Identification of target tissue glycosphingolipid receptors for uropathogenic, F1C-fimbriated Escherichia coli and its role in mucosal inflammation.

Bacterial adherence to mucosal cells is a key virulence trait of pathogenic bacteria. The type 1 fimbriae and the P-fimbriae of Escherichia coli have both been described to be important for the establishment of urinary tract infections. While P-fimbriae recognize kidney glycosphingolipids carrying the Galalpha4Gal determinant, type 1 fimbriae bind to the urothelial mannosylated glycoproteins uroplakin Ia and Ib. The F1C fimbriae are one additional type of fimbria correlated with uropathogenicity. Although it was identified 20 years ago its receptor has remained unidentified. Here we report that F1C-fimbriated bacteria selectively interact with two minor glycosphingolipids isolated from rat, canine, and human urinary tract. Binding-active compounds were isolated and characterized as galactosylceramide, and globotriaosylceramide, both with phytosphingosine and hydroxy fatty acids. Comparison with reference glycosphingolipids revealed that the receptor specificity is dependent on the ceramide composition. Galactosylceramide was present in the bladder, urethers, and kidney while globotriaosylceramide was present only in the kidney. Using a functional assay, we demonstrate that binding of F1C-fimbriated Escherichia coli to renal cells induces interleukin-8 production, thus suggesting a role for F1C-mediated attachment in mucosal defense against bacterial infections.

Epithelial linings of the host function as very efficient barriers against microorganisms. To achieve this protective effect the mucosal lining utilizes a variety of mechanisms that engage multiple signaling pathways upon bacterial exposure. The most commonly studied mechanism for the induction of the host's innate immune response in urinary tract infections is bacterial adhesion to uroepithelial cells. Although this is reported as one of the most important virulence trait of uropathogenic Escherichia coli, bacterial adhesion also leads to induction of the host's immune system (1)(2)(3). Accordingly, adhesion to the epithelium acts as a double-edged sword for bacteria.
Adhesion of Gram-negative bacteria to epithelial cells is often mediated by fimbria or pili. These rod-shaped, proteinaceous, filamentous polymeric organelles are expressed on the surface of bacteria. P and type 1 fimbriae are the two best characterized attachment organelles, both known for their central role in urinary tract infections (4). Expression of P fimbriae is mainly associated with pyelonephritogenic isolates of uropathogenic E. coli. Binding of P fimbriae to Gal␣4Gal-carrying glycosphingolipids, an epitope present in the human kidney, is of major importance for the establishment of disease (5)(6)(7). The type 1 fimbriae are mainly associated with cystitis, and confer binding to mannosylated proteins such as uroplakin that are abundant within the lower urinary tract (8 -11).
Although the biogenesis as well as binding characteristics of P fimbriae and type 1 fimbriae have been studied in detail, uropathogenic isolates of E. coli express other fimbriae that are less well characterized, in part because their target tissue receptors are unidentified. One such example is the F1C fimbriae, which are expressed by 14 -30% of all uropathogenic strains of E. coli (12,13). The F1C fimbriae are structurally related to, and genetically organized as, the type 1 fimbriae. However, comparison of the amino acid sequence reveals that F1C is more closely related to the S fimbriae (14). The S fimbriae confer binding to sialyl-␣2-3Gal␤-containing receptor molecules, and are associated to sepsis and meningitis caused by E. coli in newborn children (15,16).
The kidney has been reported to be the target tissue of F1C expressing E. coli using in vitro models and strictly biochemical approaches (17,18). In the present study, we use in vivo and in vitro model systems to identify the glycosphingolipid receptor for F1C expressing strains of uropathogenic E. coli within the human, rat, and canine urinary tract. We present evidence that the ceramide portion of the glycosphingolipid receptor confers specificity to the binding. Moreover, we report that human renal epithelial cells produce the proinflammatory chemokine IL-8 1 as a consequence of F1C-mediated attachment.

Bacterial Strains, Culture Conditions, and Labeling
The human and rat pyelonephritogenic E. coli strain ARD6 (serotype O6:K13:H1, World Health Organization designation Su 4344/41) was used in this study (19). The non-fimbriated E. coli strain HB101 was transformed with the plasmid L40 carrying the foc operon from the pyelonephritogenic strain KS71 (20). The Gal␣4Gal binding recombinant E. coli strain HB101/pPIL291-15, carrying a plasmid-borne pap gene cluster with a class II papG allele, was obtained from Dr. I. van Die (Vrije University, Amsterdam, The Netherlands). E. coli strains were cultured (37°C, 12 h) on Luria-agar plates supplemented with 10 l of [ 35 S]methionine (400 mCi; Amersham Biosciences, UK). Bacteria were harvested by scraping, washed three times in phosphate-buffered saline (pH 7.3), then resuspended to a bacterial density of 1 ϫ 10 8 colony forming units/ml in phosphate-buffered saline containing 1% mannose (w/v). The specific activity of bacterial suspensions was ϳ1 cpm per 100 bacteria.

Reference Glycosphingolipids
Total acid and non-acid glycosphingolipid fractions were obtained by standard procedures (21). The individual glycosphingolipids were isolated by repeated chromatography on silicic acid columns of the native glycosphingolipid fractions or acetylated derivatives thereof. The identity of the purified glycosphingolipids was confirmed by mass spectrometry (22), proton NMR spectroscopy (23), and degradation studies (24,25). Reference galactosylceramide was obtained from Sigma.

Glycosphingolipid Binding Assays
Binding of 35 S-labeled bacteria to glycosphingolipids on thin-layer chromatograms was performed as previously reported (28). Dried chromatograms were dipped for 1 min in diethylether/n-hexane (1:5, by volume) containing 0.5% (w/v) polyisobutylmethacrylate (Aldrich, Milwaukee, WI). After drying, the chromatograms were soaked in phosphate-buffered saline containing 2% bovine serum albumin (w/v), 0.1% NaN 3 (w/v), and 0.1% Tween 20 (v/v) for 2 h at room temperature. The chromatograms were covered with a suspension of radiolabeled bacteria. Following a 2-h incubation at room temperature, chromatograms were extensively washed (phosphate-buffered saline), and then exposed to XAR-5 x-ray films (Eastman Kodak, Rochester, NY) for 12 h. Autoradiograms were replicated using a CCD camera (Dage-MTI, Inc., Michigan City, IN), and analysis of the images was performed using the public domain NIH Image program (developed at the National Institutes of Health, and available at rsb.info.nih.gov/nih-image/).

Isolation of Binding Active Glycosphingolipids from Rat, Canine, and Human Kidneys
Acid and non-acid glycosphingolipids were isolated from rat, canine, and human kidneys by standard methods (21). In addition, non-acid glycosphingolipid fractions were isolated from samples of canine urethra, urinary bladder, the trigonum area of the urinary bladder, urethers, and kidney. The amounts obtained are summarized in Table I. HPLC separation of the total non-acid glycosphingolipid fractions of rat, canine, and human kidneys was performed using a Kromasil 5 Silica column (1 ϫ 25 cm inner diameter, particle size 5 m; Phenomenex, Torrence, CA). The fractions obtained were analyzed by thin-layer chromatography using anisaldehyde for detection, and the glycosphingolipid-containing fractions were tested for binding of F1C-fimbriated E. coli using the chromatogram binding assay.
Rat Kidney-Part of the total non-acid glycosphingolipid fraction of rat kidney (3.5 mg) was separated by HPLC eluted with a linear gradient of chloroform/methanol/water 80:20:1 to 60:35:8 (by volume) during 180 min with a flow rate of 2 ml/min. The monoglycosylceramides eluted in tubes 8 -13, and the binding-active compound was found in tubes 12 and 13. Pooling of these tubes gave ϳ100 g of pure bombardment; HPTLC, high performance thin-layer chromatography.    binding-active monoglycosylceramide (designated fraction R1).
Human Kidney-Isolation of non-acid glycosphingolipids from 520 g dry weight human kidneys has been described previously (29). The non-acid glycosphingolipid fraction was subjected to repeated silicic acid chromatography, and the mono-to triglycosylceramides were pooled, giving 700 mg. This fraction was further separated by HPLC by an isocratic elution with chloroform/methanol/water (80:25:0.5, by volume) during 180 min and a flow rate of 2 ml/min. Pure binding-active monoglycosylceramide (5.0 mg; designated fraction H1) eluted in tube 5, while tube 12 (158 mg) contained the binding-active compound migrating in the tri-to tetraglycosylceramide region. Since the latter fraction also contained several non-binding compounds, this fraction was further separated by HPLC eluted with a linear gradient of chloroform/methanol/water (80:20:1 to 60:35:8 (by volume)) during 180 min, and with a flow rate of 2 ml/min. Pooling of tubes 70 -170 resulted in 2.0 mg of pure binding-active glycosphingolipid, which was designated fraction H2.
Canine Kidney-Part of the total non-acid glycosphingolipid fraction of canine kidney (15.0 mg) was separated on a 20-g Iatrobeads column (Iatron Laboratories Inc., Tokyo, Japan) eluted stepwise with increasing amounts of methanol and water in chloroform (29). Pure binding-active monoglycosylceramide (1.0 mg; designated fraction C1) and triglycosylceramide (1.3 mg; designated fraction C2) was thereby obtained.

Enzymatic Hydrolysis
Hydrolysis of glycosphingolipids with ␤-galactosidase from Streptococcus pneumoniae (Oxford Glycosystems Ltd., Abingdon, UK) was performed according to the manufacturer's instructions. Glycosphingolipids were also treated with green coffee bean ␣-galactosidase (Glyko, Inc., Novato, CA) according to the protocol of the manufacturer.

Mass Spectrometry
Negative ion FAB mass spectra were recorded on a JEOL SX-102A mass spectrometer (JEOL, Tokyo, Japan). The ions were produced by 6 keV xenon atom bombardment, using triethanolamine (Fluka, Buchs, Switzerland) as matrix, and an accelerating voltage of Ϫ10 kV.
EI mass spectrometry was performed on permethylated aliquots of the isolated glycosphingolipids (30). The derivatized samples were analyzed on a JEOL SX-102A mass spectrometer, using the in beam technique (31). The analyses was performed with an electron energy of 70 eV, trap current of 300 A, and acceleration voltage of 10 kV. The temperature was raised from 150 to 410°C, by increases of 10°C/min.

Proton NMR Spectroscopy
1 H NMR spectra were acquired on a Varian 500 MHz spectrometer at 30°C. Samples were dissolved in dimethyl sulfoxide/D 2 O (98:2, by volume) after deuterium exchange.

Cell Stimulation
The human renal epithelial cell line A498 (ATCC HTB-44) was grown in 24-well cell culture plates in RPMI 1640 medium supplemented with 10% fetal calf serum, 25 mM HEPES, and 2 mM L-glutamine (Invitrogen, Stockholm, Sweden) at 37°C in 5% CO 2 . At confluency, cells were washed before control medium (no additives) or medium containing 2 ϫ 10 6 colony forming units of HB101 or HB101/L40 was added. Supernatants were collected 6 and 25 h post-infection and were analyzed by enzyme-linked imunosorbent assay for IL-8 (Diaclone, Besancon, France).

ARD6 Binds to Monoglycosylceramide and Tri-or Tetraglycosylceramide Isolated from Rat, Canine, and Human Kidney-
The uropathogenic E. coli strain ARD6 was originally isolated from a child suffering from pyelonephritis (19), and it has also been shown to cause pyelonephritis in rat (32). To identify renal receptor(s) to which ARD6 binds, non-acid and acid glycosphingolipids were isolated from kidneys of 20-day-old, noninfected rats. Similar preparations were also isolated from canine and human tissues. A binding assay of 35 S-labeled ARD6 to non-acid glycosphingolipids separated on thin-layer chromatogram is shown in Fig. 1. A selective interaction with a distinct minor band migrating in the monoglycosylceramide region in the non-acid glycosphingolipid fractions of rat (lane 1) and canine (lane 2) kidney is seen, together with a band migrating in the tri-to tetraglycosylceramide region in the canine kidney sample. Binding to the monoglycosylceramide region and tri-to tetraglycosylceramide region was also obtained when using non-acid glycosphingolipids from human kidney (see Fig. 2B, lane 3). Occasionally, a band migrating in the diglycosylceramide region was detected in the human and canine kidney samples. No binding to the acid glycosphingolipid fractions of rat (Fig. 1B, lane 3), canine, or human kidney (data not shown) was obtained. In comparison to this distinct binding pattern, the Gal␣4Gal-binding P-fimbriated E. coli (Fig. 1C) displayed a broader binding pattern with several binding-active compounds in each tissue sample.
F1C Fimbriae Expressed by ARD6 Are Responsible for Glycosphingolipid Binding-To investigate which attachment organelles are expressed by ARD6, PCR analysis was performed, using primers for detection of P pili, type 1 fimbriae, F1C fimbriae, and afimbrial adhesins. This analysis showed that ARD6 harbors the genes for type 1 fimbriae and F1C fimbriae but not P fimbriae or afimbrial adhesin (data not shown). This was further verified in agglutination studies. ARD6 did not agglutinate human erythrocytes due to lack of P fimbriae expression. In contrast, ARD6 induced mannose-sensitive agglu-tination of yeast, which is the hallmark for type 1 fimbriae expression (data not shown). However, type 1 fimbriae are not likely to be responsible for binding of ARD6 to glycosphingolipid receptors, because the binding assays are routinely performed in the presence of 1% mannose, which inhibits binding via the type 1 fimbriae. To investigate whether the F1C fimbriae are responsible for the observed bacterial binding to glycosphingolipids, the F1C-expressing plasmid L40 was introduced into the non-fimbriated E. coli strain HB101. When the resulting strain, HB101/L40, was tested in the chromatogram binding assay, an identical binding pattern was observed as for ARD6, i.e. binding to the monoglycosylceramide region in the rat, canine, and human kidney samples along with binding to the tri-to tetraglycosylceramide region in the canine and human kidney samples (Fig. 2C). No binding was observed using HB101. Thus, binding of ARD6 to glycosphingolipids is mediated by the F1C fimbriae.
Galactosylceramide and Globotriaosylceramide with Phytosphingosine and Hydroxy 20:0 -24:0 Fatty Acids Are Target Tissue Receptors for F1C Fimbriae-Further characterization of the binding-active compound migrating in the monoglycosylceramide region from rat (fraction R1), human (fraction H1), and canine kidney (fraction C1) identified galactosylceramide (Gal␤1Cer) with phytosphingosine and hydroxy 20:0 -24:0 fatty acids as the binding-active component. This conclusion is based on the following five observations. (i) The binding-active mono- Gal␤3GalNAc␤4(NeuAc␣3)Gal␤4Glc␤1Cer d18:1-18:0/d20:1-18:0 Ϫ Human brain a Binding is defined as follows: ϩ denotes a significant darkening on the autoradiogram when 0.2 g was applied on the thin-layer plate, while (ϩ) denotes an occasional binding at 2 g, and Ϫ denotes no binding even at 2 g. b In the shorthand nomenclature for fatty acids and bases, the number before the colon refers to the carbon chain length and the number after the colon gives the total number of double bonds in the molecule. Fatty acids with a 2-hydroxy group are denoted by the prefix h before the abbreviation, e.g. h16:0. For long chain bases, d denotes dihydroxy and t trihydroxy. Thus d18:1 designates sphingosine (1,3-dihydroxy-2aminooctadecene) and t18:0 phytosphingosine (1,3,4-trihydroxy-2-aminooctadecane). c Glycosphingolipids No. 12 was prepared from No. 18 by treatment with ␤-galactosidase.
The F1C-binding Specificity Depends on the Ceramide Composition-To investigate the specificity of F1C binding to glycosphingolipids, we next examined the binding characteristics of ARD6 and HB101/L40 to a library of pure reference glycosphingolipids (summarized in Table II). This experiment shows that in addition to galactosylceramide and globotriaosylceramide, the F1C-fimbriae mediate binding to glucosylceramide (No. 2 in Table II), lactosylceramide (No. 7), and isoglobotriaosylceramide (No. 11). The detection limit for these five compounds in the chromatogram binding assay was ϳ0.2 g, while nonbinding compounds were not recognized even when 2 g was applied on the TLC. An attempt to estimate the relative affinity of binding was made by performing densitometry of autoradiograms obtained by binding of F1C-fimbriated bacteria to serial dilutions of glycosphingolipids on TLC (Fig. 8). However, with the exception of a slightly less efficient binding to glucosylceramide, no obvious preference for any of the other binding-active glycosphingolipids was found.
A common feature of all the binding-active glycosphingolipids is the presence of a ceramide with phytosphingosine and hydroxy fatty acids. However, a ceramide with phytosphingosine and hydroxy fatty acids does not always allow F1C binding, since other glycosphingolipids such as gangliotetraosylceramide (No. 19) and blood group active pentaglycosylceramides (Nos. 21, 22, and 24), all contain this ceramide composition, but are not recognized by the F1C-fimbriated E. coli. Taken together the binding data indicates that the minimum binding epitope for the F1C fimbriae is the galactose or glucose unit linked to the ceramide part, with tolerance for some extensions of the carbohydrate chain. The requirement for a specific ceramide suggests that this ceramide gives a correct presentation of the binding epitope. Alternatively, part of the ceramide is involved in the binding process.
Binding-active Glycosphingolipids Are Distributed throughout the Ascending Urinary Tract-The tissue distribution of receptors used for bacterial attachment may be an important virulence determinant for the bacteria, facilitating their ascension through the urinary tract. When selected compartments of the canine urinary tract were analyzed for their expression of binding-active glycosphingolipids we used preparations of nonacid glycosphingolipids from urethra, urinary bladder, urethers, and kidney that were separated on thin-layer chromatograms. Binding experiments using F1C-fimbriated E. coli demonstrated the presence of binding-active monoglycosylceramide in urinary bladder, urethers, and kidney, but not in the sample from urethra (Fig. 9B). Binding-active triglycosylceramide was found only in the kidney. Again, comparative studies showed that the Gal␣4Gal binding P-fimbriated E. coli (Fig.  9C) displayed a broader binding pattern with several bindingactive compounds in each tissue sample.
F1C-mediated Bacterial Adhesion Triggers the Proinflammatory Response-To investigate whether F1C-mediated adhesion induces a proinflammatory response in renal epithelial cells, A498 cells were infected with E. coli strain HB101 and the F1C-expressing strain HB101/L40. Cellular activation was monitored by examination of the presence of the chemokine IL-8 in the supernatant 6 and 24 h post-infection (Fig. 10). Supernatants from cells infected by F1C-expressing bacteria showed an ϳ3-fold increase of IL-8 as compared with supernatants from cells infected by the isogenic non-fimbriated strain. These data suggest a similar role for F1C in inflammation as previously described for other fimbriae (2). DISCUSSION Although expression of P fimbriae are considered as one of the major determinants for the establishment of pyelonephritis, this disease can also be caused by non-P-fimbriated E. coli strains (19,32). Here, we report an alternative mechanism for bacterial adhesion. We report that the F1C fimbriae confer binding to glycosphingolipids isolated from human, canine, and rat kidney. These F1C-binding compounds were identified as galactosylceramide and globotriaosylceramide. Galactosylceramide was present in all tissues within the ascending urinary tract except for the urethra, while globotriaosylceramide was specifically expressed in renal tissue. The ceramide portion of both binding-active galactosyl-and globotriaosylceramide consists of phytosphingosine and hydroxy 20:0 -24:0 fatty acids. This structure was found to be a critical determinant for F1Cmediated binding. When screening a library of glycosphingolipids, we found that all binding-active compounds had phytosphingosine and hydroxy fatty acids, while glycosphingolipids with the same carbohydrate sequence but different ceramide composition were not recognized by F1C fimbriae (Table II). Collectively, our findings suggest a ceramide-close binding epitope for the F1C-fimbriae.
A large number of commensal as well as pathogenic bacteria preferentially bind to lactosylceramide with phytosphingosine and/or hydroxy fatty acids, while the same bacteria are unable to bind to galactosylceramide and glucosylceramide (35)(36)(37)(38). This binding deficiency is independent of phytosphingosine and/or hydroxy fatty acids in the ceramide portion of the receptors. Furthermore, globotriaosylceramide with phytosphingosine and hydroxy fatty acids is not recognized by the lactosylceramide binding Helicobacter pylori (38). These data suggest that expression of F1C fimbriae provides a unique binding capacity of uropathogenic E. coli to galactosylceramide and globotriaosylceramide containing phytosphingosine and hydroxy fatty acids, which may facilitate binding to uroepithelium in vivo for the establishment of infection. Bacterial binding has previously been shown to be a key virulence trait of uropathogenic E. coli (7,10).
The F1C fimbriae were recently reported to bind a wide variety of glycosphingolipids, i.e. glucosylceramide, galactosylceramide, lactosylceramide, globotriaosylceramide, lactotriaosylceramide, gangliotriaosylceramide, neolactotetraosylceramide, and gangliotetraosylceramide, with most efficient binding to gangliotriaosylceramide (39). We never detected binding to gangliotetraosylceramide, while occasional binding to lactotriaosylceramide, gangliotriaosylceramide, lactotetraosylceramide, and neolactotetraosylceramide was observed when high concentrations of glycosphingolipids were used on the thinlayer chromatograms. The reason for this discrepancy is unclear. Khan et al. (39) mainly used commercially obtained glycosphingolipids isolated from erythrocytes and brain, whose ceramide composition predominantely consists of sphingosine, dihydrosphingosine, and non-hydroxy fatty acids (40). The use of glycosphingolipids lacking the optimal ceramide composition might explain why the high affinity binding to certain glycosphingolipids was overlooked.
Epithelial cells located in the organs of the urinary tract utilize different mechanisms to detect and respond to bacterial infections. Bladder epithelial cells are highly responsive to E. coli infections, mainly because these cells express Toll-like receptor 4 (1). When Toll-like receptor 4 recognizes the presence of bacterial lipopolysaccharide, the major constituent of the outer membrane of Gram-negative bacteria, a signaling pathway is initiated which leads to a rapid production of IL-8. Although bladder epithelial cells also respond to bacteria that bind via the type 1 fimbriae, the elicited response constitutes only a minor fraction of the lipopolysaccharide /Toll-like receptor 4-mediated response. In contrast, renal epithelial cells lack expression of Toll-like receptor 4 and are therefore non-responsive to lipopolysaccharide. Instead, renal epithelial cells must rely on a mechanism based on microbial adhesion for initiating the proinflammatory response. F1C-fimbriated E. coli significantly induces IL-8 production in renal epithelial cells to levels that previously have been reported for adhesion mediated by the type 1 and P fimbriae (2). Considering the lipopolysaccharide non-responsive phenotype, our data suggest that the IL-8 response observed in renal epithelial cells is entirely due to attachment via the F1C fimbriae. Compared with P fimbriaemediated binding that recognize several Gal␣4Gal-containing glycosphingolipids present in rat, canine, and human kidneys, the binding pattern of F1C fimbriated bacteria is more restricted. However, the distribution of binding-active compounds within the uroepithelium suggests that the F1C fimbriae may facilitate for bacteria to ascend to the kidney, and once there, to establish pyelonephritis. It was recently reported that immunization with the FimH adhesin of type 1 fimbriae protects mice from urinary tract infections (41). Thus, proteins from the F1C fimbriae may be used as a novel vaccine candidate to confer protection against pyelonephritis caused by non-P fimbriated E. coli strains.