, like other porins, many of the loops of FepA are superficially dispensable: ferric enterobactin transport occurred without them, at levels that allowed bacterial growth.

Introduction

Bacteria acquire iron from the environment by secreting small organic molecules (siderophores) that chelate the metal and then bind to outer membrane (OM) receptor proteins. In Gram-negative bacteria, TonB-dependent ligand-gated porins actively transport iron chelates that are too large (> 600 Da) to pass through general porins. Escherichia coli transports its native siderophore, ferric enterobactin (FeEnt; Guerinot, 1994; Neilands, 1995), through FepA, a multifunctional OM protein that also serves as the surface receptor for two protein toxins, colicins B and D.

Enterobactin binds Fe3+ with remarkably high affinity (Ka = 1052; Carrano and Raymond, 1979), which derives from the nature of its iron complex: three identical catechol groups surround the iron atom, forming a hexaco-ordinate, right-handed chelate. The metal centre of the ferric siderophore interacts with FepA: substitutions or modifications to the aromatic catechol groups around the iron may severely impair recognition (Ecker et al., 1986; Heidinger et al., 1983; Thulasiraman et al., 1998). Ferric enterobactin is an acidic molecule: six negative charges contributed by the oxygen atoms of the catechols and three positive charges from the iron nucleus combine to create a net charge of minus three. Charge is a determinant of FeEnt recognition: site-directed mutagenesis showed ferric siderophore adsorption depends, at least in part, on positively charged residues (Newton et al., 1997).

The OM proteins of Gram-negative bacteria contain a distinctive, extended β-sheet that forms an amphiphilic barrel in the OM bilayer (Weiss et al., 1991; Cowan et al., 1992; Kreusch et al., 1994; Schirmer et al., 1995; Meyer et al., 1997). Large loops on the exterior and reverse turns in the periplasm join the individual strands within the β-barrel. This theme, predicted by modelling studies (Paul and Rosenbusch, 1985; Charbit et al., 1986; Vogel and Jahnig, 1986; Murphy et al., 1990; Baumler and Hantke, 1992; Killman et al., 1993), was also recently confirmed for the TonB-dependent ligand-gated porins FepA (Buchanan et al., 1999) and FhuA (Ferguson et al., 1998; Locher et al., 1998). In this study, we performed thermodynamic and kinetic characterizations of site-directed FepA mutant proteins to gain an insight into its transport mechanism in the biologically relevant situation in vivo. Our approach exploited the fact that precise elimination of porin loops may allow proper localization, folding, assembly and complete or partial retention of function (Benson et al., 1988; Klebba et al., 1994; 1995; Saint et al., 1996). Porins containing deletions in transmembrane strands, on the other hand, are often poorly expressed, proteolytically degraded and toxic to cell growth (Klebba et al., 1994; Lathrop et al., 1995). These data suggest that loop deletions may both aid in the assignment of functional activities to specific regions of an OM protein and test the validity of a folding model.

The crystallographic depiction of FepA revealed, as expected (Liu et al., 1993), the largest known OM β-barrel, covered on the exterior surface by loops that bind ferric enterobactin (Buchanan et al., 1999) and closed on the periplasmic side by its own N-terminus. In a fundamental sense, FepA fulfils the definition of a porin: it contains a transmembrane pore, through which a solute (FeEnt) passes into the cell. However, its TonB and energy dependence distinguish FepA from general and specific porins, and it is ligand gated, in that FeEnt binding stimulates conformational changes (Liu et al., 1994; Klug et al., 1998; Jiang et al., 1997) that internalize the siderophore through the transmembrane channel. The identity of loops that participate in binding and uptake, the molecular basis of their actions and the transport mechanism are unknown. We deleted the major external loops of FepA individually and analysed the ability of the resultant proteins to interact with ferric enterobactin and colicins B and D. The mutations derived from the previously proposed model of FepA folding (Murphy et al., 1990), but they were predicted with sufficient accuracy to ultimately delete all or part of each of the now confirmed major surface loops of the OM protein. The folding of the N-terminal 140 amino acids of FepA to block the periplasmic side of the β-barrel itself (Buchanan et al., 1999) was the principal unexpected feature that the constructions did not directly accommodate. Nevertheless, our N-terminal deletions removed two distinct portions of the occlusion domain. The results implicated several previously unstudied loops in transport, further differentiated siderophore uptake from colicin action and revised existing perceptions of siderophore uptake: FepA binds ferric enterobactin much more strongly than previously believed, and many loops are not needed for FeEnt binding or internalization.

Results

Ferric enterobactin binding and transport by wild-type FepA

Biochemical analysis of the mutants of interest required new assay conditions, which, when applied to wild-type FepA, showed the inadequacy of prior methods. The current experiments demonstrated that the affinity of the FeEnt transport reaction is at least 400-fold greater than previously estimated (Fig. 1), suggesting that substrate limitation at low substrate concentrations compromised the accuracy of previous measurements. Our experiments verified that exhaustion of FeEnt in binding and transport assays led to incorrect estimates of both adsorption (Kd) and transport (Km) affinity. Past studies produced Kd and Km values of ≈20 and 220 ηM, respectively, whereas improved methods revised these to 0.2 and 0.5 ηM for chromosomally expressed FepA and to 0.4 and 0.1 ηM for plasmid-expressed FepA (Fig. 1, Table 1[link]). Hence, it is apparent that the affinity of FepA for FeEnt, in both the binding and transport reactions, lies in the subnanomolar range. The actual Km values of the FepA–FeEnt interaction may be lower still, because for wild-type FepA, even our optimum conditions were not completely free of substrate limitation problems: 10–20% depletion occurred during uptake assays at the lowest concentrations of FeEnt that we tested.

Figure 1.

. Transport of ferric enterobactin by E. coli, analysed under different conditions. Top: the conditions of the 59FeEnt uptake assays with strain BN1071 (which expresses chromosomal FepA; Klebba et al., 1982) were manipulated to change the [FeEnt]/[FepA] ratio and the duration of the assays. The concentrations of FepA (ηM), uptake periods (min) and apparent uptake Km values (ηM), respectively, were: (●) 12.3, 5, 220; (●) 0.7, 0.5,25; (bsl00041) 0.9, 0.17, 1.8; (bsl00042) 0.22, 0.17, 0.5; (○) 0.11, 0.17, 0.5. Note that the apparent affinity of the uptake reaction increased as the concentration of FepA and the time of the assays decreased, to a limit of Km = 0.5 ηM. Bottom left: comparison of 59FeEnt binding by E. coli expressing chromosomal- (○; Kd = 0.2 ηM) or plasmid-derived (▵; Kd = 0.4 ηM) FepA, under conditions (see Experimental procedures ) that minimize the depletion of FeEnt. The inset shows a Scatchard plot of the data. Bottom right: comparison of 59FeEnt uptake by E. coli expressing chromosomal-(○; Km = 0.5 ηM) or plasmid-derived (▵; Km = 0.1 ηM) FepA, under conditions (see Experimental procedures ) that minimize the depletion of FeEnt. The inset shows a Lineweaver–Burke plot of the data; units on the y-axis are Min-109 cells pMol−1. At the bottom, error bars represent the standard deviation of the data used to generated the fitted curves. The mean standard errors for measurements of Km and capacity of wild-type FepA were 9% and 6% respectively.

Table 1. . Phenotypic effects of loop deletions in FepA. a. Ferric enterobactin binding and uptake: Kd (ηM); capacity (Cap; pMol 10−9 cells); Km (ηM); Vmax (pMol min−1 10−9 cells). The Hill coefficient (τ) was obtained from the slope of the central linear region of plots of log [S] versus log [v/Vmax − v], as determined by ‘grafit 4’ (Erithacus), and Km was found from the intersection of the line with log [v/Vmax − v] = 0. Asterisks mark strains analysed by 1 h uptake experiments; ‘NS’ denotes that ‘non-saturable’ binding or uptake kinetics were observed, which prevented determination of thermodynamic or kinetic constants. For all experiments, the FepA-deficient strain KDF541 was simultaneously tested to determine non-specific background levels of binding and transport, which were subtracted from the experimental samples to yield the tabulated values. For FeEnt -binding experiments, the mean standard errors for Kd and capacity were 17.4% and 6.2% respectively. For FeEnt uptake experiments, the mean standard errors for Km, Vmax and τ were 13.2%, 5.5% and 6.3% respectively. Nutrition test results (Nutr.) are expressed as the diameter of the growth halo around the ferric enterobactin disk (in mm). The tabulated values derive from a single experiment, but little variation was seen in repeated trials. A dash in a table cell indicates that the measurement in question was either not applicable to that particular mutant or not measurable.b. Colicin-binding: Kd (ηM); capacity (Cap; pMol 10− 9 cells). Killing is expressed as a percentage of wild type; ‘NS’ denotes that ‘non-saturable’ binding was observed; R, resistant. For all experiments, the FepA-deficient strain KDF541 was tested simultaneously to determine non-specific background levels of binding and killing, which were subtracted from the experimental samples to yield the tabulated values. For colicin B binding experiments, the mean standard errors for Kd and capacity were 36.6% and 17% respectively. For colicin killing, the tabulated values derive from a single experiment, but little variation was seen in repeated trials.c. The tabulated values represent mean fluorescent intensity. Each strain was tested several times; the data give the results of a single representative experiment. Antibodies against epitopes in L2 (M33), L4 (M35, M44, M45) and L5 (M24) were used in the analyses (Murphy et al., 1990). For all experiments, the FepA-deficient strain KDF541 was tested simultaneously to determine non-specific background levels of antibody binding. For each mAb, adsorption to KDF541 gave mean fluorescence values < 0.2. For analysis of β-strand deletions with mAbs 24, 35 and 45, the mean fluorescence intensity of KDF541/pITS449 was 5.9, 9.8 and 13.2 respectively.d. Susceptibility to antibiotics was determined by placing disks embedded with the test compounds (Fisher Scientific) on the surface of LB plates spread with the bacteria of interest. The tabulated values are the diameter of the zone of clearing surrounding the disk after 12 h incubations at 37°C; they derive from a single experiment, but little variation was seen in repeated trials. Chloramphenicol (Cm), erythromycin (Er), rifampicin (Rf) and bacitracin (B). Neomycin, tetracycline and novobiocin were also tested, but the deletions did not affect permeability to these compounds. Significant deviations from susceptibility conferred by wild-type FepA are designated by bold type.Thumbnail image of

Surface loop deletions

Among the 11 external loops of FepA (Fig. 2), we removed all or part of the nine largest by site-directed deletions. In many cases, we generated an otherwise duplicate deletion that incorporated a known (NSEG; Cowan et al., 1992) or predicted [NCET, DNDN, VEGIG; Wilmot and Thornton, 1988) β-turn at the deletion junction (Fig. 2, Table 1[link]) to improve the probability of correctly reversing the polypeptide. The structural autonomy of porin surface loops from the membrane-localized barrel domain stimulated these experiments (Benson et al., 1988; Klebba et al., 1990; 1994; Lou et al., 1996; Saint et al., 1996). On this basis, we used deletions to discriminate loops that are essential or non-essential to transport and, secondarily, to consider the ability of deletions to test local structural predictions in OM proteins.

Figure 2.

. Site-directed deletions within E. coli FepA. A. The structural model of FepA (Buchanan et al., 1999) shows the location and nomenclature of the deletions (blue). The amino acid sequence is in one-letter code; small letters represent residues that were not defined by the crystal structure. Residues in β-strands are in squares, and side-chains of residues marked with a double line point to the outside of the barrel. Aromatic residues on the external surface of the β-barrel that define the position of the protein in the OM bilayer are coloured green. R286 and R316, facing the interior of the pore and previously implicated in FeEnt binding (Newton et al., 1997), are coloured red. See text for further details. B. Deletions in the region between β-strands 5 and 8, encompassing residues 255–346. Transmembrane portions of strands 6 and 7 are overlined and underlined. R286 and R316 are in bold type.

KDF541 expressed all the deletion proteins (except those involving β-strands) to similar levels, with little degradation (Fig. 3). When inoculated in minimal media, strains harbouring loop deletion mutants on pUC plasmids all grew at similar rates to KDF541/pITS449, the wild-type parent, indicating that the mutant proteins were not generally toxic to the cell. Cytofluorimetric analyses of live bacteria with anti-FepA monoclonal antibodies (mAbs) recognized four distinct surface epitopes (identified by mAbs 24, 33, 44 and 45) on the mutant proteins, demonstrating their assembly in the outer membrane with at least partially correct (see Discussion ) topology (Table 1). By each of these methods, deletions in loops 2, 3, 4, 5, 7, 8, 9, 10 and 11 produced few observable deleterious effects on FepA structure. The major difference between the mutant and wild-type proteins was an enhanced susceptibility to proteolysis of proteins with deletions in loops 2, 7, 8 and 11 (Fig. 3). The cell envelope protease OmpT (Hollifield et al., 1978; Baneyx and Georgiou, 1990) initially degrades FepA near its amino-terminus, at the dibasic residues 37 (K) and 38 (R). This cleavage does not normally occur in vivo (Murphy et al., 1990), but takes place upon solubilization of the OM by detergents. The presence or absence of genetically engineered β-turns at the deletion junctions did not affect susceptibility to proteolysis.

Figure 3.

. Expression of FepA by loop deletion proteins. E. coli strain KDF541 (fepA), containing pUC plasmids expressing wild-type FepA or loop deletion mutants, was grown in MOPS media to mid-log phase, lysed and subjected to SDS–PAGE and Western immunoblot with [125I]-Protein A. A. Wild-type FepA, ΔN1, ΔN1T, ΔN2, ΔN2T, ΔL2, ΔL2T, ΔL5 and Δ5LT appear in lanes 1–9 respectively. B. Wild-type FepA, ΔS67AT1, ΔS67AT2, Δ67B, Δ67BT, ΔS6, ΔS67C, ΔS67CT, ΔS7, L3, ΔL4 and 10Tappear in lanes 1–12 respectively. C: Wild-type FepA, ΔL5, ΔL5T, ΔL7, ΔL7T, ΔL8, ΔL8T, ΔL9T, ΔL10T, ΔL11AT and ΔL11BT appear in lanes 1–11 respectively.

Functional analyses of surface loop deletions

We investigated the mutants by quantitative characterizations of ligand binding (Kd and capacity) and qualitative (siderophore nutrition and colicin killing tests) and quantitative (Km and Vmax) measurements of ligand transport (Table 1). All the deletions significantly decreased the affinity of FepA for FeEnt, as evidenced by higher binding Kd and transport Km values. The mutations also had variable effects on transport rate (reflected in Vmax) that led to three phenotypic categories: (i) slightly impaired uptake (Vmax was 45–100% of wild type; deletions in loops 3, 4, 5 and 9); (ii) strongly impaired uptake (Vmax was less than 5% of wild type; deletions in loops 2, 10 and 11); (iii) defective — unable to transport FeEnt (deletions in loops 7 and 8).

We estimated the affinity of the mutant proteins for FeEnt from the concentration dependence of its binding and uptake (Figs 4 and 5). In this respect, class i mutants (ΔL3, ΔL4, ΔL5, ΔL5T and ΔL9T) were most readily understood. For example, among the various constructs, ΔL9T had the least impact on FeEnt recognition and internalization. It still efficiently bound and transported FeEnt (Kd = 6 ηM; Km = 354 ηM), and the decrease in affinity that it did sustain was reflected in both Kd and Km, with greater effects on the latter parameter. Class ii deletions (ΔL2, ΔL10T, ΔL11AT and ΔL11BT) also diminished the affinity of FepA for the siderophore, in some cases more dramatically (ΔL2, Δ2LT and ΔL11BT). In spite of their impaired affinity, the adsorption of FeEnt suggested that the FepA ligand binding site remained grossly intact in class i and ii mutants. Class ii mutations were most distinguished from class i by the large reductions that they caused in transport rate; accurate analysis of their extremely low uptake rates required extension of the 59FeEnt uptake period to 1 h. The retention of functionality in class i and ii mutants differentiated them from class iii deletions (ΔL7, ΔL7T, ΔL8 and ΔL8T), which showed no ability to specifically bind or transport FeEnt. Deletions in loops 7 and 8 eliminated the interaction between FepA and FeEnt, preventing most characterizations of their effects on the OM protein. Although ΔL8 and ΔL8T did not bind FeEnt, they showed low-level transport of the siderophore in nutrition tests and 59FeEnt uptake studies. In the latter experiments, the uptake rate was proportional to the external siderophore concentration (data not shown); no saturation was observed. Furthermore, antibiotic sensitivity tests (Misra and Benson, 1988) indicated that class i and class ii deletions did not impart non-specific permeability to FepA, whereas all class iii deletions increased permeability to large antibiotics (Table 1). Cytofluorimetric analyses (see above) suggested that class iii mutants folded and assembled in the OM with at least partially correct topology. Thus, in several ways, class iii mutations resembled deletions that converted FepA into an open channel (Rutz et al., 1992).

Figure 4.

. Siderophore nutrition tests. The photographs show E. coli strain KDF541 (fepA) containing pUC plasmids expressing wild-type FepA or loop deletion mutants, after 24 h incubation at 37°C. A 10 mm ruler is embedded in the photos.

Figure 5.

. Concentration dependence of FeEnt uptake. Top: 59FeEnt uptake was measured in KDF541 harbouring plasmids expressing wild-type FepA (○), the class i mutations ΔL3 (▵) and ΔL5 (◊), and the class ii mutations ΔL2 (▾) ΔL10T (▪). Each mutant was initially analysed at least three times to determine its uptake properties. The plotted data represent a single experiment for each strain, measuring rates at 12 concentrations near Km, with data collected in triplicate and averaged. Left: absolute transport rates of FeEnt uptake. Right: normalized rates (to Vmax) of the same strains. Bottom: normalized (to Vmax) rates of FepA proteins carrying deletions in different loops, analysed at concentrations near their respective Km values. The concentration of FeEnt was plotted logarithmically to demonstrate the large decrease in affinity that occurred upon individual loop deletion: (○) FepA++; (*) ΔL9T; (▪) ΔL10T; (▵) ΔL3; (◊) ΔL5T; (⋆) ΔL4; (×) ΔL11BT; (▾) ΔL2.

The loop deletions in FepA produced a variety of novel effects in siderophore nutrition tests (Fig. 4). The mutations created smaller (ΔL2, ΔL2T, ΔL8 and ΔL8T), larger (ΔL3, ΔL4, ΔL5, ΔL5T, ΔL9T, ΔL10T and ΔL11AT), fuzzy-edged (all except perhaps ΔL9T) and double halos (ΔL3, ΔL4, ΔL5, ΔL5T and ΔL9T). We noted some correlations between the siderophore nutrition results and quantitative measurements of FeEnt binding and transport. Small halos always reflected a very low transport Vmax, and the ‘fuzziness’ of a halo was related to affinity: mutants with the highest affinity for FeEnt showed the sharpest halos. However, the loop deletions usually had multiple repercussions on binding and transport, which complicated the understanding of siderophore nutrition tests. No simple explanations were apparent to rationalize the enlarged, double-edged, ‘target’ halos, which were seen exclusively in class i mutants. In general, siderophore nutrition tests were not effective in diagnosing the effects of site-directed mutations in FepA. This was well illustrated by Δ11BT, which produced a halo of identical size and intensity to wild-type FepA, in spite of a 6000-fold higher Km and 25-fold lower Vmax. Unless the underlying causes of halo size, shape and definition are systematically explained, assays of this type are of little value in the delineation of mutational effects on the transport reaction.

Colicin binding and killing

Measurements of colicin adsorption to and transport through the mutant OM proteins served a dual purpose: binding affinity provided another general criterion of their cell surface conformation, and killing efficiency assessed their specific functionality. Most loop deletion mutants (except those in loops 7 and 8) were susceptible to colicins B and D. Both colicins showed the same trends of cytotoxicity in the mutants; although we collected data on both toxins, we only report our results for colicin B. Deletions of loops 3 and 4 produced the smallest effects on the interaction between colicin B and FepA. Other mutants, including deletions in surface loops 2, 5, 9 and 10, were less able to interact with colicin B. Colicin binding and killing were synonymous in one sense: productive binding to the mutant proteins always resulted in cytotoxicity, albeit with reduced efficiency (Table 1). However, as observed for FeEnt, the interaction of colicin B with the mutant proteins was usually complicated. For example, the two distinct deletions in L11 produced opposite effects on binding and transport: Δ11AT bound well but was poorly killed, while Δ11BT bound more poorly but was more effectively killed. The impairment of colicin adsorption and killing did not necessarily correlate with defects in FeEnt binding and transport: although class i mutations generally showed greater affinity towards the colicins than class ii; ΔL5, ΔL5T and Δ11AT were exceptions to this conclusion.

Deletions in the N-terminal domain

The N-terminal 140 amino acids of FepA fold into a compact structure that, in the crystal structure, lodges within and blocks the transmembrane β-barrel. Deletions in this domain (ΔN1, ΔN1T, ΔN2 and ΔN2T) eliminated FeEnt binding and transport and reduced colicin B binding to the point at which it was immeasurable. Strains expressing the N-terminus deletions were susceptible to the toxin at reduced levels, which was their only detectable functional property. Colicin killing was also the sole property affected by the introduction of β-turns at the deletion junctions: ΔN1 and ΔN1T showed a 100-fold difference in susceptibility to colicin B. Downstream surface epitopes in the FepA proteins resulting from ΔN1, ΔN1T, ΔN2 and ΔN2T were accessible to recognition by mAbs (Table 1).

Other deletions in the central domain

Previous experiments with monoclonal antibodies showed that ligands interact with FepA in the central region of its amino acid sequence. Before the solution of the FepA crystal structure, this portion was modelled as a large loop (Murphy et al., 1990) bounded by residues 258 and 339. Although several independent lines of evidence (Murphy et al., 1990; Rutz et al., 1992) supported the involvement of this region in ligand binding, especially the identification of specific residues within it that affect the affinity of the interaction with FeEnt (Newton et al., 1997), the fine structure and proposed boundaries of the proposed loop were unsubstantiated. We engineered a series of nested deletions to test the idea that this region might consist of two loops separated by a pair of transmembrane strands (as was ultimately found by the crystal structure). The deletions (Fig. 2) removed the entire region (ΔS67A: residues 257–332), the central part containing conserved basic residues and portions of strands 6 and 7 (ΔS67B: 283–316), individual β-strands 6 (ΔS6: 289–297) and 7 (ΔS7: 301–310) or parts of them (ΔS67C: 293–306). The results gave insights about the effects of deletions within different regions of FepA.

Unlike the mutations in surface loops, deletions that intruded into transmembrane strands of the FepA β-barrel affected expression and/or secretion. In the most extreme cases (ΔS6, ΔS7, ΔS67B and ΔS67C), the mutation dramatically reduced FepA levels in the OM (Fig. 3). Other deletions (Δ67A) moderately reduced FepA concentrations, whereas individual deletions of the two surface loops in the region (ΔL3 and ΔL4) had no effect on the amount of FepA (Fig. 3). For even the poorly expressed mutant proteins, some surface epitopes were accessible to binding by anti-FepA mAbs (Table 1).

Only deletions that correctly targeted the loops in this region (ΔL3, ΔL4) were mechanistically viable. While all seven other constructs were functionally defective in every way, L3 and L4 bound and transported both FeEnt with moderate efficiency and interacted with colicin B at wild-type levels. Their wild-type expression levels, relative to all other deletions in the central domain, were striking. Antibiotic permeability tests reinforced the conclusion that ΔL3 and ΔL4 had little deleterious impact on overall FepA structure: deletions that disrupted β-strands conferred increased susceptibility to large antibiotics, while ΔL3 and ΔL4 retained wild-type antibiotic resistance profiles (Table 1).

Discussion

The similarity and rigidity of β-barrels within outer membrane proteins (Weiss et al., 1990; Cowan et al., 1992; Schirmer et al., 1995; Meyer et al., 1997; Pautsh and Schulz, 1998) suggests, and evidence confirms (Benson et al., 1988; Klebba et al., 1994, Lou et al., 1996; Saint et al., 1996; Charbit et al., 1998), that the transport properties of porins derive from their surface-exposed and buried loops (Klebba and Newton, 1998). Unlike the open transmembrane channels within general or specific porins, the larger channels within ligand-gated porins are obstructed by other elements of protein structure. Loops on the outside and an N-terminal domain on the inside close the FepA pore (Buchanan et al., 1999). These and other data (Murphy et al., 1990; Rutz et al., 1992; Killman et al., 1993; Newton et al., 1997) predict the existence of a ligand binding site in the outer domain of the receptor protein, which in some way initiates transport into the normally shut pores. In vitro and in vivo evidence (Liu et al., 1994; Klug et al., 1998; Jiang et al., 1997) substantiates the idea that conformational changes actuate the FepA transport reaction. These previous findings generally outlined the mechanism of ligand-gated porin transport, but many remaining questions about the specific details of the process focus on the functions of the loops.

We used site-directed mutagenesis to assess the participation of individual loops in FeEnt uptake. Although it is more precise to consider the contributions of individual residues, the magnitude of such a task is formidable. Surface residues comprise a significant component of porin structure. Roughly 36% of general porin residues (OmpF; Cowan et al., 1992), 35% of specific porin residues (LamB; Schirmer et al., 1995) and 54% of ligand-gated porin residues (FepA, excluding the N-terminal domain; Buchanan et al., 1999) exist in surface-exposed regions. In FepA, ≈320 amino acids reside in 11 surface loops. Our main objectives with the deletions were to reduce the immensity of studying individual amino acids in such a large protein and to begin to identify the contributions of various loops to binding and transport.

Because the construction of the deletions preceded the solution of FepA structure, some uncertainty existed about their design. However, the FepA crystal structure eliminates that ambiguity, facilitating interpretation of our results. The constructs individually removed most or all of the major surface loops of FepA, without intrusion into transmembrane regions of the β-barrel (Fig. 2). Engineered mutations, especially deletions, may create abnormalities in the target protein that are unforeseeable, difficult to perceive and that may impair function in an irrelevant way. We used several phenotypic tests to address this problem, which usually provided information about the effects of the deletions on FepA structure. In most cases, the functionality of the mutant proteins towards the siderophore, and especially their binding of and susceptibility to the colicins, indicated that they were assembled in the OM with relevant tertiary structure. For certain mutations, however, the more severe loss of functionality that occurred confounded interpretation of their phenotypes. Class iii deletions in L7 and L8 eliminated the binding and uptake of all ligands, raising the possibility that loops 7 and 8 function in the recognition and transport of both FeEnt and colicins. It is also conceivable, however, that deletions in L7 and L8 disrupt FepA tertiary structure by more long-range structural effects. Germane to this point, ΔL7 and ΔL8 imparted non-specific permeability to FepA, which may arise from unshielding of a porin channel (Benson et al., 1988) or from aberrant structure in a porin β-barrel (Klebba et al., 1994; Lathrop et al., 1995). The L7 and L8 deletions were correctly limited to external residues, but were less precisely targeted than other deletions that produced more functional mutant proteins. Anti-FepA mAbs recognized the surface epitopes of ΔL7 and ΔL8, providing evidence of their proper localization in the OM and correct folding at some level. Further data are needed to fully characterize their effects on FepA structure. Similarly, class iii deletions in the N-terminal domain blocked all interactions with FeEnt and, although they retained some susceptibility to ColB, they bound it so poorly that we were unable to measure the interaction. The N-terminal pore occlusion domain contains two loops that project to the cell surface from within the barrel (Buchanan et al., 1999). ΔN1 and ΔN2 individually removed these regions, but also intruded into other structural aspects of the N-terminal pore occlusion domain, complicating their analysis.

Class i and ii deletions diminished FeEnt binding and/or transport (as little as 10-fold, as much as 20 000-fold), resulting in a variety of functional attributes. It is perhaps most enlightening that the majority of the surface loop deletions did not eliminate siderophore binding or transport. The resultant mutant FepA proteins were less proficient (Fig. 5), but still efficacious in vivo (Fig. 4). Class i deletions showed that loops 3, 4, 5 and 9 are, to a greater or lesser extent, dispensable to the overall uptake reaction. They were not needed explicitly to bind or transport the metal complex. Although the affinities of these proteins for the siderophore dropped about 1000-fold, their Vmax values were 50–100% of wild type, suggesting that, once bound, FeEnt uptake proceeded relatively normally. On the other hand, class ii mutations were more deleterious, strongly reducing both binding and transport. FepA proteins devoid of loop 2, for example, were about 10 000-fold less avid, and their Vmax values were only 5% of wild type. They recognized and transported colicin B better than FeEnt, suggesting that their deficiencies towards the siderophore are specific rather than general. All class i mutants, and most class ii mutants (except ΔL2 and ΔL2T), functioned well enough in vivo to impart substantial halos in siderophore nutrition tests.

Analysis of the central portion of the FepA polypeptide illustrates some of the problems with a genetic approach to membrane protein structure and some approaches to their solution. The model of FepA folding, proposed in 1990, predicted that residues 255–336 form a single external loop that functions in ligand binding (PL5; Murphy et al., 1990; Newton et al., 1997). Other experiments implicated the same loop in the control of permeability through the FepA channel (Rutz et al., 1992). However, later crystal structures of general (Weiss et al., 1990; Cowan et al., 1992; Kreusch et al., 1993) and specific porins (Schirmer et al., 1995; Meyer et al., 1997) portrayed PL5 as considerably larger than any other porin surface loop. Our deletions in this region, which preceded the FepA crystal structure, were originally intended to test the boundaries of PL5. The crystallographic depiction of FepA structure in this region explains the results of our experiments with these mutants. Deletions that intruded into transmembrane β-strands generated non-functional proteins that were poorly expressed and/or degraded, while other constructs in the same region that correctly targeted surface residues were functional and present at normal levels in the OM. Poor expression, in conjunction with loss of function, probably results from local or global misfolding of the OM protein (Klebba et al., 1994; Lathrop et al., 1995). Thus, the experiments reiterate that intrusion of deletions into the transmembrane regions of a porin protein may severely impair structure and function, whereas deletions within surface loops are much less deleterious.

High-affinity transport through TonB-dependent ligand-gated porins was initially characterized for vitamin B12 uptake through the protein now known as BtuB (Di Girolamo and Bradbeer, 1971; White et al., 1973), which showed an initial binding Kd of about 0.5 ηM and an overall uptake Km of about 5 ηM for cyanocobalamin. Numerous subsequent estimates of the affinity of OM ferric siderophore binding and transport systems were 20- to 5000-fold lower (Coulton and Braun, 1979; Fiss et al., 1982; Ecker et al., 1986; Zhou et al., 1995; Locher and Rosenbusch, 1997; Newton et al., 1997; Braun et al., 1999; Ferguson et al., 1998; Locher et al., 1998; Thulasiraman et al., 1998; Buchanan et al., 1999), leading to the conclusion that bacterial iron acquisition is much less efficient than vitamin B12 uptake. However, the current study suggests that existing iron binding and transport methodologies almost uniformly suffer from substrate depletion artifacts that arise from underestimation of the affinities of ferric siderophore–receptor systems. The equilibrium dialysis and rapid flow methods used for B12 binding and transport (Di Girolamo and Bradbeer, 1971; White et al., 1973) accommodated the possibility of substrate depletion, which explains their much lower Kd and Km values. Furthermore, with regard to the FeEnt uptake system, a comparison of binding affinities derived from experiments that compensate for substrate limitation either in vitro (Kd = 20 ηM; Payne et al., 1997) or in vivo (Kd = 0.2 ηM; this study) indicates that purified FepA shows ≈100-fold lower affinity than live bacteria (FepA) possess for FeEnt.

Both the binding and transport of ferric siderophores by ligand-gated porins were characterized as multicomponent or co-operative processes (Coulton and Braun, 1979; Locher and Rosenbusch, 1997; Thulasiraman et al., 1998), but the revision of our methodologies essentially eliminated the allostery that occurs during FeEnt binding and uptake in vivo (Fig. 1). The substrate depletion that occurred in previous binding and transport methods contributed to the high Hill coefficients that were reported previously and, at this time, it seems questionable that allostery occurs at all in ferric siderophore transport. Sequential technical improvements to increase the FeEnt/FepA ratio in our transport reactions progressively reduced their Hill coefficients to 1.2 for uptake by chromosomally encoded FepA, and to 1.7 for plasmid-mediated uptake. However, the latter overexpression system is most susceptible to substrate limitation artifacts. Finally, the mutant FepA proteins that were not subject to substrate depletion because of their lower affinity for FeEnt were essentially devoid of co-operativity during both binding and transport.

Experimental procedures

Strains and plasmids

Mutations were studied on the pUC19 derivative pITS449, which contains fepA under its natural, fur-regulated promoter (Armstrong et al., 1990), hosted in the E. coli strain KDF541 (fepA ; Rutz et al., 1992).

Site-directed mutagenesis

The fepA structural gene was cloned from pITS449 (Armstrong et al., 1990) into M13mp19 using Pst I and SacI sites of the multiple cloning sites. Oligonucleotide-directed deletions were generated by Kunkel mutagenesis (Kunkel, 1989) as described previously (Newton et al., 1997). Deletions were detected by appropriate restriction digestion of recombinant M13 DNA or by single-stranded sequencing and transferred back to the expression vector pUC19 by restriction fragment exchange. All deletions were confirmed by double-stranded sequencing of the exchanged fragment and about 100 bp of the upstream and downstream flanking regions, in both M13 and pUC, using an automated sequencer (ALF Express; Pharmacia Biotechnology) and a ThermoSequenase cycle sequencing kit (US Biochemicals). Deletions were designated and enumerated according to the domain of the membrane protein that they affected (Buchanan et al., 1999): the N-terminal pore occlusion domain (ΔN), the surface loops (ΔL) or the strands of the β-barrel (S); a following T indicates the addition of a β-turn (DNDN, NSEG, NCET or VEGIG; Wilmot and Thornton, 1988) at the deletion junction.

Quantification of FepA expression

Bacteria harbouring fepA and its deletion derivatives were grown in LB medium (Miller, 1972) plus ampicillin (100 μg ml−1) and subcultured at 1% into MOPS minimal medium (Neidhardt et al., 1974) containing ampicillin (10 μg ml−1) for 5.5–6.0 h at 37°C with vigorous aeration. Aliquots were harvested, boiled for 5 min in sample buffer (60 mM Tris, pH 6.8, 10% glycerol, 2% SDS), and volumes corresponding to 5 × 107 cells were applied to a 10% polyacrylamide gel. The separated proteins were electrophoretically transferred to nitrocellulose paper, and Western immunoblots were performed using monoclonal antibodies against FepA (mAbs 29, 41 and 45; Murphy et al., 1990) and developed with [125I]-Protein A. The filter paper was exposed to X-ray film overnight.

Siderophore nutrition tests

Efforts were made to perform the assay quantitatively, which required careful attention to the bacterial growth conditions. Bacteria were inoculated from frozen storage into LB broth and grown to mid-log phase (OD600 = 0.5–0.9). A sample of 100 μl of the culture was plated in 3 ml of Nutrient top agar containing 100 μM apoferrichrome A (Wayne et al., 1976) in 6-well plates. When needed, ampicillin was added to 10 μg ml−1. Filter paper disks (6 mm diameter) containing 10 μl of 50 μM ferric enterobactin were applied to the plates, which were incubated at 37°C for 24 h. Results were expressed as the diameter of visible growth around the filters.

Siderophore binding and transport

59FeEnt binding and transport experiments were performed by modifications of previous methods (Newton et al., 1997). Changes were necessary because initial comparisons of mutant and wild-type proteins revealed that FepA binds and transports FeEnt much more avidly (Kd < 0.2 M; Km < 0.5 nM) than previously thought (Kd = 20 M; Km = 220 M), which raised the possibility that at low FeEnt concentrations, where FepA was in excess, the reaction mixtures became depleted of FeEnt, violating the equilibrium assumptions of both binding and transport assays. To address this issue, we determined the transport Km in a series of uptake assays varying the ratio [FeEnt]/[FepA] by increasing the reaction volumes and decreasing the concentration of bacteria and the time of the uptake measurements (Fig. 1). These alterations revised the FeEnt uptake Km to lower values, and comparable binding studies similarly reduced the adsorption Kd, suggesting that previous methodologies for the measurement of these parameters (Coulton and Braun, 1979; Ecker et al., 1986; Newton et al., 1997; Thulasiraman et al., 1998) were subject to substrate depletion artifacts, and therefore inaccurate. These experiments also indicated that it was necessary to customize the transport conditions individually for FepA mutants that manifest either lower affinity or lower transport efficiency.

For FeEnt uptake by wild-type FepA, we ultimately adopted the following procedure. 59FeEnt was prepared at a specific activity of ≈200 c.p.m. pM−1 and chromatographically purified. All transport manipulations were performed at 37°C. A volume of mid-log bacterial culture (50–100 μl containing ≈5 × 107 cells) was deposited in a 50 ml test tube and incubated in a 37°C water bath. Without delay, 25 ml of prewarmed MOPS minimal media, containing glucose (0.2%), appropriate nutritional supplements and varying concentrations of 59FeEnt, was poured into the tube to achieve rapid and thorough mixing. The transport reactions were quenched by the addition of a 1000-fold excess of non-radioactive FeEnt, immediately filtered through 0.45 μm nitrocellulose, and the filters were counted in a Packard Cobra gamma counter. Kinetic parameters were determined from the initial rates of FeEnt uptake, which were calculated at each substrate concentration from two independent measurements made in triplicate at 5 s and 15 s: c.p.m. bound to the cells at 5 s were subtracted from the c.p.m. associated with the cells at 15 s (10 s uptakes). Independent measurements (see below) showed that binding of FeEnt to wild-type FepA reached equilibrium within 5 s. We used essentially the same procedures for class i mutants (submicromolar Km values), except that the measurements of 59FeEnt associated with the bacteria were made at 30 s and 60 s (30 s uptakes) to accommodate their slower progress to binding equilibrium and their generally lower uptake rates. For class ii mutants, of yet lower affinity (micromolar Km values), the large amounts of FeEnt required to monitor the concentration dependence of uptake were unachievable in 25 ml reaction volumes, and we performed the experiments in 1 ml volumes instead, usually collecting data at 1 and 6 min (5 min uptakes). In this case, the lower affinity of the mutants also eliminated potential distortions from substrate depletion. Finally, the low uptake rates of some class ii mutants (deletions in loops 2, 10 and 11) forced us to extend the uptake period to 60 min (Thulasiraman et al., 1998). For measurements of both class i and class ii mutants, we reduced the specific activity of the 59FeEnt two- to fourfold by dilution with non-radioactive FeEnt.

We measured the affinity of FeEnt binding independently using similar procedures that alleviated substrate depletion. For wild-type FepA, 59FeEnt was prepared at a specific activity of ≈200 c.p.m. pM−1 and chromatographically purified. All binding manipulations were performed at 0°C. A mid-log bacterial culture was chilled on ice for 1 h, and an aliquot (containing ≈5 × 107 cells) was deposited in a 50 ml test tube and incubated on ice. A 25 ml volume of ice-cold MOPS minimal media, containing varying concentrations of 59FeEnt, was poured into the tube to achieve rapid and thorough mixing. After 1 min, the binding reactions, which were performed in triplicate, were filtered through 0.45 μm nitrocellulose, and the filters were counted in a Packard Cobra gamma counter. The capacity of the bacteria for 59FeEnt was the same when binding was measured for 5 s at 37°C or for 1 or 6 min at 0°C, indicating that the initial adsorption reaction reached equilibrium within 5 s at physiological temperatures and within 1 min on ice. Again, for mutants of lower affinity, modifications were made to the basic protocol. For all mutants, the incubation period was extended to 5 min and, for class ii mutants, the experiments were performed in 1 ml volumes. For measurements of binding to both class i and class ii mutants, we reduced the specific activity of the 59FeEnt two- to fourfold by dilution with non-radioactive FeEnt. The FepA-deficient strain KDF541 was simultaneously tested as a negative control, and any non-specific adsorption of 59FeEnt by this strain was subtracted from the experimental samples. Although minimal non-specific adsorption or transport occurred in studies of wild-type FepA and most of the mutants, this control was essential when using the high concentrations of FeEnt (10–50 μM) that were needed to characterize low-affinity mutants.

Binding data were analysed using the one-site, bound versus free equation of grafit 4 (Erithacus). Transport results were analysed by both Michaelis–Menten equations and Hill plots, using grafit 4.

Colicin binding and killing

Colicin B binding was performed as described previously (Newton et al., 1997). Colicin sensitivity was tested by placing serial dilutions of purified colicins on an overlay of bacteria expressing wild-type FepA or its deletion derivatives. Titre was expressed as the last colicin dilution able to clear the lawn of bacteria to the agar.

Cytofluorimetry

Cells were grown in MOPS minimal medium, and aliquots of 108 cells were stained with monoclonal antibodies (mAbs 24, 33, 44 and 45; Murphy et al., 1990) against surface-exposed epitopes of FepA for 45 min; cells were washed once with TBS + 1% BSA and incubated with goat anti-mouse IgG conjugated to fluorescein for 45 min. After a final washing, cells were analysed in a Coulter Epics Elite cytofluorimeter/cell sorter. Results were expressed as mean fluorescence intensity. Strains KDF541 and KDF541/pITS449 were used as negative and positive controls respectively.

Permeability tests

As a measure of permeability, the strains were tested for susceptibility to a series of antibiotics [chloramphenicol (30 μg ml−1), neomycin (30 μg ml−1), tetracycline (30 μg ml−1), erythromycin (15 μg ml−1), novobiocin (30 μg ml−1), rifampicin (15 μg ml−1) and bacitracin (10 IU)], some of which are too large to pass through general porins (Er, Rf, B). Cells were grown in LB broth with ampicillin (10 μg ml−1) overnight at 37°C, and 100 μl aliquots of the culture were mixed with melted top agar and plated on LB plates. After the top layer solidified, antibiotic disks were distributed on the plate. Results were recorded after 12 h incubation at 37°C and expressed as the diameter of the killing halo around the disk.

Acknowledgements

The authors wish to thank Julie Hwang for construction of mutant ΔL8, Marjorie Montague for expert technical assistance, and Paul F. Cook and Marvin A. Payne for critical reading of the manuscript. We are especially grateful to Dick van der Helm and Johann Deisenhofer for their personal communications about the FepA crystal structure. This work was supported by NIH grant 1R01-GM53836 and NSF grants MCB 9408737 and MCB 9709418 to P.E.K.

Ancillary