Biophysical Characterization and Activity of Lymphostatin, a Multifunctional Virulence Factor of Attaching and Effacing Escherichia coli *

Attaching and effacing Escherichia coli cause diarrhea and typically produce lymphostatin (LifA), an inhibitor of mitogen-activated proliferation of lymphocytes and pro-inflammatory cytokine synthesis. A near-identical factor (Efa1) has been reported to mediate adherence of E. coli to epithelial cells. An amino-terminal region of LifA shares homology with the catalytic domain of the large clostridial toxins, which are retaining glycosyltransferases with a DXD motif involved in binding of a metal ion. Understanding the mode(s) of action of lymphostatin has been constrained by difficulties obtaining a stably transformed plasmid expression clone. We constructed a tightly inducible clone of enteropathogenic E. coli O127:H6 lifA for affinity purification of lymphostatin. The purified protein inhibited mitogen-activated proliferation of bovine T lymphocytes in the femtomolar range. It is a monomer in solution and the molecular envelope was determined using both transmission electron microscopy and small-angle x-ray scattering. Domain architecture was further studied by limited proteolysis. The largest proteolytic fragment containing the putative glycosyltransferase domain was tested in isolation for activity against T cells, and was not sufficient for activity. Tryptophan fluorescence studies indicated thatlymphostatin binds uridine diphosphate-N-acetylglucosamine (UDP-GlcNAc) but not UDP-glucose (UDP-Glc). Substitution of the predicted DXD glycosyltransferase motif with alanine residues abolished UDP-GlcNAc binding and lymphostatin activity, although other biophysical properties were unchanged. The data indicate that lymphostatin has UDP-sugar binding potential that is critical for activity, and is a major leap toward identifying the nature and consequences of modifications of host cell factors.

Enterohemorrhagic Escherichia coli (EHEC) 4 and enteropathogenic E. coli (EPEC) are enteric bacterial pathogens of worldwide importance. Both pathotypes colonize intestinal mucosa via the formation of "attaching and effacing" lesions in a manner that requires a Type III protein secretion system, as well as accessory virulence factors including surface adhesins (1). Lymphostatin (also known as LifA) is a chromosomally encoded protein with a predicted molecular mass of 365 kDa produced by most EPEC and non-O157 EHEC strains (2). We have previously demonstrated that lymphostatin is required for intestinal colonization of calves by non-O157 EHEC serogroups O5, O111 (3), and O26 (4), and it also promotes colonization of the murine intestines and colonic hyperplasia by the attaching and effacing pathogen Citrobacter rodentium (5). Chlamydia species also contain a family of proteins that have homology to lymphostatin, and which have been implied to act as cytotoxins (6).
Lymphostatin was first described as the factor required for inhibition of mitogen-activated proliferation of lymphocytes by enteropathogenic E. coli O127:H6 lysates (2). This activity has been demonstrated against lymphocytes from peripheral blood and the intestines (3,7) and is not associated with direct cytotoxicity. Peripheral blood mononuclear cells from mice, cattle, and humans are sensitive to lymphostatin (2,3,5). Lymphostatin has also been reported to inhibit the production of proinflammatory cytokines including IL-2, -4, -5, and interferon-␥ (7), and it has been suggested that it may therefore interfere with the induction of innate and adaptive immune responses. In the same year as LifA was described in EPEC, a near identical factor was described in EHEC O111:HϪ that was associated with bacterial adherence to cultured epithelial cells. The authors named the factor EHEC factor for adherence 1 (Efa1), however, it has 97.4% amino acid identity to lymphostatin, and they are likely equivalent proteins (8). Although a direct role of Efa1 in adherence has been reported using rabbit EPEC (9), mutations in some strains impair expression and secretion of Type III secreted proteins required for attaching and effacinglesion formation (3). Furthermore, it has recently also been reported that lymphostatin can be secreted via the type III secretion system, but its functions once injected into host cells are unknown (10).
Understanding of the mode of action of lymphostatin has been constrained by the instability of plasmid clones and difficulties in obtaining full-length purified protein (2). Furthermore, even plasmid-driven soluble expression of smaller fragments of lymphostatin has proven to be difficult (11). Bioinformatic analysis has identified homology between the amino-terminal of LifA/Efa1 and the catalytic glycosyltransferase domain of the LCTs (2,8). These clostridial cytotoxic molecules are large proteins whose catalytic domain glycosylates Rho-family GTPases that regulate the actin network (12). They are retaining enzymes with a GT-A fold, which belong to glycosyltransferase family 44 and are characterized by having a DXD (Asp-Xaa-Asp) motif involved in metal ion binding together with the diphosphate moiety of the UDP-sugar donor (13). It is a multistep reaction dependent on the DXD motif, where sugar binding and hydrolysis is followed by transfer of the released sugar to an acceptor protein (reviewed in Ref. 12).
In addition, recent studies show that attaching and effacing E. coli express another novel protein, NleB, which is an effector glycosyltransferase injected into host cells upon infection. NleB blocks death receptor-induced apoptosis and promotes intestinal colonization (14,15), as part of a suite of E. coli effectors that influence NF-B signaling in mammalian cells (reviewed in Ref. 16). NleB uses uridine diphosphate N-acetylglucosamine (UDP-GlcNAc) as its donor sugar and covalently N-links a GlcNAc to a conserved arginine residue in death domain containing adaptor proteins (14,15). This modification had not previously been described and most GlcNAc addition to proteins occurs via O-linkage on serine/threonine residues (14,15). Apart from the glycosyltransferase domain, searches also identified a Yersinia YopT-like cysteine protease (CP) motif in the sequence of lymphostatin (17). These features represent a small portion of the primary sequence of lymphostatin, and are restricted to the N-terminal third of the protein. Although one report claimed that deletion of the predicted glycosyltransferase and cysteine protease motifs attenuated C. rodentium in mice (5), close inspection reveals that stop codons were introduced that resulted in protein truncation at the deletion site rather than in-frame mutations, making the results difficult to interpret (4). Given its large size and the paucity in understanding how lymphostatin is able to carry out its activities, we sought to produce a full-length recombinant lymphostatin and characterize its structural and biophysical features, as well as its effects on activated T cells. Here we show that full-length lymphostatin can be expressed as a folded, biologically active recombinant protein that inhibits bovine T cell proliferation at very low concentrations. Furthermore, like other glycosyltransferases, it has sugar-binding potential, and both its biological activity and sugar binding is dependent on a DXD motif, which has been implicated in the activity of glycosyltransferases.
Although the activity of lymphostatin was identified more than a decade ago, this is the first significant step forward in understanding the mechanisms underlying its intrinsic activities.
Generation of a DTD to AAA Lymphostatin Substitution Mutant-To remove the DXD motif the amino acids DTD at position 557-559 were substituted with AAA using the QuikChange II XL site-directed mutagenesis kit (Agilent Technologies) according to the manufacturer's directions. Briefly, using the pRHAM-LifA-6xH plasmid as a template, the primers LifA-DXD-1 (Forward), GGATGTATATCCTTAAAGAGCA-TGGTGGTATTTATACAGCGGCCGCGATGATGCCTGC-ATACTCTAAACAAGTAATTTTTAAAA, and LifA-DXD-2 (Reverse), TTTTAAAAATTACTTGTTTAGAGTATGCAG-GCATCATCGCGGCCGCTGTATAAATACCACCATGCT-CTTTAAGGATATACATCC, were used to introduce AAA to the sequence, by total replication of the plasmid, followed by digestion of the parent plasmid. A NotI restriction endonuclease site (indicated in bold) was also introduced at the site of mutation. Putative mutant plasmids were screened by restriction digest with NotI, and verified by Sanger sequencing on both strands (GATC-Biotech).
Isolation of Peripheral Blood Mononuclear Cells (PBMCs) and T Cells from Bovine Blood-Access to bovine blood for these studies was approved by the local ethics committee and blood draws were carried out in accordance with the Animals (Scientific Procedures) Act 1986. PBMCs were isolated from 12-18-month-old Holstein-Friesian cows. Briefly, blood was collected into heparinized bags or syringes. Following centrifugation at 1,200 ϫ g for 15 min to generate an initial buffy coat, the white blood cell fractions were pooled, layered over Ficoll-Paque Plus (GE Healthcare), and centrifuged for 30 min at 1,200 ϫ g with the brake off. The PBMCs were collected from the interface and washed several times before use. If required, the T lymphocyte fraction was further enriched using a sterile wool column (Polysciences, Inc.), as suggested by the manufacturer. Briefly, columns were washed in sterile Roswell Park Memorial Institute medium (RPMI) supplemented with 10% (v/v) fetal bovine serum, 20 mM Hepes, 1 mM sodium pyruvate, 100 units/ml of penicillin/streptomycin, 20 mM L-glutamine (Life Technologies), and incubated for 1 h at 37°C in a 5% CO 2 atmosphere. Cells were applied at 10 8 /ml, run into the column by gravity, and incubated for 1 h at 37°C per 5% CO 2 . Unbound cells (mainly composed of T cells) were washed off the column in 10 ml of medium, spun down, and counted. The purity of T cell preparations was checked by single channel flow cytometry. Cells were stained with a commercially available anti-bovine CD3 antibody (MM1A; IgG1; VMRD, Pullman, WA). Secondary staining with a FITC-conjugated anti-IgG1 secondary antibody was carried out, and the samples were analyzed on a FACSCalibur using CellQuest (BD Biosciences) and FloJo software (Tree Star). A minimum of 10,000 events were collected, with an initial gate for live cells based on forward/side scatter parameters.
Proliferation Assay-Enriched T cells were used to test the activity of rLifA, purified partial fragments, and rLifA DTD/AAA using a standard colorimetric measurement of mitogenic activation of proliferation. Cells were plated at 2 ϫ 10 5 cells/well in 96-well flat-bottom plates (Costar) in triplicate for all conditions. rLifA, rLifA DTD/AAA , or a fragment thereof was added at a final concentration as indicated in the figure legends. Cell proliferation was stimulated using the mitogen concanavalin A (ConA, Sigma) at a final concentration of 1 g/ml in the presence or absence of recombinant lymphostatin as indicated in a final volume of 100 l/well. Cells were incubated at 37°C for 72 h. The colorimetric substrate CellTiter 96 AQueous One (Promega) was added 18 h before the end of the assay. All measurements were carried out at 492 nm on a Multiskan Ascent plate reader (Thermo Scientific). Cells and medium alone were used as negative controls. Background medium measurements were subtracted from all values. All treatments are expressed as a Proliferation Index, which is calculated by the ratio of: [absorbance(cellstreatedwithConAandrecombinantprotein)/absorbance (cells treated with ConA alone)]. Isolated T cells were tested for cytotoxic effect of rLifA using a lactate dehydrogenase release assay (Cytotoxicity Detection Test Plus, Roche) according to the manufacturer's directions, using an 8-h incubation period.
Limited Proteolysis and MALDI Mass Spectrometry-The identity of full-length recombinant lymphostatin was confirmed by in-gel protein digest and peptide analysis. Excised gel-bands were incubated at a porcine trypsin:lymphostatin ratio of ϳ1:30, in 50 mM ammonium bicarbonate overnight at 32°C (Promega). Peptides were identified by matrix-assisted laser desorption ionization (MALDI) mass spectroscopy on a Voyager DE-STR MALDI-TOF mass spectrometer (Applied Biosystems) using an ␣-cyano-4-hydroxycinnamic acid matrix. The spectral data were processed using Data Explorer software (Applied Biosystems) and the MASCOT NCBInr database searched against the peptide mass map (Matrix Science). To investigate the domain structure of lymphostatin, purified protein was incubated with trypsin at a ratio of 375:1, at 21°C, to give limited digestion. Aliquots were removed at 1, 2, 3, and 4 h and the reaction stopped by boiling samples adjusted with 2 mM EDTA and 2 mM PMSF in SDS-PAGE loading buffer. Digest products were separated by SDS-PAGE and individual bands were subjected to in-gel tryptic digestion and MALDI-TOF mass spectroscopy as described above. Peptide masses were compared with the sequence of full-length rLifA using GPMAW 9.2 software, mass tolerance 50 ppm (19). Fragment F1 was purified to homogeneity from other digest products by ion-exchange chromatography (Mono-Q 5/50 GL; GE Healthcare) as described above.
Size Exclusion Chromatography-Multi-angled Light Scattering (SEC-MALS)-Size exclusion chromatography coupled to UV, static light scattering, and refractive index detection were used to determine the molecular mass of pure rLifA in solution and to estimate the detergent load (Viscotec SEC-MALS 20 and Viscotek RI Detector:VE3580; Malvern Instruments). One hundred l of 1 mg/ml of lymphostatin was passed over a size exclusion column pre-equilibrated in 20 mM sodium phosphate, pH 7.6, 150 mM sodium chloride, 1 mM DTT, 0.1% Tween 20 (Superose6 10/300 GL, GE Healthcare) at 24°C. Light scattering, refractive index (RI), and A 280 nm were analyzed by a conjugate protein model using the following parameters: A 280 nm for lymphostatin and Tween 20 detergent, 0.97 and 0.01 absorbance unit, ml/mg, respectively; RI for protein, 0.187 ml/g; and Tween 20, 0.145 ml/g (Malvern Instrument software).
Bioinformatic Analysis-The secondary structure of lymphostatin was predicted from sequence using PredictProtein (26) and PSIPRED (27). Proteins with similar structural elements and homologues were identified with PHYRE (28) and BLASTp (29).
Intrinsic Tryptophan Fluorescence-Binding of uridine diphosphate-glucose (UDP-Glc) and UDP-GlcNAc to wildtype LifA and LifA DTD/AAA was determined by ligand-induced changes in intrinsic tryptophan fluorescence. Fluorescence measurements were performed on a SPEX Fluoromax 3 spectrometer (Horiba) in a 3-ml stirred cuvette by titrating the UDP-sugar into 0.2 M lymphostatin at 20°C in 20 mM sodium phosphate, pH 7.6, 150 mM sodium chloride, 5% glycerol, 0.1% Tween 20, 1 mM DTT. Samples were allowed to equilibrate for 5 min after the addition of each aliquot. The final volume added did not exceed 2% of the initial volume. Tryptophan was excited at 295 nm and emission spectra were recorded from 310 to 400 nm, with a 1-nm interval. Excitation and emission slits were set at 5 nm, with an integration time of 1 s. The equilibrium dissociation constant was obtained from fitting the fluorescence intensity at 340 nm to a binding model corrected for collisional quenching (Kaleidagraph, Synergy Software).
Negative Staining and Electron Microscopy-Four hundred mesh carbon-coated copper grids were glow discharged for 1 min in a Quorum Tech sputter coater with a current of 25 A and used within 1 h. rLifA (4 l at ϳ8 g/ml concentration) was applied to a glow discharged grid and incubated for 1-2 min.
Next, the grid was washed with 2 drops of water and 2 drops of 2% uranyl acetate followed by staining with a further drop of 2% uranyl acetate for 4 -5 min. Finally, excess liquid was blotted from the edge of the grid with filter paper (Whatman No. 5). Dried grids were imaged with an FEI F20 field emission gun electron microscope equipped with an 8k x 8k TVIPS CMOS camera (F816). The camera format was binned by a factor of 2 giving a calibrated binned pixel size of 3.06 Å/pixel at the specimen level. Micrographs were acquired semi-automatically with EM-tools (TVIPS GmbH) under low dose conditions (20 e/Å 2 , at 200 kV).
EM Image Processing-The defocus of each micrograph was determined with ctffind3 (30). Particles were selected semi-automatically using e2boxer (31). Selected particle images were extracted and normalized with RELION (32), with a box size of 128 pixels. The radius for background normalization was 55 pixels. Extracted particle images were classified into 100 twodimensional classes using RELION. Particle images, which did not align with an accuracy of better than 4 degrees in the twodimensional classification were excluded from further processing. This reduced the number of particle images in the data set from 27,431 to 25,244. The relative spatial orientations of two-dimensional class averages of the remaining particle images were determined by sinogram correlation in IMAGIC (33). After determining the relative orientations of the class averages a three-dimensional map was calculated by weighted back projection. The orientations of the class averages were further refined by angular reconstitution using projections of the three-dimensional map as anchor set followed by calculating an improved three-dimensional map by back-projection. The resulting three-dimensional map was used as reference in Relion for autorefinement of the data set. The resolution was estimated by Fourier-Shell correlation between two independently processed data sets (34) and was 23 Å at a Fourier-Shell correlation of 0.14 (35). The absolute hand of the map is unknown.
Small-angle X-ray Scattering (SAXS)-Synchrotron radiation data were collected at the I911-4 SAXS beamline at the MAX-II Laboratory (Lund, Sweden) as 4 ϫ 30-s exposures of a 20 -30-l sample and scattering profiles were compared with detect radiation damage. Lymphostatin was buffer exchanged into 20 mM NaH 2 PO 4 , pH 7.6, 300 mM NaCl 2 , 2 mM DTT, and 5% (v/v) glycerol (an identical buffer to the size-exclusion storage buffer but without Tween 20) prior to SAXS measurements using Amicon Ultra 0.5-ml centrifugal filters (Merck). Data were collected at 0.91-Å wavelength at 10°C with a Dectis hybrid pixel Pilatus 1M detector. To detect concentration-dependent inter-particle effects, measurements were collected at multiple protein concentrations in the range of 0.67-1.6 mg/ml, and the 1 mg/ml scattering curve was used for all subsequent analysis. Background buffer scattering was subtracted using PRIMUS (36), part of the ATSAS package (37). Pair distance distribution function P(r) and the maximum particle dimension D max were computed using GNOM (38). The Porod volume was calculated using ATSAS AUTOPOROD (39) and used for molecular weight estimation. Ab initio shape envelope was developed using 10 independent DAMMIF (40) computations in P1 symmetry. DAMAVER (41) was used to align and compare the resulting models. The most representative model was subsequently refined using DAMMIN (42). The resulting bead model was used to compute a surface envelope with Situs pdb2vol (43). The DAMMIN model was manually aligned with the EM density map using USCF Chimera (44).
Statistical Analysis of ED 50 of rLifA and rLifA DTD/AAA -The effective dose 50 (ED 50 ) for rLifA and rLifA DTD/AAA was determined using drc in R (45). One way analysis of variance was used to determine statistically significant differences between both ED 50 and dose-response curves using Minitab (46), with p values Յ0.05 taken to be significant. Post hoc Tukey test was used to calculate 95% confidence intervals and confirm significance (46).

Full-length Lymphostatin Can Be Produced as a Stable, Tagged Recombinant
Protein-Published studies have noted instability of lymphostatin clones (2,47) or inability to assemble the full-length gene from amplicons (8), possibly because of toxicity and gene size. To overcome this, an amplicon of the EPEC O127:H6 E2348/69 full-length lifA gene (9.6 kb) was cloned in pRham with a carboxyl-terminal histidine tag. In this Expresso Rhamnose cloning system, tight control of expression was achieved using a combination of glucose to repress transcription from the rhamnose-inducible promoter and a 3-h induction with 0.2% (v/v) L-rhamnose. Purification was opti-mized as described under "Experimental Procedures," and Fig.  1, A-C, illustrates the steps in the purification strategy. No evidence of lymphostatin production was detected in uninduced cultures of E. coli harboring sequence-verified lifA, however, L-rhamnose induction led to the expression of rLifA, as detected by Coomassie staining and Western blotting with an anti-His tag antibody (Fig. 1C, inset).
To confirm that the expressed protein is full-length lymphostatin, in-gel tryptic digestion and MALDI mass spectrometry were carried out. The resulting peptides aligned to 41% of the lymphostatin primary sequence, from residue 52 to 3174 (of 3229); representing good coverage for such a large protein. The predicted N terminus of the protein is relatively rich in basic residues. Lack of tryptic peptide coverage at the N terminus could be accounted for by extensive digestion at charged residues generating many low mass fragments that are not detectable by MALDI. The C-terminal end can be inferred to be intact because of detection of the histidine tag and by IMAC purification and Western blotting. The purified protein was analyzed by dynamic light scattering. The sample contained species with a Stokes radius of 6.5 nm, consistent with monomeric rLifA. There was no evidence of aggregation or insolubility (data not shown).
rLifA Is Biologically Active against Bovine T Cells-Lymphostatin-containing bacterial lysates have been reported to be capable of inhibiting the proliferation of mitogen-stimulated PBMCs using cells from humans, mice (2, 7), and cattle (3,11). Using an enriched bovine T cell population increased the signal-to-noise ratio compared with use of bulk PBMCs, giving a wider dynamic range to the assay. Using T cells obtained from four independent donors, purified rLifA inhibited ConA-stimulated proliferation, showing inhibition from the low femtomolar range with concentration-dependent titration of activity with a sigmoidal curve (Fig. 2). Measurements of the ConAstimulated response were typically 2-5-fold higher than cells alone. The carrier buffer for rLifA was determined to have no effect on ConA stimulation of cells on its own. Furthermore, a colorimetric assay to detect release of cytosolic lactate dehydrogenase found no evidence of cell lysis when primary T cells were treated with inhibitory concentrations of rLifA, indicating that the inhibitory effect of lymphostatin on T cells is unlikely to be a consequence of direct cytotoxicity.
Lymphostatin Is a Monomer in Solution-Having established that rLifA is biologically active, we examined its structural characteristics. SEC-MALS was used to calculate the average molecular weight of rLifA in solution and hence the oligomeric assembly. Lymphostatin eluted as a well resolved single peak. Light scattering, RI, and UV were analyzed by a conjugate protein model, and the mass of lymphostatin was predicted to be 330 Ϯ 15 kDa, slightly lower than the predicted molecular mass for monomeric protein of 365 kDa (Fig. 1D). Given the mapping of the N-terminal to at least 52 residues from the beginning of the putative start of the protein, and that the C-terminal is intact, the size differential to the calculated size is unlikely to be due to truncated protein, and is within acceptable error for the technique at the present time, taking into account the glycerol and detergent-rich buffer. Detergent micelle eluted later than protein with an estimated mass of 76 Ϯ 5 kDa (Fig. 1D). Taken together the data indicate that lymphostatin is a monomer in solution and that it is not necessary for the protein to fully or partially be enclosed within a micelle to remain in solution.
Limited Proteolysis Identified 3 Putative Structural Domains-With no fine structural information available on lymphostatin we set out to predict putative physical domains using limited proteolysis; the premise being that loops and flexible regions of the protein are more susceptible to digestion than buried residues. Limited tryptic proteolysis of rLifA over several hours revealed a defined and consistent pattern of cleavage products. Five major fragments were identified by mass spectrometry, F1-F5 (Fig. 3A). Mass spectrometry of tryptic peptides from each species and alignment of peptides against/with the lymphostatin primary sequence suggested 3 major domains, with the additional two species representing truncated versions of two of the major digestion products, as indicated in Fig. 3B. Identification of peptides by MALDI-TOF mass spectrometry from the in-gel tryptic digestion of fragments initially generated by limited proteolysis does not precisely identify protein domain boundaries but may be considered a useful guide. Limited proteolysis fragments are likely to be somewhat longer at both the N and C terminus than the most N-terminal and C-terminal residues identified from their tryptic digestion and MS analysis. The largest intact fragment, designated F1, represents a ϳ1435 amino acid (aa) region at the N terminus of  lymphostatin, starting about 177 aa from the N terminus, and encompassing both the putative glycosyltransferase (GT) and cysteine protease (CP) motifs. The two smaller fragments are separated from the N-terminal fragment by about 500 aa, and are approximately ϳ776 aa (F3) and 271 aa (F5), respectively. In addition, the two C-terminal proximal fragments are separated by a short stretch of 43 aa. Residues 1600 -2100 are not represented in the major early digest products but were very well represented in the full tryptic digest during mass spectrometry analysis, suggesting this region is intact in the full-length protein. The abundance of fragment F1 reduces as the digest time increases, whereas fragment F2, an N and C terminally truncated subfragment of F1, becomes relatively more abundant. Fragment F2 maps more closely to the GT domain with a C-terminal extension of ϳ140 aa. The N-terminal ϳ100 aa of F1 is richer in basic amino acids than the full-length lymphostatin. This property enabled us to separate F1 from other digest products by anion exchange chromatography.
The Predicted Glycosyltransferase Domain of Lymphostatin-To identify the boundaries of the glycosyltransferase domain of lymphostatin we performed BLAST analysis of the primary sequence. We identified homology with the N termini of LCTs (the first 541 amino acids of TcdA) encompassing their catalytic glycosyltransferase domains. The alignment indicates that the sequence of lymphostatin is most similar to that of the LCTs for residues that define the binding site of the glycosyl donor substrate, often described at the catalytic core of the GT domain (Fig. 4A, gray shading) (13). The catalytic core residues of the LCTs are not contiguous and have insertions that form helical bundles surrounding the core (Fig. 4B, Toxin A, orange  highlights). These are thought to confer specificity for the protein that becomes glycosylated by the glycosyltransferase. The most striking differences in sequences in the LCT GT domains are within these insertions. To further examine the potential of lymphostatin to form a GT fold we generated a model using the PHYRE2 server (49). The best scoring model is based on Toxin A from Clostridium difficile, which forms a GT-A-fold common to all LCTs ( Fig. 4C; model of lymphostatin Ala 242 -Arg 769 ). Lymphostatin has similar insertions that are predicted from the sequence to be coiled and helical in character. It is worth noting that differences to the helical bundles mean that even if an alignment is restricted to the catalytic core of the clostridial toxins (residues Thr 105 -Trp 534 in TcdA, which align to residues Thr 309 -Phe 860 in lymphostatin) the sequence identity is only 20%, a figure that belies their common features. Lymphostatin displays good conservation for residues that make key non-covalent interactions with the UDP-sugar (Fig.  4A, stars above sequence; bold typeface marks identity). The LCTs and lymphostatin contain the signature DXD motif, seen in most GT domains, through which the aspartate carboxylates coordinate a divalent cation and the donor substrate (Fig. 4A,  red box). Three residues at the base of the ␤-hairpin in the sugar donor binding site, XNX, are thought to confer sugar specificity (Figs. 4A, black box, and 5, A and C). Toxins A, B, and L all have INQ in this position; this means that the enzymes can accommodate UDP-Glc but not UDP-GlcNAc. ␣-Toxin and TpeL bind UDP-GlcNac; to accommodate the acetyl group, INQ is replaced by SNA and ANQ, respectively. Lymphostatin has leu-cine in position 1 and glycine in position 3, LNG. Our model suggests this would make it possible for UDP-GlcNAc to bind (Fig. 5B).
Given that digest fragment F1 spans both the putative glycosyltransferase motif as well as the cysteine protease motif; we set out to test the F1 fragment in isolation in the bovine T cell proliferation assay, to explore whether it retained its inhibitory activity against lymphocytes. Using equimolar amounts of protein in the ConA-stimulated T cell proliferation assay, comparison of the full-length protein alongside purified F1 revealed that the F1 fragment was insufficient to inhibit T cell proliferation in isolation (Fig. 6). This may be because the F1 fragment N-terminal boundary is at ϳ177 aa and the N-terminal amino acids are important for interaction with the target protein.
More likely is that the C-terminal domains are crucial for cell binding and uptake, as has been shown for the large clostridial toxins (50).
Lymphostatin Is an Ordered Protein with a High ␣-Helical Content in the Putative Glycosyltransferase Domain-To confirm that rLifA was folded and to estimate the content of secondary structural elements we measured the CD spectrum of full-length and the F1 fragment that encompasses the putative GT and the CP domains. The CD spectrum of full-length protein was consistent with a folded protein and indicated 37% ␣-helix and 17% ␤-sheet (Fig. 7, A and B); very similar to PSIPRED predictions (ϳ35% ␣-helical and ϳ22% ␤-sheet). The N-terminal half of the protein was predicted to contain the majority of the ␣-helical content, whereas the C-terminal third of the protein was predicted to be rich in ␤-sheet (Fig. 7C). The samples showed no evidence of aggregation by dynamic light scattering (ZetasizerAPS, Malvern).
Given the lack of activity of the isolated F1 fragment in the T cell proliferation assay, and to rule out that the protein was degraded and/or not folded, CD analysis was carried out on purified F1 protein. Furthermore, given its location at the N-terminal part of lymphostatin, where the primary sequence is predicted to have a higher percentage of ␣-helices, it is expected that the percentage of ␣-helical content of F1 would be higher than that of the full-length protein. The F1 fragment was predicted by PSIPRED to consist of 52% ␣-helical content and analysis of CD data inferred 51% ␣-helical content. The CD data is broadly in agreement with the secondary structure prediction for F1 but suggested a slightly higher proportion of ␤-sheet, 13% as opposed to 5%.
Three-dimensional Shape of Lymphostatin-Negatively stained rLifA showed a homogeneous distribution of elongated, slightly kinked particles (Fig. 8). From images of ϳ25,000 of these particles, we determined a three-dimensional map of lymphostatin at 23-Å resolution. The map shows an elongated, L-shaped molecule (Fig. 8). The arms of the "L" are 130 -140 Å long and between 60 and 90 Å thick. To further investigate the structure of lymphostatin, we also performed SAXS (51,52). The linearity of the Guinier region confirms the absence of inter-particle effects (Fig. 9A). The longest particle dimension D max is 218 Å, which is not dissimilar to the longest dimension of the EM map (ϳ197 Å) (Fig. 9B). The Porod volume (645 nm 3 ) indicates a particle molecular mass of 358 kDa further suggesting that lymphostatin (365 kDa based on primary sequence) behaves as a monomer in solution, and that the recombinantly purified full-length protein is stable. Next, we determined the ab initio shape analysis with DAMMIF (40) using 10 independent computations producing models related with an average normalized spatial discrepancy of 0.793. The representative and DAMMIN (42) refined model shows an elongated Lshape with dimensions (218 ϫ 116 ϫ 100 Å) slightly larger than the EM density map (197 ϫ 110 ϫ 89 Å). The alignment of the EM density map with the SAXS envelope highlights the similarities in dimensions as well as shape (Fig. 9C). The kink of the L-shaped molecule is situated in a similar position in the two envelopes, at roughly half the length of the particle, however, the kink is more defined in the EM map. In addition, the EM map contains a larger volume within one of the arms of the L.
Lymphostatin Binds UDP-N-acetylglucosamine-Lymphostatin contains 37 tryptophan residues, of which 7 are found within the putative GT domain; Trp 320 , Trp 534 , and Trp 863 most likely are in close proximity to the active site (Fig. 4). Titration of lymphostatin with UDP-Glc gave a linear Stern-Volmer plot that would be consistent with collisional quenching of solvent-exposed tryptophan rather than any specific interaction of sugar with lymphostatin (Fig. 10). In contrast UDP-GlcNAc enhances fluorescence in a non-linear concentration-dependent manner at lower concentrations; there is evidence of collisional quenching at higher concentrations. This may be explained by UDP-GlcNAc binding in the GT catalytic site and changing the environment of one or more tryptophan residues. At high concentrations of UDP-GlcNAc after the active site has become saturated there is a linear relationship between fluorescence and UDP-GlcNAc concentration that could be attributed to collisional quenching of tryptophan outside the binding site. We have fitted a mixed binding model to the UDP-GlcNAc data that takes into account specific binding and collisional quenching. The affinity of lymphostatin for UDP-GlcNAc is estimated to be 120 Ϯ 30 M. The protein was expressed in the presence of a variety of divalent cations including Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ that were present in the growth media. Good reproducibility of affinity data between batches was achieved by incubating the protein sample with 50 M MnCl 2 after purification. Manganese was selected as the divalent cation based on thermal shift experiments (differential scanning fluorimetry) that suggested MnCl 2 enhanced the thermal stability of lymphostatin and the additional stabilizing effect of UDP-GlcNac to a greater extent than MgCl 2 (data not presented) (53). The The DTD Motif within the GT Domain of Lymphostatin Is Important in Sugar Binding and Activity against Bovine Lymphocytes-Given that the DXD motif has been implicated in glycosyltransferase activity in other similar molecules (reviewed in Ref. 13), we examined the effect of a DTD to AAA substitution at position 557-559 on both sugar binding and activity of lymphostatin. rLifA DTD/AAA behaved similarly to rLifA, both during expression and purification, as well as in CD and DLS lending confidence that the residue substitutions had not disrupted the original structure of the protein.
Testing rLifA DTD/AAA using the tryptophan fluorescence assay and titrating either UDP-Glc or UDP-GlcNAc as for rLifA gave linear Stern-Volmer plots indicative of nonspecific quenching and lack of sugar binding (Fig. 11A). Furthermore, titration of the rLifA DTD/AAA protein in the bovine T cell proliferation assay shows a drastic reduction in its ability to inhibit proliferation by almost 4 orders of magnitude (ED 50 rLifA ϭ 0.014 ng/ml Ϯ 0.0015, rLifA DTD/AAA ϭ 922 ng/ml Ϯ 270) (Fig.  11B). These two observations imply that not only is the DTD motif important for sugar binding, but is also critical for the inhibitory activity of lymphostatin on T cells.

Discussion
Lymphostatin plays an important role in intestinal colonization by attaching and effacing E. coli and is therefore a potentially attractive target for intervention. Although functions have been assigned to this large molecule, its mode of action remains largely unknown. It is one of relatively few bacterial molecules reported to interfere with the function of adaptive immune cells. Historically, it has proven difficult to obtain a stable clone to express and purify full-length lymphostatin (2,8,47). Using a tightly inducible prokaryotic expression system we were able to express and purify lymphostatin with potent activity against mitogen-stimulated T lymphocytes. Mass spectrometry of tryptic peptides and SAXS indicate that the purified full-length protein is stable. Lymphostatin contains predicted non-canonical prepilin peptidase sites (8), however, it is not clear whether it is processed, at least in the laboratory-adapted E. coli K-12 strain used for expression, nor is there a predicted signal sequence at the N terminus.
Since its discovery, two predicted features have been known in lymphostatin: namely a glycosyltransferase domain and a YopT-like cysteine protease domain (8,17). Although both motifs were implicated in colonization of mice and colonic hyperplasia by C. rodentium (5), the mutations caused trunca-  tion of lymphostatin and subsequent studies with in-frame deletions suggested the motifs were not essential for intestinal colonization of cattle by enterohemeorrhagic E. coli O26:HϪ (4). However, the effect of lymphostatin and motif mutations on mucosal lymphocytes and induction of adaptive immunity was not studied and a role for the motifs in pathogenesis therefore cannot be excluded. Using purified lymphostatin we have not observed autocatalytic cleavage via the putative cysteine protease domain, including under low pH conditions or in the presence of inositol hexakisphosphate, the co-factor known to trigger autocatalytic cleavage of LCTs (reviewed in Ref. 56). If lymphostatin is able to autocatalytically cleave using the CP motif, either it requires some as yet unidentified co-factor, or it is not triggered by pH change, and merits further closer investigation.
Although we have obtained low resolution EM and SAXS envelopes for lymphostatin, a high resolution structure has been elusive. A full-length diffracting protein crystal has not yet been obtained from pilot crystallization trials. Given the behavior of the recombinant protein in solution, high resolution cryo-EM might be a promising approach to analyze the lym-phostatin structure. Alternatively, given the data from the partial tryptic digestion, an approach combining crystallization of predicted domains combined with EM could be used. Circular dichroism spectroscopy of full-length and the N-terminal putative glycosyltransferase domain has confirmed the secondary structure predictions that show the N-terminal third of the protein to be largely ␣-helical. In addition, the C-terminal third is very rich in ␤-sheet, both features shared with the LCTs. Interestingly, unlike the LCTs, lymphostatin lacks the multimodular cell wall/choline-binding repeat that form the receptor-binding domain (57).
For the first time we report that lymphostatin is able to bind UDP-GlcNAc, but not UDP-Glc in a manner dependent on a conserved DXD motif within its predicted glycosyltransferase domain and that this motif is also required for its activity against bovine T lymphocytes. Furthermore, inhibition of lymphocyte function was not associated with cell lysis as measured by release of a cytosolic enzyme or effects on the actin cytoskeleton, in contrast to LCTs (data not shown) (58,59). The target of sugar modification by lymphostatin is unknown, and is likely to be difficult to identify. For both the LCTs and NleB, there was some initial insight into rational candidate acceptor proteins or signaling pathways prior to their identification. However, this is not currently the case for lymphostatin. Furthermore, whereas potential targets of NleB activity were identified by immunoprecipitation and yeast 2-hybrid approaches, enabling specific sugar transfer onto putative recombinant targets to be tested (14,15), this approach is likely to be technically challenging for lymphostatin, because of the challenges of cloning the gene in the vectors required for yeast 2-hybrid screening. Interacting partners for lymphostatin are presently unknown and a similar FIGURE 7. Secondary structure analysis. A, far UV CD spectrum of lymphostatin and the major N-terminal fragment F1. B, secondary structure analysis from CD and predictions from primary sequence indicate that F1 is largely ␣-helical. There is good agreement between PSIPRED predictions and secondary structure analysis of the CD data for full-length lymphostatin. C, secondary structure prediction from primary sequence obtained from PSIPRED. approach may be needed to find its target(s), given the challenge of detecting addition of labeled GlcNAc to cellular proteins against the large number of endogenous modifications expected.
Although lymphostatin has recently been demonstrated to be secreted via Type III secretion (10), the fact that it is active in an isolated recombinant form, or when produced in E. coli that lack a type III secretion system, indicates that injection into cells is not vital for activity against lymphocytes. It is plausible that it may have both Type III secretion-dependent and -independent activities. Our evidence indicates that the full-length protein is required for inhibition of lymphocyte function as the isolated F1 fragment encompassing the glycosyltransferase domain showed no activity, despite being folded and soluble. As   with large clostridial toxins the C-terminal domain of lymphostatin may be required for cellular uptake and the GT domain is unable to act in isolation (60,61).
Where EPEC and non-O157 EHEC almost invariably express lymphostatin, in serogroup O157 EHEC strains, lymphostatin exists as two truncated open reading frames (z4332 and z4333, encoding proteins identical to residues 1-433 and 435-710 of full-length lymphostatin) (62,63). In addition, the pO157 plasmid encodes a putative 365-kDa homologue of lymphostatin named ToxB (L7095), which shares 28% identity and 47% amino acid similarity to LifA/Efa1 (64,65). E. coli O157 strains have a lymphostatin-like activity (3,15) that has been associated with the pO157 plasmid (2). Mutation of the truncated lymphostatin or toxB genes in a Shiga toxin-deficient E. coli O157 strain did not markedly affect the ability of bacterial lysates to inhibit bovine lymphocyte proliferation (11). However, the assay used relied on crude bacterial lysates and is insensitive compared with the highly purified protein tested herein, with lysates causing lymphostatin-independent inhibition of lymphocyte function at higher concentrations (4,11). The expression and assay systems described here are likely to be suitable to assign activities to ToxB and truncated lymphostatin, and indeed a family of homologous cytotoxins described in pathogenic Chlamydia species (16). Almost two decades after lymphostatin was originally identified we have demonstrated that it has sugar-binding potential, that it is a highly potent inhibitor of T cell proliferation in vitro, and that a DXD motif predictive of glycosyltransferase function is essential for both these activities.
Author Contributions-R. L. C. C., E. A. B., M. W., R. J., and M. P. S. contributed to conception and design of study; R. L. C. C. cloned the original expression construct and initial protein studied, carried out the T cell proliferation assays for full-length lymphostatin and the F1 fragment, analyzed data, constructed figures, and carried out statistical analysis, wrote the first draft of the manuscript, and coordinated revision and submission; E. A. B. optimized protein production and purification, carried out biophysical characterization and analysis, domain bioinformatic analysis, all CD analysis, sugar binding assays, and contributed to writing the first draft of the manuscript; H. A. carried out sugar binding assays and SAXS analysis for full-length lymphostatin; E. D. carried out SAXS analysis for full-length lymphostatin and contributed to writing the first draft of the manuscript; A. B. prepared the AAA mutant of lymphostatin, helped with optimization of production and purification, carried out proliferation and sugar binding assays using the mutant protein; B. B. carried out the EM staining, image acquisition and analysis to produce the threedimensional reconstruction of lymphostatin, and contributed to writing the first draft of the manuscript; R. J. carried out SAXS analysis and contributed to writing the manuscript; M. P. S. contributed to writing the manuscript. All authors contributed to analyzing the data, editing the manuscript, and approved the final form.